Before jumping on the ML/AI bandwagon of WebRTC media quality, make sure you’ve exhausted all of your other optimization alternatives.
TL;DR – make sure you optimize for media quality without AI before jumping to using AI…
In 2018 and 2019 at Kranky Geek we’ve started looking at machine learning. We’ve handpicked speakers and sessions who deal with these topics. We’ve done so for both voice and video technologies. The intent and idea behind this was to fit to the times. Everyone’s been doing AI so why not us in the context and domain of WebRTC and communication technologies?
It made perfect sense.
Then came 2020 and… changed everything. No one was really interested in AI or how to improve quality of experience with it. It was now used mainly for bots with the purpose of handling large loads of calls (call deflection and agent assist type technologies).
At times, it seemed like we were all back to basics. We now had to start scratching our heads and see what can be done to improve quality.Time for some quick wins
At Google and elsewhere, I am sure that a manager somewhere higher up came, saw the work that is being done, received an explanation how research into this machine learning stuff was progressing and showing promise, but in many ways required, well, more research, before it could be seen as anything that is close to being production ready.
And as managers do in these situations, they smack the table and say something like “I want quick wins”. So the developers went back to the basics. Trying to figure out what quick wins they can find to squeeze a bit more quality of that thing they had called WebRTC.
Quite surprisingly – it worked!
There seems to be ample room for optimizations. If you ask me? Someone forgot to try and squeeze this lemon properly.There’s more room for optimizations of WebRTC before we resort to machine learning Google’s optimizations of WebRTC’s code
It started somewhere with the pandemic.
One of the first indications was this tweet by Serge Lachapelle (former product manager for WebRTC at Google and leading Google Meet at the time of tweeting).April 17, 2020
Apparently, the video compositor wasn’t making the most out of the hardware it was using…
Since then we’ve seen some additional optimizations, though most of them taking place in the application level on top of the WebRTC implementation itself.
At Kranky Geek, Google discussed at length the optimization work it is working on. Mostly, making sure that video processing doesn’t take up too much CPU.Too many media format conversions in the WebRTC media pipeline
Apparently, Chrome is doing way too many video format conversions between getting the frames from the camera until it encodes and sends it out. Each conversion eats up CPU and I/O, generally killing the whole internal bus of the machine. Oh – and it means memory copies. Lots and lots of memory copies.
Video processing 101: zero copy is what you’re striving for.
We’re 10 years into WebRTC and the leading team behind WebRTC is just now starting to look at zero copying.
There are other areas and aspects where optimizations are taking place. Once the Kranky Geek videos will be ready and published, I’ll add the relevant one here.
Still got optimization juice in this lemon. Expect better performing WebRTC in the coming Chrome releases.Rushing towards 49-gallery view and 50+ group sizes
As the pandemic hit, Zoom grew. The media was filled with their gallery view.Zoom’s 49-gallery view. The holy grail of video group calls?
One use case that didn’t exist before the pandemic is large video calls. Up until today, we used to take these video meetings in the office inside meeting rooms. Cramming a few people in each room in a remote office and doing a call with 2-4 such rooms. Maybe someone joined from home or a hotel. You could see meetings with 10 participants. Sometimes. But the need just wasn’t really there.
The pandemic hit. People are now at home. And communicate with video remotely. A meeting of 4 became a meeting of 20 just because the participants are now sitting at home.
Even worse, schools are now remote. Each class has 20-40 students in it. And the teacher wants to see them all.
This made Zoom’s gallery view so popular (even if a tad useless if you ask me). It also made the magical number 49 magical. The holy grail of what is needed of a video conferencing service in a pandemic. Doesn’t matter if everyone is muting their video.
Microsoft and Google announced plans for supporting it. Then started running towards that value, each rising in the number of tiles in his gallery, reaching 49 recently.
Facebook grew from a meeting of 8 to meetings of 50.
Meetings are larger and longer now.
And again, we found the ways to make it happen with WebRTC.Best practices on group video scaling being rewritten
There are a lot of mechanisms in WebRTC that enable an application to squeeze the lemon and gain back CPU cycles as it tries to optimize for larger group calls.
But we never did have a place where all these are found and explained. A body of knowledge and understanding of how to make it happen.The larger the conference call size in WebRTC, the more complex the solution is going to be to implement it
In my recent/upcoming update to the Advanced WebRTC Architecture course there’s a lesson dedicated to this specific topic. It isn’t as if the information isn’t there in the course – it is spread all over the course. But now there’s a lesson on this alone. Because it became interesting only in 2020.
We have traded the focus on what is important to us with video communications. A video conference’s scale trumps quality at the moment. While I do understand we all want both all the time, but there is still a tradeoff between these two qualities of a system.The role of machine learning and AI in communications
Where does one fit machine learning and AI in this brave new world of large video conference calls?
Machine learning requires memory and CPU. Things we don’t have to spare at the moment in these large group calls. So we can’t just slap machine learning inference algorithms on the edge inside the web browser easily.
Edge inference in web browsers using WebAssembly is also brand new. So there’s no guide book to work with.
We won’t be using it to improve video quality or audio quality in the edge – we can’t really. Not enough CPU to spare.
There’s no real place for it on the server side either – that one requires decoding and encoding which are going to be CPU intensive and increase the costs of delivering the service. Pexip is doing that for auto zoom, but that’s because they are built as an MCU. Google decided to do this for noise suppression.
There’s packet loss concealment using machine learning now. And you can do super resolution for video to get better video quality. But in the end, all these are going to make a difference once CPUs have their own dedicated, standardized AI accelerators, like the new Apple M1 chip in them brand new Intel-less MacBooks. We just don’t have cycles to spare.
Which is why media quality has gone back to its roots. Here’s something I have in that workshop of mine:First take care of your infrastructure as much as you can to improve media quality in WebRTC
Machine learning should be added once we’re done squeezing that lemon for more performance and quality.
Google is now doing its part of optimizing the WebRTC codebase itself. It is your role to do it in your own infrastructure and application. Once done, the time will come to introduce some machine learning chops into it.
Until then? We need machine learning for two main tasks, and we see it already:
- Background blur and background replacements. We’re all humans but somehow we don’t want our kids to be in the way of our conversations
- Noise suppression. As we’re stuck at home, we can’t really control that crying kid of ours on the other side of the room
Does that mean you don’t need to invest in machine learning?
Hell no. you definitely MUST invest in machine learning.
Not for what you’ll be doing in 2021, but for what you’ll be launching in your product in early 2022. Which brings me to the heart of it all.
Machine learning is new and challenging. We’re still writing the playbook of what it means to use it for real time communications, inside a browser, using technologies such as WebAssembly.
You’ll need to decide on which use cases to invest, and what value you are going to derive of it. And you’ll need to plan for the long game here and be patient until you get results.
There’s a need to let the teams driving machine learning do the research and experimentation needed. But at the same time, they need guidance in where to look at and what to experiment with.
The post A blueprint to improving WebRTC media quality using AI appeared first on BlogGeek.me.
WebRTC growth during 2020 came in waves, just like the pandemic and its quarantines. Here how it looks and where we are all headed.
Let’s look at some interesting performance indicators of WebRTC use and adoption.
2020 is the year of video communications.
2020 is also the year of WebRTC.Table of contents
- Unified Communications & WebRTC
- Gartner’s Magic Quadrant for Meeting Solutions (& WebRTC)
- A surge in use of WebRTC
- Where do we go from here?
In my introductory slides of my WebRTC workshop 4 months ago, I had that as a very strong theme:
The slide above illustrates what the statistics at the time were for the big meetings vendors.
Since then, the numbers have grown. Microsoft Teams, for example, reached 115M DAU. That’s Daily Active Users.
While not all of the growth is in video calls, these services have a video focus to them.
Out of these 4 vendors:
- Zoom doesn’t make use of WebRTC, and like it that way
- Google Meet is “all in” with WebRTC
- Microsoft Teams has WebRTC support to it, though with pretty limited capabilities
- Cisco WebEx supports WebRTC rather nicely
Guest access growth for Microsoft Teams and Cisco WebEx can be attributed to some extent to WebRTC. With Google Meet, it is all WebRTC related.Gartner’s Magic Quadrant for Meeting Solutions (& WebRTC)
Gartner has its nice magic quadrant diagrams. Here’s the one just published for meeting solutions:
Which of the vendors in this magic quadrant diagram use WebRTC? I’ve marked the vendors in red for you:
The ones not marked might have WebRTC – I am just not aware of it. The ones marked have WebRTC support in production in their products. How central is it to their product is a different question though.
The thing here is that no matter what magic quadrant from Gartner you’ll be looking at for whatever market category that involves communications, WebRTC will be used as the underlying technology by many of the vendors.
Contemplating if WebRTC is the technology to use? Look at the reds above.A surge in use of WebRTC
I decided to leave the best for last.
Chrome collects and shares statistics of JS API calls in the browser and their “popularity”.
Lets look how getUserMedia() looks like:
Interestingly, we see an adoption curve where each round of quarantine raises the use of WebRTC to a higher level.
From a steady, boring 0.05% of use pre-pandemic, the new normal is settling well above 0.2% of the page loads.
How can we explain the rise from July to October? Is this a sustained growth happening as the pandemic found its second wave in different countries and social distancing gradually came back in force throughout the globe? Is it due to the fact that schools started opening around the world in August and September, many of them strictly remotely? Is it due to more services being introduced online that offer WebRTC based communications in them?
If you ask WebRTC, we’ve reached the peak of the second wave of the pandemic.Where do we go from here?
- A third pandemic wave. Will that raise usage even further?
- Vaccine. Even a promise of one sent collaboration stocks down
On a more serious note though, the huge surge in WebRTC traffic brought with it new use cases and a lot of learnings regarding scaling and operationalizing WebRTC.
The post WebRTC Growth – is it a back-to-school pandemic phenomena? appeared first on BlogGeek.me.
If you are planning to use WebRTC P2P mesh to power your service, don’t expect it to scale to large sessions. Here’s why.
Every once in a while someone comes in with the idea to broadcast or conduct a large scale video session with WebRTC without the use of media servers. Just using pure WebRTC P2P mesh technology.
While interesting as a research topic for university, I don’t think that taking that route to production is a viable approach. Yet.Table of contents
- What is WebRTC P2P mesh?
- Bandwidth challenges in WebRTC P2P mesh
- CPU challenges in P2P mesh
- Alternatives to WebRTC P2P mesh
- A word about WebRTC data channel mesh
If you are focusing on data only WebRTC mesh, then skip to the last section of this article.
When dealing with WebRTC and indicating P2P or mesh, the focus is almost always on media transport. The signaling still flows through servers (single or distributed). For a simple 1:1 voice or video call, WebRTC P2P is an obvious choice.From a WebRTC client perspective, a 1:1 session is similar if it is done using P2P mesh or using a media server
The diagram below shows that from the perspective of the WebRTC client, there is no difference between going through a media server or going P2P – in both cases, it sends out a single media channel and receives a single media channel. In both cases, we’d expect the bitrates to be similar as well.
Making this into a group call in P2P translates into a mesh network, where every WebRTC client has a peer connection opened to all other clients directly.WebRTC mesh architecture. Or is it mess architecture? Why use WebRTC P2P mesh?
There are two main alluring reasons for vendors to want to use WebRTC P2P mesh as an architectural solution:
- It is cheaper to operate. Since there are no media servers, the media flows directly between the users. With WebRTC, oftentimes, the biggest cost is bandwidth. By not routing media through servers as much as possible (TURN relay will still be needed some of the time), the cost of running the service reduces drastically
- It is more private. Yap. As the service provider you don’t have any access to the media, since it doesn’t flow through your servers, so you can market your service as one that offers a higher degree of privacy for the end users
If WebRTC P2P mesh is so great, with cheaper operating costs and better privacy, then why not use it?
Because it brings with it a lot of challenges and headaches when it comes to bandwidth and CPU requirements. So much so that it fails miserably in many cases.
It is also important to note here that in ALL cases of 3 users or more in a call, alternative solutions that rely on media servers give better performance and user experience. Always – at least as long as the media servers infrastructure is properly deployed and configured.Bandwidth challenges in WebRTC P2P mesh
Assume we want pristine quality. Single speaker, 10 listeners.
The above layout illustrates what most users of this conference would like to see and experience. The speaker may alternate during the meeting, switching the person being displayed in the bigger frame.
As we’re all watching this on large screens (you do have a 28” 4K display – right?), we’d rather receive this at HD resolution and not QVGA. For that, we’d want at least 1.5Mbps of the speaker to be received by everyone.Strain on the uplink
In a mesh topology, the speaker needs to send the media to all the participants. Here’s what that means exactly:In WebRTC mesh, we put a bigger strain on the uplink
1.5Mbps times 10 equals 15Mbps on the uplink. Not something that most people have. Not something that I think my strained FTTH network will be able to give me whenever I need it. Especially not during the pandemic.
In an office setting, where people need to use the network in parallel, giving every user in a remote meeting 15Mbps uplink won’t be possible.
On top of that, we’ve got 10 separate peer connections to 10 different locations. WebRTC has its one internal bandwidth estimation algorithm that Google implemented in libwebrtc, which is great. But how well does it handle so many peer connections on the client’s side? Has anyone at Google ever tried to target or even optimize for this scenario? Remember – none of Google’s own services run in a mesh topology. Winning this one is going to be an uphill battle.Bandwidth estimation on the downlink
Let’s look at the viewers/subscribers/participants/users or whatever else you want to call them.
If we pick a gallery view layout, then we are going to receive 10 incoming video streams. Reduce that to 9 for layout simplicity and we get this illustration:
There are 9 other users out there who generate video streams and send them our way. These 9 streams are competing on our downlink network resources and for our machine’s attention and CPU.
Each of them is independent of the others and have little knowledge about the others.
How can the viewer understand his downlink network conditions properly? Let alone try to instruct these sends on how and what to send. A media server has the same set of problems to deal with, but it does that with two main advantages:
- It controls all the videos that are sent to the viewer, and it can act uniformly as opposed to multiple browsers competing against each other (you can try to sync them, though good luck with that)
- You can put all incoming streams in a single peer connection from the server, which is what Google Meet does (and probably what Google is focused on optimizing for in their WebRTC implementation)
Then there’s the CPU to deal with in WebRTC P2P mesh.
Each video stream from our speaker to the viewers has its own dedicated video encoder. With our 10 viewers, that means 10 video encoders.
A few minor insights here if I may:
- If you aim for H.264 hardware encoding, then bear in mind that many laptops allow up to 3-4 encoded streams in parallel. All the rest will be black screens with the current WebRTC implementation
- Video coding is a CPU (and memory) hog. Encoding is a lot worse than decoding when it comes to CPU resources. Having 10 decoders is hard enough. 10 encoders is brutal
- 10 or more participants in a video call is hard to manage with an SFU without adding optimizations to alleviate the pains of clients and not burn their CPU. And that’s when each user has a single encoder (or simulcast) to deal with
- Your Apple MacBook Pro 2019 with 16 cores isn’t the typical device your users will have. If that’s what you’re testing your WebRTC mesh group video calling on then you’re doin’ it wrong
- I am sure you thought that using VP9 (or AV1 or HEVC, which aren’t really available in WebRTC at the moment) will save you on bandwidth and improve quality. But it eats even more CPU than VP8 or H.264 so not feasible at all
So. going for a group video call?
Want to use WebRTC P2P mesh?
You’re stuck at 300kbps or less for your outgoing video even if your network has great uplink. Because your device’s CPU is going to burn cycles on encoding multiple times.
Which also means that people aren’t going to like hearing their laptop’s fans or touch their heating smartphone (and depleting battery) on that call.Can we do better?
Probably. A single encoder would make the CPU problem a wee bit smaller. It will bring with it headaches of matching the bitrate to all viewers (each has his own network and device limitations).
Using simulcast in some manner here may help, but that’s not how it is intended to be used or how it has been implemented either.
So this approach requires someone to make the modifications to the WebRTC codebase. And for Google to adopt them. Did I already say Google has no incentive in investing in this?Alternatives to WebRTC P2P mesh
You can get a group video call to work in WebRTC P2P mesh architecture. It will mean very low bitrate and reduced video quality. But it will work. At least to some extent.
There are other models which perform better, but require media servers.WebRTC offers media server alternatives to mesh in the form of SFU and MCU
Using an MCU model, you mix all the video and audio streams in the MCU, making sure each participant receives and sends only a single stream towards the MCU.
With the SFU model, you route media around between participants while trying to balance their limitations with the media inputs the SFU receives.
You can learn more about in my WebRTC multiparty architectures article.A word about WebRTC data channel mesh
I haven’t really touched WebRTC mesh architectures for data channels.
All the reasons and challenges detailed above don’t apply there directly. CPU and bandwidth relied on the concept of needing to encode, send, receive and decode live video. In most cases, this isn’t what we’re dealing with when trying to build mesh data channel networks. There, the main concern/challenge is going to be proper creation and connection of the peer connections in WebRTC.
If what you are doing isn’t a group video call (or live video broadcast from a browser to others) then a WebRTC P2P mesh architecture might work for you. If it will or won’t is something to analyze case by case.
In the last 2 months I’ve dived into the world of CPaaS again, updating my WebRTC API focused report. Oh, and there’s a new free ebook.
There have been many changes since my last update,so this one was greatly overdue.
API platforms changed hands due to mergers and acquisitions. Vendors joining the market. Others leaving or just pivoting away from APIs.
And then we had AWS and Azure entering the CPaaS market.
What I did in these last two months was interview and review all the vendors in my report again, to see what has changed and update that part of the report. I learned a lot from the process.
As with every time where I shift focus to a certain market, I took the time to process my own thoughts by writing them down here in a series of articles.
Here are two things I wanted to share with you, as well as announce my next upcoming projects.Table of contents
- Choosing a WebRTC API report – 2020 version
- CPaaS in 2020 – a free ebook
- Advanced WebRTC Architecture Course – update & office hours
- WebRTC Insights – a new service
I finished and published the WebRTC API report last week. The result:
- 254 pages
- 24 vendors
Agora decided to sponsor this report (thanks a bunch!). They are one of the interesting vendors in this space, offering an IP video/voice focused platform with their own data centers spread across the globe and a lot of research done in machine learning to improve media processing.
If you are looking to learn more, then you can:
- Read the WebRTC API report overview
- Get the 4-pager of Agora from the report (each vendor covered in this report has a 4-pager)
- Purchase the report online
The previous 3 articles in my site here were all focused on CPaaS, looking at different angles on how CPaaS is changing.
