Running CMS software on GRID Testbeds

TL;DR

This paper reports on a large-scale test of CMS software running on the European DataGrid middleware, evaluating its performance, challenges, and solutions during a month-long distributed computing stress test involving multiple sites.

Contribution

It presents the first comprehensive evaluation of CMS software deployment on EDG middleware, highlighting procedures, encountered problems, and solutions during a large-scale distributed computing test.

Findings

Successful submission of ~10,000 jobs across nine sites.

Identification of key challenges in middleware integration and job management.

Insights into resource utilization and system performance during the test.

Abstract

Starting in the middle of November 2002, the CMS experiment undertook an evaluation of the European DataGrid Project (EDG) middleware using its event simulation programs. A joint CMS-EDG task force performed a "stress test" by submitting a large number of jobs to many distributed sites. The EDG testbed was complemented with additional CMS-dedicated resources. A total of ~ 10000 jobs consisting of two different computational types were submitted from four different locations in Europe over a period of about one month. Nine sites were active, providing integrated resources of more than 500 CPUs and about 5 TB of disk space (with the additional use of two Mass Storage Systems). Descriptions of the adopted procedures, the problems encountered and the corresponding solutions are reported. Results and evaluations of the test, both from the CMS and the EDG perspectives, are described.