Cloud Programming: CALM Lessons and a new PACT
summary
Claims that a monotonicity boundary — the CALM theorem — exactly characterizes when distributed apps can run coordination free, and proposes a practical migration path via a four-facet PACT (Program, Availability, Consistency, Targets) model. Uses verified lifting, program synthesis, a Bloom-like internal language (Hydrologic), lattices and a dataflow runtime. Builds Anna key-value store as proof: coordination-free design with multi-tier auto-scaling, delivering linearly scalable throughput and up to 700x faster than prior academic stores (and 350x DynamoDB-equivalent throughput per dollar), showing the approach works at scale.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 9:24Paxos
- 15:25CALM Theorem
- 16:50CALM Definitions
- 22:27Ana
- 26:23PACT
- 27:55Program Syntax
- 28:43Program Semantics
- 32:42The Stack
- 33:49The Paper
- 34:32More Information
- 35:06Questions
related talks
28mThe Art of the State: Serverless Computing and Distributed Data
01h 03mAdventures in Spacetime
17mKeynote: Programming the Cloud as Easily as Your Laptop
25mCloudstate—Towards Stateful Serverless
16mDeclarative vs programmatic IaC
33mCloudstate Towards Stateful Serverless
46mHow Secure Serverless Microservice Platforms Completely Change Dev & Ops
24mShaping Tomorrow's Technology: Navigating Cloud-Native, Serverless, and Polyglot Programming