Non-Cumulative Resource Analysis (Author’s version)

Abstract

Existing cost analysis frameworks have been defined for cumulative resources which keep on increasing along the computation. Traditional cumulative resources are execution time, number of executed steps, amount of memory allocated, and energy consumption. Noncumulative resources are acquired and (possibly) released along the execution. Examples of non-cumulative cost are memory usage in the presence of garbage collection, number of connections established that are later closed, or resources requested to a virtual host which are released after using them. We present, to the best of our knowledge, the first generic static analysis framework to infer an upper bound on the peak cost for non-cumulative types of resources. Our analysis comprises several components: (1) a pre-analysis to infer when resources are being used simultaneously, (2) a program-point resource analysis which infers an upper bound on the cost at the points of interest (namely the points where resources are acquired) and (3) the elimination from the upper bounds obtained in (2) of those resources accumulated that are not used simultaneously. We report on a prototype implementation of our analysis that can be used on a simple imperative language.