SRPT for Multiserver Systems

The Shortest Remaining Processing Time (SRPT) scheduling policy and variants thereof have been deployed in many computer systems, including web servers [5], networks [9], databases [3] and operating systems [1]. SRPT has also long been a topic of fascination for queueing theorists due to its optimality properties. In 1966, the mean response time for SRPT was first derived [11], and in 1968 SRPT was shown to minimize mean response time in both a stochastic sense and a worst-case sense [10]. However, these beautiful optimality results and the analysis of SRPT are only known for single-server systems. Almost nothing is known about SRPT in multiserver systems, such as the M/G/k, even for the case of just k = 2 servers.