How vSphere allocate resources during simultaneous deployement

[Flanfl]

Hello,

I'm working on a project where I have to gather some metrics on image creation for vSphere 5. I want to get the average time needed to create VMs in our infrastructure. I've got all what I want for one VM at a time. Now, I'm working on simultaneous deployement of VMs. When I'm deploying one 20Go VM (thick provisioning, I'll get metrics for thin provisioning later) it take 70 seconds. When I'm deploying tow VMs it take 140s, three VMs 210s and so on.

So I guess you saw the pattern, it's like if the resources used to deploy each VM are evenly distributed. This is rather unexpected (I don't think that infrastructure is strugling...). Therefore, it this the default behaviour? Is it because I'm deploying VMs on the same ESX server? (so the I\O is shared event if their is multiple disks on the host?) Something else?

I'm not sure if this can be a reason but I'm using VI Java SDK. (I've tried with PowerCLI and I got the same results).

Thanks!

[mcowger · Accepted solution]

If all the VMs are being deployed on the same host, then yes, it will be approx. evently disitributed.  This is ESX's way of ensuring that no VM (or VM operation) robs performance from the rest.

[Flanfl]

Thanks a lot!

Do you by any chance, know where I can find documentation which explains this behaviour? I looked at the available documentation for vSphere 4 and 5 and the SDK but I've found nothing so far.

[mcowger]

Resource allocation is part of the base of vSphere.

You might read the scheudling whitepaper.