Return to the Table of ContentsObjective: Our objective is to provide CFD programmers with a debugging tool tailored to their problem domain. The tool should be portable across a wide variety of parallel and distributed machines and should, in addition to standard debugging operations, provide assistance in isolating problems arising from the parallel or distributed nature of the target program.
Approach: We are using a client-server strategy to isolate user interface code from code dependent on architecture or operating system. This ensures that users can see the same tool on a variety of machines. The client-server interaction protocol is designed in such a way that vendors are encouraged to implement servers for their own machines using their existing debuggers as a code base.
Accomplishment: A draft version of the client-server interaction protocol has been specified and a prototype client-server pair implemented. The prototype is capable of debugging sequential Fortran programs on Sun Sparcstations.
The distributed debugger based on the client-server organization has been designed. That work is being reported professionally via a paper at Supercomputing '94.
Significance: Scientific programmers using parallel and distributed machines face two problems. The first is that debugging tools at their disposal are often not helpful enough to be worth learning and using. The second is that variance across machines makes learning about debugging tools more difficult. The consequence of these problems is that relatively few users of parallel and distributed machines use tools to help isolate problems in programs that are significantly more error-prone than their sequential analogs. What the programming community needs is a single tool that they can use on a variety of platforms that is actually useful in isolating programming and communication errors.
Status/Plans: A prototype implementation is underway. We expect to demonstrate it in November at Supercomputing '94 and to make an initial release in the summer of 1995.
Point of Contact:
Return to the Table of Contents