Hi,
I'm experiencing some strange problems with the network performance on my esx hosts. I have 4 Dell R710 servers, every one with 4x onboard 1gbit-nics and 2 Intel 10gbit-nics (all on the vmware hcl).
I have a dvswitch with 2 active 10gbit-uplinks and one 1gbit-uplink as standby adapter from each server. In addition, each host has a local vswitch with one consoleport and a vmkernel-port for vmotion. This vswitch has one active and one standby 1gbit-uplink.
I use netio to do network performance tests in my virtual machines. The VMs are connected to a portgroup on the dvswitch which uses the 2 10gbit-adapters as uplink.
When I do this tests, I am getting very slow network troughput at about 50 megabyte/s if the machines are on the same host, and only about 30mb/s if the machines are on different hosts. the guest os is windows server 2003-r2 datacenter 64.
I first thought of a hw-malfunction or maybe a wrong driver for the 10gbit-nics, but if I configure one of the 1gbit onboard-nics as the only active uplink for the vm-portgroup i am getting even lower transfer rates.
The problem seems to exist in the service console also: When I do some ftp transfers, the bitrate is also far below the value it should be.
I've already double-checked, that the nic-config matches speed/duplex-settings of my physical switching-hw, so there should be no problem. The error counters in the switch are all at 0. At the moment, the setup is: auto negotiation with no flow control - but I also tried other values (e.g. 100-full, 1000-full, etc.) In addition, when I do the tests on physical boxes connected to the same switch, I get adequate transfer rates.
Does anyone have an idea how to further investigate into this?
Thank you,
zmichl
