Brief Announcement: Acceleration by Contention for Shared Memory Mutual Exclusion Algorithms

Abstract

This paper is exploring a possibility of designing distributed algorithms accelerated by high contention. We propose a mutual exclusion algorithm with such a property for asynchronous read/write shared memory systems with N processes. In a mutual exclusion algorithm, each process executes its entry and exit sections to enter its critical section, where mutual exclusion: at most one process executes its critical section at any time, and starvation freedom: each process that executes its entry section eventually executes its critical section, are required.

We propose an efficient mutual exclusion algorithm with respect to remote memory reference (RMR) complexity. Yang et al. [1] proposed an algorithm with the worst case RMR complexity of O(logN) and Attiya et al. [2] proved the lower bound of ${\it \Omega}(\log N)$. Though our algorithm has the worst case RMR complexity of O(logN), it becomes efficient with increasing the number of processes executing concurrently. We show the efficiency using queuing theory and simulation.