Thread by @copyconstruct

The final talk of my track

The state of Serverless computing - or fixing dysfunctionnad a service

Remember this paper? This talk is basically a gist of this paper. 🍿

https://twitter.com/copyconstruct/status/1075086886610169857?s=21

There’s been a bunch of research on serverless recently - and Hacker News trashes all of it.

Yet, serverless use is on the rise. FaaS is good for embarrassingly parallel tasks as well as workflow orchestration

The limitations of current FaaS offerings. The first and last can be overcome, but the middle two are severe limitations.

what you really end up getting is dysfunction as a service.

Even if you have a purely functional model, FaaS is slow. However, not most applications can be composed in a purely functional manner.

The ideal scenario is when we can get both autoscaling and low latency.

DynamoDB has a transaction mode which gives you stronger isolation models but at the cost of latency.

Building real stateful applications is a non-starter today. @cgwu0530 at #QConNYC

Fluent is a FaaS layer on top of Anna.

db.cs.berkeley.edu/jmh/papers/ann…

But Anna as a database (albeit must faster than S3) is stilll fairly slow. Caching can be an optimization? But then introduces problems of cache coherency.

Lattice composition (db.cs.berkeley.edu/papers/UCB-lat…) can causal consistency.

The strongest consistency level that doesn’t require coordination.

How to make FaaS a fully functional.

We now have millions of cores and petabytes of RAM. We need a programming model that’ll allow us to unlock the full potential of the cloud and serverless.