The first one dealt with the future of CPaaS, especially considering the pandemic and how it affects everything and everyone.
In the second article, I looked at AWS Chime SDK and Azure Communication Services, trying to understand what their entry into CPaaS is going to change in the market.
For the third and last article, the focus went to Twilio Signal 2020. Considering how they redefined the market in the last 4 years in each such event, this event was a bit of a downer. It did bring with it many insights.
If you’re more into printing and reading, or sharing with others, then I packaged all of these 3 articles into one ebook, making it easier to consume.
I called the ebook CPaaS in 2020 – a market in transition. Because this is what it is…Download my CPaaS in 2020 ebook Advanced WebRTC Architecture Course – update & office hours
With my WebRTC API report now updated and finally launched, I can go back to focusing on other projects I am running.
My WebRTC Courses have been around for over 4 years now. I’ve been updating them regularly and I am doing it again to my main signature course – the Advanced WebRTC Architecture training.Updates
There are going to be 2 new lessons and around 10 lessons that are already being updated and recorded all over again.
The purpose is still to make this the best alternative out there to learning WebRTC.Office hours
Alongside the updates, I will be starting another round of office hours for the course. These will start in December.
The office hours is where students can come and learn online and in-person with me specific topics in WebRTC, as well as ask questions about anything related to WebRTC – and their own projects.
If you were thinking of learning WebRTC, then the best timing for it would be to enroll now and join the office hours. These are complementary to the course and open for anyone with a valid course subscription.WebRTC Insights – a new service
Following and catching up with everything in WebRTC is time consuming. It is also tedious. And you need to know where to look and what each bit of information means to you.
To make this a wee bit easier, I’ve decided with the help of Philipp Hancke to start a new service together – WebRTC Insights
In this service, you receive an email every two weeks. This email includes all the important changes to WebRTC
- Bug tracking of browser related WebRTC issues we feel are important
- Select libwebrtc code commits that we found interesting
- discuss-webrtc forum messages
- Critical PSA announcements from browser vendors
- W3C/IETF mailing list items
- Market news related to WebRTC
- Things we hear from other vendors that we can share
This gives you actionable insights to your own planning and reduces the risks in your development. Both Philipp and me have been doing this for a while, but doing it together brings it to a new level.
If you want to learn more and subscribe to this service, then check the new WebRTC Insights page.
Twilio Signal 2020 occurred virtually this year. The number of new announcements or market changing ones was low compared to previous years. I expected more from Twilio as the leading CPaaS vendor.Table of contents
- Twilio Signal – past events
- Twilio By the Numbers
- Nike and digital transformation
- Twilio Microvisor
- Twilio Video WebRTC Go
- Twilio Flex ecosystem
- Twilio Frontline
- Other announcements
- Machine Learning was missing
- The coming CPaaS fight is in the enterprise
Twilio Signal is Twilio’s yearly event where its major announcements are made. It is also a gathering place where customers, partners and even Twilio CPaaS competitors come to meet. This year, as all other events, Signal was virtual. Twilio built its own hosting platform and event experience and did a good job at that.Twilio Signal – past events
I’ve watched the keynote twice, and several of the other sessions, including all major announcement sessions. I came out of this feeling a wee bit disappointed. There was nothing really interesting or groundbreaking this year. Especially not if you compare it to some of the previous years:
- 2015 – Copilot, Video, IP Messaging, …
- 2017 – Engagement cloud; shifting from low level building blogs to higher abstractions. Especially with Twilio Studio
- 2018 – this year was about Flex, and moving into contact centers directly
- 2019 – Twilio Conversations, Media Streams, SendGrid (email), Twilio CLI, …
In 2020, we’ve seen Twilio Microservices (the Electric Imp acquisition), Frontline, Video Go, Event Streams and Verify Push.Twilio By the Numbers
The main keynote by Jeff Lawson, Twilio CEO, had 3 components to it, with 3 main messages:
- Twilio is big
- Social good
- New product announcements
I’ll focus on the big and new parts here.
Twilio is now 12 years old and it has accomplished a lot. Jeff threw the “Twilio is big” numbers too fast for my taste, not even letting some of the big numbers register in our minds properly.
Here are the numbers. I tried aligning them with last year’s numbers from Twilio 2019:20192020Interactions750B1TUnique phone numbers2.8B3BCalls/minute32,500–Peak SMS/second13,000–Email addresses3B/quarter50%Video minutes–3BCustomers160,000200,000+Developers6M– What the numbers mean
- I still don’t understand what interactions mean, but the number is growing ridiculously fast, so it must be a good thing (I’d love to know how it is calculated)
- Voice and SMS is out (no calls/minute or SMS/second numbers this year)
- Unique phone numbers indicates reachability and 3 billion is a nice number, showing decent growth from last year
- Email moved from a number to a percentage, making it even less accurate or interesting. How would one know what an email address represents? There are so many of them that are spammy or just an alias to other addresses.
- For the first time video is important to Twilio. 3 billion is a large number, but not overly so (more about this later)
- The number of customers has grown significantly
- The developers number was useless to begin with and is finally not shared at all
Jeff alluded to the new normal, forced on us due to the pandemic. In many ways, this has been the main theme of Signal and the sessions.
My gripe with the “new normal” moniker to our situation is that there isn’t anything normal about it and it isn’t really here to stay.
Yes. We are seeing an accelerated move towards digital transformation and the cloud, but some of this shift, and especially the high usage in some sectors (such as education) aren’t here to stay post-pandemic.
For me, there’s no “new normal”. Just a transition to one, which will take time. How the future is going to look is hard to say from our current position.
Which leads me to the interview Jeff did with John Donahoe, Nike CEO.Nike and digital transformation
Jeff picked John Donahoe as the first person to interview during the keynote. It is an interesting choice.
I found it a tad ironic to get an explanation about social good and how Nike in all its years promoted social causes. It got me thinking about the Nike sweatshops. Other than this little history reframing that was done, the interview was quite good.
Two sentences that John said really resonated with me:
“Every business in the world is embracing digital transformation. We all have no choice”
The shift towards making businesses more digital has been inevitable.
Just think of all the on premise contact centers and what they now have to do when all of their agents are working from home. Or how all brick and mortar stores need a digital footprint to be able to even stay in business and sell throughout the quarantines.
“There is no finish line”
I should start using it myself.
There are a lot of discussions around build vs buy that I participate in, especially when it comes to the decision to build a WebRTC infrastructure versus buying an existing one via CPaaS vendors. In many cases, the argument and focus is on the initial development effort and a lot less on maintenance. The thing about maintenance is that it is almost as hard as the initial development, especially because there is no finish line – the product team will always ask for more features and capabilities which will drive more investment.Twilio Microvisor
The first announcement made during the keynote was about a new product – Twilio Microvisor.
The Twilio Microvisor is an extension of the Twilio Super SIM and its Internet of Things initiative, which many don’t even couple and view as CPaaS (I’ve been ignoring it as well).
The world of IOT and M2M is a challenging one. It includes different networks and carriers, differences in geographies and regulation, different hardware devices and chipsets.
Earlier in the year, Twilio acquired Electric Imp. This acquisition is now the Twilio Microvisor.
Up until now, the only real touching point that Twilio had with the physical world was their Super SIM. With Microvisor (and Electric Imp) that changes, and Twilio is mucking around with microcontrollers, firmware and hardware.
It the special announcements session, Evan Cummack, GM of IoT at Twilio, explained that there was a gap in the market – as a developer you either had to begin from scratch or use readymade solutions:The gap between IOT alternatives of developers: DIY or bespoke solutions
He ignored a few of the competitors for the Twilio offering, but these are less flexible and open anyways.
What Twilio is doing with Microvisor, is taking care of a few important aspects of IOT development:Twilio Microvisor features takes care of the heavy lifting of security for developers
- Secure Boot
- Secure FOTA (Firmware Over The Air)
- Secure Debug
- Secure Communications
The secure part here is key, as it is the one thing we struggle with greatly in IOT these days. This solution will remove a lot of the headaches of IOT development and get more products released.
It is also where Twilio is competing not with other CPaaS vendors but rather with cloud vendors, who also started offering IOT tooling in recent years.Twilio Video WebRTC Go
Coming from the Video and WebRTC space, this is where I am most frustrated.The need and growth of video
With the pandemic going on, Twilio had to do something about video, an area where little investment on their part has taken place. Until 2020, this has been understandable. Growth came from elsewhere and it didn’t seem like video is that important.
All this has changed. Zoom exploded, Agora.io had a great IPO, and Twilio itself saw an increase of 500% of daily usage for its video.Twilio reiterating the need and uses of video communication
The one to talk about Twilio Programmable Video was Michelle Grover, Chief Information Officer. Her part of the keynote revolved around the market need. The main market verticals here were retail and health.
It was more a reminder that Twilio is doing video than anything else.The new WebRTC announcement
The new announcement? Twilio Video WebRTC Go
What is Twilio Video WebRTC Go?
- A free, hosted WebRTC service
- Peer-to-peer, 1:1 sessions only
- Limited to 25 GB/month of TURN for media relay
For context, pricing of 25 GB/month on Twilio’s TURN servers in the US is $10/month.
If you developed your own signaling and your own application, relying on Twilio’s TURN servers, then switching to Twilio Video WebRTC Go will save you $10.
But what you really get here is Twilio Video P2P that costs $0.0015/minute. In this configuration, you get the full infrastructure and support of Twilio’s signaling, logging and SDKs practically for free if your service is smaller than 25 GB/month of TURN media relay. How many video sessions can this accommodate? That’s something you’ll need to calculate.
For Twilio this is a win, as it gets more companies to adopt its Programmable Video at a very low price to Twilio (remember – video isn’t a serious money maker for Twilio yet, so helping these smaller users to grow their business and then have them start paying is just fine). With all the video API services out there, a free offering from a large vendor is a first. While limited, it is probably useful for many companies starting their way with 1:1 video calling.On open source and Twilio
The fact that Twilio is calling their reference apps “Open Source Video Collaboration Apps” is a bit silly. These are references/samples running on top of the Twilio Programmable Video API and are not meant, designed or easily usable on top of any other vendor or on top of any other infrastructure.
Calling a piece of code, no matter how big, open source, while forcing its user to consume other paid services in order to use it is not exactly open source.
This isn’t to say that this open source reference app isn’t useful. It surely is most useful. It gives developers a better starting point for their application, and Twilio has taken the time at Signal to offer a session titled “Accelerating Development of Collaboration Apps with Twilio Video” dedicated exactly to this.
It is a trend I see of CPaaS vendors going towards higher level abstractions. Twilio is doing that with nocode (=Twilio Studio), programmable enterprise (=Twilio Flex), reference apps for video (this one) and now with Frontline (later in this article).Nothing new under the sun here
For me this says that Twilio hasn’t invested in video as much in the last year or two. If they had, they would have announced something more thrilling and interesting. Maybe larger meetings, above 50 participants? Broadcasting capabilities? Noise suppression? Something…Twilio Flex ecosystem
The keynote and the session had a lot of Twilio Flex content in them. This is less about developers and more about contact centers.A show of force for Twilio Flex, but sharing customer logos
In this event, Tony Lama, Vice President, Contact Center Sales at Twilio mentioned in brief the fact that many features were added to Flex, but didn’t really delve into them too much. The focus was on the fact that Flex has customers and now has a thriving ecosystem of partners as well.Lots of new features, none interesting enough for the keynote
The main target for this year were the on premise contact centers – this is where Twilio is setting its sights – in the transformation these contact centers are going through as they are heading to the cloud (forced to do so earlier rather than later due to the pandemic).
This is why Twilio decided to focus on the ecosystem, making it into a big announcement:
This targets exactly the on premise contact centers, where large deployments with many agents and a lot of custom integration code and features were added over the years. An ecosystem around Flex gives Twilio the reach it needs.
It is also why Twilio introduced its latest Flex partner – Deloitte Digital – who offer system integration in this target market.
Twilio Flex and its current set of announcements is less about CPaaS and developers and more about content center as a service (CCaaS).Twilio Frontline
In that vein, the announcement of Twilio Frontline was made.
Interestingly, this was introduced by Simon Khalaf, SVP and GM, Messaging at Twilio.
Twilio Frontline is a new complete, closed, mobile application and service which enables employees in a company to directly communicate with customers through messaging channels.
The main benefits touted about Frontline? SSO (Single Sign-on) and CRM integration
- Both of these features aren’t building blocks or APIs at Twilio, which begs the question why not
- There’s nothing about programmability, APIs or building blocks here. This isn’t something by developers for developers
This is far remote from the developer roots and target audience of Twilio, so it will be interesting to see how this plays out and redefines Twilio itself. My guess is that Frontline started as a skunk works project during the pandemic, one that turned into a new product that is now looking for a home at Twilio and within its bigger storyline.
I wonder though, was this built on top of Twilio Conversations, which was introduced at Signal 2019, or is it something implemented on top of Twilio Flex?
If this was implemented on top of Twilio Flex (which I believe it was), then why is the SVP and GM of Messaging at Twilio the one introducing it? And why wasn’t it designed, developed and even introduced as a programmable solution? Part of Flex. Maybe even an “open source application” on top of Flex.
Frontline is an interesting product. But what does it have to do with Twilio?Other announcements
There was little in the keynote of Twilio about APIs and CPaaS and more about the higher level abstractions and complete applications (Flex and Frontline). This shows a maturity level at Twilio, where most of the CPaaS domains are already well covered by their APIs.
Two additional announcements of new features/products were made, though not in the keynote itself.Twilio Event Streams
That trillion human interactions? These are probably just events in the Twilio system:
This is the slide shared in the session discussing the new feature/product of Twilio Event Streams. It isn’t a trillion but it is close enough.
What Twilio did was consolidate all of its events into a single hook, calling it Event Streams, offering a single integration point for collection of events. The first sink selected for these events is Amazon Kinesis, with more to probably be added later, based on customer demand.
Moving towards consolidated data management shows maturity and an increase in the customers that are using multiple Twilio products.Twilio Verify Push
Another new product/feature is Twilio Verify Push. This enables a mobile application to be used as a trusted device/app to validate login on another device (as well as on the device itself). The end result is reduction in the SMS volume.
While nice, I am waiting here for Google and Apple to close this gap and offer their own verification mechanisms to all instead of having application developers rely on third party services.
As for Twilio, this makes for a sensible and useful addition to their Twilio Verify service.Machine Learning was missing
What was missing at Twilio Signal 2020 is AI and machine learning.
No really interesting improvements shared about Twilio Autopilot. No cool introduction of noise suppression or other media processing machine learning capability. Nothing.
There were a few mentions on how Autopilot is used by customers during the create bots in order to deflect calls and handle the volume (nice stories that we’ve heard would be the main use case for Autopilot already).
The only “real” thing around AI? At the end of the keynote, Jeff Lawson had his short “live” coding session.Jeff, coding “live”. Still magical
This time, he went for using OpenAI’s GPT-3, a per-trained natural language processing engine. He made it understand TwiML constructs (the XML format used by Twilio sometimes) so that users can write a sentence of what they want, and the service would generate the TwiML for them. A nice toy to play with. I wonder what people would do from here with it, as it opens up a lot of questions, thoughts and ideas.
Machine learning is one of the main pillars I see in post-pandemic CPaaS offerings. Twilio has the skill set inhouse to pull this off, but they need to focus there more than they are doing today. They should probably also partner or acquire in this space to keep in pace with where the industry is headed.The coming CPaaS fight is in the enterprise
The enterprise story of Twilio came at the beginning of the keynote. Jeff wanted to make sure everyone knew and understood that Twilio is ready for the enterprise and being used by the enterprise. The careful selection of guests throughout the keynote showed that as well – they were all established enterprises. No cool startup this time. No crazy garage developers. Just formidable businesses that existed for years.Twilio is ready for the enterprise, with all the relevant certificates and procedures
I decided to leave this to the end since this is where Twilio is being challenged.
The challenge comes in the form of Amazon and Microsoft going towards CPaaS. Both of these vendors are:
- Bigger, with a wider breadth of products and services targeted at developers
- Attractive programs for startups, giving them free “cash” on their platforms
- Better access and relationships with enterprises
- Global coverage and partner programs that are richer in depth, breadth and reach
Amazon will probably introduce machine learning capabilities such as noise suppression as part of its CPaaS offering soon. They have it available in Amazon Chime, so placing it in the Chime SDK is the next logical step.
Microsoft runs their CPaaS on the same infrastructure that Teams is running on. Twilio touts 3B video minutes a year while Microsoft Teams has up to 5B meeting minutes a day. I am sure that it accumulates to a considerably larger number than 3B video minutes a year.
Both Amazon and Microsoft have ways to go in stabilizing their APIs and attracting developers and attention to it. They might not be highly interested in this CPaaS business as much as Twilio is, so would probably never reach the same level of maturity and breadth of features and flexibility of Twilio. But they will surely win market share. Market share that could have easily been Twilio’s.
What is also very interesting to note is that while Amazon and Microsoft made a point of not mentioning WebRTC in the front of their CPaaS platforms (both of which are video first and use WebRTC), Twilio decided to bring WebRTC to the front with their new offering of Twilio Video WebRTC Go. I wonder which works better for enterprise sales.
Anyway, with 75% of contact centers still on premise, the enterprise market as a whole is still only starting its path towards digital transformation and with the new phrase I just adopted of “there is no finish line”, there is definitely room for growth for Twilio and its many competitors.
Interesting times ahead of our industry.
The post Twilio Signal 2020. I expected more from the leading CPaaS vendor appeared first on BlogGeek.me.
Amazon Chime SDK and Azure Communication Services mark the entrance of the cloud giants to the CPaaS space, and they are doing it from a WebRTC API angle.
Ever since Twilio became popular, a question was raised over and over again:
When will one of the large IaaS players (Amazon, Microsoft or Google) acquire them or start competing with them directly?
There was no good answer. At least not until 2020, where 3 things happened:
- The pandemic hit us and we had to stay at home and shelter, or whatever
- Video exploded
- Amazon Web Services and Microsoft Azure both launched their CPaaS offering
This. Changes. Everything.
(it doesn’t. It changes only some things, but bear with me)
I already discussed how the pandemic changes priorities for CPaaS vendors. This new development is going to make things more of a mess.Why now?
