Effect of Thread Weight Readjustment Scheduler on Scheduling Criteria

Abstract

In this paper, we propose a new approach to significantly optimize some scheduling criteria, context switches and turnaround times in this context, when running multithreaded processes concurrently. Current proportional sharing schedulers allocate CPU time based on the weights of running threads in the system and don’t take into consideration the greedy behavior of multithreaded processes (processes with more threads receive more aggregate CPU time from the scheduler relative to processes with fewer threads). In order to minimize turnaround times and context switches of running multithreaded processes in simultaneous multithreaded (SMT) architectures, we investigate the effect of adjusting the weights of running sibling threads (threads forked from the same process) using novel proportional sharing scheduler, Thread Weight Readjustment Scheduler (TWRS), a proportional share CPU scheduler designed for multithreaded processes, which aims to reduce undesirable events (e.g. context switches) and turnaround times. TWRS provides a practical solution for multitasking operating systems because it operates in concert with existing kernels. We have implemented TWRS in Linux 2.6.24-1, which represents the most prevalent scheduler design (i.e. Completely Fair Scheduler (CFS)). Our evaluation shows that our scheduler minimizes context switches and turnaround time.