Mr Igor Sfiligoi (FNAL)
The advent of the Grids have made it possible for any user to run hundreds of thousands of jobs in a matter of days. However, the batch slots are not organized in a common pool, but are instead grouped in independent pools at hundreds of Grid sites distributed among the five continents. A higher level Workload Management System (WMS) that aggregates resources from many sites is thus necessary. There are several ways to design and implement a WMS; the purpose of this project it to show how some of the most commonly used WMS-es (including gLite WMS, ReSS, and glideinWMS) behave under realistic load conditions. The results presented have been measured using the same tools for all the tested WMS-es, comparing those results against a baseline obtained by using plain Condor-G submissions. Tests were performed at various load levels and with different payloads to test for scalability and reliability issues.