Difference between revisions of "RAL Tier1 Batch Worker Nodes"
From GridPP Wiki
Martin bly (Talk | contribs) |
(No difference)
|
Latest revision as of 22:18, 14 January 2009
Introduction
The batch farm comprises a number of generations of hardware with different specifications (CPU, memory, disk). This document summarises these specifications, and gives information on how to specify that batch jobs run on nodes that provide some minimum requirement.
Worker Node Specifications
CPU Model | # Cores / node | KSI2K Rating | Total Memory (MB) | Memory /core (MB) | Pool (GB) | Pool /core (GB) | Comments |
---|---|---|---|---|---|---|---|
Intel(R) Xeon(R) CPU 5400 @ 2.83GHz | 8 | 1.994 | 16384 | 2048 | 388 | 48.5 | bigpool attribute |
Intel(R) Xeon(R) CPU 5410 @ 2.33GHz | 8 | 1.695 | 16384 | 2048 | 388 | 48.5 | bigpool attribute |
Intel(R) Xeon(R) CPU 5130 @ 2.00GHz | 4 | 2.148 | 3991.5 | 997.9 | 182 | 45.5 | bigpool attribute |
AMD Opteron(tm) 270 @ 2.0 GHz | 4 | 1.282 | 3895.3 | 973.8 | 181 | 45 | bigpool attribute |
Intel(R) Xeon(TM) @ 2.80GHz | 2 | 1.068 | 2007.2 | 1003.6 | 64 | 32 | bigpool attribute; some nodes have double the amount of memory and the bigmem attribute |
Large Memory Jobs
To submit jobs requiring a large amount of memory, just specify how much is needed:
$ qsub -l pmem=2000mb ...
and the job will be forwarded to the appropriate queue for execution on an appropriately resourced node.
Large Pool Jobs
For jobs requiring a large pool space, select the bigpool attribute:
$ qsub -l nodes=1:bigpool ...
and the job will run on a node that has at least 45GB of disk per CPU.