Hi, Norvin, In our VCL environment, we have two different VMware infrastructures (at different campuses) -- so two management nodes. One infrastructure uses plain ESX hosts with the standard license, probably quite similar to what you have. The other infrastructure pools all of the physical hosts into a single vCenter cluster, so for this, the VCL sees only one VM host. These hosts use the more expensive enterprise license and rely on a shared backend SAN for storage.
What this allows you to do (provided that each physical host uses the same datastores on a SAN) is use something VMware calls Dynamic Resource Scheduling (i.e. vMotion), which is some virtualization magic that allows the VMs to actually float across different hosts in order to balance resource loads -- for instance, if host1 starts using a lot of CPU but host2 is not, the hypervisor will move some of the VMs from host1 to host2, but no one actually notices -- certainly the VCL and any user of the VMs will neither know nor care about the change. The move takes a few milliseconds, and it buffers all network traffic while this takes place -- so not ideal for certain scenarios where consistently low latency is vital, but it is really excellent for a VCL environment. Practically, what this allows me to do is oversubscribe the vCenter cluster, which is something that I don't do (as much) with the plain ESX hosts. By oversubscribe, I mean that I can put more VMs in the host cluster than "should" fit -- that is, say I have 100 VMs, each of which uses 4GB of RAM. Ordinarily, I might think that the VM host should have ~425 GB of RAM (assuming a 250MB VMware overhead for each VM). Well, with a vCenter cluster, I may really only need 300 - 350 GB of physical RAM to run the same set of VMs (oversubscribing by 15-30%). I currently oversubscribe RAM by 20%, and I could easily be more aggressive. This assumes that, in practice, not every VM is making full use of its CPU and RAM, which generally is a safe assumption. The main requirement (aside from the license from VMware) is a shared SAN for the backend datastores. For this, if the option is available, you will want to configure the "VM Working Directory Path" to use the fastest possible part of your SAN (SSD is ideal, though that's not what we use and still the I/O latency stays sufficiently low, even when a lot of machines are booting). This will give your running VMs a big boost in performance. The "Repository Path" and "Virtual Disk Path" don't need such a fast backend. Let me know if you have any other questions. Regards, Aaron Coburn On Oct 17, 2013, at 5:27 AM, "Basilio, Norvin" <[email protected]<mailto:[email protected]>> wrote: Hello All, I’m trying to determine if vCenter is a worthy purchase for our VCL environment. I know that is a very broad statement so I will try to narrow it down. Our VCL environment currently consist of 15 HP blades running esxi 5.1 with a standard license. I was curious if the host profiles could be configured to place some machines in a resource pool that is available on a cluster of esxi host. I am hoping to use the resource pools to try and limit certain images from monopolizing the resources on the esxi host. I’ve been looking at the host profiles though and it looks like I can only assign a resource pool to a single host or cluster. Can anyone share how they have integrated vCenter in their VCL environment. Norvin
