Network performance variability in NOW clusters

Abstract

Performance management of clusters and Grids poses many challenges. Sharing large distributed sets ofresources can provide efficiencies, but it also introduces complexity in terms ofproviding and maintaining adequateperformance. Current application requirements focus on the amount of resources needed without explicitly characterizing the perfornance required from those resources. In clusters and Grids, inconsistent or highly variable application run-time is an indication of systemic inconsistency, with ramificationsfor those running the application and those managing the resources. We arefocusing on the contribution ofthe interconnection network to application run-time variability. This workpresents experimental results characterizingparallel application run-time sensitivity to communication performance variability using an Application Communication Emulator (ACE).