Difference between revisions of "High CE load"

Latest revision as of 11:32, 25 October 2006

Lancaster's CE used to suffer from a high load which caused various operational problems, most obviously a dropout from the information system. Here is a list of tweaks which can be made to alleviate the CE load.

• migrate any NFS service to another machine (we moved the exp_soft area which solved our problem)
• migrate the site GIIS to your MON box (this is now the default in glite 3 which avoids info system dropouts)
• install name service caching daemon (package nscd)
• consider using torque 2.1 and maui 3.2 packages (Torque_and_Maui)
• consider torque recommendations for large clusters (http://www.clusterresources.com/torquedocs21/a.flargeclusters.shtml)
• consider modified lcg-info-generic (http://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind0604&L=tb-support&T=0&P=6842)
• renice the gatekeeper and in doing so the problematic job-manager scripts

There was no magic bullet to decrease the load, removing the NFS server was the most effective change.