Scalable Linux Scheduling

Abstract

For most of its existence, Linux has been used primarily as a personal desktop operating system. Yet, in recent times, its use as a cost-efficient alternative to commercial operating systems for network servers, distributed workstations and other large-scale systems has been increasing. Despite its remarkable rise in popularity, Linux exhibits many undesirable performance traits. Concerned about the scalability of multithreaded network servers powered by Linux, we investigate improvements to the Linux scheduler. We focus on pre-calculating base priorities and sorting the run queue for efficient task selection. We propose an improved scheduler design and compare our implementation in terms of scalability and performance to the existing Linux scheduler. Our analysis shows that improvements can be made to the existing scheduler without introducing overhead, thus improving the scalability and robustness of the Linux operating system.