With the release of vSphere 7, there are many feature that has been announced. However, there is one feature new in vSphere 7 that wasn’t heavily marketed. That is Clustered VMDK.
So what is Clustered VMDK? Basically it is a VMDK shared by clustering solution on a VMFS volume. In the past, this was not possible especially in cluster requirement such as Microsoft Cluster Service (MSCS) or Windows Server Failover Cluster (WSFC).
The reason that this is possible is because of the support for SCSI-3 Persistent Reservation for VMDK. With this, you can fully reduce the need of Raw Device Mapping (RDM).
Of course this is not just a vSphere 7 feature. It requires the below and many is similar to a RDM LUN:
- Your array must support ATS, SCSI-3 PR type Write Exclusive-All Registrant (WEAR).
- Only supported with arrays using Fibre Channel (FC) for connectivity.
- Only VMFS6 datastores.
- Storage devices must be claimed by NMP. ESXi does not support third-party plug-ins (MPPs) in clustered virtual disk configurations.
- VMDKs must be Eager Zeroed Thick (EZT) Provisioned.
- Clustered VMDKs must be attached to a virtual SCSI controller with bus sharing set to “physical.”
- A DRS anti-affinity rule is required to ensure VMs, nodes of a WSFC, run on separate hosts.
- Change/increase the WSFC Parameter “QuorumArbitrationTimeMax” to 60.
- Windows Server 2012R2/2016/2019. SQL Server 2016/2017 were used to validate the configuration. (supported only as it is tested)
- The boot disk (and all non-shared disks) should be attached to a separate virtual SCSI controller with bus sharing set to “none.”
- Mixing clustered and non-shared disks on a single virtual SCSI controller is not supported.
- The datastore cannot be expanded or span multiple extents.
- All hosts and vCenter must be vSphere 7 or above
- A mix of clustered VMDKs and other disk types (vVols, pRDMs) are not supported
- Support for up to 5 WSFC nodes (Same as pRDMs)
- 128 clustered VMDKs per host
- Only Cluster across Box (CaB) is supported, Cluster in a Box (CiB) is not supported.
Thanks to my colleague, Liang Bing who shared with me this white paper.