Hello folks ,
I had a strange finding today ; an A800 system ontap 9.12p2 is showing strange behavior related to latency in general ; we did an ontap upgrade to 9.13p1 but the upgrade did not fix any of the issues we have seen .
The environment is as follows :
Hosts are hyper-v hosts windows 2019 build1809 HBA drivers Qlogic 9.4.3.21 .
FC Switches MDS 9396t 8.4.2b ; ports are 32gb SW ports all operational . There are 2 MDS’s one per Fabric ( zoning is done one initiator ports to all targets of the FC SVM ) .
MDS BB Credits are showing :
transmit bb credit 64
Received bb credit 32
The system has 18x 7.68 NVME drives FW NA50
ODX is not working at all and when it is enabled the latency on the netapp system on a normal copy or running a iometer can be from 20-50 or even 150 ms ( that is extremely high ) .
When odx is disabled ( via powershell ) latency is a bit better but still too high 2-7 even 8ms with 15ms spikes .
We have done a lot of LUN tests on single host hyper-v and on an esx host with direct RDM LUN .
Some tests show this :
For a workload iometer ; 256K Blocks 50%read sequential we have the following latency :
A800::*> qos statistics volume latency show -volume test_vol2 -vserver FC2_SVM
Workload ID Latency Network Cluster Data Disk QoS Max QoS Min NVRAM Cloud FlexCache SM Sync VA AVSCAN
-total- - 16.36ms 516.00us 0ms 15.81ms 0ms 0ms 0ms 31.00us 0ms 0ms 0ms 0ms 0ms
Other test using normal file copy ( windows file copy ) on the LUN using 50gb file induces more than 2ms on the system :
There are no error in the log ; the MDS switch does not show out of order packages resent or anything strange / discards etc .