Re: Bonded gigabit cards
On Tue, 2003-08-12 at 12:11, Josh Lauricha wrote:
> On Mon 13:49, Michael Heldebrant wrote:
> > Then I could set up routing for all cluster traffic over bond0 by ip
> > address in the routing tables.
>
> I'm not familiar with routing tables, but this sounds like what I want
> to do. So, using the examples:
>
> eth0 on 192.168.0.1
> eth1 on 192.168.1.1
> bond0 on 192.168.2.1
>
> Would the calls to route be like:
>
> route add -net 192.168.0.0 netmask 255.255.255.0 dev eth0
> route add -net 192.168.1.0 netmask 255.255.255.0 dev eth1
> route add -net 192.168.2.0 netmask 255.255.255.0 dev bond0
> route add default dev bond0
>
I believe so. The ifup/down tools and /etc/network/interfaces should be
able to set those for you with the proper configuration.
> Then with DHCPD and tftp setup to listen on eth0, PXE has half a chance
> in hell of actually working?
>
Yep, DHCPD, tftp, FAI, anything that needs to talk to a new machine not
on the bonded network should be able to communicate over just eth0 on
the 192.168.0.0 network until it's ready to join the 192.168.2.0
network.
The only other thing I can think of is that it will be tough to have
both a bond0 dhcpd and an eth0 dhcpd running simultaneously, I guess the
cluster machines having a static ip is not a bad thing necessarily.
Depending on how linux handles the incoming packets you may be able to
get away with bond0 and eth0 not ever reading each other's packets so
you could technically have the tftp and dhcpd working simultaneously. I
don't know how this would work though with a broadcast packet unless
some sort of magic happens. Maybe something with iptables could be done
if someone was really hard up to implement it.
> > I've wanted to give it a shot but I don't have any hardware to test this
> > theory at the moment.
>
> Well, if the funding goes through I'll be setting up a 30-40 CPU cluster
> with dual bonded gigabit cards. And will definately be testing this out.
>
> Thanks
Cool, I hope it works.
--mike
Reply to: