Performance prediction on multi-core processors

Abstract

Contention for shared resources between co-running programs on a multi-core processor causes degradation of performance. Previous research work made efforts to characterize and classify memory behavior of programs to predict their performance. Such knowledge may be used to design workloads for performance studies on multi-core processors. It could also be utilized to formulate policies that alleviate contention for shared resources at a system level. In this work we apply machine learning techniques to predict the performance on multi-core processors. The main contribution of the study is enumeration of solo-run program attributes, which can be used to predict paired-run performance. The paired run involves the contention for shared resources between co-running programs. We observed that machine learning techniques could be utilized to predict the paired run performance with reasonable accuracy. © 2010 IEEE.