A new mechanism of the message aggregation is proposed for debugging large-scale parallel programs. A remote source level debugging prototype system that is suitable for debugging large-scale parallel programs is created. The core of the message aggregation strategy includes: a tree structure that is used to aggregate and transfer the messages
and the resulting message returned by debug combination commands. Through the improved tree structure with fixed number the tree can be partitioned into two layers: a tree formed by nodes
and a tree formed by controllers within nodes
so that the amount of message passing between nodes can be greatly reduced. The serial debug tool GNU gdb is integrated into the bottom of the system to carry out the single process debug. A method of message combination is designed for satisfying the four kinds of gdb results. Meanwhile
the original output data of gdb are retained in the resulting information. Experimental results show that the debugging prototype system satisfies user's basic demands
simplifies the remote calculation
and shields the complicated hardware structure and processing flow.
关键词
Keywords
references
LUMETTA S S, CULLER D E. The mantis parallel debugger [C]∥Proceedings of the 1st Symposium on Parallel and Distributed Tools. New York, USA: ACM, 1996: 118-126.
GRAG W. Parallel tools platform-parallel debugger[EB/OL].[2008-08-12]. http:∥dev.eclipse.org/viewcvs/index.cgi/~checkout~/org.eclipse.ptp/doc/presentations/CDT_2006.pdf?cvsroot=Tools_Project.
EDGAR G, GRAHAM E F, GEORGE B, et al. Open MPI: goals, concept, and design of a next generation MPI implementation [C]∥Proceedings of 11th European PVM/MPI Users' Group Meeting. Budapest, Hungary: Springer, 2004: 97-104.
CHEN Yong, HE Kedong, LU Zaichao, et al. Design and implementation of a parallel debugger for dawning cluster system [J]. Computer Engineering, 2004, 30(9): 50-52.
HOOD R. The p2d2 project: building a portable distributed debugger [C]∥Proceedings of ACM SIGMETRICS Symposium on Parallel and Distributed Tools. New York, USA: ACM, 1996: 127-136.
BALLE S M, BRETT B R. Extending a traditional debugger to debug massively parallel applications [J]. Journal of Parallel and Distributed Computing, 2004, 64(5): 617-628.
PHAM C, ALBRECHT C. Optimizing message aggregation for parallel simulation on high performance clusters [C]∥The 7th IEEE International Symposium on Modeling, Analysis, and Simulation of Computer and Telecommunications Systems. Piscataway, NJ, USA: IEEE, 1999:76.
Message Passing Interface Forum. MPI2: a message passing interface standard [J]. International Journal of High Performance Computing Applications, 1998,12(1/2):1-299.