Processor-shared time-sharing models in heavy traffic

Abstract

Processor sharing (PS) is a mathematically tractable approximation to time sharing, a procedure followed in many actual computer systems. In effect, PS assigns to each job of the i, (i = 1,2,...) present for processing 1/ith of the total processing effort; equivalently, a single job with Markovian service rate micron completes processing in (t, t+dt) with probability (micron/i)dt + o (dt). One advantage of PS is that short jobs are not trapped behind long jobs, as is possible in a FC-FS discipline. In this document probability models are presented for computer systems with processorshared(time sliced) service discipline. The response (sojourn) time of an arriving job that requires T units of processing time is shown to be approximately Gaussian/normal under moderately heavy traffic conditions, e.g. when the number of terminals becomes large. Keywords: Tables(data)