The Distributed Application Debugger

摘要

In this paper we introduce a tool for debugging parallel programs which utilize the popular MPI (message passing interface) C library. The tool is called The Distributed Application Debugger and introduces distinct components around the code being debugged in order to play, record, and replay sessions of the user's code remotely. The GUI component incorporates the popular GDB debugger for stepping through code line by line as it runs on the distributed cluster as well as an analysis tool for stepping through the executed portions of code after the session has completed. While the system is composed of multiple components, the logistics of coordinating them is managed without user interaction; requiring the user only to provide the location of the program to debug before starting.