Size-based scheduling policies with inaccurate scheduling information

Size-based scheduling policies, such as shortest remaining processing time (SRPT), have been studied since the 1960s and have been applied in various areas, including packet networks and Web server scheduling. SRPT has been proven to be optimal in the sense that it yields - compared to any other conceivable strategy - the smallest mean value of occupancy and therefore also of waiting and delay time. One important prerequisite to applying size-based scheduling is to know the sizes of all jobs in advance, which are unfortunately not always available. No work has been done to study the performance of size-based scheduling policies when only inaccurate scheduling information is available. We study the performance of SRPT and fair sojourn protocol (FSP) as a function of the correlation coefficient between the actual job sizes and estimated job sizes. We have developed a simulator that supports both M/G/l/m and G/G/n/m queuing models. The simulator can be driven by trace data or synthetic data produced by a workload generator we have developed that allows us to control the correlation. The simulations show that the degree of correlation has a dramatic effect on the performance of SRPT and FSP, and that a reasonably good job size estimator makes both SRPT and FSP outperform processor sharing (PS) in both mean response time and slowdown.