SQL 2012 AlwaysOn with three node cluster non-shared storage


  • Hi Guru

    We are achitecturing Windows 2008R2 clustering with non-shared storage model. Here is our design:

    DC1(Primary)                                     DC2(DR)                                DC3


    Server A                                           Server C                         Fileshare Witness

    Server B

    • A to B is synchronous with automatic failover
    • A to C is asynchronous with manual failover
    • A,B and Fileshare Witness each have one vote but C has zero vote

    In the situation we lost DC1 entirely then we would need to follow steps from Joe Sack's white paper to bring server C for production use. The disagreement from Windows Engineers is we CAN NOT force quorum to bring cluster up and running because there would be a “Tied Vote” one for server C and another vote for fileshare witness.

    We need to have a majority of the votes in DR. In other words, we need to build one more node (OS only) in DR to have one vote to successfully bring SQL up and running.

    My question is do we really need the 4<sup>th</sup> node in DR adding to cluster?

    Thanks so much,


    Tuesday, January 08, 2013 6:40 PM


All replies

  • Hello,

    As you have only one node in second DC, after bringing the cluster online, you can change the quorum model to node majority. As there is only one in second DC, so obviously you cannot expect HA. For details please refer:

    Thanks, Sohail ~Please mark answered if your question is completely answered; vote it as helpful to help others find a solution quicker. This posting is provided AS IS with no warranties.

    Tuesday, January 08, 2013 10:24 PM
  • Hi KongDBA,

    Probably the biggest flaw in the current design (as it stands) is that the Fileshare Witness is really not giving you any extra HA beyond that which you would have got should you have ran the 3 node cluster (without removing the vote from Server C). Obviously if you lose the link between DC1 to DC2 then you lose HA in either model. Adding the extra node to DC2 and having full node votes using this design would over come this.

    In terms of being able to force the quorum in the event of connectivity lost between DC1 to DC2, yes this is possible.

    I would also like to draw your attention to Windows 2012 that introduces a new concept known as dynamic quorum, so you should seriously consider moving to that OS for any new cluster implementations for that and a multitude of other reasons.

    Mark Broadbent.

    Contact me through (twitter|blog|SQLCloud)

    Please click "Propose As Answer" if a post solves your problem
    or "Vote As Helpful" if a post has been useful to you
    Watch my sessions at the PASS Summit 2012

    Wednesday, January 09, 2013 9:01 AM