NLB across datacentres

[wharlie]

Hi,

Is it possible to have a multicast NLB cluster across datacentres.

We are running ESX 4.1 hosts across 2 datacentres connected by 10GB fibre.

At each site we have cisco 6509 switches that the ESX hosts are plugged straight into.

We have configured static arp entries on each of the 6509 switches.

We have stretched VLANs between both of the datacentres.

Notify switches is configured on all the ESX hosts.

Basic diagram attached.

If I have NLB nodes running at each site and I disconnect one of the nodes will the switch that the node is connected to redirect incoming requests at that site to the other switch at the other site, even with the static arp entry.

At the moment everything seems to run fine if we have both the nodes at the same site (connected to the same switch).

Traffic is load balanced between nodes and failover occurs when a node is taken offline.

If I distribute the nodes across sites/switches then fail one of the nodes then they traffic is not re-directed to the other node.

Is there anyone else that has successfully implemented this configuration?

Thanks

[bulletprooffool]

Wharlie, have you ha d a look at this article:

http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=100655...

[wharlie]

Thanks for that but I have already seen that document and the link http://kb.vmware.com/kb/1006525.

Also http://www.cisco.com/en/US/products/hw/switches/ps708/products_configuration_example09186a0080a07203...

They refer to only a single switch.

My setup works fine when both nodes are on hosts connected to a single switch.

When I distribute the nodes across 2 switches the NLB has issues.

I was wondering if anyone had successfully done NLB distributed across different physical switches.

[lefthandshake]

Hi Wharlie,

I assume your 6509 switches in each site are connected via a trunk port of some sort? If so, have you tried setting the static mac-address-table entry on the switches to forward packets to the trunk port as well as the interfaces directly connected to your ESXi hosts?

For example, to expand on the configuration described in the Cisco article:

mac-address-table static 0300.5e11.1111 vlan 200 interface fa2/3 fa2/4 Port-channel1 disable-snooping

where Port-channel1  is the trunk link between your two sites. In theory, this would forward your Multicast NLB traffic to both the locally connected ESXi hosts and the hosts in the remote site, though I'm not sure it would work that simply in practice.

We are in a similar situation where we would like to use Multicast NLB across two sites connected at Layer 2 but I'm not sure we'll get an opportunity to test the above configuration as the network switches are not under our control.

I'd be interested in hearing whether you managed to get this working or dropped back to Unicast NLB.

[sfurness]

lefthandshake,

Adding the trunk (port channel) interface worked in our situation.  Thanks for the tip!  Even though we are using physical servers this was a network switch issue, not a VM / server issue.

We have 2 Win2008 servers with Multicast NLB setup.  Each server has two Broadcom NICs that are teamed (in failover mode, not load balanced) and each server is plugged into a separate 6509 switch.  The issue we were having is that depending on which server you took offline, the cluster would go offline as well.