NIKHEF Local Guide for Administrators

All of the NIKHEF Grid service systems are contained in the NDPF (NIKHEF Data Processing Facility). These services include all the batch systems and storage elements. User interface clients, however, can be installed anywhere. There is a distribution based on apt (for RH73 only), and there is a menu-driven installer available as part of the VL-E Proof-of-Concept programme. Please look at the PoC Installer for details and a Readme. The PoC Installer works on many different RedHat Linux version (including RHEL3 and CentOS3).

Scheduling

• Torque RPMs with the "transient TMPDIR patch"
  The TMPDIR part means that a per-job scriatch directory is created and the environment variable TMPDIR is set to it, according to POSIX 1003.2d
• Maui RPMs and source
• Example configuration file for Maui (with fair-shares)

The current LCG software can put severe load on a pbs server due to the times it calls qstat (we have seen as many as 25 qstat calls per second). This load can eventually bring the entire system to an halt. In order to reduce the impact of this problem (a full solution requires rewriting part of the grid job manager), we wrote a set of utilities that wrap the original pbs commands and provide caching:

• PBS Caching Utilities

Misc stuff

• firewall.php