Multiple ESX Clusters

[mittim12]

I just wanted to get an idea from the community how people handle different types of VM's in their environments. Do you place SQL VM's in their own cluster, leave Exchange in it's own cluster, or just lump everything into one big cluster?

It's always great to hear what other people do in their environment so thanks in advance for any info.

[Troy_Clavell]

For us, we leave all VM's regardless of the app in the same cluster. Now, we usually have a cluster of 8 Hosts, so if we go beyond 8 hosts, we'll create a new cluster. But for me, I feel as though DRS is there to do the job of load balancing your cluster, and setting a 3 star recommendation is very sufficient. No need to over complicated administration by creating resource pools, setting limits, etc.

just my two cents.

[Rob_Bohmann1]

To echo Troy, I think you want to have a heterogeneous group of workloads to be able to more fully use all of the 4 resources. Having a group of vms that all maximize one or two types of resource (i/o or mem) means you hit that bottleneck earlier and the other resources are not being utilized like they could be.

Now the example you list sounds more about making some organizational decisions to make mgmt easier. You can use resource pools and folders potentially for those needs, I do not think it is best to use clusters for what you've described.

Another question not stated is do you mix dev/test-qa/prod in the same clusters or do you separate them that way. I've seen this done both ways and there are plusses and minuses with each. For example, when we had some problems with capacity and a crash, it sure was convenient to be able to shut down some dev vms in order to make sure prod was not impacted. So this depends alot on how your networks are setup and a lot of the policies in place in your org.

[bulletprooffool]

We have multiple clusters split as follows:

Remote office have small 2 host clusters hosting only a handful of VMs each (8 - 15 VMs)

Main office has Production Cluster and a UAT/Dev cluster (4 - 6 hosts per cluster)

DR Site has Production Cluster

Traffic is split by VLans, server types by resource pools

this makes it easier to script failovers etc

[mnasir]

I define my clusters on their role - All servers and the enterprise apps are in one cluster - all desktops related vms/linked clones are in a seperate cluster.

[mittim12]

Thanks everyone this is great information. I have heard people before say that they tried to keep large multiprocessor VM's away from their general VM Clusters. This was awhile back and may have been before the co scheduling was redone for multi processor VM's.

[PGRSteve]

We have multiple clusters set up for each differnet environment in multiple data centers. This is a better way to mannage the large vm environment

1. General production population

2. General Test/ Dev/ Qa population

3. SQL Production

4. SQL Test/ Dev/ Qa

5. e-commerce prod

6. e-commerce test/dev/qa

7. Hosted Desktops production

8. PCI Prod

9. etc...

If you feel this is helpful please award points.