Survey of Parallel A* in Rust

TL;DR

This paper surveys various parallel implementations of the A* algorithm in Rust, comparing their performance to identify the most efficient approach for leveraging multi-threading in graph search tasks.

Contribution

It provides a comprehensive implementation and comparison of centralized, decentralized, and massively-parallel A* algorithms in Rust, highlighting their performance differences.

Findings

Massively-parallel A* shows significant speedup over traditional methods.

Decentralized methods offer better scalability in multi-core environments.

Rust implementations demonstrate the practicality of parallel A* algorithms.

Abstract

A* is one of the most popular Best First Search (BFS) techniques for graphs. It combines the cost-based search of Breadth First Search with a computed heuristic for each node to attempt to locate the goal path faster than traditional Breadth First Search or Depth First Search techniques. However, A* is a sequential algorithm. The standard implementation only runs in one thread. There are a few attempts to get A* to leverage multiple threads. Centralized (SPA*) and Decentralized (DPA*, HDA*) methods are the most standard attempts, with the most unique and modern method being massively-parallel A* (MPA* or GA*). We will attempt an implementation of each in Rust to determine if there is a performance boost, and which one has the best performance.