On the Fairness of Linux O(1) Scheduler

The scheduling algorithm of Linux operating systems has to fulfill several conflicting objectives: fast process response time, higher throughput for background jobs, avoidance of process starvation, reconciliation of the needs of low and high priority processes etc. The set of rules used to determine when and how to select a new process to run is called scheduling policy. Current Linux kernel uses Completely Fair Scheduler(CFS) which does not meet the complete requirements of a good scheduler, since its complexity is O(log n) and its scheduling policy can cause starvation of optimally threaded process in the presence of greedy threaded process. Previous scheduler was O(1) scheduler, which has a complexity of O(1). The O(1) scheduler had limitations like poor interactive performance and low throughput for background jobs and hence was replaced by CFS. In this paper we illustrate the reasons for poor interactive performance and low throughput for background jobs and also provide a simple solution to the above problems. Finally we explain an enhancement to the solution for better performance of the modified scheduler.