5

Setting up a new Hyper-v cluster. Got the following setup;

  • Network storage presenting iscsi
  • Hypervisor01
  • Hypervisor02
  • Physical DC

Want to set it up using windows failover clustering. However my question is I'm going to have 6 VMs running in the cluster and will be splitting loads evenly between the two. Am I right in thinking each VM will need it's own iscsi LUN on the network storage. If they're all on the same LUN they wont be able to move VMs around one at a time?

MissCoder87
  • 415
  • 2
  • 4
  • 13
  • 2
    You don't really need any physical domain controllers these days. Virtualize ;) https://technet.microsoft.com/en-us/library/virtual_active_directory_domain_controller_virtualization_hyperv(v=ws.10).aspx – BaronSamedi1958 Nov 16 '16 at 11:21

3 Answers3

7

Per Microsoft recommendations https://technet.microsoft.com/en-us/library/jj612868(v=ws.11).aspx you should consult your storage vendor that may recommend configuring each LUN with one partition and place one CSV volume on it.

Generally, I would recommend having iSCSI LUN and CSV numbers equal to hypervisor hosts. You can bound each cluster shared volume (iSCSI LUN) and virtual machine to one hypervisor host to be sure that if both cluster nodes are running all the resources are spread equally across the cluster and migrate only in case of failover.

Going a bit further Microsoft recommends having different CSVs for virtual machine’s system partition and data drives that may result in 4 CSVs in your case which is an overkill I guess.

Since each of iSCSI LUNs and cluster shared volumes is accessible by all hypervisor hosts (as shared storage) you can easily migrate your virtual machines between hosts and between CSVs if needed.

I would strongly recommend you to create one more virtual machine as a secondary DC. If your physical DC fails (or goes down for maintenance) there is a great probability of losing the whole cluster too. And don’t forget to configure failover cluster witness drive (or share). Your physical DC is a good option for this purpose however Microsoft recommends having witness clustered/cloud to be highly available too.

Net Runner
  • 5,626
  • 11
  • 29
3

IMHO granular LUN configuration is still the best way to properly distribute workload within an environment. This will really make the management easier, and I think it makes sense for any environment. But in you case, I would just distribute the VMs by type of workload or something.

batistuta09
  • 8,723
  • 9
  • 21
  • It is totally the wrong, sadly, once you do that in practice. It also goes against best practices. – TomTom Nov 17 '16 at 15:43
  • 3
    it may go against best practices, but as early @Net Runner pointed out, having 1 CSV per each VM would be an overkill for this small environment. Also that does not really matter if the system is designed properly and you selected right backup method. – batistuta09 Nov 18 '16 at 10:15
  • Who taks about 1 CSV "per machine"? I talk about ONE CSV. It is called Clustered SHARED volume. That ie ONE CSV, One ISCSI LUN, for all the VMs unless they have VERY speial requirements (and need a separate LUN to address separate discs). – TomTom Nov 18 '16 at 10:27
  • 3
    Keeping just a single iSCSI LUN for all production VMs is a quite risky move. It fails - you lose everything. 2 LUNs, 2 CSVs distributed across 2 Hyper-V hosts is an optimal way to set it up. – batistuta09 Nov 18 '16 at 11:02
  • Yeah. It sort of assmed you not being stupid and knowing what you do, I agree. – TomTom Nov 18 '16 at 11:05
  • while having multiple LUNs can have advantages for managing some aspects, having a single LUN has the advantage of the most efficient use of space, and free space is the most frequent issue I see on small deployments where multiple LUNs have been used... – shouldbeq931 Aug 08 '17 at 23:56
-2

Am I right in thinking each VM will need it's own iscsi LUN on the network storage

  • Yes. IF YOU SET IT UP THIS WAY.
  • No. If you do not set it up this way, but set up CSV (Cluster Shared Volume) on the ISCSI.

The later is the recommended way, btw., in most cases.

And the C means Cluster as in Cluster aware. It is QUITE smart. So, yes, you can move the VM's around.

TomTom
  • 50,857
  • 7
  • 52
  • 134