Hi, I'm trying to investigate a general performance problem with our ESX 3.5 cluster.
I was checking raw tcp speed when I found something with the windows (2003 server) machines.
First what seemed OK with some linux machines using vmxnet 2:
* between 2 vms on differents ESX, linked by a gigabit switch: I get >110MBs on the VM, ~900Mbs on the switch
=> the gigabit link is saturated
* between 2 vms on the same ESX: I get >400MBs on the VM, 0Mbs on the switch
=> the virtual switch does its job
Now what's wrong when connecting to the same linux machine with a windows VM, also using vmxnet 2:
* between 2 vms on differents ESX, linked by a gigabit switch: I get ~30MBs on the VM, ~250Mbs on the switch
=> not even a third of the link is put to use
* between 2 vms on the same ESX: I get ~30MBs on the VM, 0Mbs on the switch
=> the virtual switch works, the VM is still limiting...
To be sure I made some other tests:
* with the windows image installed directly on a physical host => it was able to use the full gigabit link.
* with a windows image using flexible/vmxnet nic => same problem / one third
I guess I need to find some kind of bottleneck in the system, or some incompatible setting in windows networking...
Does anybody has an idea ? Where can I look next ?
Thanks for your answers.