Description

Maintaining performance guarantee of applications running in Grid environment has always been a demanding research area. Grid provides scalable, secure pool of virtual high-end resources. This work proposes an agent-based integrated framework for automated performance analysis and tuning of compute intensive applications running in computational Grid environment. Objective of this work is to design and implement the framework with two definite components - the first one is an automated performance analysis system that gathers performance-monitoring data and on the basis of this data detects performance problems in the application. The other one is adaptive execution of the application based on the performance problem detected in previous phase during runtime. The work demonstrates the effectiveness of the system for managing execution of multiple concurrent jobs by extending the features of an existing tool, PRAGMA (Performance based Resource Brokering and Adaptive execution in Grid for Multiple concurrent Applications). This work also evaluates the efficiency of this tool by performing different sets of experiments using batches of jobs (benchmark codes) on a local Grid test-bed.