Re: [dhcwg] Muti-homing support.

[Ted Lemon <Ted.Lemon at nominum.com>]

In the case you are describing, it seems as if you have created a  
somewhat artificial setup.   Normally I would expect that in a  
situation where multihoming would be beneficial, it would be because  
you were connecting to two separate networks, independently managed  
and on independent physical hardware.

In that case, the protocol would progress slightly differently than it  
does in the case that you describe.   However, because of the server  
selection process in DHCPv6, you are correct that the DHCP client  
would only get addresses on one prefix.

Having agreed to that, I wonder how much it matters.   Historically,  
we've recommended against using DHCP to configure servers.   DHCP is  
seen more as a way to automatically configure client machines.   The  
distinction between client and server is somewhat artificial, but I  
think the main point is that a server has a well-known address, and  
tends to be predominantly the recipient of incoming connections,  
rather than the source of outgoing connections.

This distinction makes a difference.   A multi-homed server with AAAA  
records pointing at all its addresses will be reachable on all those  
addresses, and thus an outage on one link will not cause it a  
problem.   Clients will presumably cycle through the addresses, so  
that when one connection is down the other will be used.

To do the same thing on a client, you would have to have a clever  
source address selection policy.   You'd also have to have clever  
routing: suppose both routers advertise a default route.   Through  
which router do you send the outgoing packets?

I have to admit I don't know much about current IETF practice on  
source address selection.   If a client is multihomed, will it in fact  
be able to take advantage of its multiple egress routes, or will it,  
in practice, wind up either choosing randomly or else always using the  
same router?   Will the performance of the system in the presence of  
an outage on one of the two routers be better or worse if it gets  
prefixes for both?   Will it be easier or harder to debug the problem?

My intuition is that if we solve this problem by making DHCP support  
multi-homing, we'll wind up with a system that fails less predictably,  
and is harder to debug, and we won't get any measurable benefit.

_______________________________________________
dhcwg mailing list
dhcwg at ietf.org
https://www.ietf.org/mailman/listinfo/dhcwg