id3as:为出错而设计的流媒体技术

id3as directors Steve Strong and Adrian Roe have been in business together for 30 years. 罗伊的职业生涯始于零售业，后来进入了金融科技领域. He admits that they knew nothing about the media space until a decade ago when they joined forces with video tech specialist (and 流媒体 contributing editor) Dom Robinson.

"What became obvious when we looked at media clients embarking on streaming live events was that when things went wrong there was very little you could do about it," Roe says. “如果你有摔倒的日志记录，那你就很幸运了.

"Since we came from dev background, we thought that we could make that pain go away and id3as 结果就产生了."

The UK-based outfit set itself the task of delivering streaming solutions which are more reliable than any one link in the chain. The aim was to drive out costs associated with downtime, overcapacity, and undercapacity.  

When Roe helped build online banks for companies like Northern Rock he did something similar by using back-to-back service level agreements (SLAs). 他说:“现在这不是一个可行的做法。. "The approach needs to be that you construct a system so that if any single element of your system goes wrong nobody notices.

“已经有人尝试在通用容器和Kubernetes上做到这一点, 但它们本身并不是真正的解决方案. Our technology can self-orchestrate or coexist within Kubernetes and Docker environments, 几乎可以在任何芯片组上运行, OS, and platform. We use Erlang, a programming language developed 25 years ago to deliver Carrier Class telecommunications."

真正的高可用性, he continues, 是接受事情会出错，并确保出错, 它不会影响下游的客户.

“我们对解决方案在实验室条件下的表现不感兴趣. 我们更感兴趣的是什么时候出问题."

The company developed a virtualised video pipeline that can be tailored and installed to deliver premium streaming models for broadcasters and operators

Advanced Video Pipeline (AVP) is a modular architecture that spans ingest through transcoding and enrichment to packaging and CDN. 所有功能和数据捕获都通过发布的openapi公开. 因此，AVP可以集成到现有的控制和监测系统中.

“我们并不寻求成为下一个Wowza、Unified或eme，”罗伊说. "We understand our niche and want to partner with a comparatively small number of high value enterprise customers."

For Intrado (formerly Nasdaq, formerly Thomson Reuters), id3as enables 75,000 live events a year. These are mostly financial fair disclosure events and a "perfect example of how we can help scale business and manage cost by looking carefully at where in practice things can go wrong."

In this case, Nasdaq was finding that phone calls were dropping out regularly during teleconference events. Roe says introducing AVP immediately took the impairment rate down by an order of magnitude and today, 损伤程度几乎可以忽略不计. 这甚至包括“没有人注意到”(特别是没有纳斯达克的客户)!)当亚马逊U.S. East was down for several days, despite 168 live events being delivered from there when it went down.

For DAZN, id3as provide both mezzanine encode and ABR ladder creation for a number of services, 以及一套丰富的音频处理能力. One of these is a remote commentary solution for DAZN's live event production which began half a dozen years ago and was one of id3as' forays into WebRTC.

“当我们管理直播的超高质量编码时, we could also do a decent quality low latency feed pushed to a commentator based anywhere else on the internet. This decreased the costs for DAZN of having to send multiple commentators to the event and increased the number of languages they could produce in."

另一个id3as“现实世界中实用的脏东西”的例子,而不是临床实验室测试的性能, 这是交付的工作 Arqiva混合电视功能. “他们收购了一家公司，启动了他们在该地区的能力, 使用基于现成组件的技术," Roe says. "The problem they faced was that many of the TV manufacturers had paid at best scant regard to international standards—the solution only worked on 16% of target devices.

"Six weeks into an engagement with us, we had a solution running on 96% of target devices. 很多设备不符合标准, 所以我们为它们提供了兼容的(但不是标准的)流."

It has also worked with Limelight to replace its real-time streaming platform with one based on WebRTC. Limelight实时流媒体 提供可靠的, 宣传, real-time video streaming using the UDP data transfer protocol and is integrated with Limelight's global CDN. Although the focus of the project was on improving reliability and being "good on a bad day," it also managed a full order of magnitude increase on the amount of traffic LLNW’s existing hardware could deliver.

id3as, 这家公司仍然只有8名全职员工, 是否打算让像Limelight这样的cdn达到他们的sla.

“你可以忽略99%及以下的SLA，”斯特朗说. "Frankly, you've got so much time to respond to stuff that you can just have a single box set in your cupboard doing the job. At 99.9%，你的时间更有限. 如果你很有条理的话, you can probably still get away with a fairly simplistic approach to delivering that sort of number.

"Once you get up to four nines, it starts getting out of the realms of human control. 你必须有多个系统全程运行. You might get away with something like an n+1 model where you've got a bunch of live systems and a couple of hot spares that are sat there running. But you're certainly in a world where you're going to have to have some form of distributed system."

sla为99.999或更多，“现实是你得到了0.86秒的反应时间，”他说. "And that's not just to respond; that's to detect, fix, and have the service back up and running. 这是非常少的时间做任何事情，特别是在广域网.Our focus has always been on tackling those sorts of challenges and our tool set has been built with precisely those requirements in mind. 这种级别的可用性不可能是事后才想到的."

[Editor's note: 流媒体 contributing editor Dom Robinson is one of the principals in id3as. 他与这篇文章的写作没有任何关系, and he does not write about the company or its competitors for 流媒体.]