A Comparison of Fast and Low Overhead Distributed Priority Locks

Abstract

Distributed synchronization is necessary to coordinate the diverse activities of a distributed system. Priority synchronization is needed for real-time systems or to improve the performance of critical tasks. Practical synchronization techniques require fast response and low overhead. In this paper, we present three priority synchronization algorithms that sendO(logn) messages per critical section request on average and useO(logn) bits of storage per processor. Previous algorithms required 2(n− 1) messages per critical section entry andO(n) bits of storage per processor. Two of the algorithms are based on Li and Hudak's path compression techniques, and the third algorithm uses Raymond's fixed-tree structure. Since each of the algorithms have the same theoretical complexity, we make a performance comparison to determine which of the algorithms is best under different loads and different request priority distributions. We find that when the request priority distribution is stationary, the path-compression algorithm that uses a singly linked list is best overall, but the fixed-tree algorithm requires fewer messages when the number of processors is small and the load is high (100% or greater). When the request priority distribution is nonstationary, the fixed-tree algorithm requires the fewest messages when the load is 100% or greater.