Topic
IC4NOTICE: developerWorks Community will be offline May 29-30, 2015 while we upgrade to the latest version of IBM Connections. For more information, read our upgrade FAQ.
6 replies Latest Post - ‏2013-01-09T18:04:50Z by SystemAdmin
janfrode
janfrode
20 Posts
ACCEPTED ANSWER

Pinned topic Upgrade path for old cluster (3.2/3.1)

‏2013-01-08T08:05:09Z |
I have an old 7 node GPFS cluster with a mix of RHEL4 running GPFS 3.1.0-25 and RHEL5 running 3.2.1-20. Could someone suggest what the best upgrade path would be to get this up to a supported level? My current plan is to re-install the RHEL4 servers with RHEL5-latest + GPFS 3.2.1-29 as the first step. Then look into how to upgrade from v3.2.1-29 to 3.3/3.4/3.5.. Does that sound sensible ?
Updated on 2013-01-09T18:04:50Z at 2013-01-09T18:04:50Z by SystemAdmin
  • vpaul
    vpaul
    79 Posts
    ACCEPTED ANSWER

    Re: Upgrade path for old cluster (3.2/3.1)

    ‏2013-01-08T18:10:16Z  in response to janfrode
    Hello,

    Please see the GPFS Concepts, Planning and Installation guide (http://publib.boulder.ibm.com/epubs/pdf/a7604137.pdf), Chapter 8, section "Migrating to GPFS 3.5 from GPFS 3.3, GPFS 3.2, GPFS 3.1, or GPFS 2.3". Also follow the instructions in Chapter 5 (page 59) and the GPFS FAQ to make sure the OS kernel is at the prerequisite level before installing any version of GPFS.

    Regards,
    The GPFS team
    • SystemAdmin
      SystemAdmin
      2092 Posts
      ACCEPTED ANSWER

      Re: Upgrade path for old cluster (3.2/3.1)

      ‏2013-01-08T19:07:44Z  in response to vpaul
      In other words, we recommend you upgrade your GPFS software to 3.5. A cluster of nodes running 3.5 software can mount a 3.1 filesystem, and will "promise" not to upgrade the disk info or formats unless and until you tell it to do so.
      • janfrode
        janfrode
        20 Posts
        ACCEPTED ANSWER

        Re: Upgrade path for old cluster (3.2/3.1)

        ‏2013-01-08T21:34:08Z  in response to SystemAdmin
        I'd prefer to minimize downtime and do a rolling upgrade, so I need to go to fully to v3.2.1-x first.
        The gpfs faqs doesn't seem to go back to v3.2.1 versions anymore. Is there a list of RHEL5 kernel levels tested with v3.2.1-29 ? Will the latest RHEL5-kernel (2.6.18-308.24.1.el5) be OK ?
        • dlmcnabb
          dlmcnabb
          1012 Posts
          ACCEPTED ANSWER

          Re: Upgrade path for old cluster (3.2/3.1)

          ‏2013-01-08T21:40:23Z  in response to janfrode
          You can only do rolling upgrade to the next version. So you would have to roll up from 3.2 to 3.3, then another roll up to 3.4, then another roll up to 3.5. And you have to have the 3.x.0.0 install media for each release level. At some point it is just simpler to stop GPFS everywhere and take the time to upgrade all nodes at once. That depends on your staffing and the size of the cluster to decide, however.
          • SystemAdmin
            SystemAdmin
            2092 Posts
            ACCEPTED ANSWER

            Re: Upgrade path for old cluster (3.2/3.1) to most recent release (3.5)

            ‏2013-01-09T15:51:10Z  in response to dlmcnabb
            How many GPFS nodes/machines do you have in your cluster? If you can take half of them offline - upgrade this first half to 3.5 - then find a few moments to take quickly take the remaining 3.1 nodes offline and bring up the first set of 3.5 nodes, etc, etc. you may be able to miniminze your downtime.

            You may want to practice with a small test cluster...
            • SystemAdmin
              SystemAdmin
              2092 Posts
              ACCEPTED ANSWER

              Re: Upgrade path for old cluster (3.2/3.1) to most recent release (3.5)

              ‏2013-01-09T18:04:50Z  in response to SystemAdmin
              (It has been pointed out to me by more experienced GPFS pros ...)

              I left out the additional steps that you'd have to mess about with redefining quorums, also I'm assuming you have multiple paths to each disk, etc, etc.
              So... Unless you really really need to minimize downtime AND you are really a GPFS expert AND you are willing and able to prepare and rehearse on a test system...

              Fahgeddaboutit ;-) Just tell your bosses and users that you must ...

              And promise us you will will never fall so far behind in your updates again! ;-)