Serverless and RTC Panel Discussion.
At TADSummit EMEA 2019 we had a number of FinTech start-ups presenting and they expressed surprise at the lack of adoption of serverless across the RTC community. Referring to the approaches described by several RTC service providers as “yak-shaving”.
In this session we hear from a number of thought-leaders on why RTC adoption of serverless is lagging other industries, what it means wrt industry competitiveness, and ways to overcome those barriers.
- Tim Panton, Co-founder and CTO at Pipe
- João Camarate Silva, Founder & CTO GoContact
- Simon Woodhead, CEO Simwood
- Jason Berryman, Freelance Google Cloud Architect and Google Developers Expert at 418.dev
- Sebastian Schumann, Network Production Development, Technology & Innovation at Deutsche Telekom
- Marten Schoenherr, CEO/Founder at Automat Berlin GmbH
Each panelist kicked off a quick 5 (or so) min intro to their views on the topic. You can review them in this Serverless Opinions weblog. It’s a great resource, and well-worth your time. In this weblog we focus on the panel discussion, and in the comments section break into a Q&A.
Again an apology, I was aiming for 30 mins, and failed miserably as the discussion got interesting on the future of RTC.
I focused on 5 questions:
- Does everyone agree with Tim’s view that: the lifecycle of serverless functions is short, max 9 mins. Which is not compatible with the longer sessions of communications?
- One issue brought up is the limited set of functions available on serverless today, is that significant for RTC?
- Less serverless more cloud: What’s your experience with VoIP/SIP servers and containers?
- Timing of serverless for RTC? Now, part of a hybrid solution, or never as the world will have moved on?
- Will some technology companies skip serverless and move to a decentralized event drive stack using HTTP/3 RIPP?
I apologize for grossly simplifying the answers / discussion below. My aim is to give you the main gist, in the hope you invest your time and watch the video. The answers were:
- Generally yes. Jason highlighted there are some tools that can extend the lifetime to 15 mins. General consensus was Serverless for production RTC services is not necessarily appropriate. Now you’d think that would be the end of the panel discussion 😉
- There are tools like Knative and Cloud Run, which extend serverless to containers. So serverless functionality is not an necessarily an issue.
- Simon mentioned Simwood has been running its services in containers for 5+ years. As part of a hybrid approach across bare metal, containers, and bursting into the public cloud. While Sebastian highlighted telcos are taking an incremental approach, in part limited by their planning cycles. And open source telecom projects are moving into containers as well. Containerization should be considered BAU by now.
- Some workloads are ideal for serverless, particularly in development and testing, strong recommendations there and excellent examples. Also for workloads like data analysis, machine learning, transcription, make sense. So severless is being adopted just not for the RTC core.
- The last question was where things got interesting, and the reason we overran. Briefly: because Amazon, Google, Microsoft dominate developer fashion, technology companies have no choice by to adopt serverless, its inevitable. The discussion above shows the focus is for specific workloads, not necessarily the RTC core. The continued progression of abstraction from AGM, where rather than CPaaS APIs or WebRTC SDKs, calling simply becomes a function received much discussion and recognition of its inevitability given development trends. The role of edge compute was a little more contentious, but I think decentralization and edge compute are a reaction to the inevitable abstraction and silo lock-in AGM want to achieve.
Overall, an excellent panel discussion thanks to the world-class panelists and lots of items to discuss into next year on the emerging communications dichotomy.