MAximum disk latency

[m1kkel84]

Hey!

What is a "okay" maximim disk latency in vmware to a 12 SAS disk iscsi storage, with 3 arrays with 4 disks ?

Just for fun, i installed a program called "Vkernel kapacity view" and it told me all my vm's had too high disk latency. When i check quened write latency, write latency, and the others. I never see latencies above 60 ms.

For a few luns, i see "read latency" about 110 ms at MAX, and average at 15 ms.

I guess that is okay, and the program is just crap.. or ??

[eeg3]

You want your disk latency to be as low as you can get it. How far away on your network is your SAN from your cluster? If the iSCSI network traffic is routing improperly, you could have your iSCSI traffic flowing all through your network which will increase this latency.

____________

blog.eeg3.net | Useful VMware-related Links

If you found this or any other post helpful, please consider the use of the Helpful/Correct buttons to award points.

[mittim12]

Have you monitored the storage counters in esxtop? It's hard to give a certain number to look for since the threshold for what your business deems bad performance could be different than mine but I'm ok with device latency being around 15 to 20 MS, though it's typically much lower. If I see it higher than 20 MS I start to investigate to see if there is a potential problem.

Here is a great storage performance monitoring document, http://communities.vmware.com/docs/DOC-5490.

If you found this or any other post helpful please consider the use of the Helpful/Correct buttons to award points

Twitter: http://twitter.com/mittim12

[m1kkel84]

In each esx server, i have 1 dual port intel nic, connected to a switch just for iscsi. port a in vlan2, port 2 in vlan 3, port 3 in vlan2, port 4 in vlan 3, etc. so it is pretty seperatede, with no brodcast traffic from others.

I just wanted to make sure that UP TO 50 ms was ok, in our current setup?

Its a dual controller ibm DS3300

[m1kkel84]

Ive checked queue command latency, and it is zero on all our luns! Thats a good sign.

[mittim12]

No queing is definitley a good sign. To be honest if I have sustained 50MS device latency when monitoring from esxtop than I'm not happy. That might be ok in your environment but I would want it lower. I don't mind if I have spikes every now and then but it's the sustained values that I'm really looking for.

If you found this or any other post helpful please consider the use of the Helpful/Correct buttons to award points

Twitter: http://twitter.com/mittim12