SCALEA:Performance Instrumentation, Measurement, Analysis and Visualization Tool for Parallel Programs.

Introduction

SCALEA is a performance instrumentation, measurement, analysis and visualization tool for parallel programs. SCALEA currently focuses on performance analysis for OpenMP, MPI, HPF, and mixed parallel/distributed programs. It computes a variety of performance metrics based on a novel overhead classification. SCALEA also supports multiple experiment performance analysis that allows to compare and to evaluate the performance outcome of several experiments.

Based on SCALEA, we are currently developing SCALEA-G which is a unified monitoring and performance analysis system in the Grid.

SCALEA architecture

Above figure shows the architecture of SCALEA which consists of several components. The instrumentation system(SIS-Scalea Instrumentation System) allows the users to instrument their programs. The input can be Fortran OpenMP, MPI, HPF, hybrid programs such as OpenMP/MPI( of course sequential program also support). Currently, SIS is integrated with VFC compiler. SIS enables the user to select (by directives or command-line options) code regions and performance metrics of interest. Based on pre-selected code regions, SIS automatically inserts probes (instrumentation code) in the code which will collect all relevant performance information in a set of profile/trace files during execution of the program on a target architecture. SIS also generates an instrumentation description file that enables to relate all gathered performance data back to the input program. SIS is based on some instrumentation libraries: SISPROFILING, PAPI,etc. SISPROFILING supports profiling (timers, counters, and hardware parameters) and creates the dynamic code region call graph. Hardware parameters are determined through an interface with PAPI.

The SCALEA performance analysis and visualization module analyzes the raw performance data which is collected post-mortem or online and stored in the performance data repository. It computes all user-requested performance metrics, and visualizes them together with the input program. Besides single-experiment analysis, SCALEA also supports multi-experiment performance analysis. The visualization engine provides a rich set of displays for various metrics in isolation or together with the source code.

The SCALEA performance data repository holds relevant information about the experiments conducted.

Detail information can be found in:

SCALEA provides instrumentaion and measurement system, performance data repository, performance analysis and visualization component and other utilities for the AKSUM, the ZENTURIO tool in the context of the ASKALON toolset.

SCALEA is a part of ASKALON - a programming environment and tool set for cluster and grid computing. This research is funded by the Austrian Science Fund as part of Aurora Project under contract SFBF1104, and the IST Working Group Apart under Contract No. EP 29488.

Back to homepage of Aurora Project 4
This page is under construction. Last modified: Mon Nov 17 13:25:08 MET 2003 by Hong-Linh Truong.