An OS-Oriented performance monitoring tool for multicore systems

Abstract

Hardware performance monitoring counters (PMCs) have proven effective in characterizing application performance. Because PMCs can be only accessed directly at the OS privilege level, kernel-level tools must be developed to enable the end user and userspace programs to access PMCs. A large body of work has demonstrated that the OS scheduler can perform effective runtime optimizations in multicore systems by leveraging per-thread performance-counter data. Notably, while existing tools greatly simplify collecting PMC application data from user space, they do not provide a simple mechanism making it possible for the thread scheduler to use performance counters for its own purpose. To address this shortcoming we present PMCTrack, a novel tool for the Linux kernel that provides a simple architecture-independent mechanism making it possible for the OS scheduler to access per-thread PMC data. Despite being an OS-oriented tool, PMCTrack still allows gathering PMC values from user space, enabling kernel developers to carry out the necessary offline analysis and debugging to assist them during the scheduler design process. In addition, the tool provides both the scheduler and the userspace PMCTrack components with other insightful metrics available in modern processors that are not directly exposed as PMCs, such as cache occupancy or energy consumption. In this paper, we analyze different case studies that demonstrate the potential benefits of PMCTrack.