SMW 17: id3as's Dom Robinson Talks Muxing, Multicasting, and Architecting CDNs
Tim Siglin: Welcome back to Streaming Media West 2017. I'm here with Dom Robinson from id3as. Dom, tell us quickly what's id3as.
Dom Robinson: id3as is a software company. We've been around for about eight years now. Very focused on micro-service architected delivery of video workflows. About eight years ago we took what was then Reuters and is now Nasdaq's video workflow into the cloud, which was quite a big, radical move for them. Big savings and we now deliver 35-40,000 live events a year for those guys. It's all fully automated and all cloud based.
Tim Siglin: That's kind of the fair disclosure types of events?
Dom Robinson: Fair disclosures, corporate announcements. It's a very cyclical, goodly business, as a case study in its own right.
Tim Siglin: Oh, I'm sure. Yeah, because everybody wants to cram it into about two weeks of earnings
Dom Robinson: That's it, and then we have three months where there’s massive under-utilization. They came to us, and we suggested virtualizing it. This was before anyone else was doing it at that scale. It left us with a really interesting stack. We use an unusual telecoms language, which will be familiar to the WhatsApp engineering team or the Twitter engineering team, but most other people haven't heard of Erlang. It's really designed for developing carrier-class video workflows. It's self-healing, which is really fundamental. We don't have to spend a lot of time engineering for the failure cases. So that's where we as a very small company get a very strong competitive edge over much, much bigger companies.
Tim Siglin: Very interesting, and of course you’re a Streaming Media contributing editor.
Dom Robinson: I'll put my writer's hat on.
Tim Siglin: Yeah, exactly. You and I and Eric have started SMAF, Streaming Media Advanced Forum, which we're not really allowed to talk about here, but people can come to SMAdvancedForum.com and catch up on what we did on the show-floor today.
And you have a book, Content Delivery Networks: Fundamentals, Design, and Evolution. I'm going to hold the book up in my best Vanna White impersonation, and I'll put it here so that it's part of the close-up, but this is Content, Delivery Networks, Fundamentals, Design and Evolution. Tell me briefly about the book.
Dom Robinson: As an outcome of editing my first book for Wiley, Advanced Content Delivery, Streaming, and Cloud Services with Mukaddim Pathan and Ramesh Sitaraman, I noticed that the academic books were always about current best practice. There was very much a competition between people to say, "This is the best way you can possibly do this,” which is absolutely valid. But there was a sense that people coming into the industry, moving into the sector for the first time, or moving across from a different role of responsibility, really wanted to understand streaming content delivery networks. They didn't have context. They didn't have a picture of the cycles in the industry and so on. The sections I'd written for that book were much more of a short concise history of the sector. Basically, what I've done is I've done is I've elaborated on that and expanded that to this book, which is much more, to an extent, my voyage into the industry over 20 years.
It goes into technical detail about things like; I've got a real belief in service philosophies, as the fundamental driver between most technical decisions and how you should be architecting content delivery networks, and streaming workflows, and probably other aspect of your business workflow.
Tim Siglin: Let's do a slight tangent off of that. We'll come back to the book at the very end, but cellular bonding, cellular muxing. This was something you got into early on and now I was just on a panel with Chris Knolton where we were talking about that as a nice redundancy to house ethernet. Where do you see that some of these technologies are going in terms of the ability to deliver live from almost anywhere? It used to be, if you went remotely, if you went out into the "field," you had to roll a satellite truck. Now most people are saying, "Oh yeah, we can do this with a couple of cellmuxes."
Dom Robinson: Yeah, if you're very puritanical about it, the capability's been in the Linux kernel for 12 years, maybe even a little bit longer. Link aggregation control panel, LACP, was doing channel bonding a good decade ago.
Tim Siglin: It makes sense, because those of us, like me who came out of video conferencing, we would do ISDN bonding of B channels and then have the back-channel for control.
Dom Robinson: When I started writing for Streaming Media, no one was really focused on that space six or eight years ago, so when Eric was asking me to write a little bit more about it, I realized it wasn't good enough just to talk about link aggregation or channel bonded link aggregation. I started creating these tiers.
