Specifying performance properties of parallel applications using compound events

Authors:
Felix Wolf;Bernd Mohr
Affiliations:
Forschungszentrum Jülich, Zentralinstitut für Angewandte Mathematik, Germany;Forschungszentrum Jülich, Zentralinstitut für Angewandte Mathematik, Germany
Venue:
On-line monitoring systems and computer tool interoperability
Year:
2004

Citing 9
Cited 0

Debugging programs in a distributed system environment

Debugging programs in a distributed system environment
Using MPI: portable parallel programming with the message-passing interface

Using MPI: portable parallel programming with the message-passing interface
Practical programming in Tcl and Tk (2nd ed.)

Practical programming in Tcl and Tk (2nd ed.)
The Unified Modeling Language user guide

The Unified Modeling Language user guide
Performance analysis of distributed applications using automatic classification of communication inefficiencies

Proceedings of the 14th international conference on Supercomputing
Advanced PERL Programming

Advanced PERL Programming
Python; Essential Reference

Python; Essential Reference
The Paradyn Parallel Performance Measurement Tool

Computer
EARL - A Programmable and Extensible Toolkit for Analyzing Event Traces of Message Passing Programs

HPCN Europe '99 Proceedings of the 7th International Conference on High-Performance Computing and Networking

Quantified Score

Hi-index	0.00

Visualization

Abstract

In this article we present a novel approach to analyzing performance properties of parallel applications based on event traces. We define a framework for formally specifying compound events that characterize performance behavior. The framework allows the creation of abstract building blocks that represent concepts of the underlying programming model and therefore provide an easy means to specify complex compound events. We show that using these specifications it is possible to automatically locate inefficiencies in parallel applications by looking for occurrences of the corresponding compound events in event traces. We demonstrate how our ideas apply to both the Message Passing Interface (MPI) programming model and to the Open specifications for Multi Processing (OpenMP).