Automatic Tuning in Computational Grids 1

Abstract. The possibility of have available massive computer resources to users as a market commodity opens the ideas for the future of the interoperability between the multiple infrastructure systems. This wide system should be composed of multiple high performance resource clusters and their users should share them to solve big scientific problems. These resources have a dynamic behavior and to reach the expected performance indexes it is necessary to tune the application in an automatic and dynamic way. The MATE environment was designed to tune parallel applications running on a cluster. This paper presents the key ideas for tracking down application process in a wide distributed environment like Computational Grids, and enables the use of MATE for dynamic application optimizations in such systems.