CAMP: Cost-Aware Multiparty Session Protocols
summary
Extends multiparty session types with runtime cost annotations and extracts per-participant cost equations that account for message sizes, node/CPU counts and network latencies. Builds an EDSL in Haskell that instantiates protocols, runs an analyzer, and proves the extracted equations are an upper bound on operational traces; defines bounded cost and a latency metric Cω for recursive protocols. Evaluations on benchmarks show predictions within less than 15 percent error for most cases and identify communication contention as the main source of deviation.
AI-written summary · about
chapters (12)
- 0:00Multiparty Session Types
- 1:35Example: A Divide-and-Conquer Protocol
- 3:22How much time is required for running a process?
- 3:51Cost-Aware Multiparty Session Protocols
- 7:17Latency of Recursive Protocols: NX.G
- 8:05Example: Recursive Pipeline
- 8:49Asynchronous Optimisations
- 10:25Fast Fourier Transform
- 10:45Correctness (Bounded Cost)
- 11:13Correctness (Recurrences)
- 11:46Implementation
- 12:31Evaluation
related talks
35mEvent-Driven Multiparty Session Actors
25mAutomatically Deriving Cost Models for Structured Parallel Processes(...)
32mBenchmarks for Multiparty Session Types
39mProtocol engineering for communicating actors
37mSession Types: A History and Applications
18mHybrid Multiparty Session Types: Compositionality for Protocol Specification th...
32mMulticompatibility for Multiparty-Session Composition
20mEfficient, Portable, Census-Polymorphic Choreographic Programming