Processor Cache Misses are Reported as Execution

When refactoring a serial application into a parallel application the minimum bar of acceptable performance is that the parallel application must run from beginning to end in less wall clock time than the serial application. If it doesn’t run faster, there’s no point in doing the refactoring. I was working on making a certain application…

0

False Sharing: Why You Can’t Afford To Ignore It

Many of us are aware that caches are our friend, reducing memory latency by exploiting locality. A performance problem known as false sharing can occur in multithreaded scenarios when locality backfires, leading to costly performance degradation. Igor Ostrovsky, Stephen Toub, and Huseyin Yildiz, all members of the Parallel Computing Platform team, have written this article,…

0