Primary Storage

  • 1.  Spectrum Virt. Extent migration

    Posted Mon December 07, 2020 03:40 AM
    Hello all,

    When I remove mdisks from a regular pool, the system will start to migrate extents to the remaining mdisks in the same pool.
    How fast will this go?
    Is there a rule of thumb to calculate the duration of this extent migration, like 4 extents in parallel?

    Thanks.

    ------------------------------
    TMasteen
    ------------------------------

    #StorageManagementandReporting
    #PrimaryStorage
    #Storage


  • 2.  RE: Spectrum Virt. Extent migration

    Posted Thu December 10, 2020 06:04 AM
    Hi,
    To my knowledge, there is no specific rule of thumb. The speed of the migration will depend on the technology of the disks you read from and the disks you write to and on the workload on the system.
    From my experience (not too recent though): a customer removed a 7TB Mdisk from a 55TB pool (SSD drive arrays) and it took around 8.5 hours, on a V9000, so around 800GB/hour (~2Gbps), pretty slow... I think it also depends on the number of Mdisks you write to.

    ------------------------------
    Erwann Auffret
    ------------------------------



  • 3.  RE: Spectrum Virt. Extent migration

    Posted Fri December 11, 2020 11:34 AM
    Hello T Mastee,

    I think for that case there is no special rule. But this means not that there is no limit. Such kind of "migrations" are "priority very low"-tasks and first of all they depend on the overall workload on the pool. As far as I know there are no options to tune the settings of such a migration. For example at migratevdisk you can at least decide to migrate one or up tu 4 extents at a time (threads). Maybe the backend syncrate can help if yo change it for on volumes on a pool, so their extents could be faster migrated. But as I said the priority is very low for that kind of workloads and the host I/O's has one of the highest priorities.

    If the mdisk is not to big, just start it in the background and after some time it will be finished.To get it faster done do it during a low host workload time. And to prioritize the migration of the volume-extents try to change the syncrate for the volumes. I sometimes tried different things and you can see the result immediatily using the performance-tab in the web GUI.

    A last one, I'm not sure but there must be also a "ls*progress" command and some commands also show the expected finishing time.

    Regards,

    Dorde

    ------------------------------
    Dorde Knezevic
    ------------------------------