How does "Virtual Disks Latency" get computed?

[Navalynt]

Can anyone explain to me exactly how the Virtual Disks Latency is being calculated? I'm trying to figure out how accurate of a picture these numbers are painting for me.

We had some performance issues with one of our key virtual servers over the past few months. We found that our SAN configuration was not actually compliant with the vendors best practices guidelines so we added a new expansion array and dedicated some high performance drives to a single pool with a single VMFS datastore that's dedicated to running just that one server.

Once the server was migrated over we started to see immediate performance gains, so this also backed up our belief that it was latency causing the performance issues. Looking at the graphs in VeeamONE for IOPS there is no change (as expected) but when looking at Virtual Disks Latency for that server the improvements are very small. Taking a sample of three days from before the new storage then comparing against three days from after the upgrade (both Mon-Wed) the latency only dropped by about 15%. We were expecting to see more than that.

[Vitaliy S.]

Hi Michael,

What time intervals do you choose to view on the graphs? Veeam ONE have 2 hours and 5 minutes data samples in the database, just make sure you compare the intervals that have the same aggregation of performance data points. As regards your initial question, then we retrieve this data via VMware API, so all manipulations are done on the vCenter Server side.

Can you post any screenshots so I could take a look at the graphs as well?

Thanks!

[Navalynt]

Here's a graph for 8/11/2014:


And here's a graph for 8/25/2014:


This compares a Monday before the upgrade to a Monday after the upgrade between 7am and 6pm. The graphs themselves look a little difference, which I don't understand, and the latency is not significantly reduced.

If Veeam ONE is getting the latency numbers from vCenter, are you at all familiar with how vCenter is coming up with these? Is it actively polling the drive at specific intervals and is it averaging in 0's for those times polled where the servers were idle?

[Vitaliy S.]

Ah...that's what I thought. The reason why you do not see high reduction in latency values, is that on your first graph data is aggregated by 2 hours data stamps, while on the most recent screenshot our graphs are using 5 minutes aggregation intervals.

Please be aware that Veeam ONE uses 5 minutes data stamps for the past 7 days and after that data is aggregated into 2 hours data points. vCenter Server is less granular and stores 5 minutes just for the past day (as far as I remember).

If you want to compare apples to apples, then I would either recommend waiting till 5 minutes data points get transitioned to 2 hours or use Veeam ONE reporting (custom reports) engine that display 2 hours data points in the reports.

Hope this helps!