Towards Architecture Independent Metrics for Multicore Performance Analysis

The prevalence of multicore architectures has made the performance analysis of multithreaded applications an intriguing area of inquiry. An understanding of locality effects and communication behavior can provide programmers with valuable information about performance bottlenecks and opportunities for optimization. Unfortunately, most performance analyses are architecture dependent, and hence insights gleaned from an application's behavior on one platform may not apply when the application is run on another. The prevalence of multicore architectures has made the performance analysis of multithreaded applications an intriguing area of inquiry. An understanding of locality effects and communication behavior can provide programmers with valuable information about performance bottlenecks and opportunities for optimization.