Amazon Chime SDK was already announced and launched close to the end of 2019. They already have customers and success stories under their belt. Why am I just now getting to look at how IaaS vendors are changing the market?
Probably a bit because I am doing the update to my WebRTC API platforms report this month. But also because of Microsoft’s announcement of their Azure Communication Services.Amazon Chime SDK
Amazon started the work to video communications by the introduction of Chime a few years back. Chime is an enterprise communication service (in the UCaaS space), which is akin to Zoom, Google Meet and Microsoft Teams. It enables companies to communicate internally and externally via video and voice with a better set of collaboration tools than just phone calls.
For some time now Amazon Chime was also offered as a whitelabel solution that vendors could “make their own” and integrate it with their service. But it doesn’t allow for much flexibility in terms of the workflow, business logic and user authentication. This has led Amazon to introduce the Amazon Chime SDK.
The Chime SDK is one rung lower in the stack. It enables a developer access to the logical building blocks of communications, offering a pure communication API that can be used to connect to any other service. A direct competitor to the other CPaaS vendors offering video capabilities.
What Chime SDK did to really disrupt the market was lower the price point per minute. It comes at a rate of $0.0017 per user per minute. Twilio answered with its own price drop in September 2020:A 60% reduction in Twilio Programmable Video price points
The new rates are still above the Amazon Chime SDK price points, but they are 40% their previous price points.
It should be noted that peer-to-peer calling available in Twilio Programmable Video is at $0.0015, lower than the Amazon price, but of a slightly different service and feature set.
What Amazon is “selling” here? The AWS story. From the main Chime SDK page:
AWS Lambda is already there. Connectivity to other AWS services are also part of the bigger spiel.Azure Communication Services (AKA ACS)
Microsoft just announced Azure Communication Services in a public preview. This is a full CPaaS offering that includes Video, Chat, SMS and Telephony calling. The interesting tidbits alluded to in the announcement:
- Azure enabled, with all the knobs and pieces to connect it to other Azure services; along with the security and compliance of the Azure cloud
- Connectivity with Microsoft Teams, which isn’t available yet in the public preview
Watch that video above. There’s a visual explanation of remote visual assistance. I’d never think of explaining embedded video communications or programmable video communications this way – because I am in the industry for this long. What Micsoroft is doing here is educating the market in the most basic way possible. Something we were missing in our market without even knowing it. This type of an approach can work well in the enterprise space, which hasn’t adopted such services in droves just yet.
What makes this so interesting is this:
- Microsoft is the only CPaaS vendor who has a huge UCaaS offering. Huge as in up to 5B (or more) meeting minutes a day. Starting off with the same underlying scalable infrastructure means resilience, reliability and scale
- This is part of Azure and not tied to Teams. Like the AWS Chime SDK offering, the tie in with machine learning in their compute cloud brings value to developers using Azure already
- Microsoft has Office as another huge asset. If they can make the connection to it here, this is another great differentiator
On pricing, Microsoft was a bit more traditional and less bold than Amazon, sticking to the $0.004/minute price point the market seems to have adopted.The new model for Video CPaaS?
Even before Amazon and Microsoft joined this space, there were two objectives you could see in the mid-term and long-term roadmap for video CPaaS vendors:
- Add support for machine learning
- Introduce higher level of abstraction
These map where the new video CPaaS is headed, and the fact that Amazon and Microsoft both come with this “built-in” will accelerate things further.Machine Learning
Everyone’s doing machine learning these days, and it is part of the future of communications and WebRTC.
Amazon Chime SDK will be offering their noise suppression capabilities. Connect to Kinesis and enable access to all their other machine learning services.
Microsoft in their launch already mentioned Azure Cognitive Services as something that plays/will play nice with ACS.
Other CPaaS vendors are figuring out their way in this space as well, but part of their offering is usually how to gain access to the media for… sending it to the cloud for machine learning analysis. That cloud is going to be AWS and Azure more often than not. Being in that cloud to begin with is going to be an advantage for these cloud vendors and their CPaaS offerings.
Also remember that cloud vendors live and breath machine learning already. CPaaS vendors? Less so.Higher abstractions
Everyone in this space is talking about simplicity now.
How can I get developers to do their work in hours versus days. Days versus weeks. Weeks versus… no… weeks is too long already.
While this is unrealistic for a full fledged, polished service, it is something that works well towards an MVP or a first stab at a ready product.
Some do this by offering open source or reference applications on top of their CPaaS APIs. Others by offering this as a set of ready-made and highly configurable widgets.
It doesn’t seem like anyone has cracked the code of what is needed here, but the growing focus shows there’s something missing. Especially if we want developers to need to know less about WebRTC and media routing and more about their application logic.
I think that Amazon and Microsoft joining this market will speed up the efforts in this domain, as companies search for differentiation and quick onboarding.Why telephony is dying and communication is growing
Both Amazon and Microsoft are leading here with video, adding chat and telephony later. Later can be immediately after the initial launch, but it is still later.
In the past it made sense to do the opposite. Lead with PSTN and SMS as money makers, and add WebRTC voice and video, waiting for them to grow in adoption.
Taking the opposite approach shows where the future of consumption is.Winners
Who are the winners when CPaaS is done by the cloud vendors?Users
If cloud vendors are joining this game, it means there’s enough $$$ in this market to make it interesting, which means more users are consuming such services.
The market education that these cloud vendors are capable of doing and their reach is higher than the other CPaaS vendors, excluding maybe Twilio. This will end up with more enterprises and businesses offering such services and end users using them.Tier 1 cloud vendors
Amazon and Microsoft. Their timing couldn’t have been better.
If I haven’t known that Bill Gates is causing the pandemic so he can chip us all when his vaccine comes to market and causes all birds to fall from the sky due to 5G, I’d might end up saying that Jeff Bezos is to blame because he wanted the Chime SDK to grow in market share.
In all seriousness though, this gets both Amazon and Microsoft in front of the developers that use them for additional types of services that these developers are going to consume.Smaller cloud vendors
Digital Ocean and Oracle.
Why are they winners? I am not sure how Twilio can continue running Programmable Video on top of AWS and compete with AWS Chime SDK on price and geographic spread.
Same for the other CPaaS vendors who might be using AWS or Azure. They will be thinking hard if they want to keep their media stacks on these platforms or move them elsewhere. They can move them to Google Cloud, but Google just might introduce the same capabilities and become a competitor. Next in line will be Digital Ocean and Oracle, both cloud vendors that are carrying real time media traffic already. If I were a sales person there, I’d pick up the phone today and call the CPaaS vendors one after the other…Developers
A definite win. More choice. In clouds they already use. With a price war coming up.
What’s there to lose?Losers
Who are the losers when CPaaS is done by the cloud vendors?CPaaS vendors
They now have more competition. And not from smaller startups, but rather from the leading cloud vendors.
Cloud vendors already cater to developers, and a larger audience of developers.
Things are going to get interesting for these vendors, as they need to rethink differentiation, their own infrastructure and their pricing.Twilio
Twilio is the leading CPaaS vendor today.
They are using AWS. Everywhere.
This is definitely hurting them and will hurt them more moving forward.
Out of all the threats to Twilio, having cloud vendors competing head to head with them was the biggest one, and it is now happening.
It made sense for someone like Amazon to acquire them and use them as the communication stack for AWS. now it won’t happen.
Maybe Google will acquire them, though this seems far fetched to me.Google
3 leading cloud vendors.
- Now has AWS Chime SDK
- Lots of adjacent services for developers
- Now with Azure Communication Services
- Lots of adjacent services for developers
- Owner of Microsoft Teams, used as the underlying technology and media stack, with the ability to connect ACS to Teams if and when needed
- Got Office 365 as another huge asset
- Nothing in communication APIs
- Owner of Google Meet and Google Duo
- Leveraging RCS with carriers and in Android
- Has G Suite and Android as huge assets
- Has Chrome and Chrome books as assets
- Did I say no communication APIs?
Google is left behind in its communication APIs for developers, which is sad, considering they are the main driving force behind WebRTC.
I wonder if and when will Google close this gap.Developers
This will definitely rattle the existing vendors. Some of them might not make it through. So choice will again get a wee bit limited as this plays out.
While cloud vendors are great, their support isn’t the best. They tend to offer support to the smaller developers and companies through third parties and not directly, so there’s going to be less of that available. And that for a domain that is still very complex in its nature.
Developers both win and lose from this development.Updating my WebRTC API report
There’s a lot of change in the CPaaS domain. I mostly look at these vendors from a WebRTC prism, but not only.
This past month I’ve been working on updating my Choosing a WebRTC API platform report. I had a lot of briefings with the various vendors, researched their websites, added vendors, removed vendors. Grueling work.
The updated report will be published during October. It will include ~25 vendors, and touch everything from build vs buy, selection KPIs, vendor listing and pricing.
If you are looking to understand this domain better or need to select one vendor over another for an important project, then this report is for you. From today and until the report gets published, there’s a wee bit over 25% discount using coupon code API2020LAUNCH. Purchasing the report now will give you access to the current report as well as the fresh update once it is available.
The post Cloud giants joining the WebRTC API game. How is that changing the CPaaS landscape? appeared first on BlogGeek.me.
The pandemic is changing everything. CPaaS providers need to change their priorities and focus as well.
It is around this time of the year that I start thinking about where the CPaaS market is headed.
The pandemic is an epochal event. It caught the CPaaS industry somewhat ready, with gaps found in their video offerings. Behind the pandemic, a few other market changes are taking shape, affecting how CPaaS providers need to plan ahead.
I’d like to look at a few of these trends and outline what I see as the basis of CPaaS competition for the future.CPaaS features map CPaaS marketecture and features map
The diagram above shows the CPaaS features map. It is a kind of a marketecture diagram of the various bits and pieces that make up CPaaS.
I’ve layered it from Infrastructure, through Communications Building Blocks and Higher Abstraction to the Simplified Runtime domain. While not all CPaaS vendors will fill all building blocks in this map, they all see it in front of them one way or another.
Here are a few things to note:
- I’ve decided not to place Email or IoT in here though I could without much effort
- The importance of each block will be different for different customers and will change over time. The pandemic certainly changed priorities shifting them towards Video for example
- I am using the term Studio, though Flow is the one that is used by most of Twilio’s competitors
- ML stands for Machine Learning and it has its place throughout the CPaaS product stack. More on that later
If I had to map priorities for 2021, I’d probably create this heatmap:CPaaS areas of investment in 2020-2021 The pandemic and CPaaS vendors
In many ways, the pandemic is accelerating the need for CPaaS providers. The world switched en masse from one of physical interactions to a virtual one. This, in turn, exposed a few aspects in the CPaaS market.Digital transformation fast forward
The image above circulated on Twitter some time in March-April this year. It is spot on.
Digital transformation is here and it is here to stay. It came about a few years faster than expected and to get by, companies are relying more on communications and a lot of it comes today from vendors who use CPaaS or by developing the solutions needed on top of CPaaS platforms.
The thing is, in many cases, the increase is also catching businesses off guard, with call centers and support teams being overwhelmed with incidents. And that at a point in time where everyone is forced to work from home – including the call center agents.
This in turn, increases the requirements around technologies that assist in automation of processes and communication channels. Call deflection and agent assist solutions are taking center stage. This changes a bit how CPaaS vendors need to treat communication APIs, and especially what these APIs need to enable.
Are we looking now for more or less Uber-like solutions of matching a customer to a service provider? Or are we more about getting hold of the interaction’s content in real time and injecting insights into it, with or without a human agent?
I don’t have the answers, but I have a feeling that they are different than they were 9 months ago.CPaaS vendors totally missed video Video growth was unexpected, catching most CPaaS vendors unprepared
Yap. We had CPaaS vendors doing video. A few of them. And they’re just fine. Up until the point that video becomes important for everyone and that totally new use cases pop up in our market on almost a daily basis.
Zoom doesn’t mean a magnifying glass anymore. Nor is it talking about getting a closer look.
During the pandemic?
- Daily officially launched. And raised money
- Dolby.io launched
- Agora raised some $350M in their IPO
All of the above? Focus on video communications. None of them have any telephony roots or strong telephony capabilities. No phone numbers or SMS capabilities to speak of.
AWS decided it would be nice to join the frey, so they launched their own Chime SDK. With price points that challenge the existing players.
Twilio decided this month to lower their video price points. Cutting them down by some 60%.
8×8’s Jitsi is coming up with its own managed video API service, pricing it around MAU as opposed to the more common per minute pricing.
There’s a minor price war coming up around video APIs. It will be interesting to see how this plays out.Lack of WFH tooling in CPaaS
WFH = Work From HomeWorking from home isn’t just working from a different location
Welp… we’ve built all these nice communication services, but we’ve designed them mostly to work for the office.
On premise call centers moved to the cloud by adopting CPaaS, which is great, but the workforce itself still came to the office. All calls and communications took place from a controlled and managed environment.
The pandemic has forced call centers of tens of thousands of agents to stop coming to the office while continuing to work. From home. How do call center managers know anything about the environment of the home employee? How can he make sense of the quality of experience his agents and his customers are getting?
From the interest we see at testRTC in our qualityRTC service, there’s a real gap there.
Call this self promotion, but it is one of many areas where CPaaS vendors need to improve in order to offer a suitable WFH solution. Giving APIs is nice. Giving backend network insights and quality related dashboards is nice. Giving pre-call tests capabilities is nice. But I am not sure it is enough anymore.
Other aspects of WFH that aren’t catered for by CPaaS vendors? The need for noise suppression and background blurring/removal – to fit into the current work environments of call center agents and other workers.The pandemic will pass, but digital transformation won’t Are we really in a new normal?
It was supposed to be a quick 2 months thing. Maybe 6. A year tops.
Then came Google and Facebook (not governments, because they can’t seem to be so realistic and pessimistic with their citizens), and simply let anyone work from home at least until July 2021. At least.
Fujitsu? Decided to cut office space by 50% in 3 years as the new normal.
LivePerson, an Israeli company with 1,300 employees decided to give up on its offices altogether and go 100% WFH. This saves money and apparently most employees prefer it while management doesn’t see enough of a degradation in production output.
This obviously isn’t the case everywhere. In a recent interview with the The Wall Street Journal, Reed Hastings, CEO of Netflix had this to say about remote work:
“I don’t see any positives. Not being able to get together in person, particularly internationally, is a pure negative. I’ve been super impressed at people’s sacrifices.”
To some degree, he is correct. It greatly depends on the type of industry and company.
Dean Bubley says it best about business events:
In-person business events will rise again, although I’m less certain about office work.[…]
The #NewNormal will not be 100% remote. Once a vaccine is available, I hope that it isn’t even 50% #WFH.
My wife is a Pilates and Salsa dance teacher. She needs to work remotely now from time to time, with Zoom and recorded lessons. Her students? They’re fine with it, but whenever they can come over or do a face-to-face-in-the-flesh lesson – they’d take the opportunity.
This means that whatever it is CPaaS vendors are seeing as requirements may well stay and stick with them for the long run. What we have now isn’t a new normal, but there’s no going back to the old normal either.3 pillars of CPaaS competition and differentiation in 2021
When I had to decide what are the main areas of investment for CPaaS when it comes to differentiation and competition towards 2021, I came to these 3 domains: machine learning, video and diagnostics.
There are two reasons why I chose these domains:
- Renewed focus on IP based communications. WebRTC and VoIP are becoming paramount to the growth and future of CPaaS. SMS and phone numbers are great money makers, but they’re not the future. The pandemic threw us a few years into the future, accelerating this trend
- Competing with in-house development. Phone numbers are complicated. Not because they are technically complex, but because they require haggling and contracting with multiple carriers around the globe, which gives an immediate advantage to CPaaS providers. With WebRTC that doesn’t exist anymore, and in-house becomes a bigger competitor to CPaaS providers. The domains below will increase the gap between build and buy for potential clients and also increase the perceived value of a solution
Noise suppression. Background replacement. Super resolution. Bandwidth estimation. Packet loss concealment. …
All these are algorithms in the media processing domain affecting the user experience in communications. Like everything else they are now shifting towards using a lot more machine learning than in the past.
The current forerunner in importance and mindshare is noise suppression, with a lot of partnerships and M&A activities around it.
When it comes to machine learning in media quality, what are CPaaS vendors doing today?
Almost nothing at all.
- Dolby.io have noise suppression and other audio enhancements
- Agora actively invests in machine learning. They’ve spoken about it at Kranky Geek multiple times, around super resolution and video encoding
- AWS added noise suppression to Chime, which is most likely part of the new Chime SDK. They even won an award for their noise suppression capabilities
The rest? Not doing much about machine learning, researching or doing bots.
This cannot last.
We’ve already seen how WebRTC is being unbundled for the purpose of differentiation. That differentiation will come in the form of optimizations, mostly done by use of machine learning.
What will vendors do? Especially when we see the leading UCaaS vendors actively investing in machine learning media processing capabilities? This sets the bar to what a communication service needs to look like, and without such capabilities, why should I as a developer use that CPaaS vendor?2# – Video, Video, Video Tony Robbins going virtual. Is this a CPaaS implementation???
Did I already say we’re in the year of the video?
A billion have been indoctrinated over a period of 1 month this year on how to use Zoom. don’t nitpick me on the exact number please. My mother now users Zoom in her daily life of a variety of activities, including a book reading club she joined
Many CPaaS vendors had video capabilities but they usually amounted to 1:1 interactions or small group sizes. There isn’t a day going by where I don’t get a new requirement from someone that CPaaS providers can’t cater for today. Many of these are in the domain of broadcasts and large groups (100 or more participants). Using CPaaS for them today feels like hacking at best. Impossibly challenging at most.
There are many areas where CPaaS providers are lacking when it comes to video. Here are the few that immediately come to mind:
- What we are seeing is a rapid growth in the feature set and requirements of video centric use cases. These needs to be addressed. As a simple example, how do you do a live session with one presenter streaming to a large audience and the audience in turn sending their own video to the presenter, so that the presenter sees them all at the same time (or can alternate between them)?
