Forum Discussion

CesarJimGon's avatar
6 years ago

ZFS BACKUP SLOW THROUGHPUT

Hi.

I have a Netbackup 7.7.3 environment with a Solaris 11 SPARC Master Server which backups an Oracle ZFS storage, besides another applications.

The ZFS shares are mapped to the Master Server by NFS, so this shares are seen as FS. The information of these ZFS shares goes to LTO5 tapes. I use 2 LTO5 tape drives for this backup operation.

The communication from the ZFS to the Master Server goes through Infiniband.

Two months ago, the throughput of the backups have been too slow. Before that, I registered throughtputs of 180 Mb/sec. That was the average throughput. Now, the throughput is 50 Mb/sec max.

I haven't done any changes on the backup configuration.

Could you help me with some tips to improve the throughput?

  • Thanks Marianne.

    I checked the NFS configuration in the Master Server and compared it to another Master Server with the same ZFS backup solution. I found there were missing options:

    root@uxvtbck1 # more /vfstab
    /vfstab: No such file or directory
    root@uxvtbck1 # more vfstab
    #device         device          mount           FS      fsck    mount   mount
    #to mount       to fsck         point           type    pass    at boot options
    #
    /devices        -               /devices        devfs   -       no      -
    /proc           -               /proc           proc    -       no      -
    ctfs            -               /system/contract ctfs   -       no      -
    objfs           -               /system/object  objfs   -       no      -
    sharefs         -               /etc/dfs/sharetab       sharefs -       no      -
    fd              -               /dev/fd         fd      -       no      -
    swap            -               /tmp            tmpfs   -       yes     -

    /dev/zvol/dsk/rpool/swap        -               -               swap    -       no      -
    192.168.10.50:/export/respaldo1 -       /respaldo1      nfs     -       yes     rw,bg,hard,nointr,rsize=1048576,wsize=1048576,proto=tcp,vers=3,forcedirectio
    192.168.10.51:/export/respaldo2 -       /respaldo2      nfs     -       yes     rw,bg,hard,nointr,rsize=1048576,wsize=1048576,proto=tcp,vers=3,forcedirectio
    192.168.10.50:/export/respaldo3 -       /respaldo3      nfs     -       yes     rw,bg,hard,nointr,rsize=1048576,wsize=1048576,proto=tcp,vers=3,forcedirectio
    192.168.10.51:/export/respaldo4 -       /respaldo4      nfs     -       yes     rw,bg,hard,nointr,rsize=1048576,wsize=1048576,proto=tcp,vers=3,forcedirectio
    192.168.10.50:/export/respaldo5 -       /respaldo5      nfs     -       yes     rw,bg,hard,nointr,rsize=1048576,wsize=1048576,proto=tcp,vers=3,forcedirectio
    192.168.10.51:/export/respaldo6 -       /respaldo6      nfs     -       yes     rw,bg,hard,nointr,rsize=1048576,wsize=1048576,proto=tcp,vers=3,forcedirectio
    192.168.10.50:/export/respaldo7 -       /respaldo7      nfs     -       yes     rw,bg,hard,nointr,rsize=1048576,wsize=1048576,proto=tcp,vers=3,forcedirectio
    192.168.10.51:/export/respaldo8 -       /respaldo8      nfs     -       yes     rw,bg,hard,nointr,rsize=1048576,wsize=1048576,proto=tcp,vers=3,forcedirectio

    The missing options were proto=tcp,vers=3,forcedirectio

    These options were added to the /etc/vfstab file of the Master Server. Before that, the ZFS mount points were unmounted. After the configuration these mount points were mounted again.

    I'm running a full backup right now. The throughput is 138 Mb/sec. The backups run faster.

12 Replies

  • CesarJimGon

    Have you tried to test read speed outside of NBU?

    Something like :

    time tar cBf  /dev/null </mountpoint>
    Or instead of mountpoint, a subdirectory that is about 5 - 10GB.
    Get the size before you start in order to work out read speed. 

    • CesarJimGon's avatar
      CesarJimGon
      Level 4

      Hi Marianne.

      I'm testing your suggestion, but I don't think I'm doing it right.

      Checking size of test directory

      du -as /usr/openv/netbackup
      35739031        /usr/openv/netbackup

      Read test

      time tar cBf /usr/openv/netbackup
      tar: Missing filenames

      real    0m0.020s
      user    0m0.003s
      sys     0m0.008s

      • Marianne's avatar
        Marianne
        Level 6
        You missed /dev/null.
        You need to specify /dev/null as backup device, meaning you are simply reading the files and not really writing to anything.
        Please choose a folder on the NFS mount.
  • First you need to isolate if it is a read problem or write problem. In your scenario media sever and client is same so we can try NOSHM touch file. If file is present remove it, if not then create it.

