Toward Integration of Data Race Detection in DSM Systems

摘要: We present a distributed algorithm, called djit, for detecting data races in dsm systems. djit is designed as add-on, race condition soon one created. It instantly displays to the user precise place program where occurred. There are no false detections, and missed. have implemented on top of millipage?a fine granularity, page-based system. Our implementation makes novel use operating system protection mechanisms. In particular, we propose cache, which can be used local logging accesses variables. As result, our does not increase message complexity execution, piggybacking all its communication activity dsm-related messages. The performance figures show that detection mechanism has only minor influence performance. measured overheads, averaging few percent, two orders magnitude smaller than those achieved previous work. Thus, technique integration on-the-fly during regular execution feasible first time.