- There’s a blurring of the lines between voice, video, broadcast and streaming. There’s a need to seamlessly switch from one to the other. Broadcast and streaming comes today predominantly from non-CPaaS vendors. There’s a growing pressure for these to be wrapped into CPaaS for interactive use cases
- Price points of video services need to be adjusted. With the change brought by AWS Chime SDK, and the pricing model of 8×8 JaaS, there are bound to be changes for other CPaaS vendors. This is imperative, especially when build vs buy decisions rely so heavily on back of the napkin calculations of minutes use multiplied by a static number
- Location of data centers and the latency brought about due to it. Most CPaaS vendors have 10 or less data centers they operate from. Now that everyone is using video, this just isn’t enough. It might be nice for voice calls in call centers, but video calls the world over are different – and they take place a lot more locally within regions and countries now, so having data centers closer to users is becoming more important than ever
The investment in video communications in all its facets will be important to stay competitive in this space.#3 – Diagnostics and analytics
It is great that you can communicate, but what happens when things go haywire?
In my recent round of updates I am doing for my Choosing a WebRTC API Platform report, many of the vendors made sure I know they have a dashboard for quality and network monitoring. Different vendors give it different names, but they all understood that unlike telephony, there’s a need for insights here, especially since networks are unmanaged.
It isn’t about me as a client understanding if the CPaaS vendor is doing a good job, but rather about me understanding my users’ networks and experience. Current dashboard solutions will need to evolve further to give the insights their customers are looking for.Didn’t you miss anything?
In my future of CPaaS article from last year I mentioned a few additional trends. Some of them have been reiterated here, though from a different angle and with a different narrative that fits better with the changing times.
There were three topics that weren’t mentioned here yet, and I want to give them a bit of room and explain where I see them in 2021 with CPaaS.nocode / low code
Still a thing. Serverless, Flow, Zapier integration, drag and drop tools. All there. All needed.
For the most part, CPaaS vendors seem to be content with the current state of affairs and the current tools they have. Investment in this domain in 2020 didn’t yield anything vastly different, new or interesting.
The domain of nocode is still relevant and interesting. For now, it seems to be mostly limited to the telephony (and voice) aspects of CPaaS.CCaaS and UCaaS
The lines are blurring elsewhere as well. Areas of IoT (below), messaging and notifications, live streaming – are all suitable adjacencies for expansion of CPaaS vendors.
The largest areas though are CCaaS and UCaaS: contact centers and unified communications
Acronyms will be tricky here. So bear with me.
- CCaaS and UCaaS are investing heavily in ML. A lot of it now is around #WFH
- CPaaS is going up the foodchain, mainly after CCaaS. Some do it directly (Twilio Flex), others pivoting sideways to conversations (MessageBird Omnichannel Chat Widget)
- UCaaS is vying towards CPaaS, introducing their own APIs and even CPaaS offerings
In another world, just next by, other SaaS solutions are blurring their lines. Gist (the chat widget I am using on my WebRTC course site) announced to its customers that it is releasing a full fledged CRM. From conversations to CRM.
CRMs in turn, can use CPaaS vendors directly to build up their own CCaaS offering. With the higher level abstractions geared towards customer engagement, CPaaS vendors now offer a simple route for CRMs in this direction.
This will continue, though I don’t see it as direct competition or real differentiation within the CPaaS domain itself.IoT
Twilio seems to be the only CPaaS vendor investing in the Internet of Things. It acquired Electric Imp earlier this year. The acquisition wasn’t made with much fanfare, as this isn’t the main focus of Twilio and the current market is interested less in IoT than it is in video calls.
Is IoT part of CPaaS? Time will tell.
I believe that it is, but for now, only Twilio seems to be investing in that domain where none of its other immediate CPaaS competitors have the appetite for it. This will not change in the next couple of years as focus for CPaaS is elsewhere at the moment.Updating my WebRTC API report
There’s a lot of change in the CPaaS domain. I mostly look at these vendors from a WebRTC prism, but not only.
This past month I’ve been working on updating my Choosing a WebRTC API platform report. I had a lot of briefings with the various vendors, researched their websites, added vendors, removed vendors. Grueling work.
The updated report will be published during October. It will include ~25 vendors, and touch everything from build vs buy, selection KPIs, vendor listing and pricing.
If you are looking to understand this domain better or need to select one vendor over another for an important project, then this report is for you. From today and until the report gets published, there’s a wee bit over 25% discount using coupon code API2020LAUNCH. Purchasing the report now will give you access to the current report as well as the fresh update once it is available.
The post What should CPaaS providers do today to prepare for the “post pandemic”? appeared first on BlogGeek.me.
Communication vendors are waking up to the need to invest in ML/AI in media processing. The challenge will be to get ML in WebRTC.
Two years ago, I published along with Chad Hart a report called AI in RTC. In it, we’ve reviewed the various areas where machine learning is relevant when it comes to real time communications. We’ve interviewed vendors to understand what they’re doing and looked at the available research.
We mapped 4 areas:
- Speech Analytics
- Computer Vision
- RTC Quality and Cost Optimization
That last area was tricky. Almost everyone was using rule engines and heuristics at the time for all of their media processing algorithms and only a few made attempts to use machine learning.
My argument was this:At some point, applying more heuristics to media processing algorithms loses its appeal
There’s so much we can do with rule engines and heuristics. Over time, machine learning will catch up and be better. We are now at that inflection point. Partially because of the technology advances, but a lot because of the pandemic.ML in media processing is challenging
When looking at machine learning in media processing, there’s one word that comes to mind: challenging
Machine learning is challenging.
Media processing is challenging.
These are two separate and far apart disciplines that need to be handled.
The data you look at is analog in its nature, and there’s often little to no labeled data sets to work with.
A few of the things you need to figure out here?
- How do you find machine learning engineers, or whatever they are called in their titles this day of the week?
- Do these engineers know anything about media processing? How do you get them up to speed with this technology? Or is it the other way around? Getting media engineers trained in machine learning
- Can you generate or get access to a suitable data set to use? Do you even have access to enough data?
- Where do you focus your efforts? Audio or video? Maybe just network? Should you go for server side implementation or client side one? What about model optimizations?
- When do you deem your efforts fruitful? Ready for production?
This isn’t just another checkmark to place in your roadmap’s feature list. There’s a lot of planning, management effort and research that needs to go into it. A lot more than in most other features you’ve got lined up.The noise suppression gold rush
If I had to pick areas where machine learning is finding a home in communications, it will be two main areas:
- Video background processing (more on that at some future point in time)
- Noise suppression
Both topics were always there, but took centerstage during the pandemic. People started working from weird places (like home with kids) and you now can’t blame them. One of the best games I play in workshops now? Checking who’s got the most interesting room behind him…
Video background is about stopping me from playing. Noise suppression is about you not hearing the lawn mower buzzing 16 floors below me, or the all-too-active neighbor above me who likes to home renovate whenever I am on a call – with a power drill.How I think my neighbor looks like whenever I am on a conference call
This need has led to a few quick wins all around. The interesting 3 taking place in the domain of WebRTC (or near enough) are probably the stories about Google Meet, Discord/Krisp and Cisco/BabbleLabs.Google Meet Google Meet built its own noise suppression technology
In June, Serge Lachapelle, G Suite Director of Product Management was “called to the flag” and was asked to do a quick interview for The Verge on Google Meet’s noise suppression. Serge was once the product manager for WebRTC at Google and moved on to Google Meet a few years back.
You can watch the short interview here:
The gist of it?
- Google decided to implement it in the cloud
- They use “secure” TPUs for that (Tensorflow Processing Units, a specialized chip in the Google Cloud for machine learning workloads)
- The feature is optional. It can be enabled or disabled by the user
- Noises it cancels are almost arbitrary. It is something that is really hard to define as initial requirements. It is also something that will be fine tuned and tweaked over time by Google
As I stated earlier, Google isn’t taking any prisoners here and contributing this back to the community freely as part of WebRTC. They are making sure to differentiate by making sure their machine learning chops are implemented outside of the open source WebRTC library. This is exactly what I’d do in their place.Discord Discord “bought” its way to noise suppression by partnering with Krisp
Krisp is one of the few vendors tackling machine learning in media processing and doing that as a product/service and not a feature. They’ve been at it for a couple of years now, and things seems to be going in their favor this year.
Krisp managed to do a few things:
- Focus on noise suppression. They started “all over the place” with voice related media algorithms, and seem to be finding product-market-fit in noise suppression
- Won a deal/partnership with Discord
- Got their technology to work inside the browser (see here)
The Discord story was first published in April on Discord’s blog. Noise suppression was added in beta to the Discord desktop app. Was that done using the browser technology used in Discord’s Electron app or by the native implementation that Krisp has is an open question, but not the most relevant one.
Three months later, in July, Discord got noise suppression into iOS and Android. This was also done using Krisp and with a spanking short video explainer:Ongoing success
Here are my thoughts here:
- Adding this to mobile means they got positive feedback on the desktop integration
- Especially considering how they phrased it:
As we continue to improve voice chat, Krisp is an integral part of making Discord your place to talk. No matter how stressful the world around us may be, Krisp is here to help every one of our 100 million monthly active users feel more connected to our far-away friends.
- 100 MAU is what Discord now has, and this is a vote of confidence in Krisp and in ML-based noise suppression technology
- Discord shouts to the world that they are using Krisp. Something not many companies do for their suppliers
- This may mean that they got this on the cheap (or free)
- Or it means that they are cozying up to Krisp
My read of it? Krisp might be acquired and gobbled up by Discord to make sure this technology stays off the hands of others – if that hasn’t happened already – just look at this page – https://krisp.ai/discord/ (and then compare it to their homepage).Cisco Cisco gobbled up BabbleLabs to own noise suppression technology
In the case of Cisco, the traditional approach of reducing risk by acquiring the technology was selected – acquihiring.
Last week, Cisco issued a press release of their intent to acquire BabbleLabs.
BabbleLabs was in the same space as Krisp. A company offering machine learning-based algorithms to process voice. The main algorithm there today as we’ve seen is noise suppression. This is what Cisco were looking for and now they will have it inhouse and directly integrated into WebEx.
Cisco devices not to self-develop. They also decided to own the technology. The reasons?
- Google owns it
- Zoom has its own implementation
- That left… WebEx
Will BabbleLabs stay open? No.
In his recent post about the acquisition, Chris Rowen, CEO of BabbleLabs, explains what lead to the acquisition and paints a colorful future. The only thing missing in that post is what about existing customers. The answer is going to be a simple one: They will be supported until the next renewal date, when they will simply be let go.
A win to Krisp. If it isn’t in the process of being acquired itself already.Who’s next?
This definitely isn’t the end of it. We will see more vendors taking notice to this one and adding noise suppression. This will happen either through self-development or through the licensing of third party solutions such as Krisp.
The challenge with these third party solutions is that they feel more like a feature than a product or a full fledged service. On one hand, everyone needs them now. On the other hand, they need to be embedded deep in the technology stack of the vendors using them. The end result is relatively small companies with a low ceiling to their potential growth (=not billion dollar companies). This puts a strain on such companies, especially if they are VC backed.
On the other hand, everyone needs noise suppression now. Where do they go to buy it? How do they build it?Noise suppression is just the beginning
Noise suppression is just the beginning here. In the workshop I did last month on WebRTC innovation and differentiation, I’ve taken the time to focus on this. How machine learning is now finding a place in bringing differentiation to the actual communication. Noise suppression was one of the topics discussed, with many others.
There were 3 main areas that we will see growing investment in:
- Voice treatment – noise suppression, packet loss concealment, voice separation, etc
- Video treatment – video compression, super resolution, etc
- Background blur/replacement – I am placing it on its own, as it seems to be the next big thing
Each of these domains has its own set of headaches and nuances.Server, native or browser? Should you employ ML in WebRTC in the cloud or on the edge?
This is a big question.
If you look at the examples I’ve given for noise suppression:
- Google Meet chose cloud
- Discord is native and browser
- WebEx is native as far as I can tell
Going for native or browser means you’re closer to the edge and the user. You can do things faster, more efficiently and with a lower cost to you (you’re practically employing the user’s device to bear the brunt of running the machine learning inference algorithm). That also means you have less resources for other things like the actual video and you’re limited in the size of the model you can use for your algorithm.
Cloud means a central place where you can do training, inference, A/B testing, etc. It is probably easier to maintain and operate in the longer run, but it will add some delay to the media and will definitely cost you to run at scale.
Each company will choose differently here, and you may see a company choosing for one algorithm to run it in the cloud and for another to run on the edge.Are you planning for this ML/AI future?
Machine learning and artificial intelligence is in our future. Both in the communication space and elsewhere. It is finally coming also to media processing directly. In a few years, it will be a common requirement from services.
Are you planning for that in any way?
Do you know how you’re going to get there?
Will you be relying on third parties or on your own inhouse technology for it?
There are no open source solutions at the moment for any of it. At least not in a way that can be productized in a short timeframe.
If you need assistance with answering these questions, then check my workshop. It is recorded and available online and it is more relevant than ever.
2020 marks the point of WebRTC unbundling. It seems like the new initiatives are the beginning of the end of WebRTC as we know it as we enter the era of differentiation.
Life is interesting with WebRTC. One moment, it is the only way to get real time media towards a web browser. And the next, there are other alternatives. Though no one is quite announcing them the way they should.
We’re at the cusp of getting WebRTC 1.0 officially released. Seriously this time. For real. I think. Well… maybe.Towards differentiation
If I were to chart our path through this crazy world of WebRTC, it would look something like this:2020 marks the beginning of the differentiation stage for WebRTC
Towards the end of 2019, and at greater force during the pandemic, we’ve seen how the future of WebRTC looks like. It is all about differentiation.
Up until now, all vendors had access to the same WebRTC stack, as it is implemented by Google (and the other browser vendors), with the exact same capabilities in the browser.
I’ve alluded to it in my article about Google’s private WebRTC roadmap. Since then, many additional signals came from Google marking this as the way forward.
Today, there are 2 separate WebRTC stacks – the one available to all, and the one used internally by Google in native applications. While this is something everyone can do, Google is now leveraging this option to its fullest.
The interesting thing that is happening is taking place somewhat elsewhere though. WebRTC is now being unbundled so that Google (and others) don’t need to maintain two separate versions, but rather can have their own “differentiation” implemented on top of “WebRTC”.Unbundling WebRTC
At this point, you’re probably asking yourselves what does that mean exactly. Before we continue, I suggest you watch the last 15 minutes from web.dev LIVE Day Two:
That’s where Google is showing off the progress made in Chrome and what the future holds.
The whole framing of this session feels “off”. Google here is contemplating how they can bring a solution that can fit Zoom, that when 99% of all vendors have figured out already how to be in the browser – by using WebRTC.
The solution here is to unbundle WebRTC into 3 separate components:The components set to unbundle WebRTC
- WebTransport – enables sending bidirectional low latency UDP-like traffic between a client and a “web server”, which in our context is a media server
- WebCodecs – gives the browser the ability to encode and decode audio and video independently of WebRTC
- WebAssembly – a browser accelerator for running code and an enabler for machine learning
While these can all be used for new and exciting use cases (think Google Stadia, with a simpler implementation), they can also be used to implement something akin to what WebRTC does (without the peer-to-peer capability).
WebTransport replaces SRTP. WebCodecs does the encoding/decoding. WebAssembly does all the differentiation and some of the heavy lifting left (things like bandwidth estimation). Echo cancellation and other audio algorithms… not sure where they end up with – maybe inside WebCodecs or implemented using WebAssembly.What comes after the unbundling of WebRTC?
This isn’t just a thought. It is an active effort taking place at Google and in standardization bodies. The discussion today is about enabling new use cases, but the more important discussion is about what that means to the future of WebRTC.
As we unbundle WebRTC, do we need it anymore?
With Google, as they have switched gears towards differentiation already, it is not that hard to see how they shift away from WebRTC in their own applications:Google Stadia Does Stadia has a reason to use WebRTC?
Google Stadia is all about cloud gaming. WebRTC is currently used there because it was the closest and only solution Google had for low latency live streaming towards a web browser.
What does Google Stadia need from WebRTC?
- The ability to decode video in real time in the browser
- Send back user actions from the remote control towards the cloud at low latency
That’s a small portion of what WebRTC can do, and using it as the monolith that it is is probably hurting Google’s ability to optimize the performance further.
Sending back user actions were already implemented in Stadia on top of QUIC and not SCTP. That’s because Google has greater control over QUIC’s implementation than it does over SCTP. They are probably already using an early implementation of WebTransport, which is built on top of QUIC in Stadia.
The decoding part? Easier to just do over WebTransport as well and be done with it instead of messing around with the intricacies of setting up WebRTC peer connections and maintaining them.
For Stadia, unbundling WebRTC will result moving away from WebRTC to a WebTransport+WebCodecs combo is the natural choice.Google Duo & Google Meet Meet & Duo. Will moving away from WebRTC improve their competitiveness?
For Duo and Meet things are a bit less apparent.
They are built on top of WebRTC and use it to its fullest. Both have been optimized during this pandemic to squeeze every ounce of potential out of what WebRTC can do.
But is it enough?Differentiation in WebRTC
Google has been adding layers of differentiation and features on top and inside of WebRTC recently to fit their requirements as the pandemic hit. Suddenly, video became important enough and Zoom’s IPO and its huge rise in popularity made sure that management attention inside Google shifted towards these two products.
This caused an acceleration of the roadmap and the introduction of new features – most of them to catch up and close the gap with Zoom’s capabilities.
These features ranged from simple performance optimizations, through beefing up security (Google Duo doing E2EE now), towards machine learning stuff:
- Proprietary packet loss concealment algorithm in native Duo app
- Cloud based noise suppression for Meet
- Upcoming background replacement for Meet
Can Google innovate and move faster if they used the unbundled variant? Instead of using WebRTC, just make use of WebTransport+WebCodecs+WebAssembly?
What advantages would they derive out of such a move?
- Faster time to market on some features, as there’s no need to haggle with standardization organizations on how to introduce them (E2EE requires the introduction of Insertable Streams to WebRTC)
- Google Meet is predominantly server based, so the P2P capability of WebRTC isn’t really necessary. Removing it would reduce the complexity of the implementation
- More places to add machine learning in a differentiated way, instead of offering it to everyone. Like the new WaveNetEQ packet loss concealment was added outside of WebRTC and only in native apps, it could theoretically now be implemented without the need to maintain two separate implementations
If I were Google, I’d be planning ahead to migrate away from WebRTC to this newer alternative in the next 3-5 years. It won’t happen in a day, but it certainly makes sense to do.Can/should Google maintain two versions of WebRTC?
Today, for all intent and purpose, Google maintains two separate versions of WebRTC.
The first is the one we all know and love. It is the version found in webrtc.org and the one that is compiled into Chrome.
