Parallel Depth First Search on the Ring Architecture

This paper presents the implementation and analysis of parallel depth-first search on the ring architecture. At the heart of the parallel formulation of depth-first search is a dynamic work distribution scheme that divides the work between different processors. The effectiveness of the parallel formulation is strongly influenced by the choice of the work distribution scheme. In particular, a commonly used work distribution scheme is found to give very poor performance on large rings( < 32 processors). We present a new work distribution scheme that is better than the work distribution scheme used by other researchers, and gives good performance even on large rings (128 processors). We introduce the concept of iso-efficiency function to characterize the effectiveness of different work distribution schemes.