[c-nsp] Microsoft multicasted cluster vs. Cisco IOS

[Niels Bakker]

* olivier.dauby at scarlet.be (Olivier Dauby) [Fri 04 Nov 2005, 11:46 CET]:
>I have a 3 Microsoft servers running in cluster facing 2 Cisco routers 
>on an Ethernet segment. The routers allow remote sites to gain access to 
>the cluster.
>
>Problem is that remote users can't gain access to the cluster, but local 
>users can.
>
>The Microsoft cluster is supposed to listen on a virtual ip address 
>using a multicast mac address. Local users see the virtual ip of the 
>cluster (10.106.49.6) with a mac address 03:BF:0A:6A:31:06. I'm already 
>a bit confused here since I expected the multicast mac address to begin 
>with 0x01 instead of 0x03.

3 = 1 (multicast) + 2 (local).

Cisco routers won't enter MAC addresses with the multicast bit set into 
their ARP tables.  Foundry routers and Linux hosts will, I believe 
Force10 will too.  JunOS won't.

(Aside: when a Foundry router sends packets to this MAC address with the 
 multicast bit set, only the unicast interface counter will increase.)

As a workaround, you'll have to, as you already found out:

>I quickfixed this with a dirty 'arp' command on both routers:
>	arp 10.106.49.6 03BF.0A6A.3106 arpa
>
>Is there a nicer (read : more dynamic) way to solve this issue?

No.  Except to use a hub instead of a switch to connect the "highly 
available" cluster members, so they see all packets.

One final thing: Make sure the switch that currently connects them 
forwards multicast frames in hardware.  Not all do by default.


	-- Niels.

--