The other one is the one Google uses and promotes, where it invests in differentiation through the use of machine learning. This is where their WaveNetEQ can be found.
Do you think Google will be putting engineers to improve the packet loss concealment algorithm in the WebRTC code in Chrome or would it put these engineers to improve its WaveNetEQ packet loss concealment algorithm? Which one would further its goal, assuming they don’t have the manpower to do both? (and they don’t)
I can easily see a mid-term future where Google invests a lot less in WebRTC inside Chrome and shifts focus towards WebTransport+WebCodecs with their own proprietary media engine implementation on top of it powered by WebAssembly.
Will that happen tomorrow? No.
Should you be concerned and even prepare for such an outcome? That depends, but for many the answer should be Yes.The end of a level playing field and back to survival of the fittest
WebRTC brought us to an interested world. It leveled the playing field for anyone to adopt and use real-time voice and video communication technologies with a relatively small investment. It got us as far as where we are today, but it might not take us any further.Recent changes marks the shift from a level playing field in WebRTC towards survival of the fittest
For this to be sustainable, browser vendors need to further invest in the quality of their WebRTC implementations and make that investment open for general use. Here’s the problem:Apple (Safari)
Doesn’t really invest in anything of consequence in WebRTC.
- They seem to care more about having an HEVC implementation than in getting their audio to work properly in mobile Safari in WebRTC
- To date, they have taken the libwebrtc implementation from Google and ported to work inside Safari, making token adjustments to their own media pipelines
- I am not aware of any specific improvements Apple made in Safari’s WebRTC implementation to quality via the media algorithms used by libwebrtc itself
Apple cares more about FaceTime than all of that WebRTC nonsense anyways…Mozilla (Firefox)
Actually have a decent implementation.
- While Firefox uses libwebrtc as the baseline, they replaced components of it with their own
- This includes media capturer and renderer for audio and video
- They have invested a lot in improving the audio pipeline in Firefox, which affects quality in WebRTC
Their latest Edge release is Chromium based.
- They aren’t doing much at the moment in the WebRTC part of it is far as I am aware
- They could improve the media pipeline implementation of Chromium (and by extension Edge) for Windows 10
- Do they have an incentive?
- Would they contribute such a thing back to Google or keep it in their Edge implementation?
- Would Google take it if Microsoft gave it to them?
And then there’s Microsoft Teams, which offers a sub par experience in the browser than it does in the native application. All of the investment of Teams is going towards improving quality and user experience in the app. The web is just an afterthought at the momentGoogle (Chrome)
Believe WebRTC is good enough.
- There are some optimizations and improvements that are now finding their way into WebRTC in the browser
- But a lot of what is done now is kept out of the web and the open source community. WaveNetEQ is but an example of things to come
- It is their right to do that, but does this further the goal of WebRTC as a whole and the community around it?
Now that we’re heading towards differentiation, the larger vendors are going to invest in gutting WebRTC and improving it while keeping that effort to themselves.
No more level playing field.Prepare for the future of WebRTC
What I’ve outlined above is a potential future of WebRTC. One that is becoming more and more possible in my mind.
There’s a lot you can do today to take WebRTC and optimize around it. Making your application more scalable. Offering better media quality as well as use experience. Growing call sizes to hundreds or participants or more.
Investing in these areas is going to be important moving forward.
I’ve recently created a workshop covering the present and future of WebRTC, along with techniques and best practices employed by vendors in this space. If you want to learn more, you may want to take that workshop.
The post WebRTC unbundling: the beginning of the end for WebRTC? appeared first on BlogGeek.me.
WebRTC IP addresses and port ranges can be a bit tricky for those unfamiliar enough with VoIP. I’d like to shed some light about this topic.
A recent back and forth discussion that I had with one of the people taking my online WebRTC course made it clear to me that there are still things I take for granted because I come with a VoIP heritage to what it is I am doing today. Which is why this article here.
Connecting a WebRTC session takes multiple network connections and messages taking place over different types of transport protocols. There are two reasons why that decision was made for WebRTC:
- There was a desire to have it run peer to peer, directly exchanging real time media between two browsers. This requires a different look at how to handle network entities such as NATs and firewalls
- Real time media is different from other data sent over the internet in browsers. The transport and signaling protocols already available were just not good enough to preserve high quality and low latency
Lets see how connections get made over the internet and how WebRTC makes use of that.A quick explainer to internet connections
We will start by looking at the building blocks of digital communications – TCP and UDP.
The table below summarizes a bit the differences between the two:TCP and UDP are two extremes of how transport protocols can be expressed TCP connections
TCP is a reliable transport protocol. As such, it has a built-in retransmission mechanisms that is meant to make sure whatever is sent is received on the other end and in the same order of sending.
To do that properly, a TCP connection needs to be created. A TCP connection is a set of 4 values:
Source IP:Source port + Destination IP:Destination Port
How does one establish a TCP connection?
On your local machine you “bind” one of the local IP addresses of the machine to a local port number. That IP and port needs to be available and not taken for something else already. Then you need to try and connect to the destination IP:port.
Let’s say I want to connect to google.com.
For me, google.com resolves to the IP address 22.214.171.124. Assuming I want to connect to port 80 (a “randomly” picked port), I’d do the following: Bind a local IP:port (arbitrary local port), and connect it to 126.96.36.199:80.
Knowing the IP:port on source and destination of the connection means knowing the connection – there cannot be two such connections. Once you bind a local port to connect it to a remote address over TCP, that port cannot be reused until the connection is closed and done with.
If I want to open another TCP connection from my machine to the same address, I will need to bind yet another port on my local address and connect it to the destination IP:port,
Obviously, there are some caveats and edge cases I am ignoring here, but for our needs, the above is enough of an explanation.UDP “connections”
Since UDP is connectionless, there’s no real connection with UDP. No context whatsoever.
To send a message over UDP, I need again the quad of values:
Source IP:Source port + Destination IP:Destination Port
But this time, there’s no real connection. What happens here is that I open a local IP:port, and whenever I want to send out a message, I just tell it the destination IP:port and be done with it.WebRTC signaling connections and addresses
WebRTC signaling is just like any other web application connection.
In order to send and receive the SDP blobs to make the connection, I need to be able to communicate between the browsers and that is done using traditional networking means available in the browser: either HTTP or WebSocket. Both (ignoring HTTP/3) are implemented on top of TCP.
What does that mean?
- When my browser connects to the signaling server, it connects to an HTTPS or a Secure Websocket address (because… security)
- The destination address will be whatever the DNS will resolve for the name of the server I connect to; and for the most part, this connection will be done towards port 443
- The local address will be whatever local address I have on my machine
- The local port will be an arbitrary local port that the operating system will allocate
The end result?The signaling server has a static IP and port, while the client is “dynamic” in nature
Local ports are arbitrary (and ephemeral). Destination port is 443 (or whatever advertised by the server).WebRTC media connections and addresses
Media in WebRTC gets connected via SRTP. Most of the time, that would happen over UDP, which is what we will focus on in this section.
In naive SRTP implementations from before the WebRTC era, each video call usually used 4 separate connections:
- RTP for sending voice data
- RTCP for sending the control of the voice data
- RTP for sending video data
- RTCP for sending the control of the video data
What happens though is this –
- You create a peer connection
- Then you add media tracks to it, effectively instructing it on what it is about to send or receive (or at least what you want it to send or receive)
- WebRTC will then allocate and bind local IP addresses and ports to handle that traffic. As with outgoing TCP connections, the local ports are going to be arbitrary and ephemeral
- The allocated IP:port addresses are now going to be used in the SDP being negotiated. These will be used as the local candidates
Since these addresses and ports are local, there’s high probability that they will be blocked by firewalls for incoming traffic.
Media servers work in the exact same way. In most cases, the addresses that they will use will be public IP addresses, but the ports will be arbitrary. That’s because media servers usually prefer handling each incoming device separately, by receiving its traffic on a dedicated socket connected to a specific port.STUN “connections” and addresses
Since we’re all behind NATs with our private IP addresses, we need to know our public IP address so we can connect to others directly (peer-to-peer).
To do that, STUN is used. WebRTC will take the media local IP:port it created (in that section above), and use it to “connect” over UDP to a STUN server.
This is in concept somewhat similar to how our signaling works – the local IP address has an arbitrary port, while the remote IP:port is known – and configured in advance in our peer connection iceServers. My advice? Have that port be 443.
Why do we do all that with STUN? So that we create a pinhole through the NAT which will allocate for us a public IP address (and port). The STUN server will respond back with the IP address and port it saw, and we will publish that so that the other side will attempt reaching out to us on that public IP:port pair. If the NAT allows such binding, then we will have our session established.The STUN server has a static IP and port, while the client (and NAT) operate with “dynamic” IP addresses and ports
The above shows how Google’s STUN server works from my machine in AppRTC:
- My private address was 192.168.1.100:57086
- The STUN address in the iceServers was 188.8.131.52:19302 (the 19302 is the static port Google decided to use – go figure)
- My public IP address as was allocated by the NAT was 184.108.40.206:57086 (it managed to maintain and mirror my internal arbitrary port, which might not always be the case). This is the address that will get shared with the other participant of the session
With TURN, the server is relaying our media towards the other user. For that to happen, my browser needed to:
- Connect to a TURN server
- Ask the TURN server to allocate an address for the relay (and let me know what that address is)
- Use that address for incoming and outgoing traffic for the remote participant of the session
The above shows how Google’s TURN server works from my machine in AppRTC:
- My private IP for this session was 192.168.1.100:57086 (it was a different port, but I was too lazy to look it up, so bear with me)
- The STUN address in the iceServers was 220.127.116.11:19305 (as with STUN, the 19305 is the static port Google decided to use – still not sure why)
- The TURN server replied back with an allocation address of 18.104.22.168:28798. This was then placed as my ICE candidate. If the remote participant were to send media towards that address, the TURN server would forward that data to me
UDP, TCP and TLS work similarly in TURN when it comes to address and port allocation. What is important to notice here is how the TURN server opens up and allocates ports on the its public IP address whenever someone tries to connect through it.Understanding port ranges in WebRTC configurations
WebRTC makes use of a range of addresses, ports and transport protocols. Far more than anything else that we run in our browsers. As such, it can be quite complex to grasp. There is order and logic in this chaos – this isn’t something inflicted on you because someone wanted to be mean.
In WebRTC the addresses and ports that get allocated by the end devices (=browsers), media servers and TURN servers are dynamic. This means that in many cases we have to deal with port ranges.
Go to any voice or video conferencing service running over the Internet. Search for their address and port configuration. They all have that information in their knowledgebase. A list of addresses and ports you need to open in your firewalls, written nicely on a page so that the IT guy will be able to copy it to his firewall rules.
Should these ranges be large? As in 49,152 to 65,535? Should this range be squeezed down maybe?
I’ve seen vendors creating a port range of 10 or 100 ports. That’s usually too little to run in scale when the time comes. I’d go with a range of 10,000 ports or more. I’d probably also try first to estimate the capacity of the machine in question and figure out if more ports might be needed to maintain the sessions per second I am planning on supporting (allocated TCP ports take some time to clear up).
Is this “wholesale” port range a real security threat or just an imaginary one? How do you go about explaining the need to customers who like their networks all clamped down and closed?
If you are looking to learn more about WebRTC, check out my WebRTC training courses. In the near future, I will start working on a new course about TURN installation and configuration – if you are interested in early access – do let me know.
The post WebRTC ports: Understanding IP addresses and port ranges in WebRTC appeared first on BlogGeek.me.
WebRTC TURN servers are an essential piece of almost any WebRTC deployment. If you aren’t using them, then make sure you have a VERY good reason.
Connecting a WebRTC session is an orchestrated effort done with the assistance of multiple WebRTC servers. The NAT traversal servers in WebRTC are in charge of making sure the media gets properly connected. These servers are STUN and TURN.3 ways to connect WebRTC sessions
When connecting a session between two browsers (peer-to-peer) in WebRTC, there are 3 different alternatives that might happen.Connect directly, across the local network Connecting WebRTC over a local network
If both devices are on the local network, then there’s no special effort needed to be done to get them connected to each other. If one device has the local IP address of the other device, then they can communicate with each other directly.
Most of the time and for most use cases, this is NOT going to be the case.Connect directly, over the internet, with public IP addresses Connecting WebRTC directly using public IP address obtained via STUN
When the devices aren’t inside the same local network, then the way to reach each other can only be done through public IP addresses. Since our devices don’t know their public IP addresses, they need to ask for it first.
This is where STUN comes in. It enables the devices to ask a STUN server “what is my public IP address?”
Assuming all is well, and there are no other blocking factors, then the public IP address is enough to get the devices to connect to each other. Common lore indicates that around 80% of all connections can be resolved by either using the local IP address or by use of STUN and public IP addresses.Route the media through a WebRTC TURN server Connecting WebRTC by using TURN to relay the media
Knowing the public IP address is great, but it might not be enough.
There are multiple reasons for this, one of them being that the NAT and firewall devices in use are not allowing such direct traffic to take place. In such cases, we route the data through an intermediary public server called TURN.
Since we are routing the data, it is an expensive endeavor compared to the other approaches – it has bandwidth costs associated with it and it is why you Google won’t ever offer a free TURN server.Transport protocols and WebRTC TURN servers
TURN comes in 3 different flavors in WebRTC (6 if you want to be more accurate).How testRTC checks and explains connectivity alternatives of TURN servers in qualityRTC
You can relay your WebRTC data over TURN by going either over IPv4 or IPv6, where IPv4 is the more popular choice.
Then there’s the choice of connecting over UDP, TCP or TLS.
UDP would work best here because WebRTC knows best when and how to manage network congestion and if to use retransmissions. Since it doesn’t always work, it might require the use of TCP or even TLS.
Which type of a connection would you end up with? You won’t really know until the connection gets established, so you’ll need to have all your options opened.When is a TURN server needed in WebRTC?
That’s easy. Whenever there can’t be a direct connection between the two devices.
For peer to peer, you will need to install and run a TURN server.Try direct, then TURN/UDP, then TURN/TCP and finally TURN/TLS
The illustration above shows our “priorities” in how we’d like a session to connect in a peer to peer scenario.
If you are connecting your devices to a media server (be it an SFU for group calling or any other type of a server), you’ll still need a TURN server.
Why? Because some firewalls block certain types of traffic. Many just block UDP. Some may even block TCP.
With a typical WebRTC media server, my suggestion is to configure TURN/TCP and TURN/TLS transports and remove the TURN/UDP option – since you have direct access to the public IP address of the media server, there’s no point in using TURN/UDP.Try direct to server, then TURN/TCP and finally TURN/TLS
The illustration above shows our “priorities” in how we’d like a session to connect with a media server.What about ICE-TCP?
There’s a mechanism called ICE-TCP that can be used in WebRTC. In essence, it enables a media server to provide in the SDP a ICE candidate using a TCP transport. This means the media server will actively wait on a TCP port for an incoming connection from the device.
It used to be a Chrome feature, but now it is available in all web browsers that support WebRTC.
This makes the use of TURN/TCP unnecessary, but will still leave us with the need of TURN/TLS.Try direct UDP to server, then direct ICE-TCP to server and finally TURN/TLS
The illustration above shows our “priorities” in how we’d like a session to connect with ICE-TCP turned on.The elusive (mis)configuration of TURN servers in WebRTC
Configuring TURN servers in WebRTC isn’t an easy task. The reason isn’t that this is rocket science. It is more due to the fact that checking a configuration to ensure it works properly isn’t that simple.
We are used to testing things locally. Right?
Here’s the challenge – in WebRTC, trying it on your machine, or with your machine and the one next to it – will ALWAYS WORK. Why? Because they connect directly, across the local network. This means TURN isn’t even necessary or used in such a case. So you never test that path in your code/configuration.
What can you do about it?
- Be aware of this
- Use the sample provided by Google for Trickle ICE testing. It won’t check everything, but it will validate that you’ve at least installed and configured the TURN server semi-properly
- Block UDP on the machine in your local network and then try to connect a session to another machine on your local network. Make sure it went over TURN/TCP relay (check webrtc-internals dump for that)
The above things can be done locally and repeatedly, so start there. Once you get this to work, move towards the internet to check it there.Quick facts Do you need a TURN server if you connect your sessions to a WebRTC media server?
Yes. WebRTC media servers don’t support TLS type of transport. Sometimes they do support TCP via ICE-TCP. For the cases where calls can’t connect in other ways, you will need to use TURN/TCP or TURN/TLS.Do media servers need to have WebRTC TURN server configuration?
Usually not. In most cases you will be installing media servers with direct internet access on a public IP address. This means that having TURN configured only on the WebRTC client side is enough.How do you test a TURN server configuration for your application?
An easy way is to block UDP traffic and see if your WebRTC client can still connect. Another one is to use Google’s Trickle ICE sample.
The post WebRTC TURN: Why you NEED it and when you DON’T need it appeared first on BlogGeek.me.
Once a month, I will be publishing along with Philipp Hancke a WebRTC fiddle of the month as a free lesson in my WebRTC Codelab. This continues an old tradition of Mozilla’s Jan-Ivar, the fiddle ot the week.
Time for a new experiment. If all goes well, we will be making it a monthly thing.
Somehow, a week or two ago, we came to the conclusion that it would be nice to do a short video explainer of something that people are trying to figure out with WebRTC.
To make this happen, we decide together what to do the short lesson about, then Philipp Hancke writes a jsFiddle piece of code to implement it. And then we sit and record the explanation of it, creating a new free lesson in our joint WebRTC Codelab course.What does each WebRTC fiddle of the month include?
Each WebRTC fiddle of the month has these 3 resources:
- A short video explaining the problem and how we solved it
- The link to the jsFiddle code
- Transcription of the video
Creating the WebRTC Codelab was fun. We’re thinking of recording a new course at some point, but until we wrap our heads around that one, we’ve decided to continue recording some more lessons.
Seems like we work good together, so finding yet another excuse to do something made enough sense.
Oh, and if you happen to decide that you need to learn more about WebRTC, and end up enrolling to our course, then that’s a definite winTwo requests I have for you #1 – Check our first “fiddle of the month”
Our first explainer fiddle is about creating a peer connection that includes screen sharing and an audio microphone stream wrapped nicely together. You might get zoom-fatigue (or the WebRTC equivalent of that term), but you almost always want to talk when trying to screen share and collaborate.
