Split Vm between different hosts on same cluster

[sappomannoz]

Hello Community,

I have two different kind of servers with GPU in my farm, some optimised for core numbers and density and some for core frequency and single three performance. I 'm also using vSAN as storage. From a storage point of view it makes sense for me to put all nodes in a single vsan cluster, in this way I am more flexible in allocating disk space, but how can tell Horizon on which kind of server a VM should run? Maybe using fault domains? But how?

[BenFB]

Are you referring to different GPU profiles (e.g. memory size) on the hosts in the same cluster? My understanding is that HA/DRS is smart enough to only place VMs on hosts that first off have a GPU and secondly are configured for the profile you need.

[sappomannoz]

Hi BenFB​,

unfortunately I will have Tesla P40 on all nodes, so I won't be able to split the VMs based on the vGPU profiles.

I need to put general purpose VMs on one set of hosts and virtual workstations on another set of hosts.

I can't use DRS VM/HOST rules since the VMs are created dynamically.

[BenFB]

Why not just let DRS balance the cluster? It's OK to mix GPU and non-GPU enabled VDI on the same host.

[sappomannoz]

Also general purpose VMs have GPUs, with Windows 10 it's almost mandatory. The hosts for workstation have a higher CPU frequency and less cores for single threaded applications. But I could use you suggestion and replace the 2 P40 with 6 P4 for the virtual workstations hosts.

[BenFB]

Do you have a mixed cluster with two flavors of hosts that have different CPU models?

We've had mixed results with Windows 10 and GPU. Everyone tells us that we need them but so far we've deployed 1,000 Windows 10 linked clones without GPU and haven't identified any issues. We are planning to add P4 soon for a specific application that needs GPU and will do some additional testing on the user experience.

[sappomannoz]

Right now I have a single cluster with identical machines that I'm using as virtual lab for our students , we have replaced all PC with thin clients notebooks.

I had to use GPUs for things like video playback and simple 3D apps that we are using.

Now I'm planing to offer a Workstations as a Service for our researcher. Here comes the need for a different configuration. I would like to have all hosts in the same VSAN Cluster to better manage the storage and host maintenances. The problem is the I can't span VSAN across multiple  DRS clusters, it has to be one, so I need a way to logically distribute different kind of VMs onto the different kind of hosts. P4 vs p40 could be a possibility as you suggested.

[BenFB]

The only other option I can think of is that you let them provision to any host in the cluster. Then periodically you either manually or schedule a script to run automatically that adds the new machines to DRS groups that pin them to the correct hosts.