View 5.2 Compose linked clone pool expected to be distributed evenly across multiple hosts/datastores

[EricNichols]

Lets say we have 3 pools of linked clones and 3 hosts/datastores. When we create a small pool of say only 4 VM's we expect it to try and evenly distribute them across the 3 hosts. Instead it is putting all four VM's on one host. How do we remedy that?

Also, if we had 20 VM's in each of three pools, we would expect each pools VM's to be evenly distributed across the 3 hosts. Instead we are seeing the majority of one pool, the last to compose, have its VM's unequally placed mostly on one of the three hosts.

We need equal distribution (or as close as possible) on a per pool basis of VM's across hosts. How can we accomplish this? We arent looking for overall performance balancing, more like redundancy between data centers where each host is in a different physical location.

[Linjo]

I think it would be easier to setup 3 pools, one for each location.

View will not distribute the desktops the way you expect but is designed to be moved around with DRS.

Do you have shared storage?

// Linjo

[EricNichols]

Answering my own question:

Unless we upgrade to 5.3 and set the datastore overcommitment levels to unbounded we would need to reduce the number of concurrently composed VM's to the lowest denominator of the smallest pool we envision.If we want a pool of 6 linked clones spread across 3 hosts, we would need to set the concurrent number to 2 and even then, the composer might think it has a better idea.

http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=204749...

Can anyone confirm this?

[EricNichols]

The replica is on shared storage, the cache and disposable disk are on local spinning disks and the linked clone is on local fusionIO/SSD. Separate pools is not a solution for us. We have not enabled DRS, only HA. We have not found any documentation on enabling DRS with Horizion View Linked Clones Pools.

This document best describes our setup:

http://www.vmware.com/files/pdf/view/VMware-stateless-virtual-desktops-ref-arch.pdf

[mpryor]

At the time the linked clones are provisioned, they are making a datastore selection based on free space. Nothing is decided based on VMs per host, the expectation is that if you have a cluster then vSphere will handle DRS. When a VM is provisioned, its datastore is picked and it is created - if you have multiple provisioning operations happening at the same time, they may each pick the same datastore based soley on its current state as the other VMs don't exist yet. For larger environments this will even out.

Because you are using local storage, the datastore selection then dictates which VMs will run on which hosts. Reducing the concurrency will improve the spread for smaller pools across the datastores (and therefore hosts). Ensuring the VMs are all set up on shared storage of some form would allow vSphere to distribute VMs across the cluster.