Go watch our fiddle – Sharing screen + microphone together
Future WebRTC fiddles won’t be announced here, but only on social media and in the WebRTC Weekly (so subscribe to it). These fiddles will all be available in the WebRTC fiddle of the month section of the WebRTC Codelab course.#2 – Suggest some more ideas for such fiddles
Got ideas or requests for fiddles?
I can’t promise we will record your request, but I can promise you we will seriously think about it once we sit down to decide what to record.
2020 offers an interesting viewpoint to WebRTC browser support. Where exactly is it available in desktop and mobile, and what can you do about it as a developer?
This is almost a yearly article that I now write, each time with a slightly different focus to it. We’re now halfway into 2020, and things are changing fast.
Here’s a quote that I am seeing a lot this year:
Sometimes this quote is quite literally true. pic.twitter.com/SkVooF9Fez— The Long Now Foundation (@longnow) January 1, 2020
It rings true for the last few weeks when it comes to WebRTC, but somehow, in the domain of WebRTC browser support, we’re still standing in place.
My most up to date slide on WebRTC browser support?
We will get back to it in detail a bit later.
For now I’d like to look at the “Can I use” website, filtered for WebRTC. It gives a good starting point (although somewhat misleading). I will use that as the basis of looking at WebRTC on desktop and mobile.WebRTC support on desktop
On the desktop today, all modern web browsers support WebRTC.
This has been the case for quite some time now. I’ve announced that this means that WebRTC is ready towards the end of 2018.
Because the consumption model in the desktop today is done through web applications, while on mobile, it is predominantly based on native applications. So the moment all desktop browsers are nicely represented and supported, things look bright.
This isn’t to say that there aren’t challenges with WebRTC browser support – obviously there are.
I can list a few of them here out of the top of my head:
- No support in Internet Explorer (and there won’t be any for this dying browser)
- Edge still has its old version and the new Chromium based one. They behave differently, with the new Chromium one acting just like Chrome for anything WebRTC related
- Safari still suffers from its need to differentiate by not doing what other browsers are doing. It has no VP9 support, and it is still somewhat buggier than the rest of the crowd
- Things like hardware acceleration support and proper rendering, antivirus CPU issues and CPU consumption in general still plague the main implementation (=Chrome) today
When it comes to mobile, support for WebRTC is a bit more complicated.WebRTC iOS Safari support Is WebRTC really available on iOS Safari?
iOS Safari has been supporting WebRTC since Safari 11.
We’re now in Safari 13.5 and things are still rather grim when it comes to true support of WebRTC.
iOS Safari WebRTC is such a broken mess that my going suggestion to clients unfortunately is to not support it and redirect users to a native app installation. I had to manually go through all open WebRTC bugs in webkit to figure out how to explain this to my clients and help them in reaching that conclusion and even conveying that to their customers.
There are nasty bugs in iOS Safari that have been opened since 2019 or earlier relating to media handling of WebRTC. These aren’t just edge cases, but rather things you’ll have users bump into in regular use. Some of them have finally been fixed in the latest 13.5.5 beta earlier this month.
Oh – and if you plan on using any OTHER browser on iOS then WebRTC won’t be supported there. Why? Because Apple hasn’t made WebRTC available in its Webkit Webview on iOS and they aren’t allowing anyone to build a mobile iOS browser that doesn’t use Webkit as its rendering engine. So much for freedom and choice.
Up until now, there was no serious way to run a WebRTC web application in iOS Safari in production at scale. Hopefully, this is now mostly solved…Android browsers support for WebRTC How well is WebRTC supported in the fragmented Android world?
Android has its own set of headaches when it comes to WebRTC. That’s because there’s no single Android out there, but rather a slew of them.
Here’s what we can glean from a close look at that “can I use” list above.
- Android Browser, Chrome for Android and Firefox for Android are the gold standard. If these are what your browsers “bump” into, then you’re in good shape
- Opera… it depends. Opera Mobile is just fine. Opera Mini doesn’t support WebRTC
- UC Browser for Android has no WebRTC support
- Samsung Internet, quite the popular option you may run into, has WebRTC, but requires a webkit prefix. What does this mean? That it probably runs an old version of the WebRTC implementation, never a good thing with WebRTC
- QQ Browser and Baidu Browser, both the China alternatives to Chrome also have WebRTC support but also with a webkit prefix. Support there would also be tricky for web applications without some serious regression testing
While WebRTC is nicely supported in Android, it is going to be hard sometimes to decide what that support exactly means. Knowing that is a mix of understanding the device and the browser the web application is being executed on top.Where do we go from here?
If you read everything until here, then understand this: WebRTC is a work in progress.
It is the best (and only) alternative you have for real time communications that works in the browser without any installation. It works well enough for large companies to release applications (web and native) that attract massive user bases.
As with many other technologies, starting to use it is simple. Getting it to a professional level requires a lot more investment and commitment.
Next week I’ll be starting off my “future of WebRTC” workshop. This workshop is going to cover many aspects of the changing landscape of WebRTC. I’ll be touching issues of infrastructure, optimization and differentiation. All with a view of the current best practices as well as the latest trends.
There are 2-3 more seats available in the workshop. If you are interested in joining, check here.
VP9 is the best unused codec today that can improve video quality and media experience in your WebRTC application. Lets see who is this codec good for.
Last year there were 3 video codecs available in browsers for WebRTC: VP8, H.264 and VP9. Now there seem to be 5, with the addition of HEVC and AV1. Let’s put some sense into what is really going on, and where does each of these fit, focusing on VP9.WebRTC video codec support by browser All modern web browsers today support WebRTC
In the good old days, WebRTC video codec support was “simple”. The industry was bickering and fighting between VP8 or H.264 until resolving the matter by mandating both VP8 and H.264 codec support in web browsers.
Then Google went ahead, adding VP9 into the mix. Mozilla went along with it and added it to Firefox.
After that, we’ve got to the point where the Alliance of Open Media was created with AV1 as its video codec, prepping us nicely into the next codec war we’re going to face – the upcoming AV1 codec versus HEVC – both of which are now available (or eminently available, or soon to be available) in web browsers with WebRTC.
I’ve created this simple table for you to understand in which web browser which video codec is available for WebRTC:
To make things simple, if you need to launch something in 2020, then the video codecs available to you are:
- H.264 and VP8 on ALL browsers
- VP9 on all browsers other than Safari
Which leads us to the next question…HEVC & AV1. Should you join the experiment? Be sure you want to be part of WebRTC’s future video codec(s) experiment(s)
Let’s check the new video codecs that are sprouting in WebRTC to understand where we stand with them.HEVC
It seems that Apple is adding HEVC to Safari. It is available to some extent in the Safari Technology Preview, so developers can tinker with it without knowing when it will be publicly available in Safari. And there is no indication or an inkling of an indication that other browser vendors are going to join – Google won’t. Mozilla definitely won’t. Microsoft just might, but that would mean forking away a bit from Chromium which is now the engine inside their Edge browser – not the right focus in my mind for Microsoft here.
HEVC is like VP9 in the same way that H.264 is like VP8:
- You need to deal with patent royalties when using HEVC (which are a litigation mess in the making)
- VP9 has less hardware acceleration available than HEVC
The only difference is that HEVC doesn’t exist in any browser yet and will only be available on Safari while H.264 is available in all browsers.
To understand where we’re at with H.264 vs VP8 we only need to read the stats shared by Google in their recent semi-celebration for 10 years of WebM and WebRTC:
“These technologies have succeeded together, as today over 90% of encoded WebRTC video in Chrome uses VP8 or VP9.”
The bolded marking is my own doing – and just so we’re clear:
- This is Chrome only
- It shows H.264 has less than 10% “market share” in Chrome WebRTC
- It also insinuates that VP9 doesn’t have a large “market share”, otherwise, a ballpark figure would have been provided. More on that later
- This is why I think HEVC doesn’t really stand a chance against VP9 in the context of WebRTC
Now with AV1 coming up and the huge backing behind it, the HEVC track is all but dead. At least for the majority of WebRTC developers.
If you want to use HEVC in WebRTC, then you limit yourself to future Safari releases and native applications (where you modify the WebRTC codebase to add HEVC). Don’t expect it to work in any other web browserAV1
AV1 is the best next thing in video coding. The best invention since sliced bread. The best unlikely cooperation amongst industry co-opetitors moving away from royalty bearing video codecs towards an open video codec.
It is supposed to be better than both VP9 and HEVC from a compression standpoint.
And it is supposed to be a coombaya experience where everyone is supporting it. The members list of the Alliance of Open Media foundation behind it is impressive. It includes all browser vendors and many chipset vendors. How can you go wrong here?
The only problem for me is adoption time. It takes a long time to get a video codec to market.CodecYear startedAgeH.264200317VP8200812HEVC20137VP920137AV120191Video codec maturity
To get a video codec to market properly time is needed. From specification, to implementation, to modifying the implementation to work for real time communications, to optimizing the implementation to work reasonably on available CPUs.
In Chrome it doesn’t officially exist. It is there behind a flag, making sure users can’t really enjoy it and web developers don’t have meaningful enough access to it.
Getting a codec that came out of the oven a year or two ago to production is risky business.
If you need this article to learn about codecs, then AV1 should NOT be in your roadmap in 2020. You better wait this one out a little bitWho is using VP9 codec today?
Not Google it. Google.
They use VP9 in Google Meet. That’s a large traffic source using VP9, but it says a lot about the adoption of VP9 so far.
There are also a few instances where VP9 is used in streaming or live streaming use cases. Nothing major though.Adoption challenges of using VP9 codec Photo by Mathias Jensen on Unsplash
Why so low an adoption after being out on the market for 7 years?
I can only guess…
- It takes time. 7 years just isn’t enough when we’re still all figuring out WebRTC
- VP8 and H.264 is good enough for almost everyone
- VP9 requires more CPU than VP8 or H.264, and there are complaints about CPU use in WebRTC with these codecs already, so VP9 won’t help alleviate that problem – only worsen it
- Not enough hardware acceleration for VP9. There are more hardware decoders for VP9 today than they used to, but not much in the way of encoders. It does exist on Intel however, which is great
- Not enough knowledge and understanding on how to utilize VP9, so no one’s really trying it properly
- The popular open source media servers are all configured for VP8 or H.264 by default even if they do support VP9. And no one changes default settings. I would also assume and expect these open source platform to not optimize for VP9 anyways – not enough uptake yet
- The Alliance of Open Media and its success to put out such a strong cadre of supporters. By doing that, many large vendors in the industry are making the decision to “wait it out” with VP9 and skip this video codec generation directly to AV1
I’ve written about the role of VP9 in WebRTC before.
The premise of VP9 is improving encoding compression over VP8.Compression rate
That comes at a cost of expending more CPU, giving us the option to balance between using network and CPU resources.VP9 gives you either less bitrate for the same quality or more quality for the same bitrate than VP8
When looking at the higher end of the bitrate equation, one may prefer using VP8 or H.264 – we have enough network resources so we couldn’t really care on that front while saving on CPU might be beneficial
On the lower end of the bitrate equation, we’d want to squeeze every bit we have running on the network on higher quality. And then using VP9 might make more sense: since the bitrate is limited, we can spare more CPU on that and use VP9.
Sad thing is we can’t really know this in advance, at least not always.Scalability
Implementing a workable large scale video group call with WebRTC isn’t trivial. There are a lot of aspects to deal with both from a network perspective as well as from a CPU perspective.
The name of the game in this case is optimization, and this comes by having more flexibility in the tools you can use for optimizing the hell out of your video experience.
The flexibility here comes from VP9 SVC implementation in WebRTC:
- With H.264 and VP8 you can use simulcast – sending multiple video streams in multiple bitrates for the same content
- VP8 also supports Temporal Scalability – sending multiple frame rates in a single video stream
- With VP9 you can use SVC (Scalable Video Coding) – sending a single video stream with multiple layers for different resolutions, frame rates and quality levels
- To be clear, getting SVC to work in WebRTC isn’t easy, requires “reverse engineering” some of the workings of Google Meet and its proprietary SDP munging, but it might actually be worth the effort
If I had to chart the flexibility of a WebRTC video codec for large group sessions based on the tools it gives developers, this is what I’ll get:
We expend more CPU on VP9 but we win in network performance and scalability of a video group call by doing that.Plotting a route towards AV1
AV1 is the future.
Should you skip VP9 and just head to AV1 once it is ready? I don’t know.
The thing is, we had 2020. A pandemic that got us all cooped up at home doing video calls like crazy. The world has changed and with it priorities in our industry.
We’ve fast forwarded roadmaps by 5 years, so the future is already here. Can you wait a year or two more before you introduce a better video codec? If yes, then go straight to AV1. If you can’t, then you should seriously consider starting off with VP9 adoption.A quick recap Who is using VP9 codec in WebRTC applications?
Google Meet makes use of VP9 codec. Sadly, there is no other popular, large scale WebRTC application that makes use of it.Does VP9 codec support SVC in WebRTC?
Yes. In fact, VP9 is the only codec today that supports SVC (Scalable Video Coding) in WebRTC. This gives developers more flexibility in large group video calls and even live broadcasts than other video codecs.
The challenge is that VP9 SVC support in WebRTC isn’t official or well documented.
Better compression rate compared to VP8 and H.264 which are mandatory to implement in WebRTC.
Better scalability. It has flexible tools that assist in scaling video group calls.
HEVC and AV1 don’t yet exist in WebRTC browser implementations. At least not in a way you can utilize in a production service.
VP9 is available and usable in Chrome, Firefox and Edge.
If you are looking to improve video quality or reduce bitrates in your WebRTC application, then you should seriously look at VP9.
The post VP9 Codec: Is it time to adopt it in your WebRTC application? appeared first on BlogGeek.me.
Enabling large group video calls in WebRTC is possible, but requires effort. WebRTC CPU consumption requires optimizations and that means making use of a lot of different techniques.
Cramming more users in a single WebRTC call is something I’ve been addressing here for quite some time.
The pandemic around us gave rise to the use and adoption of video conferencing everywhere. Even if this does slow down eventually, we’ve fast forwarded a few years at the very least in how people are going to use this technology.It all started with a Gallery View
What’s interesting to see is how requirements and feature sets have changed throughout the years when it comes to video conferencing. When I first joined RADVISION, the leading screen layout of a video conference was something like this:
It had multiple names at the time, though today we refer to it mainly as gallery view (because, well… that’s how Zoom calls it).
Somehow, everyone was razor focused on this. Cramming as many people as possible into a single screen. Some of it is because we didn’t know better as an industry at the time. The rest is because video conferencing was a thing done between meeting rooms with large displays.
It also fit rather well with the centralized nature of the MCU, who ruled video conferencing.Enter Speaker View
At some point in time, we all shifted towards the speaker view (name again, courtesy of Zoom):
There were a few vendors who implemented this, but I think Google Hangouts made it popular. It was the only layout they had available (up until last month), and it was well suited for the SFU technology they used. It also made a lot of sense since Hangouts took place in laptops and desktops and not inside meeting rooms.
With an SFU, we reduce the CPU load of the server by offloading that work to the user devices. At the same time, we increase our demand from the devices.Hello 2020
Then 2020 happened.
With it came social distancing, quarantines and boredom. For companies this meant that there was a need for townhall meetings for an office, done on a regular interval, just to keep employees engaged. Larger meetings between room meetings because even larger still as everyone started joining them from home.
The context of many calls went from trying to get things done to get connected and providing the shared goals and values that are easier to achieve within the office space. This in turn, got us back to the gallery view.
Oh… and it also made 20+ user meetings a lot more common.3 reasons why WebRTC is a CPU hog
The starting point is challenging with WebRTC CPU use when it comes to video calling. WebRTC has 3 things going against it at the get go already:#1 – Video takes up a lot of pixels
1080p@30fps is challenging. And 720p isn’t a walk in the park either.
The amount of pixels to process to encode 1080p?
62 million pixels every second… I can’t count that fast
You need to encode and decode all that, and if you have multiple users in the same call, the number of pixels is going to grow – at least if you’re naive in your solution’s implementation.
What does this all boils down to? WebRTC CPU use will go over the roof, especially as more users are added into that group video call of yours.#2 – Hardware acceleration isn’t always available
Without hardware acceleration, WebRTC CPU use will be high. Hardware acceleration will alleviate the pain somewhat.
Deciding to use H.264?
- Certain optimizations won’t be available for you in group calling scenarios
- And on devices without direct access to H.264 (some Android devices), you’ll need to use software implementations. It also means dealing with royalty payments on that software implementation
Going with VP8?
- Hardware acceleration for it isn’t available everywhere
- Or more accurately, there’s almost no hardware acceleration available for it
What about VP9?
- Takes up more CPU than H.264 or VP8
- You won’t find it in Safari
- Not many are using it, which is a challenge as well
- Hardware acceleration also a challenge
So all these pixels? Software needs to handle them in many (or all) cases.#3 – It is general purpose
WebRTC is general purpose. It is a set of APIs in HTML that browsers implement.
These browsers have no clue about your use case, so they are not optimizing for it. They optimize for the greater good of humanity (and for Google’s own use cases when it comes to Chrome).
Implementing a large scale video conference scenario can be done in a lot of different ways. The architecture you pick will greatly affect quality but also the approach you’ll need to take towards optimization. This selection isn’t something that a browser is aware of or even the infrastructure you decide to use if we go with a CPaaS vendor.
And it all boils down to this simple graph:Complexity vs group size in WebRTC conference calls
The bigger the meeting size, the more WebRTC CPU becomes a challenge and the harder you need to work to optimize your implementation for it.Why now?
I’ve been helping out a client last month. He said something interesting –
“We probably had this issue and users complained. Now we have 100x the users, so we hear their complaints a lot more”
We are using video more than ever. It isn’t a “nice to have” kind of a thing – it is the main dish. And as such, we are finding out that WebRTC CPU (which people always complained about) is becoming a real issue. Especially in larger meetings.
Even Google are investing more effort in it than they used to:April 17, 2020 3 areas to focus on to improve performance in group video conferences
Here are 3 areas you should invest time in to reduce the CPU use of your group video application:#1 – Layout vs simulcast
Simulcast is great – if used correctly.
It allows the SFU to send different levels of quality to different users in a conference.
How is that decision made?
- Based on available bandwidth in the downlink towards each participant
- The performance of the device (no need to shove too much down a device’s throat and choke him with more data that it can decode)
- The actual frame resolution of where that video should appear
- How much is that video important for the overall session
Think about it. And see where you can shave off on the bitrates. The lower the total bitrate a device needs to deal with when it has to encode or decode – the lower the CPU use will be.#2 – Not everyone’s talking
Large conferences have certain dynamics. Not everyone is going to speak his mind. A few will be dominant, some will voice an opinion here and there and the rest will listen in.