You've got link aggregation, which you can just do with a standalone modem. It's a router. You've got five different IP connections, and if you're doing little bursts of data like web data, an office can get a sense that they're all aggregated, because between all of the offices, there's always availability. You can't treat it all as a single pipe if you're trying to stream over it. What you need to do that, is you need something that will multiplex and demultiplex the signal at both ends. Then you get a channel bonding link aggregator, which actually bonds the different channels together with a counterpart system at the remote end, which creates a bit of a buffer basically, and will take your, if you've got three one-megabyte connections, it will turn it into a, as far as everything else is concerned, a three-megabyte pipe.
Tim Siglin: This is similar in some ways to the home load balancers you can get if you have a DSL and a cable modem.
Dom Robinson: Yeah, some of those vendors are dominant vendors. Vislink and Mushroom Networks, those guys, were all about having multiple redundant pipes between offices, and actually being able to use them all as a single pipe to do high bandwidth file transfer, that sort of thing.Then, there's another generation on top of that, which is what you start seeing in backpacks and increasingly on camera top systems.
Tim Siglin: Teradek, LiveU, etc.
Dom Robinson: Exactly, and those guys, what they've done is they've actually got the specific application of the encoding, is able to apply back-pressure, if you like in the buffer, with the link aggregator You can understand whether you want to drop the bit rate of your video, because your availability's gone wrong. If your 3 MB drops to 1 MB, you can actually signal to the encoder, drop from 3 MB encoding to a 1 MB encoding. I call those cellmuxes, cellular multiplexes.
Tim Siglin: That leads naturally into the very last topic I want to talk to you about, which is reliable UDP, because one of the things we were talking about in this panel was, you've got the cell bonding. You've got the muxing as you say, but ultimately, if the network's not stable and if you can't aggregate together well from all these different service providers, the need to have forward error correction to have reliable UDP, where do you see that going?
Dom Robinson: Reliable UDP, I certainly was using it 18 years ago within global mix, within multicast. We were interested in scaling multicasts in the public consumer networks, crazy projects, different story, it's in the book, but basically, when you're pushing a packet out with a connection in this environment, which is UDP, but don't forget, TCP is UDP with back-channel, TCP is like a specific use case of UDP. It's just a very, very good one and very established. The problem is, it's designed to be a good net citizen, so it shares the session over the network and if two people doing separate things are trying to use the same link, it will be fair to both. What's happened with a lot of the more recent entrants to the space, is they're saying, "Well, if this is your IP network and you don't want to be a good net citizen to other people, then force the UDP down. Saturate the links and don't have a TCP to back your network.
Tim Siglin: I think of TCP and cartoon terms, with Chip and Dale, the two chipmunks--"after you, no after you; no, after you”--where UDP is sort of Hulk coming in and just blasting through.
Dom Robinson: Just drives through regardless. Then the problem is, when you start losing packets, you need to have some intelligence to actually manage that. Where we went, we built our own link aggregation and contribution stack. I suppose something akin to what SRT's been about, but also, our outcome's a little bit more like maybe quick, as an evolution, speedy has been. We tried to turn TCP into something that looks like UDP, rather than turn all of the TCP connections into UDP. What we're interested in is very, very, very small fragments of data over many paths.
Tim Siglin: Doesn't that increase the chattiness though, or are you decreasing the chattiness?
Dom Robinson: It also allows the back-off processes to happen in lots of different managed ways, and you don't have to implement a reliability protocol over the top. It is a reliability protocol. It just works. Particularly for the Nasdaq guys, they use that extensively and the field encoders we develop for them just because it's quick and easy and they don't have to think about firewalls. It also brings link aggregation, cellular bonding as a byproduct.
Tim Siglin: It does that traversal, that kind of thing?
Dom Robinson: It's TCP.
Tim Siglin: Because it's TCP, you’re right, fair point. All right, where can people get the book?
Dom Robinson: Go to Amazon. That's the fastest way to get the copies delivered. You can go to the Wiley.com site, and if you message me directly during November, I can give you a little 20% off token as well.
Tim Siglin: The Streaming Media 2017 discount. Awesome. Thank you Dom, as always, and I like the hat.
Dom Robinson: Yeah, it's for my writing role only. Back to id3as.
Companies and Suppliers Mentioned