The Diffusion Method

Abstract

With the diffusion method, each processor would transfer fractions of its workload to some of its neighbors while simultaneously receiving workloads from its other neighbors at each iteration step. In this chapter, we analyze the convergence and the efficiency properties of the diffusive load balancing process. The focus is on the design of efficient diffusion algorithms in the higher-dimensional tori and meshes including their special cases, the ring, the chain, and the k-ary n-cube.