DIRAC Workload Management System

DIRAC (Distributed Infrastructure with Remote Agent Control) is the Workload and Data Management system (WMS) for the LHCb experiment. The DIRAC WMS offers a transparent way for LHCb users to submit jobs to the EGEE Grid as well as local clusters and individual PCs. This paper will describe workload management optimizations, which ensure high job efficiency and minimized job start times. The computing requirements of the LHCb experiment can only be fulfilled through the use of many distributed compute resources. DIRAC provides a robust platform to run data productions on all the resources available to LHCb including the EGEE Grid. More recently, user support was added to DIRAC that greatly simplifies the procedure of submitting, monitoring and retrieving output of Grid jobs for the LHCb user community. DIRAC submits Pilot Agents to the EGEE Grid via the gLite WMS as normal jobs. Pilot Agents then request jobs from the DIRAC Workload Management System after the local environment has been checked. Therefore DIRAC realizes the so-called PULL paradigm, which ensures a high efficiency for LHCb Grid jobs. The possibility of using generic VO Pilot Agents is very exciting and DIRAC is ready to exploit tools such as glexec in order to optimize workloads. This would allow DIRAC to work in a ‘filling’ mode, by which multiple jobs may be requested for execution by Agents deployed to Grid Worker Nodes in a secure way.