LinuxCon + CloudOpen + ContainerCon NA 2015

In 2004, "The Cursor Wiggles Faster: Measuring Scheduler Performance", presented at the Ottawa Linux Symposium, patches were introduced to create some objective metrics by which to judge scheduler performance. Revisions of these patches were eventually accepted into the kernel and persist to today.

But eleven years later, the scheduler has changed dramatically. Some of those original measurements are still in the code. Additional, equally useful metrics have been introduced outside of the 2004 framework. The need to evaluate scheduler changes has not decreased, but workloads and system capabilities have shifted. Do we still have effective tools to evaluate current work? This paper discusses the aging of scheduler statistics, and discusses whether the old patches need to be gently merged, subtly enhanced, or even completely removed.