Optimizing Resource Utilization in Grid Batch Systems
Andreas Gellrich · Journal of Physics Conference Series · 2012
On Grid sites, the requirements of the computing tasks (jobs) to computing, storage, and network resources differ widely. For instance Monte Carlo production jobs are almost purely CPU-bound, whereas physics analysis jobs demand high data rates. In order to optimize the utilization of the compute node resources, jobs must be distributed intelligently over the nodes. Although the job resource requirements cannot be deduced directly, jobs are mapped to POSIX UID/GID according to the VO, VOMS group and role information contained in the VOMS proxy. The UID/GID then allows to distinguish jobs, if users are using VOMS proxies as planned by the VO management, e.g. 'role=production' for Monte Carlo jobs. It is possible to setup and configure batch systems (queuing system and scheduler) at Grid sites based on these considerations although scaling limits were observed with the scheduler MAUI. In tests these limitations could be overcome with a home-made scheduler.