Detecting unaffected race conditions in message-passing programs

Abstract

Detecting unaffected race conditions before which no other races causally happened is important to debugging message-passing programs effectively, because such a message race can affect other races to occur or not. The previous techniques to detect efficiently unaffected races do not guarantee that all of the detected races are unaffected. This paper presents a novel technique that traces the states of the locally-first race to occur in every process, and then visualizes effectively the affect-relations of all the locally-first races to detect unaffected races. © Springer-Verlag 2004.