VMware Cloud Community
JDizzle84
Contributor
Contributor

Network Load Balancing issue

I have a Cisco UCS Blade Chassis with 5 blades running ESXi 4.1. The chassis is connected to a pair of 6120 interconnects, and the interconnects are connected to our 6509e core router/switch.

Im running 2 Windows 2008 R2 Ent servers for my guest OS. They both have 2 NICs assigned to them using the VMXNET3 adapters. The first nic is the production NIC used for normal network access to the servers. The second is a NLB nic which is specifically for use with the NLB built into windows. The prod NIC is connected to a port group called 42-PROD-NET on my vSwitch, and has the following properties:

1990144.png

The NLB NIC is connected to a port group called 42-NLB-NET on the same vSwitch with the following properties:

1990144_1.png

My problem is that I can't access anything over the NLB port group. It was half working before i changed the "Notify Switches" option to No as recommended by a VMware article I found regarding using NLB in unicast mode on vSphere. It was working on some computers in our network, but not from others. There reason was because the 2 VMs were on different hosts and that isnt supported. So when I moved the virtual machines to the same host, and changed the Notify Switches to No on the NLB port group (which was how the VMware article said to do it), now there are no computers in the network that can access the NIC on the NLB port group. I feel like im overlooking something simple, but I don't know what it is. If anyone has any ideas on what could be the issue, or has experience in using NLB in unicast mode on vSphere, I'd greatly appreciate the help as I'm about to pull my hair out!

Tags (4)
0 Kudos
2 Replies
Berg18930
Contributor
Contributor

Hello, We are on 5.0 using Cisco B200 blades through a fiber pair of fiber interconnects and are having the same issue.

We can ping the NLB cluster from the same host but not from any other.  Did you get any resolution to this?

Thanks,
Kristine

0 Kudos
JDizzle84
Contributor
Contributor

Sorry for the delayed response! There really is only 1 fix for this that we have found. After doing tons and tons of research, as well as going through lots of pain, the only way we found to do this is with a static ARP entry in the switch infrastructure. I'm assuming that the 2 VMs that are hosting the NLB service are on separate physical hosts, and the port groups and vSwitches that those VMs are attached to have multiple uplinks. This will freak out the physical switch as the NLB mac address will flap between different logical uplinks.

Here are the 2 articles that we used in resolving this.

http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=100658...

http://www.vmware.com/files/pdf/implmenting_ms_network_load_balancing.pdf

I'd be glad to help out any way I can if you still have not found a solution. Again, sorry for the delayed response!

0 Kudos