Reducing Heavy-Traffic Response Time with Asymmetric Dispatching

Abstract

Reducing mean response time has always been a desirable goal in queueing systems. If job sizes (a.k.a. service times) are known to the scheduler, the policy that minimizes mean response time of a single-server queue is SRPT (Shortest Remaining Processing Time). This is true even for queues that are part of a larger system, such as immediate-dispatch systems where jobs are sent to one of multiple single-server queues upon arrival.