The Art of the State: Serverless Computing and Distributed Data
summary
Berkeley RISELab presents a stack that makes serverless functions stateful by combining a monotonicity-based theory (the CALM theorem) with a key‑value store and serverless runtime: Anna KVS and Cloudburst. Anna uses lattice, coordination‑free data types to scale linearly and shows up to 700x faster under contention and roughly 10x faster than Cassandra geo‑distributed. Cloudburst serves ML prediction workloads with latency near local execution and enables cache‑consistent, auto‑scaling stateful serverless execution.
AI-written summary · about
chapters (17)
- 0:00Intro
- 0:52New Platform + New Language = Innovation
- 2:35Industry finally woke up: Serverless Computing
- 2:50Serverless 101: Functions-as-a-Service (FaaS)
- 4:02The State of Commercial Serverless Computing
- 4:56Three Limitations of Current FaaS (e.g. AWS Lambda)
- 6:47A First Step: Embracing "State"
- 7:45The Challenge: Consistency
- 8:47Classical Consistency Mechanisms: Coordination
- 9:28Coordination Avoidance (a poem)
- 10:33Why Avoid Coordination?
- 16:01That's interesting. Who cares?
- 18:25Hydro: A Platform for Programming the Cloud
- 19:30Anna Serverless KVS
- 20:51Anna Performance
- 23:06Cloudburst: A Stateful Serverless Platform
- 25:17Applications on Cloudburst
related talks
25mCloudstate—Towards Stateful Serverless
40mThe State of Serverless Computing
33mCloudstate Towards Stateful Serverless
28m8 Keys for Successful Serverless Architectures
41mCloud Programming: CALM Lessons and a new PACT
39mServerless at the end of the Universe
51mConfusion In The Land Of The Serverless
17mCloud Computing Made Better: An AWS View on Serverless Computing