Hi everybody,
we are running still a vSphere 4.1 U2 infrastructure with both static server-/desktop-virtualization and View 4.6 implemented on two different VCs on a total of 32 hosts (ESX for static environment and ESXi for View). The assciated blockbased storage arrays are two HP EVA 6100 (HSV210) and on 3Par F400 FC-Storage-systems addressed via 2 fabrics.
As long as we can think back we see some grave performance on cloning VMs. At first we thought this would result from the hardpressed EVA-storage, but adding a complete new 3Par we got the same grave throughput-values. We confronted VMware five months ago with this problem, collected a lot of logs, made a lot of tests, took HP into the boat as well, making HP and VMware talk to each other directly, but no one came up with a cause, not to talk about that someone might have found a solution for this problem. Probably everyone of you might still know the support-standard-sentence: "We have never seen something like this before. You are the first customer showing us something like this. This problem has never been reported." We tested further on our own besides VMwares and HPs efforts, and narrowed the issue to a single operation-type: Cloning VMs with snapshots.
After we informed VMware about our findings they "tested" this on their own infrastructure (some IBM storage) and finaly (after 4 months) they affirmed, that they can see the issue we mentioned but not in the same intensity. They mentioned, that some slight penalty is normal for this kind of operation (strange that nobody has checked that as cause for the issue in the first place), but again they said the impact is not supposed to be that high. Since then we wait, to get some input from engineering about this issue and we still got the information: "This has never been reported before. We have never heard about it".
Let me give you some values:
For all tests a VM with the following parameters has been used:
Base-vmdk: 10GB
2 snapshots: each aprox. 17MB
(we used a former W2K3-template-VM to make sure the disk are not just filled with zeros)
The calculated throughput results are the following
Moving or cloning the VM without snapshots (just to have something to compare to):
on the EVAs: 50 to 70 MB/s
on the 3Par: 130 to 140 MB/s (with VAAI enabled 170 to 180 MB/s)
Cloning the VM with snapshots (and of this with VAAI disabled):
on the EVAs: 8 to 14 MB/s
on the 3Par: 30 to 32 MB/s
So here comes our question or better our plea to whoever might want to help. VMware still claims this to be an completely isolated issue, wich we find hard to believe. Whoever uses blockbased FC-storage-arrays regardless of which vendor, could you please provide us in this thread with information about your storage device and the throughput-values you can generate under the conditions specified earlier. If you still use vSphere 4.1 it would be great, but vSphere 5 values would be welcome too. Probably we can make out a pattern.
Thanks to everyone who is going to help us in doing this in advance.
Kind regards,
Mike