30 seconds is not enough!

Abstract

The basic system timer facilities used by applications and OS kernels for scheduling timeouts and periodic activities have remained largely unchanged for decades, while hardware architectures and application loads have changed radically. This raises concerns with CPU overhead power management and application responsiveness. In this paper we study how kernel timers are used in the Linux and Vista kernels, and the instrumentation challenges and tradeoffs inherent in conducting such a study. We show how the same timer facilities serve at least five distinct purposes, and examine their performance characteristics under a selection of application workloads. We show that many timer parameters supplied by application and kernel programmers are somewhat arbitrary, and examine the potential benefit of adaptive timeouts. We also discuss the further implications of our results, both for enhancements to the system timer functionality in existing kernels, and for the clean-slate design of a system timer subsystem for new OS kernels, including the extent to which applications might require such an interface at all.