The Grid ate my cluster: Tier-1 gets more CPU

Tue 6 Feb 2007

The Tier-1 service at RAL recently had its CPU capacity upgraded by 50% with the addition of 64 Intel dual core "Woodcrest" servers. Within 30 minutes, all the new CPU were busy with jobs from the Grid. The new hardware was put into service in batches of 16 starting at about 17:30 on the 10th January. All four batches were in production by 18:00, and the service had filled with additional work opportunistically flowing onto it from the Grid.

Martin Bly, Fabric Manager for the Tier-1 commented, "Physicists need the resource - it's good to see them making full use so quickly. It's just like plug and play on my laptop - only much bigger."

The process of integration of the new servers into the cluster went smoothly. Delivery was on 14 November and by 3 January a 28 day acceptance test had completed successfully. Once the servers are added to the cluster, the additional resource is automatically published by the Grid information system. Middleware, running on resource brokers around the world, matches the increased availability with demand, starting additional jobs until once again the service is full.

Jobs running at the Tier-1 from each of the VOs on 10 January are shown in the chart below. The rise in jobs can be seen at around 18:00.

The upgrade added an additional 256 job slots to the existing 1000, increasing the overall capacity from 1035K to 1578K SpecIn2000. The servers are based on two Intel dual core "Woodcrest" 5130 CPUs in a Supermicro chassis with X7DUL-E motherboard. They have 4GB of memory and dual 1Gb network cards.

Jeremy Coles, the GridPP production manager, told us, "We're very pleased with the way the CPUs filled up with work very quickly once they were connected to the Grid. It shows how well the information services and the resources brokers now work, and that an efficient, well-used Grid is starting to take shape."


© Copyright GridPP
If you wish to reproduce this piece please credit GridPP and contact Neasan O'Neill to say you are using it