Automated Grid Probe System to Improve End-To-End Grid Reliability for a Science Gateway

In 2010, the science gateway nanoHUB.org, the world’s largest nanotechnology user facility, hosted 9,809 simulation users who performed 372,404 simulation runs. Many of these jobs are compute-intensive runs that benefit from submission to clusters at Purdue, TeraGrid, and Open Science Grid (OSG). Most of the nanoHUB users are not computational experts but end-users who expect complete and uninterrupted service. Within the ecology of grid computing resources, we need to manage the grid submissions of these users transparently with the highest possible degree of user satisfaction. In order to best utilize grid computing resources, we have developed a grid probe protocol to test the job submission system from end to end. Beginning in January 2009, we have collected a total of 1.2 million probe results from job submissions to TeraGrid, OSG, Purdue, and nanoHUB compute clusters. We then utilized these results to intelligently submit jobs to various grid sites using a model for probability of success based in part on probe test history. In this paper we present details of our grid probe model, results from the grid probe runs, and a discussion of data from production runs over the same time period. These results have allowed us to begin assessing our utilization of grid resources while providing our users with satisfactory outcomes