Hoping that someone encountered this issue before. We have 15 hosts BL460G8 ( with 554FLB adapter - Emulex ) in C7000 chassis connected to EMC VNX storage via software iSCSI and we haven't had issues . We added 2 new Gen9 ( 536FLB - QLogic 57840S chipset ) which are in a new HPE C7000 chassis and we have seen , from time to time, huge datastore latency 3000-4000 ms spikes .
We have an SQL on that datastore that is backed up from hour to hour . When the SQL runs on one of the 2 new hosts and the backup is running, we see during that 1-2 minute backup window 4000 ms datastore READ latency. When the SQL runs on an older blade, the latency is a lot smaller compared with the gen9.
The Gen9 blades are up to date with the latest SPP, latest HPE custom build ( Dec 2016 ) , there are no newer bnx2x drivers. The new C7000 chassis appears to be configured ok, networking colleagues checked and mentioned that they don't see any difference between the uplink configuration of the new C7000 compared with the old ones .
Did anyone had some sort of issues like this ? Maybe a "magic" parameter for the driver or whatever ?
We have a case opened to VMware , provided the logs etc for quite a while now , but so far no recommendations . Case opened also to HPE, also without result.