More tasks being allocated to a compute node than the number of cores available RRS feed

  • Question

  • I have a small cluster (2 * 8 Core Compute Nodes running on HPC 2008 R2 Edition). Initially, the compute nodes were added to the cluster with hyperthreading enabled so in the cluster manager they reported as 16 (not 8) cores each. The nodes were taken offline and manaully restarted, with the bios settings being changed to disable hyperthreading. The cluster (HN and CNs) were rebooted and the Cluster manager then correctly reported that there were 8 cores available on each CN. However, when a job was submitted (set to use all available cores with the default job template), it allocated 16 processes to the first compute node and none to the second. On trying to submit a job with 20 tasks it allocated 16 to the first compute node and 4 to the second, so it seemed to thing the CNs still had 16 available cores (or behaved that way). Do I need to remove the Compute Nodes from the cluster and re-add them. If so how best to I removed them?
    Tuesday, January 25, 2011 7:19 PM