locked
Mirror Very Slow After Failover - Pls Assist RRS feed

  • Question

  • I'm testing database mirroring as a high availability solution for our application (no witness) and so far have been very pleased.  The xfer time from principle to mirror has been great.  Failover and failback work without issue

    The problem I'm having occurs with the original mirror once we fail over and it becomes the new principle.  When I run the test script (1000 insert statements, total data for each statement < 200 bytes) the new principle takes 5 times as long (50 secs) to process the data as it did on the original principle (10) seconds.  When I try the inserts on another test database in the same instance (still new principle) they only take 1 second, so I know that the problem is in network latency between the new principle and the former principle.

    How can I diagnose this to get a better idea of what's going on.  I have the endpoints configured using certificate authentication using the procedure documented in the msdn online help.  http://msdn.microsoft.com/en-us/library/ms191477.aspx

    Both machines are 64bit Windows Server 2008, 64bit SQL Server 2008 Standard

    Thx very much,
    Kathy

    Could someone pls assist me with getting started on this issue?  We'd like to roll out to production once we get this reconciled.

    Thx again,
    Kathy
    Tuesday, July 14, 2009 5:36 PM

All replies

  • Did you check mirroring perf counters (they have their own page in BOL)? Standard Edition can run only in high safety mode, that is, basically a two-phase-commit happens for every transaction. It might happen that your mirror server is too busy with other tasks and applies the log slower.
    -- Erik -- http://blog.rollback.hu
    • Proposed as answer by Ajay.G Thursday, November 18, 2010 10:06 AM
    Friday, July 17, 2009 12:18 AM
  • Agreed to what eric said. It also happens because of log size as well as the synchronous commit of the transactions across principle and mirror. You may have to start thining of options to have high availability mode rather than high performance mode for faster failover.
    Phani
    Friday, August 7, 2009 6:38 PM