    The location of the file is either <install_path>\veritas\netbackup\ for Windows, or /usr/openv/netbackup/ for UNIX.
    The touch file name is NOSHM with no extension.

    https://www.veritas.com/support/en_US/article.000026448

    There is another simple way to figure out the issue, configure null storage unit on media server and send backup to it. Though the process looks like too many steps, it is very easy. (1)Copy libstspinulldiskMT.so & libstspinulldisk.so (2) restart nbrmms service (3) Configure storage server, disk pool & storage unit (4) Take backup.

    https://www.veritas.com/support/en_US/article.000013708

    If above two suggestion do not help then its time to get into logs, bpbkar (from client) and bptm (from media server) logs. 

     

  • Hello Cesar,

    During the backup use below command to check the disk utilization. 

    iostat 5

    Also you can try hdparm command to check the disk performance in Linux:

     

    $sudo hdparm -Tt /dev/sda

    where sda is your disk where your backup is writing. Please refer the below two links and share the output of disk performance. 

     

    Reference:

    https://docs.oracle.com/cd/E23824_01/html/821-1451/spmonitor-4.html#scrolltoc

    https://askubuntu.com/questions/87035/how-to-check-hard-disk-performance

     

     

    • CesarJimGon's avatar
      CesarJimGon
      Level 4

      Thanks Marianne.

      I checked the NFS configuration in the Master Server and compared it to another Master Server with the same ZFS backup solution. I found there were missing options:

      root@uxvtbck1 # more /vfstab
      /vfstab: No such file or directory
      root@uxvtbck1 # more vfstab
      #device         device          mount           FS      fsck    mount   mount
      #to mount       to fsck         point           type    pass    at boot options
      #
      /devices        -               /devices        devfs   -       no      -
      /proc           -               /proc           proc    -       no      -
      ctfs            -               /system/contract ctfs   -       no      -
      objfs           -               /system/object  objfs   -       no      -
      sharefs         -               /etc/dfs/sharetab       sharefs -       no      -
      fd              -               /dev/fd         fd      -       no      -
      swap            -               /tmp            tmpfs   -       yes     -

      /dev/zvol/dsk/rpool/swap        -               -               swap    -       no      -
      192.168.10.50:/export/respaldo1 -       /respaldo1      nfs     -       yes     rw,bg,hard,nointr,rsize=1048576,wsize=1048576,proto=tcp,vers=3,forcedirectio
      192.168.10.51:/export/respaldo2 -       /respaldo2      nfs     -       yes     rw,bg,hard,nointr,rsize=1048576,wsize=1048576,proto=tcp,vers=3,forcedirectio
      192.168.10.50:/export/respaldo3 -       /respaldo3      nfs     -       yes     rw,bg,hard,nointr,rsize=1048576,wsize=1048576,proto=tcp,vers=3,forcedirectio
      192.168.10.51:/export/respaldo4 -       /respaldo4      nfs     -       yes     rw,bg,hard,nointr,rsize=1048576,wsize=1048576,proto=tcp,vers=3,forcedirectio
      192.168.10.50:/export/respaldo5 -       /respaldo5      nfs     -       yes     rw,bg,hard,nointr,rsize=1048576,wsize=1048576,proto=tcp,vers=3,forcedirectio
      192.168.10.51:/export/respaldo6 -       /respaldo6      nfs     -       yes     rw,bg,hard,nointr,rsize=1048576,wsize=1048576,proto=tcp,vers=3,forcedirectio
      192.168.10.50:/export/respaldo7 -       /respaldo7      nfs     -       yes     rw,bg,hard,nointr,rsize=1048576,wsize=1048576,proto=tcp,vers=3,forcedirectio
      192.168.10.51:/export/respaldo8 -       /respaldo8      nfs     -       yes     rw,bg,hard,nointr,rsize=1048576,wsize=1048576,proto=tcp,vers=3,forcedirectio

      The missing options were proto=tcp,vers=3,forcedirectio

      These options were added to the /etc/vfstab file of the Master Server. Before that, the ZFS mount points were unmounted. After the configuration these mount points were mounted again.

      I'm running a full backup right now. The throughput is 138 Mb/sec. The backups run faster.