G-PPI: Discovery of Protein-To-Protein Interactions On the Grid
Presented by Eray OZKURAL on 23 Sep 2008 from 16:24 to 16:24
Session: Demos and Posters
Description of the activity
We have gridified an existing parallel application for discovering protein-to-protein interactions in a protein database using the P-GRADE portal as part of the SEE-GRID 2 project.
We have made use of the Parameter Study support in P-GRADE, and we have shown how to gridify a real-world application using the high-level interface of P-GRADE alone, by implementing a map/reduce style computation using the Generator/Collector jobs in P-GRADE's PS feature. The simplicity of the application and the way we have managed to encapsulate the gridification in the P-GRADE workflow should be of appeal to grid application developers. The aim of the gridification is to speed up the massive computations required by large protein databases by distributing the jobs to grid sites, making use of multiple sites for the computation.
The common grid applications seem to be MPI applications encapsulated in a JDL file. However, we think that decomposing the parallel task to several small non-MPI jobs can be more effective and suitable for the Grid. Our application demonstrates that this is indeed possible, using the high level GUI of P-GRADE. We think that our application should encourage better infrastructure for supporting a large number of jobs and fault recovery, and for designing applications in this vein. In particular, we have shown how useful P-GRADE can be for application developers.