CAMP: Cost-Aware Multiparty Session Protocols

TL;DR

CAMP is a static analysis framework that uses multiparty session types with annotations to predict communication and computation costs in message-passing systems, aiding performance optimization.

Contribution

It extends multiparty session types with cost annotations and provides a method to estimate execution costs with high accuracy, including asynchronous communication analysis.

Findings

Predicts execution costs with less than 15% error.

Applies to various protocols in multiple programming languages.

Effectively identifies performance bottlenecks.

Abstract

This paper presents CAMP, a new static performance analysis framework for message-passing concurrent and distributed systems, based on the theory of multiparty session types (MPST). Understanding the run-time performance of concurrent and distributed systems is of great importance for the identification of bottlenecks and optimisation opportunities. In the message-passing setting, these bottlenecks are generally communication overheads and synchronisation times. Despite its importance, reasoning about these intensional properties of software, such as performance, has received little attention, compared to verifying extensional properties, such as correctness. Behavioural protocol specifications based on sessions types capture not only extensional, but also intensional properties of concurrent and distributed systems. CAMP augments MPST with annotations of communication latency and local computation cost, defined as estimated execution times, that we use to extract cost equations from protocol descriptions. CAMP is also extendable to analyse asynchronous communication optimisation built on a recent advance of session type theories. We apply our tool to different existing benchmarks and use cases in the literature with a wide range of communication protocols, implemented in C, MPI-C, Scala, Go, and OCaml. Our benchmarks show that, in most of the cases, we predict an upper-bound on the real execution costs with < 15% error.