Compensating for perturbation by software performance monitors in asynchronous computations
Theodore F. Hehr · 1990
Designing performance efficient programs for parallel computing systems can benefit from running experiments for retrieving information about how computations execute. Retrieving this information requires some kind of data generation/collection facility that monitors computations as the computations execute. Unfortunately, monitoring an application almost always perturbs its execution profile. The general problem is not the elimination of perturbations, but the detection, measurement and compensation of them. The perturbations considered here are those due to the presence of software sensors and independent monitoring threads that compete for processor resources. The best case results show that the difference between compensation estimates of the average maximum, minimum and mean executions are within 1% of their real values. The results of the work have potential applications for general research on performance prediction. ftn*This research was sponsored in part by the Defense Advanced Research Projects Agency (DOD), Arpa Order No. 4864, monitored by the Space and Naval Warfare Systems Command under contract N00039-85-C-1034, and in part by the National Science Foundation, Grant No. CCR-86-02-143. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of the Defense Advanced Research Projects Agency, the National Science Foundation or the U.S. Government.