Managing Parallel Performance Experiments with PPerfDB

M.E. Colgrove, C.L. Hansen, and K.L. Karavanic (USA)


Performance Evaluation, Parallel Performance Tools, Experiment Management, Heterogeneous Data


Effective performance tuning of large scale applications on today's clusters and supercomputers requires the ability to integrate data gathered with a variety of monitoring tools, stored in different formats, and possibly in geographically separate data stores. In this paper we present the PPerfDB Experiment Management tool. PPerfDB incorporates comparative analysis techniques, enables direct comparison of performance data collected using a variety of types of performance tools, and provides the user with a simple interface to navigate what is often a large amount of data. PPerfDB includes an experiment launch feature that allows unavailable performance data to be gathered in fresh measurement runs. PPerfDB's PPerfXchange module allows performance data stored in different locations and formats to be integrated and used in a single performance analysis session.

Important Links:

Go Back