stretching HACMP

[Patrick Agsten]

Hi there,

I want to stretch a HACMP cluster. Assuming a FFFFFF00 netmask my
current configuration consists of two nodes A and B. A has boot address
10.23.10.11 and standby address 10.240.10.11. B has boot 10.23.10.12 and
standby 10.240.10.12. The service addresses are from the same subnet as
the boot addresses are, say .15, .16, .17 for A and .25, .26, .27 for B.
The cluster members are directly connected through layer-2 networking
elements. Heartbeat is made over RS-232.

My theory is that the same should work within different subnets if I
tell the layer-3 networking elements to set the appropriate routes. I
discussed that with our networking staff and they see no problem with
the routing. So the question is what happens to the cluster if the
layout is boot 10.23.10.11 for A and boot 10.23.20.11 for B. Standby are
10.240.10.11 for A and 10.240.20.11 for B. All with netmask FFFFFF00.
The service addresses are in the same subnet as the boot addresses of
the primary cluster member. One problem at far distances is the RS-232
but this can be solved using SSA adapters with optical extenders. For
all communication within the cluster should be TCP/IP based this layout
should work if the network makes sure that all addresses are reachable.

Has anyone done something like this before or is there any opinion to
the theory that would help me?

Patrick

[Patrick Agsten]

Hi Peter,

[email]peter.glover@dsl.pipex.com[/email] wrote:

> Patrick Agsten <p.agsten@freenet.de> wrote in message news:<vHvYa.2$Ua3.509@psinet-eu-nl>...
>

[snip: story o far]

> We have several clusters using fibre optics to extend the geographical
> distance. I would suggets that if fibre is available why not simply
> extend your network also making the setup the same as you have now
> just over a greater distance?

I really thougt about that but this would be very expensive because the
two sites will be 9 kmtrs away from each other. We will have a
cross-connected core of four layer-3 switches and I had in mind to use
them directly as a very straight forward solution. Furthermore, using
the network core gives us every flexibility to connect all our users to
the cluster. Last, using layer-3 switches will prevent the cluster
members from talking to each other within one layer-2 domain because
vlan connections are not provided over more than one core switch.

> You will also need to consider changing some of the configuration of
> the HACMP as you should mrrir between sites and you need to consider

You're right, of course. This will be the 2nd step after I have verified
that the principle works.

Patrick