[296] in linux-net channel archive
Re: multiple IP# for one IF ?
daemon@ATHENA.MIT.EDU (Robert Sanders)
Tue May 9 14:08:20 1995
Date: Tue, 9 May 1995 12:29:34 -0400
From: Robert Sanders <rsanders@interbev.mindspring.com>
To: submit-linux-dev-net@ratatosk.yggdrasil.com
In article <3om39a$too@smurf.noris.de>,
Matthias Urlichs <urlichs@smurf.noris.de> wrote:
>Choosing the best source address is a definite plus. Sometimes. Other
>times, I don't _want_ any "best" source address, I want the address I'm
>always using... the only reason I have to add a second address is that the
>_other_ guys don't understand multiple networks and I have to do some
>routing there by hand.
Okay. I guess I'm not really following your point. The ipalias patches
certainly don't get in the way of that. If you don't want a "better"
source address to be chosen, you can a) bind to that address specifically
or b) just add the aliases on the loopback device. Is there a specific
case you'd like to mention? I don't think the ipalias patches ever
*prevent* you doing something you couldn't do without them.
>But do we need the added complexity for the IP router to search not a
>single list (interfaces) for possible local matches, but a list of chains,
>many of which will have only one link?
Actually, the ipalias patches retain the dev->pa_addr field, and only
search the alias list if there are entries. I don't think that's a real
performance (or complexity) cost at all. There are many more areas that need
attention, like the linear list routing table.
Anyway, the list of addresses approach doesn't seem to hurt the BSD
networking code.
-- Robert
