2

We plan to build a failover storage with Windows2012r2. There will be 2 or 3 physical machines with 2 virtual machines on each server. One virtual machine building a failover, second one being storage machine (storage + failover machine on each physical server).

The network layout looks like this: enter image description here Click for bigger image.

On storage machines I plan to create storage pools (with tiering and mirrored) and iSCSI virtual disks that will be provisioned to failover machines (one disk on each storage machine). I have created such cluster in my laboratory with vmware (failover cluster working), but I am posting here in hope to get some advice on following considerations:

  • If network looks like above, with Dell Force managing all the traffic and 6 NICS on each physical server node (2x10 Gb and 4 x 1Gb) what would be the best way to utulize them? Or even is that setup a reasonable one (I know there will be SPOFs)?

  • How would you set up the storage pools and failover later, to have data mirrored on each storage node, and how would it affect perfromance? What would be the most optimal setup for such setting (not to loose too much usable space)?

  • If I have 64 gigs of ram on each physical node, which virtual machine should have more ram. The one with storage or the failover ones to achieve good performacne?

I'll appreciate all insight from people who've done similar things.

-->

(please move the comments or answers, thanks)

You can indeed cluster Microsoft iSCSI on top of an existing shared storage.

http://technet.microsoft.com/en-us/library/gg232632(v=ws.10).aspx

http://techontip.wordpress.com/2011/05/03/microsoft-iscsi-target-cluster-building-walkthrough/

What you do you make a config which will die with every node of cluster died. Because target images are not synchronized between each other.

...or draw a correct interconnection diagram :)

<--

To answer above:

Actually I've managed to create Scale Out File Server (clustered) with those 3 virtual ISCSI disks connected to cluster server. I've created mirrored virtual disk on the pool (formed from the virtual iSCSI) and added that to cluster. Then created SOFS. Then I switched the server with one of the virtual iSCSI's off (plugged the power off)- the cluster and the share on it was working. The cluster reported one disk missing but the cluster was not offline. Then I repowered the storage and it reconnected to cluster.

-->

I see what you do now finally! OK, the problem is Microsoft does not support anything except SAS in production within a Clustered Storage Spaces for a reason. See:

http://blogs.msdn.com/b/clustering/archive/2012/06/02/10314262.aspx

"The clustered storage pool MUST be comprised of Serial Attached SCSI (SAS) connected physical disks. Layering any form of storage subsystem, whether an internal RAID card or an external RAID box, regardless of being directly connected or connected via a storage fabric, is not supported."

Aidan had did it ages ago and we did it as well. See:

http://www.aidanfinn.com/?p=15145

What you do is not going to work reliably. Unfortunately. It's a pity Microsoft had removed non-SAS filter with R2 and did forget to provide warning.

<--

4
  • 1
    Config looks weird to me. You show a lot and show nothing. Few remarks: 1) Windows has no fault tolerant iSCSI (you can cluster MS target but that would require some external shared storage and a Windows cluster - makes no sense). How are you going to workaround this? Clustered Storage Spaces with iSCSI cannot work (doable with R2 but not working). So... ? 2) Your switch is alone. That's a SPOF. What you going to do with a fryed PSU on it? Need a second one and second set of cabling for redundancy. That's for the beginning... Please draw a complete interconnect diagram. Dec 5, 2013 at 0:08
  • this is Dell Force 10 it has 2 + 2 PSU. I've set up Failover Storage Cluster with this iSCSI method and it was ok (in lab setting), yet you write it cannot work, can you explain what won't work?
    – piotrektt
    Dec 5, 2013 at 10:17
  • About iSCSI - I've done it this way. On two machines I've created storage pools, on that pools iSCSCI virtual disks - and those disks are connected to each failover cluster node. You say this setup won't work? When I test the cluster storage I've got all green, no errors.
    – piotrektt
    Dec 5, 2013 at 11:43
  • What happens with iSCSI virtual disk when the node hosting it goes down? Or have you done RAID1 on top of iSCSI volumes? Well, what will happen after short node outage, re-syncing many-many TB in degraded RAID1? If you read about internals of DRBD you can be surprised how much should be done for replicated storage to be working well.
    – Veniamin
    Dec 5, 2013 at 13:26

0

You must log in to answer this question.