On Tue, 19 Jun 2007 05:14:02 -0700, unixgeeked wrote:

> On Jun 18, 2:32 pm, "andrew.for...@gmail.com"
> <andrew.for...@gmail.com> wrote:
>> On Jun 18, 1:55 pm, unixgee...@gmail.com wrote:
>>
>> > Greetings all,
>>
>> > I am looking for any documentation that will give tips on how to set
>> > up Cisco switches for IP aliasing on multiple subnets. I found the
>> > HACMP cookbook, but that isn't giving me the needed information. My
>> > networking group wants exact information on how they need to set up
>> > their switches to allow multiple subnets over single interfaces.
>> > They are very apprehensive in allowing this type of behaviour, and I
>> > want to give them some solid guidance on what is needed.
>>
>> > Thank you in advance!
>>
>> > Eddie
>>
>> Hi Eddie,
>>
>> Basically you don't really need to tell them anything. All we did was
>> set up a boot IP that was local to the cluster vlan and add persistent
>> IP addresses for the hosts themselves.
>>
>> Basically, en0 on node 1 will have an ip of 172.16.0.1 (in mktcpip)
>> and a persistent IP (an HACMP resource) of whatever DNS resolves to
>> for that host itself and node 2 will have en0 be 172.16.0.2 and a
>> persistent IP of whatever DNS resolves to for node 2.
>>
>> 172.16.0.x doesn't need to be routed off of that VLAN. The boot IP
>> and the service IPs can reside on the same NIC as long as they're on
>> the same VLAN. Make sense?
>>
>> Andrew
>
> Well, our network switches are set in such a way that they only allow
> traffic for a pre-determined subnet. I drew this out on a piece of
> paper yesterday, and determined that I could ask for 3 VLANS, one for
> my boot addresses, one for my admin addresses, and one for my services
> address(es). By leveraging the layer 3 capabilities of the
> switchwork, I have the ablility to fail over to another interface
> instead of moving the entire resource group in the case of adapter
> failure. Normally network switching doesn't care about the IP scheme
> that you use, ours does.

The only problem I see with your scheme is that some forms of in-line
diagnostics require the two local/boot IP addresses to talk to each other.
As I remember, there are some cases where HACMP puts in a route to make
the two interfaces on a single node talk to each other...


-Chris