When I was sitting my VCP4 training our trainner told us that a good practice would be to keep VMs of the same vCPU count in the same cluster.
EG. 1-2 vCPU VM in a Cluster
4 vCPU VM in a Cluster
8 vCPU VM in a Cluster
The thinking behind this was to give each VM a fair chance to get on the phiscal cores because there must be the same number of cores free as the VMs vCPU count before the VM will be schduled on. Based in this advise we built our clusters in this manner because it makes perfect sense.
While on a call today with VMware discussing vCD I made the statment that this is how we have built our clusters....... I was a little supprised to hear that this was the first time they had heard of this type of configuration.
I am interested if anyone has thoughts on this.....is anyone concidering building clusters based on vCPU? Or is one big mix vCPU count cluster the way to go?
Cheers folks
In my opinion it sounds like a bad decision.
In general the servers with more vCPUs are the most important ones. If you create 3 small clusters, instead of 1 large cluster you may find the wait times on your most important servers to be too high. Trust the scheduler.
I find with vSphere/vCenter it is best to just keep it simple. Don't create more virtual datacentres or virtual clusters then you need.
Was this suggested back when 3.x was current? vSphere has a bunch of improvements.
http://www.vmware.com/files/pdf/perf-vsphere-cpu_scheduler.pdf
vSphere4 Enterprise License provides upto 4 way virtual SMP
vSphere4 Enterprise PLUS License provides upto 8 way virtual SMP
The "good practice" as it is said, is rightly said by your trainer, but it holds true mainly for an ESX\ESXi host in a HA cluster enabled with Host Failures Cluster Tolerates policy.
What this policy does is, that it creates "Slot" sizes for each vm based on the configuration of the VMs.
Hence, if the slot size is highly variable among the vms in a single host, you would face issues powering on new vms even though you have the resources.
Having said that, there is no need to maintain a common vcpu count across the cluster but maintaining a common vcpu count for an ESX\ESXi host in a HA enabled cluster with Host Failures Cluster Tolerates policy is a GOOD PRACTICE.
Hope this helps.
Regards,
Kushal Sukhija
VCP (vSphere 4.1)
it should be a mix so that there would not be an overuse of vCPU. and let DRS does the automatic loan balancing. you do not want to have a cluster over CPU subscribe and the other left with very idle CPU.