Efficiently computing derived performance data

摘要

In this paper, we examine the problem of generating instrumentation that efficiently computes derived performance data of program execution. The instrumentation computes these derived data at run time as the directly accessible data is collected. It tests conditions defined on the collected data and eliminates the data that does not satisfy the stated conditions, thus, avoiding recording irrelevant data. We examine this problem in terms of generating instrumentation for computing the answers to monitoring questions. Our solution is based on the use of temporal conditions defined on the events in the questions to determine which data to record and to be used as relevant run time filters on the events producing that data.