MSCS : Cluster Design Considerations

I am busy with a design doc on a VMware Cluster to be used exclusively for Microsoft 2 node Clusters. MSCS does have it’s place when you need to have Application availability over and above VMware HA. VMware HA does not protect against OS failure (OS Corruption/Patching..ect) but by using MSCS you can ensure that Applications (Like SQL, Exchange..ect) is available when and OS fails.

However if you do want to run MSCS Clusters on VMware(or any other Hypervisor for that matter) there is some addition design considerations that you need to take and be aware of. Here is a list of some of the things that I took into account with my design for the client.

Supported OS and Hypervisor Version

  • ESXi 5.1 is supported with up to 5 nodes on MSCS
  • Windows 2008R2 is Supported (Windows 2012 is not yet supported)

Application Clustering support

  • List is here of supported Applications

Cluster HA/DRS Settings

  • Using Anti-Afinity rule keep the Nodes of a Cluster Pair on separate ESXi Hosts.
  • If you have Shared Disks vMotion is not supported. Thus ensure that you don’t place all the “active nodes” of the Clusters on one host. You need to ensure that you place the Nodes on the ESXi Hosts so that not all of the active Nodes is on one host. You don’t want all the Active Nodes on the same host.
  • DRS is in “Partially Automated” mode to ensure that no VM’s are V-motioned.
  • Disable HA for all Cluster Node VM’s. In the event of a host failure the Application will fail over by means of the MSCS. But if the VM starts up somewhere else you might (I say MIGHT) have a an issue where the Application is still failing over but the Failed over VM is starting up already. Might have issues with that.

Storage Settings

  • Do not use Round Robin Path Policy…not supported. Fixed or MRU only.
  • U can use VMDK disks on VMFS volumes. They must be Thick Eager Zero VMDK’s. I would suggest to create a folder on the VMFS datastore, give it a descriptive name and then created your shared VMDK inside this folder. If you created the shared VMDK’s  with one of the Node VM’s it create a folder with the same name of the Node VM…not ideal.
  • RDM is supported with FC only. Must be in Physical Mode if Cluster Nodes is on separate ESXi Hosts.

Backup Considerations

  • When using Shared VMDK or RDM you cannot make snapshots of the VM’s. So you need to use in guest MSCS cluster aware Backup Software.
I think one have to be sure that when you design for MSCS Clusters that is Virtualized you need to take a lot more into consideration. When all is taking in consideration I ask my self :
  • Is it worth all the additional configuration ?
  • Will HA also work ?
  • Can the Application not natively do HA ?
  • What is the Risk of Image Lost in the Media if we have a failure/down time ?
  • How much money will we loose in an event of Failure ?

Leave a Reply