A New Model for Cluster Communications Optimization

摘要

Performance losses of cluster applications can arise from various sources in the communications network of computer clusters. Typically, CPU intensive applications generate a small amount of network traffic the overall influence of the network subsystem is minimal. On the other hand, a dataintensive and network aware application generates a large amount of network traffic and the influence of the network subsystem is significantly greater. This paper presents a model that aims to improve the cluster’s network performance by reducing the data transfer time, this solution having the advantage that doesn’t imply modifications of the original applications or of the kernel.