Statistical methods for automatic performance bottleneck detection in MPI based programs

Authors:
Michael Kluge;Andreas Knüpfer;Wolfgang E. Nagel
Affiliations:
Technische Universität Dresden, Dresden, Germany;Technische Universität Dresden, Dresden, Germany;Technische Universität Dresden, Dresden, Germany
Venue:
ICCS'05 Proceedings of the 5th international conference on Computational Science - Volume Part I
Year:
2005

Citing 4
Cited 1

CATCH - A Call-Graph Based Automatic Tool for Capture of Hardware Performance Metrics for MPI and OpenMP Applications

Euro-Par '02 Proceedings of the 8th International Euro-Par Conference on Parallel Processing
Scalable analysis techniques for microprocessor performance counter metrics

Proceedings of the 2002 ACM/IEEE conference on Supercomputing
Compressible memory data structures for event-based trace analysis

Future Generation Computer Systems
A new data compression technique for event based program traces

ICCS'03 Proceedings of the 2003 international conference on Computational science: PartIII

Knowledge based automatic scalability analysis and extrapolation for MPI programs

Euro-Par'05 Proceedings of the 11th international Euro-Par conference on Parallel Processing

Quantified Score

Hi-index	0.00

Visualization

Abstract

With increasing number of processors the visualization of trace data for MPI communication primitives becomes harder due to many limitations. An automatic analysis of the recorded data should be able to identify some clues for the optimization of the source code. The approach presented with this paper is able to divide the communication time for each communication relationship into a necessary part and the overhead. We use statistical methods to define a time for each single communication event that we will accept as the ‘usual' time for this event. Additional runtime will be considered as overhead. At the end of this article, the benefits from this method are demonstrated on the Sweep3D benchmark.