We then use clustering algorithms (Note 4) to regroup nodes that can be executed on the same processor. The task nodes within a cluster are executed by the same processor of the grid. The order of execution is specified by the clustering algorithm.

After obtaining the clusters, we evaluate the performance gain obtained by looking at the amount of computation required to execute the program over the grid as compared to what it was when the program was running on a single processor.

The framework currently works on C and Java code and is being extended for C++. We are further extending the framework to include other parameters that govern the grid application migration effort to create a holistic migration methodology. 

We have taken some sample applications and applied our framework to determine the performance gain achieved through parallelization. We found that the proposed framework is quite effective in analyzing applications for grid enablement and predicting performance gain.

Conclusion
Migrating application systematically and successfully to grid environment is a critical factor in grid adoption. In the note, we introduce GAMF – a framework that systematically analyzes application for grid suitability and predicts performance gains. The initial results look quite promising.

The author is research associate at Infosys Technologies Limited.