A multithreshold scheduling policy for readers and writers

Abstract

We consider a multiserver queueing system processing "readers" and "writers." A writer must be processed singly, as if it is using all servers in the system, while readers can be processed concurrently, each using one server. Higher throughputs and better response times for both readers and writers can be achieved by increasing the degree of concurrency in processing readers. The threshold fastest emptying (TFE) policy stops processing new readers in a system closed with respect to readers when the number of writers which have arrived at the system exceeds a certain threshold K. In a system with reader as well as writer arrivals the TFE policy achieves a higher maximum throughput than the FCFS policy (even when K = 1). We use the solution of a system closed with respect to readers to determine an "optimal" value for K from the viewpoint of minimizing the mean overall response time (Ro) in this case. Activating several readers is more advantageous than activating a single writer from the viewpoint of reducing waiting times and hence Ro, but the issue of efficient processing of readers has to be taken into account at higher loads. A threshold scheduling policy based on a threshold J (resp. J′) to start (resp. stop) the processing of readers is intended to maximize the fraction of time the system is available for processing writers. A queueing analysis of a system with readers only shows that significant improvements in reader processing efficiency can be attained with the single threshold J and a further improvement with the additional threshold J′. Finally, a double-threshold policy with a threshold K for writers and thresholds J and J′ for readers is proposed. The processing of additional readers is stopped when there are less than J′ readers and the threshold K for the processing of writers is attained. The processing of writers is stopped when there are J readers in the system and a certain number of writers are served in the current cycle for processing writers. It follows from simulation studies that this policy provides more flexibility than the aforementioned policies in minimizing Ro. © Elsevier Science Inc. 1998.