We are facing the following problem:
We have an Exchange Cluster and the Access Server are running a Multicast NLB.
Before we replaced our old core (9700), we just need to add the "arp <VIP> <MAC of VIP>" command and all is fine.
But now we have problems to access the answering node on TCP, when he is on the same switch as the client. (ping is working fine).
The problem is, that not only Alcatel is involved. Our Exchange Servers are running in a HP BladeCenter with 2 HP-Switches in it, LACP teaming is enabled.
So I'm not sure who causes this problem (beside the, eh, not so good NLB
), but it began after replacing the alcatel-core.For understanding, the physical topology:
[Core] <--LACP-->[IRF-BladeSwitches1]<--LACP--> [Exchange1]
|
LACP---------- -->[IRF-BladeSwitches2]<--LACP-->[Exchange2]
logical:
Exchange1 and 2 are on two subnets (2 configured and tagged VLANs on the HP NIC Teaming software)
2 subnets because there are also 2 server running the exchange databases, and they are in the different vlan.
(IRF stands for Intelligent Resilient Framework, the proprietary HP virtual-chassis solution)
The problem only occurs when the answering node is on the same Bladeswitch as the client (and there are many of them, a few "normal" blades and blades running Vmware).
I have searched many pages in the web to find a solution.
One of them is to add a multicast entry for every involved port, so I have created this for the linkaggs on the core, and for all linkaggs and the underlying ports on the bladeswitches, but nothing changed.
The NLB is running on Windows Server 2008 R2
The core switch is running 7.3.4.248.R02.
Any hints? Or do you need more infos/config snippets?
Thanks
Jens
). But look in guide for HA-feature. This is in OS6900 the best way for NLB.