Brief Announcement: Suffix Analysis

We present the suffix analysis problem: a method for analyzing the performance of cache replacement algorithms in the presence of pauses and restarts. Our model looks at how such algorithms perform when forced to empty their contents and then serve the remainder of a trace by comparing their performance on a full trace to their performance on any suffix of that trace. We provide upper and lower bounds for the Landlord (AKA Greedy-Dual Size) algorithm, and show that those bounds depend on internal implementation choices (ranging from Theta(1) to Theta(cachesize)). To the best of our knowledge, this is the first formal theoretical analysis of cache performance that focuses on the penalties of restarting.