Can you mute the participants not speaking? Is there an elegant way for you to do it in your application without sacrificing the user experience?
There are different ways to handle this. Anything from a dominant speaker, through the use of DTX towards automatic muting and unmuting of certain users.#3 – UI implementation
How you implemented your UI will affect performance.
The way you use CSS, HTML and your JS code will eat up the CPU without even dealing with audio or video processing.
Look at things like the events you process. Try not to run too much logic that ends up changing UI elements every 100 milliseconds of time or less on each media track – you’re going to have a lot of these taking place.My eBook on the topic is now available
If you are interested in how to further optimize for video conference sizes, then I just published an eBook about it called Optimizing Group Calling in WebRTC. It includes a lot more details and suggestions on the above 3 areas of focus as well as a bunch of other optimization techniques that I am sure you’ll find very useful.
The post Surviving WebRTC CPU requirements in large group calls appeared first on BlogGeek.me.
If you are going to start a WebRTC project that requires a media server, you better be sure you know how frequently as well as when was the last time the code got updated.
A WebRTC media server is a type of server that is required to build applications that offer group calling capabilities among other things. There are other types of WebRTC servers that are needed, but this is not the place or time to discuss them.Rising interest in WebRTC media servers
Here’s a discussion I had multiple times in the last month: People asking about this or that WebRTC media server or project, wanting to know if they should adopt it for their own application.
This rise stems from the increased interest in video conferencing due to the pandemic. Video has shown its usefulness in the biggest possible way. We’re all stuck at home, and the only way to communicate is by “calling”. Video adds context and meaning to voice only calling so it is becoming widespread.
In some cases, as in India, the government decided to put out funding for a video conferencing challenge, where vendors are invited to build applications for the local market. In others, remote-something is becoming a thing where the existing generic solutions don’t cut it (there are many such verticals).
As it so happens, a lot of teams are now trying to figure out which open source WebRTC media server they should pick and use.
There’s an article I wrote almost 3 years ago on 10 Tips for Choosing the Right WebRTC Open Source Media Server Framework. I took the time today to update it as well as the selection worksheet in it.
One thing that developers seem to miss is how easy it is to understand the freshness of the code – how up to date a WebRTC media server code really is.
So here we go.How do the most popular WebRTC media servers compare to each other?
What I like doing is using the insights feature in github. It gives a nice initial perspective of a project besides the popularity metrics of watches, starts and forks (they are nice, but just to get me interested – not it making an opinion).
For that purpose, I like looking at the Pulse, Contributors and Code frequency metrics provided by github.
Doing such a check is useless without context. And context is built by looking at alternatives. In this case, I decided to look at some of the most popular WebRTC media server alternatives out there: Janus, Jitsi, Kurento and mediasoup
Why these 4? Because they are mentioned in almost every conversation I have about open source WebRTC media servers.
Some of these projects are built out of multiple github repositories. For the purpose of this comparison, I tried looking at the main repo holding the media server itself. Here’s the ones I’ve used for each:
The github pulse lists recent activity of the project. I’ve looked at a period of 1 month here on all 4 projects to get the following picture:JanusJitsiKurentomediasoupAuthors141014Total commits7635842Files65569514Additions3,1522,0912922,670Deletions3631,8622141,993
A few thoughts:
- Jitsi and mediasoup seem to be doing a lot of code optimizations based on the ratio between additions and deletions
- Janus had a lot of additions versus deletions, so a lot of new code this past month?
- All projects are skewed towards a single developer at their core. Jitsi skewed towards two main developers
- Kurento is way behind the others
- mediasoup seem to have a smaller community of active contributors around it (it is the youngest of these projects)
The github contributors view gives us a nice time perspective of these projects, with a focus on who are the main contributors over time.
A few thoughts:
- Activity in Janus has been stable over time, but leans heavily on a single developer, Lorenzo Miniero
- For Jitsi, activity is leaning on 3 developers, 2 of whom have been with the project for a very long time
- Kurento’s activity fell since the acquisition by Twilio and never really recuperated. Most activity since then can be attributed to Juan Navarro
- mediasoup’s activity has a kind of a cyclic nature to it that is hard to explain from outside. Most of the work there can be attributed to Iñaki Baz Castillo
This chart on github shows the additions and deletions made to a project throughout its lifetime.
For the image below I tried aligning the projects as well as I could on the 10k range, which might have distorted them a bit, but should place them nicely in the context of each other. Notice that I couldn’t do that for mediasoup – see my thoughts below. Also note that the X axis of the timescale in each is different, but that wasn’t interesting for me for this comparison.
A few thoughts:
- Janus and Jitsi show healthy code contributions throughout their existence. I attribute the spikes outside the chart area I’ve selected as on time efforts or just mistakes. Over time, they show stability
- Kurento’s activity has seen better days
- mediasoup is interesting. It shows bouts of activity, mainly due to heavy use of branching across its version releases
- These projects are rather small in nature. If you search github for popular projects outside the WebRTC space, you’ll see a lot higher levels of activity
This looks like an obvious question, but it really isn’t.
Does this make JsSIP a dead project? Or is it just that there’s nothing much to add besides code fixes here and there?
(Interestingly, SIP.js shows totally different behavior)
When it comes to WebRTC, the same cannot be said. WebRTC is “work in progress” at its core. Browsers deprecate and introduce new features with each release, new codecs are introduced and the slew of use cases using WebRTC is still growing strong. This means that to keep pace with these changes, WebRTC media servers need to be updated as well. Otherwise, they wither and die, with time being unable to offer the level of quality and connectivity developers and users expect.What other criteria should you be looking for?
Freshness of code is only one criteria, but there are many more. The first one should probably be does this media server fit my requirements? Not all media servers are built equal or for the same purpose, and deciding which one is most suitable is important as well.
Other criteria include usage, maintainability, support, documentation, etc.
You can find the full list in my article about it – 10 Tips for Choosing the Right WebRTC Open Source Media Server Framework
And if this is of real interest to you, then you should look at your selection process itself. For that, I can suggest my free media server selection KPI sheet.
Oh, and once you’re there, think about scaling as well. I have an existing eBook about best practices in scaling WebRTC applications, and an upcoming eBook on Optimizing Group Video Calling in WebRTC (available for pre-purchase).
The post How to know if an open source WebRTC media server is kept up to date? appeared first on BlogGeek.me.
AV1 is coming to WebRTC sooner rather than later. Apparently so is HEVC. It is an AV1 vs HEVC game now, but sadly, these codecs are unavailable to the “rest of us”.
WebRTC codec wars were something we’ve seen in the past. During the early days of WebRTC there have been ongoing discussions if the mandatory video codec in WebRTC should be VP8 or H.264. The outcome was to have both of them mandatory to implement in browsers.
Fast forward to today, and life is simply. We have ubiquity and support across all browsers that have WebRTC in them, which is great.
COVID-19 is causing all communication vendors to fast forward and accelerate their roadmaps by 6-18 months. Those that don’t are going to be left behind on the other side of this pandemic.COVID-19 is fast forwarding all roadmaps and plans related to WebRTC, including codec improvements
This isn’t an attempt to scare anyone or to FUD people into doing things. It is just the way things are.
If you want to see how serious things are, just check what’s going on around you:
- Zoom is rolling out new features on almost a daily basis. They are plugging in their security gaps faster than most vendors can plan their roadmap, let alone develop anything
- Google is releasing features on Duo and Meet to drastically improve them. A lot of it hinges on machine learning but also on the latest coding technologies (more on that later, when we get to AV1 again)
- Most UCaaS vendors have launched their own video meeting service in the past 6 months. A lot of them in the last month. Many now offering it for free
- Many vendors in the video space from all verticals are seeing a 10x or more increase in use
- There’s a race towards filling different gaps when comparing these meeting services versus Zoom
The AV1 vs HEVC angles here are VERY interesting.
HEVC requires royalties and is a licensing mess.
AV1 is so new it hasn’t even had an opportunity to cool down a bit after being taken out of the oven. Frankly? It is still half baked and requires a bit more cookin’ – and yet… it is now being rolled out in Google Duo.
The thing is, that 6 months back, video was nice to have. A feature that needs to be ticked in a long requirements list.
Today? Video first. All the rest comes later.
Zoom’s stock price and market cap is the best indicator of that change.A brief history of WebRTC video codecs
In less than 10 years, we’ve witnessed 3 codec generations in WebRTC:
- VP8 / H.264
- VP9 / HEVC
With each generation of codec introduced, CPU and memory requirements grow along with the complexity of the codec and the resulting quality for a given bitrate increases.VP8/H.264
I’ve been working with H.264 since 200x. Probably somewhere in 2005. It was brand new at the time and was about to replace H.263 and all of its extensions.
Fast forward to around 2010, when you started it being deployed in almost all video conferencing room systems.
VP8 came to our lives along with WebRTC, in around 2012. It is comparable to VP8.
Both are probably at their peak right now when it comes to video calling:
- They are ubiquitous
- Readily available
- Understood and known, with vibrant ecosystems
- They can run on most CPUs
This is the tipping point, where a new video codec is being sought after.
If you are using it today, you should be just fine. If you seriously want to be at the forefront of technology, right on the bleeding edge (and you will bleed – time, money and blood), then read on to your next alternatives.
And if you need to decide between VP8 and H.264, check out this free video course: H.264 or VP8?VP9
It should have been a VP9 vs HEVC thing and not an AV1 vs HEVC thing.
The next best thing in video codec was supposed to be VP9. VP9 is the replacement to HEVC. HEVC is what comes next after H.264, and the intent was always for VP9 to be the alternative to HEVC.VP9 gives you either less bitrate for the same quality or more quality for the same bitrate than VP8
As things go, VP9 advantages are just what you’d expect in a new codec generation:
- Compression efficiency
- Higher complexity
- Scarcity of hardware acceleration (an issue still)
What VP9 was supposed to bring to the world is SVC – scalability. With VP9 SVC we were supposed to improve resiliency of video as well as the ability to scale large group video calls better than ever before.
Need a boost and have a very good grasp at who is in a call before everyone joins? VP9 might be a good alternative for you.AV1
- AV1 is the new kid on the block. An impossible dream coming true: vendors working together in a new Alliance of Open Media, working on a royalty free video codec. Something that was never heard of a few years ago and now feels like the new norm
- Starting with 7 founding members, this changed the dynamics of the WebRTC codec wars. Instead of having Google with VP9 on one side of the ring and the rest of the world on the other side with HEVC, it brought a team of large players to the royalty free side, standing behind the AV1 video codec
- Today, the alliance includes 48 members, including all browser vendors and most chipset vendors
- The focus in terms of comparisons are now AV1 vs HEVC
I’ve written at length about AV1 when the specification got released. You can learn about AV1 there.
There are those who believe AV1 is ready and have been ready for quite some time. Reality says otherwise. It isn’t for the faint of heart at this point. More on that – below.
Adventurous? Go AV1!Where in the world is WebRTC VP9 video call?
VP9 shipped in Chrome 48 for WebRTC. That was January 2016. 4 years later and it is safe to say that not many are using VP9 in WebRTC.Adoption of VP9 is slow
The two main places where VP9 is making sense?
- Google. Google Meet for example has been using VP9 for quite some time in its own calls
- Peer-to-peer calls. Just because it is easy to achieve
Once AV1 was announced, the debate began if one should even try and adopt VP9 or wait for AV1 instead. The majority are waiting for AV1. Laziness at its best (and what I would have selected as well if you’re wondering).
The other reason for delaying and skipping a generation is investment in VP9. Since everyone’s looking at AV1, VP9 is left with less eyeballs and developers improving it. Add to that the slow release of SVC support to it in Chrome and the fact that Safari still doesn’t support VP9 and you can understand the reluctance of going this route.Apple’s appetite for HEVC in WebRTC
The big Apple is insatiable. Apple has been banking on HEVC for many years now, and where HEVC & WebRTC fits in Apple has been a topic here in the past as well.Apple is banking on both royalty bearing (HEVC) and royalty free (AV1) video codecs
On Apple’s release notes for Safari Technology Preview 104 there’s a bullet point that shows where things are headed:
Added initial support for WebRTC HEVC
I wonder whatever for?
- Apple is a founding member of the Alliance of Open Media, so it is banking on AV1 as the future video codec
- In iOS 11 (2017), Apple introduced HEVC to its devices. That was done with the addition of hardware acceleration
- Android devices usually don’t have HEVC hardware support, and licensing being as tough and expensive as it is, this is a continued differentiator for Apple
- Google will be reluctant to add HEVC to Chrome. So would Mozilla. Not sure what Microsoft’s stance would be on this one
- Apple isn’t playing an AV1 vs HEVC game, but rather an AV1 and HEVC game, and they are alone in that at the moment
- Apple isn’t especially strong or dominant in the WebRTC space. Safari is the worst browser these days in terms of WebRTC support, with users already used to switching to Chrome on Mac. What would adding HEVC to WebRTC Safari add? Especially when there are so many other, more basic things to fix and improve in Safari WebRTC support…
To me, this is the biggest conundrum at the moment. A piece of this puzzle is missing. What would make developers use HEVC if it is only available in Safari and nowhere else? This isn’t the app store. It is the web.
Time will tell.WebRTC AV1 support in Google Duo
I said it before and I’ll iterate it again. AV1 is too new. Too early to be adopted in WebRTC or real time communications. And yet… Google just announced supporting AV1 in Google Duo:[…] in the coming week, we’re rolling out a new video codec technology to improve video call quality and reliability, even on very low bandwidth connections.
They made sure to add a nice moving GIF so you can see the difference between “a video codec” and AV1 in the same bitrate.
Is that other codec VP8? VP9? H.264? HEVC? Maybe H.261…
Are they using it for all Duo calls? In all devices? In all network conditions?
The only thing I could find is that this rolls out to Android with iOS 2 weeks behind in the roll out. There are more things left unsaid.Some thoughts here
- AV1 doesn’t have hardware acceleration on smartphones. Maybe on 1 or 2 very new ones (I doubt it), and even then, the hardware would still be buggy as hell – especially for real time video, which is different than just camera recording or playing YouTube videos
- This means that going to HD resolutions with AV1 on smartphones is going to be brutal to CPU, battery life and device temperature. This isn’t where AV1 support in DUO is going
- This leaves us with the low bitrate scenario – probably anything from VGA or lower. Maybe even a quarter of that (QVGA)
- It is where AV1 is going to shine in 2020 and into 2021
We’re all stuck at home burning the networks. The large streaming vendors are lowering resolutions (and bitrates) for their default players in certain countries. This reduces the CPU load, making room for improving quality on lower bitrates. And that leads to the ability (and need) of better video codecs.Why not VP9?
Google Duo most probably already makes use of VP9. Maybe even HEVC on iOS devices due to hardware acceleration benefits. When it comes to 1:1 sessions, there’s no real reason to stick to a single video codec for all sessions.
With Apple working publicly now on HEVC in WebRTC, it put pressure on Google, and getting AV1 into Duo in order to bolster their side in the AV1 vs HEVC debate became a pressing matter. Google Duo’s 1:1 call scenarios were the most suitable candidate for Google to make that stand.Enter AV1
When a new video codec generation was introduced, the thinking was simple: “we are expecting it to support a higher resolution, at a higher bitrate, with a higher CPU consumption”
- Higher resolution, because let’s face it – QVGA sucked in 1995 and we were still using it in 2000 in video conferencing. So each generation had to get 4 times the pixels the previous one was capable of dealing with
- Higher bitrate, because at 4 times the pixels we couldn’t really get 25% the size, so there was an expectation of needing more bandwidth for the content we wanted to use
- Higher CPU consumption, because we were adding more work to the encoder and decoder
In 2020, things are changing.Sometimes, all you need is a better fit into smaller spaces (like low bitrate) Bigger is no longer better with video codecs
I have 4K resolution on my desktop and laptop. 1080p on my phone and TV. I am happy with 720p content most of the time. I hate fonts on a 4K screen that aren’t enlarged (the damn characters are just too small to read).
What is the value of higher resolution? HDR content? 8K? 360? VR? If all I need is just plain video, no higher resolution is required. We’re all content most of the time with 720p resolutions for business meetings anyway.
Resolution requirements for most content types and use cases are not going to get higher any time soon.
We are probably at peak resolution already.
So we are free to think of next gen video codecs as ones that help consume lower bitrates.
There’s a distinction here. While any new video codec generation consumes lower bitrates for the same resolution/quality, the main purpose of these new video codecs was almost always in increasing the resolution as well.
AV1 on mobile makes perfect sense here. Especially for low resolutions – since we can have some CPU to spare for that scenario.A quick FAQ on the latest WebRTC video codecs Is HEVC (H.265) supported in WebRTC?
No. Not officially.
Apple is adding support for it in Safari, but no other browser has added support for it or indicated plans to add support for it
Yes, but not in browsers.
Apple will introduce HEVC in Safari, but no other vendor will. If you build your own native application for either PC or mobile you can add HEVC as another supported codec and use it in your application.
That depends. If you want to add AV1, you need to make sure your use case fits well, as well as the devices you expect your users to have.
You will also need to put a considerable investment of time and money to make it happen.
My suggestion for most vendors would be to wait with AV1 support.
That is a good question with no good answer.
I believe it is a matter of timing. When the time came to adopt VP9, AV1 was already announced and on its way, so vendors preferred to wait and jump directly to AV1 instead of going for VP9.
VP9 doesn’t enjoy much hardware acceleration, which also makes it CPU intensive, requiring companies to tweak, fine tune and optimize their systems to use it. That kind of work is something many prefer not to do.
We’re at war again. The video codec war of WebRTC. And this time, each vendor needs to pick a strategy to play.Is there a single video codec today that will answer all of your WebRTC needs?
We’ve got multiple codecs in our warchest: VP8, H.264, VP9, AV1 and sometimes even HEVC now.
Which one will we be using?
Which ones will we be using?
Here, scenarios matter. Different scenarios will call for totally different video codec selection to optimize for quality, CPU use, performance, bitrate, cost, etc.
In 1:1 sessions, you may want to keep your options open – use the best one dynamically just by making a decision as the session is set up.
