Our VCL deployment uses a set of standalone VMware ESXi vhosts, each
with its on LUN on a shared common SAN.  Due to license restrictions we
aren't able to use VMware's copy-on-write technology (Linked Clones), so
every time we load an OS image to a VM, it's a full copy.

This has the impact you'd expect.  The worst case we've been able to
generate so far involves a set of VM's loaded with a Linux image, all
reloading with our 60G Windows image.  Up to about eight or 10 systems,
reloads and block allocations work fine.  But much above that number,
and resource contention slows down the deploys to the point that VCL
will time out waiting for some of the VM's to come up.  Between 15 and
20 instances will result in all or most of the loaded VM's being marked
"failed".

So the first question is obvious: Is anyone else using a similar setup,
and do you have any advice on dealing with this issue?

One idea I had: I believe that VCL already has a delay between VM
starts, and we might squeeze a little more life out of our setup by
raising that factor.  Assuming I'm right and that parameter does exist,
where is it?

Thanks,
-m

-- 
Michael Jinks :: [email protected]
University of Chicago IT Services

Reply via email to