NIC load balancing

[gvbd9255]

Hello,

I installed a VMWare Hypervisor 5.5 server and put 2 virtual machines on it.

The physical server has 2 NICs and connected to the LAN.

When I go to the virtual machine performance tab and look at the network data I see that most traffic is going over 1 vmnic.

The stats of the other NIC are most of the time plain 0 values.

see attachment...

Isn't this supposed to be spread over vmnic0 & vmnic1 ?

The vSwitch has loadbalancing on, if I turn loadbalancing on for the VMNetwork it doesn't seem to change much either.

When I add another virtual NIC and leave it unteamed I get the same result.

Will this only work when teaming 2 virtual NICs in the virtual machine or will it only use the second NIC when it reaches almost full load?

The problem is that I have issues with teaming the microsoft way, the server becomes unreachable for some reason (previous HP soft messed up something?)

I'm using the VMXNET3 NICs.

[gvbd9255]

I just read that Windows only supports native NIC teaming since 2012.

This is a Windows 2008 R2 server.

So I don't have a clue where this teaming feature comes from as I removed all the HP software already.

So this raises another question...

is teaming useless inside a VM server?

Edit: sorry for the confusion, it was bridging not teaming.

[bharathl]

We do not normally do the teaming from Virtual machine level but at the vswitch level. What is the load balancing policy you are using at the vswich level.

[gvbd9255]

thanks for the reply.

actually the standard as you see in the attachment.

not sure why it limits the subnet to .15 tho

Edit: as you noticed you don't see the .15 in the screenshot, I got 3 different end address when re-opening it ? and we don't even have a 192.168.63.x subnet here.

[bharathl]

you are using the default load balancing policy where the traffic from a VM vnic always goes from a single NIC. But if you have two VMs then you can send the traffic from both and see if the traffic is going from the both NICs. We use the IP hash load balancing in our environment(Requires static link aggregation from physical switch side) which balances the load based on the hash of source and destination IPs.

[gvbd9255]

does that mean that the traffic of 1 VM can't be spread over 2 or more NICs?

won't that pull down a heavy load RDP server when it all goes over one physical NIC?

[bharathl]

Yes, the traffic from a vnic can not spread over two physical NICs in the default load balancing policy. We normally use 1G NICs for the management network and 10G for the VM network and yes it is possible that one VM can use all the bandwidth of one NIC.

[bharathl]

Also if you are concerned about one VM taking all traffic you can define the traffic shaping policy at vswitch level or port group level.

[gvbd9255]

It seems that the ip hash is doing what I hoped ESX(i) did as default but I doubt the switches (allied telesis) support that, I'll do some searching for it.

[gvbd9255]

I checked another host.

this has a mail and rdp server as 2 virtual machines.

the only issue I have under rds is that sometimes outlook freezes for a few seconds and gives this white/blurred screen.

so I wonder...

as these 2 VMs are on the same VM host does the traffic leave the host or is it kept internal?

if it is internal then it couldn't be a physical nic overload issue, right?

then I need to look elsewere (the vm cpu/nic stats don't give a reason for the slow down either)

[bharathl]

If both VMs are on the same vlan and connected to the same vswitch then ESXi route the traffic internally. I think the issue with your VM is different and you may need to check with MS for the issue.