we have some trouble with a new vSAN cluster. It's on all Flash.
We already have three other vSAN cluster (14 hosts) in production which are really similar to the new one.
The new one is still under investigation because we see in the performance charts some weird results.
We don't know exactly it is a false-positve information or do we really have an issue which will impact the vms.
We already did some HCIBench tests an the results didn't look so bad. Over 600.000 IOPS
The cluster is on vSphere 6.7 U3. Dedup and Unmap is enabled.
On some of the hosts we can see massive spikes regarding Unmap Latency on the VM.
There are only few vms on the hosts. If we compare these values with our existing vSAN cluster we never see unmap latency over 2ms.
Did you guys use unmap on the vsan cluster? Or any ideas what can cause this latency?
It's a streched cluster but both datacenters are on the same campus and the ISL is very fast.
The unmap latency in the graphs is for the actual unmap process itself which isn't priority traffic and thus may be throttled on and off to some extent frequently so as not to impinge on the performance of traffic that is priority (e.g. VM IOs) and thus I wouldn't be too concerned about this - if you are observing problematically high VM read/write latency at the same time, then this is a different story.
If you want more insight into why you are seeing this in one cluster but not another then you would have to look in more detail into exactly what unmap activities are occurring at times of larger spikes - this isn't necessarily always going to look or operate in the same way for every VM as different Guest-OS and file-systems handle this differently.