An Efficient Multi-level Trace Toolkit for Multi-threaded Applications
Abstract
Nowadays, observing and understanding the behavior and performance of a multithreaded application is nontrivial, especially within a complex multithreaded environment such as a multilevel thread scheduler. In this report, we present a trace toolkit that allows a programmer to precisely analyze the behavior of a multithreaded application. A application's run generates several traces that are merged and analyzed offline. The resulting super-trace contains not only classical information such as the number of elapsed cpu cycles per functions but also details about thread scheduling at multiple levels.