Load balancing for heterogeneous clusters of PCs

Abstract

With commercial supercomputers and homogeneous clusters of PCs, static load balancing is accomplished by assigning equal tasks to each processor. With heterogeneous clusters, the system designers have the option of quickly adding newer hardware that is more powerful than the existing hardware. When this is done, the assignment of equal tasks to each processor results in suboptimal performance. This research addresses techniques by which the size of the task assigned to a processor is a suitable match. Thus, the more powerful processors do more work and the less powerful processors perform less work. We find that when the range of processing power is narrow, some benefit can be achieved with asymmetric load balancing. When the range of processing power is broad, dramatic improvements in performance are realized — our experiments have shown up to 92% improvement when asymmetrically load balancing a modified version of the computationally intensive NAS Parallel Benchmarks’ LU application on a heterogeneous cluster of Linux-powered PCs.