locked
shared cluster volume - failure RRS feed

  • Question

  • I am trying to figure out how I can make a clustered shared volume redundant. I understand that if I have a CSV and one cluster node goes down other nodes will still be able to access the shared cluster volume. But what if the CSV itself goes down? How do I get some redundancy for it? 
    Thursday, April 8, 2010 6:58 AM

Answers

  • Hi,

     

    Good question!

     

    Cluster Shared Volumes(CSV) is a new feature available in Windows Server 2008 R2. With Cluster Shared Volumes, multiple clustered virtual machines can use the same LUN (disk) while still being able to fail over (or move from node to node) independently of one another.

     

    In a way, CSV is a mount point for the LUN. So what you want to do is actually to ensure your shared storage device(maybe SAN or something like it) keep living. I am not a professional of SAN, but I can give a little example.

     

    Some guy can’t afford(or they don’t want) to buy a SAN in a test environment, so they can install some iSCSI target applications such as Microsoft iSCSI Target or StarWind on a normal computer with Windows Server system. If this computer only have one hard disk, you may lost everything if you get a disk failure, in this situation, the cluster nodes will die.  Under a permittable condition, we can configure a RAID 5 to ensure the safety on the disk level.

     

    Cluster Shared Volumes Support for Hyper-V

    http://technet.microsoft.com/en-us/library/dd630633(WS.10).aspx

     

     

    Best Regards,

    Vincent Hu

     

    • Marked as answer by Vincent Hu Monday, April 12, 2010 8:11 AM
    Friday, April 9, 2010 9:36 AM

All replies

  • Hi,

    You should do backup on each VM. DPM 2010 RC support backup for CSV.

    You seldom will have CSV down as CSV is tied to SAN storage. If you want, replicate to another SAN storage as an alternative.


    laiys
    Thursday, April 8, 2010 9:13 AM
  • Maybe its not really direct answer to your question, but still:

    I would use SAN software  (software is cheaper, and such reliable as hardware in case of reliable vendors, such as StarWind or Lefthand ), that supports High Availability. That will definitely provide redundant failover to your system.

     

    Thursday, April 8, 2010 3:03 PM
  • I appreciate all your answers. 

     

    I would like to do a little more digging if you don't mind.

    DPM: 

    1) With DPM would I be 'live continuous' copying the cluster shared volume from SAN A to SAN B.

    2) What automation is there if the SAN that the source cluster shared volume is on dies? 

     

    SAN software:

    Note: I am familiar with software SAN vendors. I am using Open-E.

    1) Isn't SAN 'mirroring' slower than regular software or hardware mirroring that takes place inside a server e.g. MS software mirroring, RAID controllers

     

    2) MS clustering seems all about VM failover e.g. VM in computer A dies, VM in computer B kicks in. It doesn't address SAN failing e.g. it's all based on an ISCI single disk model - if that disk goes down, MS clustering doesn't seem to do anything. That part seems to be left up to the SAN vendor. Is that correct?

     

    Or is there any way I can 'mirror' ISCSI drives in Windows just using Windows software that will look and feel and perform like the regular Windows software mirroring that is available for internal drives.

     

    Is there a way to 'mirror' two VHD's in different SAN's/computers with MS product?

     

    I have always had mirroring in my systems and it seems all I'm buying if I do SAN mirroring is that the two drives are on two different devices but with that comes a lot of SAN baggage and at the end of the day is there any dual SAN system that will mirror as fast as internal mirroring and if there is what is the cost of it?

     

    Your response is appreciated in advance.

     

     

     

    Thursday, April 8, 2010 8:15 PM
  • Hi,

     

    Good question!

     

    Cluster Shared Volumes(CSV) is a new feature available in Windows Server 2008 R2. With Cluster Shared Volumes, multiple clustered virtual machines can use the same LUN (disk) while still being able to fail over (or move from node to node) independently of one another.

     

    In a way, CSV is a mount point for the LUN. So what you want to do is actually to ensure your shared storage device(maybe SAN or something like it) keep living. I am not a professional of SAN, but I can give a little example.

     

    Some guy can’t afford(or they don’t want) to buy a SAN in a test environment, so they can install some iSCSI target applications such as Microsoft iSCSI Target or StarWind on a normal computer with Windows Server system. If this computer only have one hard disk, you may lost everything if you get a disk failure, in this situation, the cluster nodes will die.  Under a permittable condition, we can configure a RAID 5 to ensure the safety on the disk level.

     

    Cluster Shared Volumes Support for Hyper-V

    http://technet.microsoft.com/en-us/library/dd630633(WS.10).aspx

     

     

    Best Regards,

    Vincent Hu

     

    • Marked as answer by Vincent Hu Monday, April 12, 2010 8:11 AM
    Friday, April 9, 2010 9:36 AM
  • I would love to have a clear answer on this myself. Speaking from experience VHD's can become corrupted and if you have  no redundancy then two expensive servers full of virtual machine configs attached to a failed SAN you are hosed. Restoring an entire failed array or even 100 gb of corrupt VHD is not a highly avalable solution. I think in additon to backed up VHD content you need a way to fail over to a mirrored version of your SAN. Does that exist? I just watched two hors of Technet presentation that essentially answer no.....

     

    Friday, June 25, 2010 5:14 AM
  • I've asked this same question and the answer I got is to buy Symantec Veritas Storage Foundation product for Windows. This way it could be possible to mirror the complete CSV.

    Another way I've found to mirror a complete CSV is to mirror all the VM individually. Let me explain it:

    When you make a CSV with 2 disk from 2 different located SANs you have a directory (C:\ClusteredStorage) wich has 2 directories (Volume1 and Volume2) each one belonging to each disk you gave to the CSV. Then, you could give each VM 2 disks that would be stored in each volume and then mirror it (since Windows 2000 it's possible to mirror and other RAID modes (using 2k3 disk files) and most of the *nix machines can also mirror). With this, you'll have an exact copy of all machines in both SANs.

    The problem is that if one SAN fails, all the machines would stop working and you should manually (or automatically with an script that is what I've done) change the .XML configuration of all VMs to take the failed disk with a new blank from the working SAN (if not, the VMs won't boot by Hyper-V error to find one of the VHDs).

    Of course this is not an automated failover for the SAN, but think that if you have one SAN and it fails because a water flood or burned in flames, with this method you'll have an exact copy to the last second, whereas using only 1 SAN it would be a complete disaster.

    With scripting is very easy to get it working in a reasonable way and time when it fails and you don't need to buy additional Symantec products.

    Cheers,

    Akuma.

    • Proposed as answer by HAkuma Saturday, December 18, 2010 7:01 PM
    Saturday, December 18, 2010 7:01 PM
  • HAkuma - would you mind sharing this script plz?

     

     


    • Edited by CSing Friday, October 21, 2011 4:46 PM
    Friday, October 21, 2011 4:45 PM