VCAP-DCA Mock Lab Q6
November 07, 2010 
Sean Crookston 
Question
You have been asked to virtualize and existing infrastructure that consists of 30 production servers. 6 of these server happen to be servers that require Microsoft Clustering. You have been asked to keep these clusters intact while also providing HA and DRS capability where possible to all servers. All servers must be virtualized as part of this transition.
Describe how you would lay these servers out into a VMware cluster and your methodology for using MSCS, including where your virtual machines will be stored.
Tags: 
For the non-MSCS virtual machines, follow standard best practices for build and placement. For the MSCS based servers (assuming they are clustered pairs as vSphere only supports two node clusters) I would place each VM in the cluster on a different ESX host (Cluster Across Boxes) and configure a Physical Mode RDM (FC Only) for the storage of the data. The VM’s will need to be hardware version 7, and disable vMotion for these systems as it is not supported.
Adding to the above post, you set the machine to not participate in DRS by setting them to manual or disabled. setting them to manual will allow HA to place them on different hosts when starting up using the DRS algorithms, disabled will never move it from the host its sitting on.
Also set up affinity rules to keep MSCS nodes apart.
vMotion is not suported for Nodes in a MSCS cluster
The boot disks should be on local storage, not shared
1. Not necessary to use local storage – as you stick the cluster to fixed hosts you CAN use local storage if there is any.
2. Virtual RDM for W2k3 Clusters is sometimes the BETTER choice, its supported and runs smoother.
3. Again more a W2k3 thingy – present the LUNs with the Cluster Ressources carefully on higher LUN numbers only to these hosts hosting the MSCS nodes. The device reservation cause sometimes deep trouble on all hosts which are not holding the active resource.
W2k8 Clusters are NOT using any device reservations and are quite painless.
4. Again W2k3 pain – You will have a “one legged window” during conversation – because you need to disable clusdisk.sys until Converter has captured the “non-cluster-ressource” partitions. Consider coldcloning the passive node.
5. W2k8 – PAIN!! .. be aware of the status of the LUNs before adding them as RDMs to the virtual machines.
There is a minor chance that ESX catched them with RO Status and represent this status when you show them first to the W2k8 machine. The volumes will be marked as RO and hidden – its not fun to repair this all over the Windows registry afterwards. Check VPD pages before adding them – rescan the HBA in doubt.