Improving online performance diagnosis by the use of historical performance data
SC '99 Proceedings of the 1999 ACM/IEEE conference on Supercomputing
Formalizing OpenMP Performance Properties with ASL
ISHPC '00 Proceedings of the Third International Symposium on High Performance Computing
Automatic Performance Analysis of MPI Applications Based on Event Traces
Euro-Par '00 Proceedings from the 6th International Euro-Par Conference on Parallel Processing
Autopilot: Adaptive Control of Distributed Applications
HPDC '98 Proceedings of the 7th IEEE International Symposium on High Performance Distributed Computing
A Monitoring Sensor Management System for Grid Environments
HPDC '00 Proceedings of the 9th IEEE International Symposium on High Performance Distributed Computing
Specification of Performance Problems in MPI Programs with ASL
ICPP '00 Proceedings of the Proceedings of the 2000 International Conference on Parallel Processing
HiPC '02 Proceedings of the 9th International Conference on High Performance Computing
Logging kernel events on clusters
ICCS'03 Proceedings of the 2003 international conference on Computational science
TAUg: runtime global performance data access using MPI
EuroPVM/MPI'06 Proceedings of the 13th European PVM/MPI User's Group conference on Recent advances in parallel virtual machine and message passing interface
Hi-index | 0.00 |
Performance analysis for applications on teraflop computers requires a new combination of concepts: online processing, automation, and distribution. This article presents the design of a new analysis system that performs an automatic search for performance problems. This search is guided by a specification of performance properties based on the APART Specification Language. The system is being implemented as a network of analysis agents that are arranged in a hierarchy. Higher level agents search for global performance problems while lower level agents search local performance problems. Leaf agents request and receive performance data from the monitoring library linked to the application. Our online analysis takes also into account design patterns for parallel applications. These patterns make the analysis more effective and the output more application-related. The analysis is currently being implemented for the Hitachi SR8000 teraflop computer at the Leibniz-Rechenzentrum in Munich within the Peridot project.