Right metric for vROps storage performance

[TheVMinator]

I'm looking for the right vROps metric to use for storage performance on a virtual machine.  What I need is the metric that gets me as close to the OS as possible when measuring command latency.  there are virtual machine metrics for "storage" "disk" "virtual disk" and "datastore".  Ideally, I'd like a metric that utilizes VMware tools to measure command latency from the time the OS sends the command to the SCSI controller until the time the OS receives it back.  Is there such a metric that measures storage latency from that high up in the OS?  If not what is the metric that gets up the farthest and closest to the OS?  disk, storage, virtual disk, datastore, or one particular metric within these categories?

Thanks!

[MichaelRyom · Accepted solution]

Take a look at the documentation http://pubs.vmware.com/vrealizeoperationsmanager-64/topic/com.vmware.ICbase/PDF/vrealize-operations-...

The reference guide has a lot of metrics and the explanation of them.

I guess these are what you are looking for:

disk|totalReadLatency_average

Disk Read Latency

The average amount of time taken for a

read from the perspective of a Guest

OS. This is the sum of Kernel Read

Latency and Physical Device Read

Latency.

disk|totalWriteLatency_average

Disk Write Latency

The average amount of time taken for a

write from the perspective of a Guest

OS. This is the sum of Kernel Write

Latency and Physical Device Write

Latency.

disk|totalLatency_average

Disk Command Latency (ms)

The average amount of time taken for a

command from the perspective of a

Guest OS. This is the sum of Kernel

Command Latency and Physical Device

Command Latency

Hope this helps

Best regards

Michael Ryom

MichaelRyom.dk

[TheVMinator]

ok thanks again - does "disk" latency figure come from vmware tools?  Is it equivalent to the disk latency figures from something like perfmon?

[MichaelRyom]

It should be the same number you get from perfmon, but it is not gathered from VMware tools afaik.

[TheVMinator]

OK but If it comes the virtual hardware that VMware controls instead of from inside the OS, then if there is an issue inside the OS, it might not show the additional latency introduced by an OS layer issue.  In that case it could vary at least slightly from the perspective that perfmon sees.  Right?