Monitoring datastore latency inside a VM (VMware)

[JWester]

Hi,
we have different performance storage classes in our cluster, e. g. SAS-datastores = "performance" storage, SATA = "mass".

1. I want to get an alarm if the datastore latency inside a VM is above a defined trigger level. For example: 10 ms for "performance" and 40 ms for "mass". I can define a VM datastore latency alarm, but it's not possible to define different triggers per storage class. I also tried to solve this through "Business Views" - I have datastores grouped there - but I cannot assign a VM latency alarm for a datastore class.
Is there a workaround for this?

2. When I define a VM datastore read latency alarm I can define that e. g. the alarm should only be triggered if the latency is above 10 ms for 5 minutes. But I get immediate alarms if the latency is higher than 10 ms - not after 5 minutes of more than 10 ms. A bug, feature or Layer 8 problem?

Many thanks in advance!

Regards
Joern

[Vitaliy S.]

Hi Joern,

1. Well...if you could group VMs based on the their datastore location (via BV), then you would be able to set different thresholds for VM latency alarm.

2. The alarm is triggered only if you the average latency for the past 5 min is above 10 ms. What kind of performance chart do you have for VM latency? Are there any spikes?

Thanks!

[JWester]

Hi Vitaliy,

1. Well...if you could group VMs based on the their datastore location (via BV), then you would be able to set different thresholds for VM latency alarm.

Yes, but I don't see any way other than to group them manually - and with almost 1k VMs that would be no fun.
And we have VMs with mixed datastore classes...

2. The alarm is triggered only if you the average latency for the past 5 min is above 10 ms. What kind of performance chart do you have for VM latency? Are there any spikes?

Ah ok, I didn't know that the average value is used. Yes, there are spikes which raise the average.
Thanks!

[Vitaliy S.]

Joern, so if we add a "datastore name" as an additional parameter to the dynamic group options it should do the trick for you, right?

[JWester]

Joern, so if we add a "datastore name" as an additional parameter to the dynamic group options it should do the trick for you, right?

Yes, that should work for the most VMs. Still no solution if one VM has disks in different storage classes, but that is low prio.

[Vitaliy S.]

Ok, thanks for the feedback I will ask our R&D team to pre-create these groups in the next updates.

[JWester]

Btw, "host name" and "cluster name" as additional parameters would also be fine.
Use case (from last week): Find out the ratio of Windows/Linux VMs according to RAM usage on a specific datastore.

[Vitaliy S.]

Great suggestion, thanks!