Abstract:

In this paper, a scalable parallel debug tool (SSTD) was proposed. By capturing and dumping process stack traces when the target parallel program exits abnormally, the SSTD can help users fast locate program faults and improve parallel debug efficiency. The system architecture and some key technologies of the SSTD, including scalable communication infrastructure and process equivalence classification, were discussed in detail.

Key words: parallel debug;stack trace analysis;scalable