Deconstructing Queue-Based Mutual Exclusion

TL;DR

This paper introduces a modular approach to designing queue-based mutual exclusion algorithms for shared memory multiprocessors, achieving optimal remote memory reference complexity and providing rigorous correctness and complexity analyses.

Contribution

It presents a generic mutual exclusion framework using a new MutexQueue data structure and offers two O(1) RMR implementations based on common synchronization primitives.

Findings

Achieves O(1) RMR complexity for mutual exclusion algorithms.

Provides rigorous correctness proofs for the proposed algorithms.

Demonstrates implementations based on existing synchronization primitives.

Abstract

We formulate a modular approach to the design and analysis of a particular class of mutual exclusion algorithms for shared memory multiprocessor systems. Specifically, we consider algorithms that organize waiting processes into a queue. Such algorithms can achieve O(1) remote memory reference (RMR) complexity, which minimizes (asymptotically) the amount of traffic through the processor-memory interconnect. We first describe a generic mutual exclusion algorithm that relies on a linearizable implementation of a particular queue-like data structure that we call MutexQueue. Next, we show two implementations of MutexQueue using O(1) RMRs per operation based on synchronization primitives commonly available in multiprocessors. These implementations follow closely the queuing code embedded in previously published mutual exclusion algorithms. We provide rigorous correctness proofs and RMR complexity analyses of the algorithms we present.