For group calls, will you be using a single, static video codec? Or allow for multiple ones? Will you have multiple codecs in a single group session? Are you going to have an SFU tweaked and tuned for that? Will you pick the best video codec for a session and then dynamically switch over as the nature of the session changes (=someone joins and leaves who has certain limitations)?
What about consumers? What kind of video codec selection strategies are going to be prevalent there? How are they going to be different than the ones we see in enterprise solutions? What will be the difference for mobile first or application based versus web based solutions?WebRTC differentiation: the next battlefield lines are being drawn WebRTC differentiation is back in focus
We live in interesting times.
Codec selection has never been more interesting or important.
While WebRTC offers 2 codecs (H.264 & VP8), most browsers support VP9 and now we’re seeing browser vendors either adding HEVC or using AV1 in their own apps.
If media quality is at the core of your service (think carefully about your answer to this question), then rethinking your video codec selection strategy might be in order.
It is going to require research and investment. But this is where the future lies for video codecs in WebRTC.
When someone says WebRTC Server – what does he really mean? There are 4 different WebRTC servers that you need to know about: application, signaling, NAT traversal and media.
WebRTC is a communications standard that enables us to build a variety of applications. The most common ones will be voice or video calling services (1:1 or group calls). You can use it for broadcasts, live streaming, private/secure messaging, etc.
To get it working requires using a multitude of “WebRTC servers” – machines that reside in the cloud (or at least remotely enough and reachable) and provide functionality that is necessary to get WebRTC sessions connected properly.
What I’d like to do here is explain what types of WebRTC servers exist, what they are used for and when will you be needing them. There are 4 types of servers detailed in this article:There are 4 types of WebRTC servers you need to know about
- WebRTC application servers – essentially the website hosting the service
- WebRTC signaling servers – how clients find each other and connect to each other
- NAT traversal servers for WebRTC – servers used to assist in connecting through NATs and firewalls
- WebRTC media servers – media processing servers for group calling, recording, broadcasting and other more complex features
More of the audio-visual type? I’ve recorded a quick free 3-part video course on WebRTC servers.
Not exactly a WebRTC server, but you can’t really have a service without it
Think of it as the server that serves you the web page when you open the application’s website itself. It hosts the HTML, CSS and JS files. A few (or many) images. Some of it might not even be served directly from the application server but rather from a CDN for the static files.
What’s so interesting about WebRTC application servers? Nothing at all. They are just there and are needed, just like in any other web application out there.WebRTC signaling servers WebRTC signaling servers are in charge of connecting users to one another
Signaling servers for WebRTC are sometimes embedded or collocated/co-hosted with the application servers, but more often than not they are built and managed separately from the application itself.
While WebRTC handles the media, it leaves the signaling to “someone else” to take care of. WebRTC will generate SDP – these are fragments of messages that the application needs to pass between the users. Passing these messages is the main concern of a signaling server.A WebRTC signaling server passes signaling messages between the users to establish a session
There are 4 main signaling protocols that are used today with WebRTC, each lending itself to different signaling servers that will be used in the application:
- SIP – The dominant telecom VoIP protocol out there. When used with WebRTC, it is done as SIP over WebSocket. CPaaS and telecom vendors end up using it with WebRTC, mostly because they already had it in use in their infrastructure
- XMPP – A presence and messaging protocol. Some of the CPaaS vendors picked this one for their signaling protocol
- MQTT – Messaging protocol used mainly for IOT (Internet of Things). First time I’ve seen it used with WebRTC was Facebook Messenger, which makes it a very popular/common/widespread signaling server for WebRTC
- Proprietary – the most common approach of all, where people just implement or pick an alternative that just works for them
SIP, XMPP and MQTT all have existing servers that can be deployed with WebRTC.
The proprietary option takes many shapes and sizes. Node.js is quite a common server alternative used for WebRTC signaling (just make sure not to pick an outdated alternative – that’s quite a common mistake in WebRTC).
If you are going towards the proprietary route:
- Don’t use apprtc as the baseline for your WebRTC signaling server
- Consider my WebRTC the missing codelab course
To work well, WebRTC requires NAT traversal servers. These WebRTC servers are in charge of making sure you can send media from one browser to another.
STUN is used to answer the question “what is my public IP address?” and then share the answer with the other user in the session, so he can try and use that address to send media directly.
TRUN is used to relay the media through it (so it costs more in bandwidth costs), and is used when you can’t really reach the other user directly.
A few quick thoughts here:
- You need both STUN and TURN to make WebRTC work. You can skip STUN if the other end is a media server. You will need TURN even if your other end of the session is a media server on a public IP address
- Don’t use free STUN servers in your production environment. And don’t never ever use “free” TURN servers
- If you deploy your own servers, you will need to place the TURN servers as close as possible to your users, which means handling TURN geolocation
- TURN servers don’t have access to the media. Ever. They don’t pose a privacy issue if they are configured properly, and they can’t be used by you or anyone else to record the conversations
- Prefer using paid managed TURN servers instead of hosting your own if you can
- Make sure you configure NAT traversal sensibly. Here’s a free 3-part video course on effectively connecting WebRTC sessions
WebRTC media servers are servers that act as WebRTC clients but run on the server side. They are termination points for the media where we’d like to take action. Popular tasks done on WebRTC media servers include:
- Group calling
- Broadcast and live streaming
- Gateway to other networks/protocols
- Server-side machine learning
- Cloud rendering (gaming or 3D)
The adventurous and strong hearted will go and develop their own WebRTC media server. Most would pick a commercial service or an open source one. For the latter, check out these tips for choosing WebRTC open source media server framework.
In many cases, the thing developers are looking for is support for group calling, something that almost always requires a media server. In that case, you need to decide if you’d go with the classing (and now somewhat old) MCU mixing model or with the more accepted and modern SFU routing model. You will also need to think a lot about the sizing of your WebRTC media server.
For recording WebRTC sessions, you can either do that on the client side or the server side. In both cases you’ll be needing a server, but what that server is and how it works will be very different in each case.
If it is broadcasting you’re after, then you need to think about the broadcast size of your WebRTC session.A quick FAQ on WebRTC servers Can I run WebRTC without any server?
Not really. You will need somehow to know who to communicate with and in many cases, you will need to somehow negotiate IP addresses and even route data through a server to connect your session properly.Will WebRTC servers spy on me and my data?
That depends on the service you are using, as different implementations will put their focus on different features.
In general, signaling and NAT traversal servers in WebRTC don’t have access to the actual data. Media servers often have (and need) access to the actual data.
Yes. You can host your WebRTC servers on AWS. Many popular WebRTC services are hosted today on AWS, Google Cloud, Microsoft Azure and Digital Ocean servers. I am sure other hosting providers and data center vendors work as well.Can I run WebRTC on my PHP WordPress site?
WebRTC can be added to any WordPress, PHP or other website. In such a case, the PHP WordPress server will serve as the application server and you will need to add into the mix the other WebRTC servers: signaling server, NAT traversal server and sometimes media servers.Know your WebRTC servers
No matter how or what it is you are developing with WebRTC, you should know what WebRTC servers are and what they are used for.
If you want to expand your knowledge and understanding of WebRTC, check out my WebRTC training courses.
I am launching a new WebRTC codelab course, created together with Philipp Hancke. This goes into the intricacies of WebRTC signaling and best practices.The State of online WebRTC resources
Where should you start with WebRTC? There’s not enough information about it and at the same time too much information about it. Most of it is old and outdated. Needle in a haystack.Up to date WebRTC code is hard to come by
Sift through discuss-webrtc, stackoverflow and the W3C WebRTC mailing list? All great. But there’s no explanation besides pieces of code. It lacks context.
Read the spec and work your way from there? If you don’t fall asleep, you might just find that browsers aren’t exactly spec compliant yet anyways.
Books? None from 2020. None from 2019. Less than 10 in total. A handful.
How about online courses? There are a few on udemy and pluralsight, but they are also old and broken by now.
My own Advanced WebRTC Architecture course? While great, I purposefully haven’t gone through the APIs of WebRTC, understanding they change so frequently.
A codelab then? Sure. find one that works and explains signaling properly.
The way I see it, there are 3 types of WebRTC codelabs today:
- The outdated ones
- The broken ones
- The ones that sell you a managed streaming service
I’ve been working with Philipp here and there in the last couple of years. He is fun to work with and knows everything there is to know about WebRTC and the intricacies of its APIs along with the status of browsers.
Here’s a bit about him:
I don’t quite recall how, but we got to the point a couple of months back that it would make sense to create a solid WebRTC codelab that covers the signaling aspects of WebRTC. Simply because there is nothing out there that does the trick properly.WebRTC: The missing codelab
Fast forward to today, we have a codelab course for you:
The codelab has 4 parts to it:
- Github repo and course introduction module, both publicly available
- Codelab walkthrough. Over 2 hours of recorded lessons covering the source code and explaining all you need to know to grok it
- Exercises. We’ve started creating useful exercises on the codelab. These include the challenge, the solution and the explanation. Along with the additional code for the solution and best practices
- Resources library. If you are new to WebRTC, this will come handy in closing some gaps for you. They are lessons from my other courses that fit nicely here
We’ve made sure the lessons aren’t boring by making them interactive. I’ll be there with Philipp, taking the part of the student asking questions. The intent is to try and get into his head, understand his thought processes. What do you gain out of it? You will understand why things are implemented that way and not only how.Interested?
Head to the WebRTC developer courses – there’s a 20% discount until the end of April.
If you are uncertain, then you are invited to join this week’s WebRTC Live webinar where I’ll be talking with Arin Sime about the codelab. Alternatively, just go through the introduction module and make a decision.
If you know that this is for you, then there’s a 50% discount if you enroll in March (that’s in the next 2 days) by using FASTMOVER as your coupon code on checkout (for any of the WebRTC course bundles).
If you’ve enrolled in my All Included course in 2020, then you are already automatically enrolled to the codelab. If you enrolled throughout 2019, then you are eligible for a 50% discount during April.
The WebRTC market global key players may be different than you think. They include the browser vendors, a few CPaaS vendors, dominant “creators” of WebRTC sessions and… open source projects.
The title for this post came from one of the many lead generating headlines I see for reports that mention companies that no longer exist in our market. It is sad in a way: The same market report released every year, with the main difference between the last one being X+1 where X denotes a year mentioned in the report.
(if you are about to purchase a market report on WebRTC – make sure the companies mentioned there actually do something meaningful in WebRTC – check against the companies listed here – or just ask me)
My intent here is to actually ask the question – Who are the WebRTC Market Global Key Players? – and then also answer it.
I’d like to segment the key players in WebRTC into 4 main groups:
- Browser vendors
- CPaaS vendors
- Customer facing services
- Open source projects
There are exactly 4 browser vendors that are interesting. The rest? Less so.
I will list them here in the order of their importance to WebRTC.Google Chrome
This one is obvious. Google is the main driving force behind WebRTC. They aren’t alone in being there, but they are the dominant browser player in market share AND they host the most popular implementation of WebRTC (that would be libwebrtc).
In many ways, Google decides where WebRTC is headed. It does that in 3 different angles:1. libwebrtc and Chromium
Maintaining the most popular WebRTC implementation (did I say that already?).
So much so that all browser vendors use this implementation in one way or another (either directly or indirectly by copying the pieces of it that they want/need).
I’ve added here also Chromium. This is the open source components of the Chrome browser and it is used in MANY important projects:
- Microsoft now operates Edge (its new browser) on top of Chromium
- Electron, which is quite popular in packaging web apps as installable apps is built on top of Chromium
- Many of the “other” browsers out there are implemented on top of Chromium
THE market leader in browsers.
Taken from StatCounter (and yes. It also includes mobile)
Need I say more on how that influences WebRTC adoption and implementation?
3. Google Apps
We used to have only Hangouts using WebRTC at Google.
Now we’ve got a lot more.
The shortlist I am aware of include:
- Google Hangouts / Google Meet
- Google Duo
- Google Stadia
- Chrome Remote Desktop
- YouTube Live
For these, Google has their own agenda with their own WebRTC roadmap. This means that the requirements that fall into any of these services end up either directly as part of the open source WebRTC implementation distributed by Google – or Chrome alone.Apple Safari
Apple has been quiet about WebRTC. That’s true about Apple and a lot of other technologies as well.
That said, Safari now supports WebRTC in Mac and iOS for a couple of years now (there are those who missed that fact).
While Apple has no other direct involvement around WebRTC, all developers and entrepreneurs are affected by their decisions about WebRTC. A lot more than in any other case.
Point in case is WebRTC iPhone support:
- You can build your own native app with WebRTC on an iPhone. This was always true
- You can run a web app using WebRTC on an iPhone inside Safari
- You can’t run a web app using WebRTC on an iPhone in any other iOS browsers – just because WebKit on iOS still doesn’t support WebRTC properly
While Edge didn’t enjoy any growth or market share, this may soon change.
Microsoft decided a bit over a year ago to stop investing its resources and focus on building a browser engine, and instead took Chromium “as is”, building their Edge browser on top of it.
Somehow, I am assuming (hopefully) that this change also means that more Microsoft engineers are involved in the inner workings of Chromium itself, optimizing it to run on Windows and elsewhere for their own scenarios.
Apple is more important than Microsoft when it comes to WebRTC simply due to the current state of market share of their browsers.Mozilla Firefox
Firefox is the 4th important browser and decision maker in the table of WebRTC.
It isn’t as important as the rest since its only “contribution” to the game is Firefox itself whereas the other browser vendors here come with operating systems and applications of their own as well.
Nevertheless, it is way more important than all other browsers not mentioned here combined.2. CPaaS vendors CPaaS vendors offer the tools for others to embed WebRTC communications
CPaaS vendors enable others to build their applications without delving into the communication technology stack too much. In many cases, that includes support for WebRTC as well.
To me, they are key players within this industry, and I want to mention those that are the most important when it comes to WebRTC adoption.Twilio
Twilio is the leader in CPaaS. It is also one of the dominant players when it comes to WebRTC in CPaaS.
When it comes to voice calls via WebRTC done through CPaaS, Twilio are probably the largest player. Twilio has a lot of visibility to issues and requirements related to voice use cases, especially ones related to call centers.
Twilio is also growing in their video use of WebRTC.Vonage
Vonage is the owner of TokBox, now part of its API platform.
As one of the leading video API platforms using WebRTC, TokBox is important. As with Twilio, this stems from their visibility to issues and requirements, but in this case, related to video use cases.Others?
There are other interesting CPaaS vendors in the WebRTC space, but none of them are dominant enough.
The ones worth mentioning in this context?
- Video/Enghouse – due to their use and focus on SVC and VP9/AV1
- Agora – due to their research of ML/AI in audio and video domains
- Dolby – due to their work around 3D voice
Customer facing services are the end products. What users interact with when it comes to WebRTC. This title won’t get a vendor to be a global key player unless there’s a real reason…Facebook
Facebook is huge. Doesn’t matter if you look at Messenger, WhatsApp or Instagram.
Messenger uses WebRTC.
Instagram uses WebRTC for its live chat.
WhatsApp doesn’t use WebRTC directly, but there’s ongoing effort to consolidate the infrastructure of all these messaging platforms at Facebook. Will we be seeing 2 billion WhatsApp users able to conduct voice and video calls by way WebRTC in 2020?
Then there’s the Portal device.
Anyways, the sheer size of Facebook, along with their work with WebRTC places them as a market leader in WebRTC use. And due to their size, they are also largely alone here.No one else
There are vendors that contribute large WebRTC traffic. But other than Facebook, I am not sure who else to include here.
Probably Amazon. Due to multiple products and that minor thing called AWS.
I decided not to put any of the enterprise vendors. I think they should matter more, but got a feeling that they don’t at the moment.4. Open source projects Open source projects are at the heart of the WebRTC ecosystem
These are sometimes neglected when discussing market leaders, which is rather sad. A lot of the development and WebRTC traffic out there ends up going through some of these open source projects, which is why I’ve added them here.
It is apparent that some projects should have a seat at the table. When Kurento got acquired by Twilio, there was a year when many of the discussions I had was about finding a suitable replacement for them.
If WebRTC open source projects fail to make progress, upgrade or just die, they affect those using them. The popular open source projects matter. A lot. They are at the heart of the WebRTC ecosystem.Janus
I’ve decided to put Janus first because:
- Wherever it is used, there is use of WebRTC
- It is very versatile and flexible. I see it in a lot of different use cases
- It is quite popular
Jitsi is now owned by 8×8 after switching hands.
It is still run as an independent open source project and is widely deployed.FreeSWITCH
FreeSWITCH comes from the telephony world.
Not in all of the deployments of FreeSWITCH out there WebRTC is used, but when companies need to connect telephony to WebRTC, they often go with FreeSWITCH for that.
Since FreeSWITCH is so common in so many of these instances, they are important for the WebRTC ecosystemKurento (to some extent)
To some extent, Kurento still matters.
Kurento is a shadow of its former glory prior to the acquisition. It isn’t part of Twilio – Twilio acqui-hired the developers, but not the name. Kurento lives on, but different.
It is being developed alongside OpenVidu, but the progress made to both seems somehow slower than the other open source projects mentioned here.
I am not sure they are a key player anymore.Others?
There are other important projects that can/should make the title of key players in WebRTC. The 3 that immediately come to mind are mediasoup, PION and Asterisk.
Why haven’t I added them? Because their popularity is lower than the others.
For mediasoup and PION it is just that they are newer. They are growing, so I believe they will become key players if they continue in the current adaption trajectory.
For Asterisk, it is because they are used in a similar fashion to FreeSWITCH. I just don’t see them as much in the conversations around WebRTC that I have.What does it mean to be a “WebRTC Market Global Key Player”?
This is where I started the article, and I think it bears thinking about.
When one coins a company as a “WebRTC Market Global Key Player” what does that mean exactly?
For me that means that they have the ability and potential to affect what happens with WebRTC moving forward.
While the standardization work is done in the W3C, a lot of the work happens elsewhere. In what Google places into Chrome as experiments and later as features. In what Apple decides to implement or not implement in Apple. In what CPaaS vendors deliver to their many customers and the feedback they get. In the companies that build large scale products and in the open source projects that make up a large portion of these products.
There are steps one can take to become a more dominant player. To be able to join the conversation and affect where WebRTC is headed. While I’ve conversed with many who want to become dominant players, only a few have the courage and the willingness to invest the time and resources needed.
Want to know who the global key players are? Don’t read it in a copy+paste research paper that is poorly updated…