Monitoring of Parallel Programs in Clusters and in Grids

Podhorszki Norbert <pnorbert@sztaki.hu>

MTA SZTAKI

A parallel application running in a computer cluster can be examined by the GRM monitoring and by the PROVE performance visualisation tools. These tools are currently integrated parts of the P-GRADE graphical parallel program development environment. In the framework of the EU-DataGrid (Esprit Framework V) project these tools are being extended for the examination of grid applications and to be standalone tools.

In this paper, GRM and PROVE are shortly introduced as they are used for the performance analysis of a P-GRADE application running in a cluster. The new requirements of their use in a grid environment are analysed. Two different ways of the development are described.

In the first case, the GRM monitoring tool is modified to be a standalone grid tool while in the latter case, it is connected into the relational monitoring and information architecture of the DataGrid project.

The hardest problem of the first direction is the start-up of the GRM and attaching it to the running application. In the other direction, the GRM should be connected to a Java-based tool (R-GMA) supporting the relational data model. In this case, the instrumentation library should be modified as well, as the global trace data collector component of GRM.

The P-GRADE environment can be found at: http://www.lpds.sztaki.hu/projects/p_grade