Top-Down or Bottom-Up? Complexity Analyses of Synchronous Multiparty Session Types

TL;DR

This paper compares the complexity of top-down and bottom-up approaches in multiparty session types, introducing new algorithms and analyzing their efficiency and computational costs.

Contribution

It provides detailed complexity analyses of both approaches, introduces new algorithms using graph representations, and demonstrates the efficiency differences through theoretical bounds.

Findings

Top-down approach generally more efficient than bottom-up in realistic scenarios.

Type inference in bottom-up approach has exponential complexity.

Liveness checking in bottom-up approach is PSPACE-hard.

Abstract

Multiparty session types provide a type discipline for ensuring communication safety, deadlock-freedom and liveness for multiple concurrently running participants. The original formulation of MPST takes the top-down approach, where a global type specifies a bird's eye view of the intended interactions between participants, and each distributed process is locally type-checked against its end-point projection. A more recent one takes the bottom-up approach, where a desired property $\varphi$ of a set of participants is ensured if the same property $\varphi$ is true for an ensemble of end-point types (a typing context) inferred from each participant. This paper compares these two main procedures of MPST, giving their detailed complexity analyses. To this aim, we build several new algorithms missing from the bottom-up or top-down workflows by using graph representation of session types. We first propose a subtyping system based on type graphs, offering more efficient subtype-checking than the existing (exponential) inductive algorithm. Next for the top-down, we measure complexity of the four end-point projections from the literature. For bottom-up, we develop a novel type inference system from MPST processes which generates minimum type graphs, succinctly capturing covariance of internal choice and contravariance of external choice. For property-checking of typing contexts, we achieve PSPACE-hardness by reducing it from the quantified Boolean formula problem, and prove membership in PSPACE. Finally, we calculate the total complexity of the top-down and the bottom-up approaches. Our analyses reveal that the top-down based on global types is more efficient than the bottom-up in many realistic cases; liveness checking for typing contexts in the bottom-up has the highest complexity; and the type inference costs exponential against the size of a process, which impacts the total complexity.