Abstract
We study hierarchical configuration of distributed systems. A distributed system consists of processes that are distributed over a network of processors and cooperate to perform some functions. A hierarchical approach is to group and organize the distributed processes into a logical hierarchy to improve the overall system performance. In this work, we focus on distributed computing jobs that require collecting information from all processors. By limiting the levels of hierarchy to two, we propose optimal hierarchical configurations for meshes and hypercubes. Based on analytical results, algorithms for multiple-level partition are proposed which achieve near-optimal communication cost attributed to information collection.