What is GPFS

IBM General Parallel File System (GPFS) is a high performance parallel filesystem  featuring storage virtualization, high availability and is designed to manage large amounts of file data,  You can find out more about GPFS in this introduction.

 

  • Checking your GPFS quotas on the Atlas cluster

    By default all users of the atlas gpfs space get 100 GB in the u directory and 2 TB in the g dir.

    On hosts running native gpfs like the rhel6-64 cluster you can issue the following 2 commands to see your quota and space used:

    df -h /gpfs/slac/atlas/fs1/d/$USER
    df -h /gpfs/slac/atlas/fs1/u/$USER

    On other hosts that dont run the gpfs code, but do have nfs access, you can issue:

    df -h /nfs/slac/atlas/fs1/d/$USER
    df -h /nfs/slac/atlas/fs1/u/$USER

 

  • GPFS building block

          Below is a schematic of a typical SCS GPFS storage building block.  It includes two file servers, two storage servers and two storage arrays.   The two sets of servers operate as ACTIVE/ACTIVE, but also provide failover capability if needed. This example would provide 320 TB of space.  Local  iozone tests show max write ~4GB/sec, max read ~6GB/sec, using large block sequential I/O.

  • No labels