On Tuesday 09 December 2008, Micha Feigin <email@example.com> wrote
about 'the cost of ethernet bridging':
>I was thinking of using ethernet bridging to simplify naming and routing
> on my system. This system consists of four computers all interconnected.
> The question is whether there is any hidden cost, as I need the cards to
> work at full throughput (three ethernet cards). So, will this just give
> me the same ip for all three cards or will it also start copying data
> from one port to the other?
I think you need to be more specific.
If your network is like:
* * *
(* = computer, /, | or = network cable)
ethernet bridging won't buy you much speed, but it won't hurt either.
Normally a forwarded packet goes though the iptables stack, with ethernet
bridging is goes though the lighter ebtables stack. Turn off STP for
maximum available bandwidth.
If It's more like:
| X |
(- = also network cable, X = two separate network cables)
bridging still won't hurt too much, but a solution involving OpenBGP  or
something that routes around congestion would help. It would allow up to
3x max mic speed to travel between two nodes in an otherwise idle network.
Using bridging, you'll need STP, you won't pool bandwidth across the
multiple, non-overlapping links, but packets will still bypass iptables in
favor of ebtables.
It could also be like:
| |/ | |
* * * *
(+ = switch port; '=' = switch backplane)
In which case bridging should not be used and instead bonding should be
used if supported by the switch. If bonding is not supported, you should
use round-robin routing, I think.
I can see "four computers all interconnected" + "three network cards" as
any of the above.
 OpenBGP might not be the best idea, as I think it works at the IP
level. You'd really want to pool the links at the ethernet layer but I
don't think bridging will do everything you'd need for that.
Boyd Stephen Smith Jr. ,= ,-_-. =.
firstname.lastname@example.org ((_/)o o(\_))
ICQ: 514984 YM/AIM: DaTwinkDaddy `-'(. .)`-'