Monitor datastore throughput

[tsohst]

Hey, is it possible to monitor the datastore throughput (MB/s)? Would be a great extension to the exsisting IOPS monitoring.

[rschitz]

would the hba traffic be enough for your need?

[tsohst]

If we can break-down the traffic for each datastore individually, that would be nice. As the datastores are shared over multiple hosts, I don't know if the hba traffic can be "assigned" to which datastore it belongs.

[rschitz]

no it's a per hba metric indeed. can you tell me your use case for that metric if you have any?

[tsohst]

We mostly saw a latency issue on our datastores in the past, based on reaching the maximum (allowed) throughput which was configured in the underlaying storage qos. VMware Operations Manager can provide those metrics, which in my understanding, just graps metrics from the vCenter.

[rschitz]

Interesting! And the QOS is bandwidth based not IOPS based like vmware storage policies ?

[tsohst]

Its actually IOPS based, but the storage auto-calulcated the "max" bandwith based on 4K IOPS. Due to different io-sizes we're having less IOPS but higher throughput on the storage, so we're getting latency when reaching the max-throughput even if we're far away from the max-iops. We had a lot of (headache creating) disscussions with the vendor whats right and what is wrong and how it works (or nor ;) ). So we really would like to keep an eye on that value.

https://docs.vmware.com/en/vRealize-Operations/8.6/com.vmware.vcom.core.doc/GUID-A77F1497-A21F-40A9-B240-446A66A174DD.html

[rschitz]

Since we got the vm bandwidth, we could aggregate those values per datastore, let me see what we can do without increasing the load too much

[tsohst]

Sounds good 👍

[rschitz]

can't do it using the vms since the storage usage is per vm and not per vmdk, will post pone this one