VMware Cloud Community
maneeshlutz
Contributor
Contributor

vROPS: Storage and VM Latency

Hi All,

This question is regarding the analysis of performance data collected from vROPS. Specifically storage performance on an Infrastructure where HP DL360G9 servers and MSA 2052 SAN (1+2 Shelves with a combination of SSDs and SAS | 10G iSCSI ) are used.

While running a performance test, which is write and read intensive, I was looking at the vROPS metrics. The Datastore is showing a latency of 10-15 ms and the VMs running on that shows very high latency than the data store (Say around 150 ms). How can we interpret this data, and how to co relate the values. VM Latency are captured from 'virtualDisk | totalReadLatency_average' and 'virtualDisk | totalWriteLatency_average'

Any thoughts on this?

Thanks

Maneesh

Reply
0 Kudos
2 Replies
sk84
Expert
Expert

vROps is not really good for real time performance monitoring. It collects every 5 minutes 15 data points of 20 second samples for all configured counters and averages them to one single 5 minute data point. This means that peaks that last for less then 5 minutes are averaged and flattened. In addition, vROps also only collects average values from vCenter. So, at the end you get with the mentioned counters the average of the average of the last 5 minutes.

Why is a discrepancy between the datastore view and the virtual disk view?

But it could be that SIOC limits some of your VM disks (is SIOC enabled?) or that the performance counters of the datastore consider all virtual disks and therefore it is the average for all vm disks in that datastore. This would explain why the datastore average values are lower than some virtual disks values.

--- Regards, Sebastian VCP6.5-DCV // VCP7-CMA // vSAN 2017 Specialist Please mark this answer as 'helpful' or 'correct' if you think your question has been answered correctly.
Reply
0 Kudos
maneeshlutz
Contributor
Contributor

Thanks sk84​ for response.

Actually I am not using this as real time monitoring. In fact, we use this for analysing performance statistics for particular time frame when we run tests. Also I have  reduced the sampling interval from 5 minutes to 1 minute. Storage IO Control is not enabled. As the sampling interval is 1 minute and the latency of VMs were consistent for particular time period, why the data store level average latency does not match. can't we co relate both data at any way?

Reply
0 Kudos