Stretched Cluster with Partially Shared Storage

[vmkw]

Hello all,

We're planning to introduce VPLEX to our virtual environment to benefit from cross-site high availability for critical VMs. I have three sites (A, B & C) with three different vCenters. Will be presenting VPLEX for sites A & B.

Which design do you suggest to follow:

1. create three clusters. One for Site A shared Storage. One for Site B shared Storage and a third one for Stretched (metro) Cluster?

2. Merge Site A & B clusters and have one cluster with partially shared storage? (only VPLEX datastores will be visible to all hosts, site datastores will remain visible only to site specific Hosts). See table below:

Cluster
Site A Hots		Site B Hosts
Site A Datastore	VPLEX Datastores		Site B Datastores

My main questions are related to scenario 2. What are the pros/cons?

• Would having Partially Shared Storage (used ONLY for VMs for specific storage profile) affect DRS/HA? and How?
• As far as I know, HA will only benefit from cross site hosts ONLY for VPLEX hosted VMs. Other HA VMs will work as if they are two separate cluster (due to non-shared storage and the fact HA is reactive). Please correct Me if I'm wrong
• would this setup make DRS (vMotion) benefit from both sites for load balancing and disaster avoidance (vMotion across L3 for VMs on Site specific Storages)? again, all I know is enhanced/xVmotion is a manual process and DRS will not use it for load balancing. If so, how would DRS balance the cluster?

Notes to consider:

1. We will be upgrading to ESXi 6.0 within 6 months to 1 year, So Cross-vCenter vMotion can be taken into consideration. But I would appreciate listing scenarios with/without v6.
2. VPLEX will only be used for highly critical VMs as it consume twice amount of storage space as regular ones. I have no plans to make all VMs on VPLEX (thus having just one cluster)
3. OTV/VXLAN is supported for sites A&B, but not for C

Appreciate your contributions.

[CoolRam]

DRS:

1.If the virtual machine is receiving the resources it requires then there is no need to move the virtual machine to another hosts.

  2. To avoid wasting resources, DRS calculates two metrics, the current host load standard deviation and the target host load standard deviation. based on the these matrix it will take the decision.    

In a same scenario you want the cluster that will protect against your critical VM even if the site is down. Yes your VM will be protected in terms of vmdk in the stretched metro cluster but it will not failover and poweron because they are not part of same Cluster.

About Q :

• Would having Partially Shared Storage (used ONLY for VMs for specific storage profile) affect DRS/HA? and How?

               You want the HA/DRS would across the different site than you have to use the SRM(Site recovery manager). but if you just want to protect the data in LUN located on different site than you use this LUN but it can't help you against the site failure.