On Tue, Jan 27, 2004 at 05:08:11PM -0500, Alan Altmark wrote:
> I'm pretty sure that Linux doesn't do IP takeover.  The VIPA is registered
> in the OSA filters, but the OSA won't respond to ARPs.

And then what Linux believes--in terms of what addresses are configured
on the interface--and what the OSA believes are two different things and
everyone gets unhappy.  Which leads to....

> Don't confuse VIPA with IP takeover.  They are two different things.  The
> concepts are mixed together on z/OS from an implementation point of view.

Which is why I prefer to do VIPA like this:

Set aside a block of IP addresses not in the same subnet as anything
else.  Use these as your VIPA addresses.  Give each interface on your
host a real address with its real netmask and so on, just like normal.
On each host, configure a dummy interface (really: dummy0, for example)
with your VIPA address.  Use OSPF (Zebra or Quagga) to advertise a
*host* route to your own particular VIPA address; OSPF will figure out a
route to get it there over whichever of the physical (from Linux's
viewpoint--may well, indeed probably is, a virtual QDIO device) adapters
it likes better.  Turn on OSPF on VM too, obviously.

This way, the VIPA address can be your canonical IP address, but which
interface the packet gets to that address with can change.

Adam

Reply via email to