Prediction and bounds on shared cache demand from memory access interleaving

Abstract

Cache in multicore machines is often shared, and the cache performance depends on how memory accesses belonging to different programs interleave with one another. The full range of performance possibilities includes all possible interleavings, which are too numerous to be studied by experiments for any mix of non-trivial programs. This paper presents a theory to characterize the effect of memory access interleaving due to parallel execution of non-data-sharing programs. The theory uses an established metric called the footprint (which can be used to calculate miss ratios in fully-associative LRU caches) to measure cache demand, and considers the full range of interleaving possibilities. The paper proves a lower bound for footprints of interleaved traces, and then formulates an upper bound in terms of the footprints of the constituent traces. It also shows the correctness of footprint composition used in a number of existing techniques, and places precise bounds on its accuracy.