[c-nsp] preformance problems with nesux5596

[Arne Larsen / Region Nordjylland]

Hi all

Has someone experienced something like this?
We are having performance problems with sql call between application servers and database servers.
We are seeing delay up to 500 msec between 2 app and db server.
The virtual 2 servers are located on the same broadcast domain on to different esx hosts.
We are running wmware 6.0 on the esx hosts. And the 5K' are running   version 6.0(2)N2(3).
These two esx hosts are connected to a Nexus5596 with 2 10G each. We are using FEX 2248PQ 10G with 8x10G uplink, but there is no difference in behavior whether the hosts are connected to the FEX or direct into the nexus5K's
We are not running LACP on the links towards the hosts.
We can send icmp packets with 15000 Bytes or more without any problems.
The esx hosts are running on Hitachi blade servers with 10G pass-through modules. So there aren't any switches in the blade chassis.
I 'we tried installing a 3850 on the 10G links in front of the Nexus5K, so all layer2 switching was done on the 3850, and that works fine, but when the traffic is switched on the Nexus5K's it get slow again.

/Arne