Section 2 describes related work. Section 3 presents the distribution of lifetimes of jobs on the Intel Paragon at SDSC, and proposes the conditional lifetime model (CLM) we use to predict the remaining lifetimes of jobs. In Section 4 we use this CLM to predict the wait time for a given cluster size. Section 5 describes a trace-driven simulator we used to evaluate these techniques and presents our results.
In Section 6 we apply the proposed techniques in another environment (the IBM SP2 at the Cornell Theory Center). Section 7 summarizes our findings and describes our plan for continuing work.