Near-critical path analysis of program activity graphs

Abstract

Program activity graphs can be constructed from time-stamped traces of appropriate execution events. Information about the activities on the k longest execution paths is useful in the analysis of parallel program performance. In this paper, four algorithms for finding the near-critical paths of program activity graphs are presented and compared, including an efficient new algorithm that utilizes slack values calculated by the critical path method to perform a best-first search in linear space. The worst-case time and memory requirements of the new algorithm are in O(ke) and O(k+e), where e is the number of edges in the graph. Results confirming the efficiency of the algorithm are presented for five application programs. A framework for utilizing the near-critical path information is also described. The framework includes both statistical summaries and visualization capabilities.<>