Using two ethernet controllers to improve throughput ?

I've got a few Dell boxen running various permutations of Red Hat
Enterprise Edition. These are basically used as a build farm; one is an
NFS server, and it together with the others all do compile jobs in
parallel using distcc.

At the moment these communicate through a 100Mb ethernet switch onto the
main LAN. Since the boxes have two ethernet ports, and since the
ethernet ports are gigabit ports, I'd like to try to come up with a
scheme such that packets destined for one of the machines in the build
farm are routed through one of the ethernet ports, and packets destined
for elsewhere on the LAN are routed through the currently used port. The
tricky requirement is that ideally I'd like to refer to all the machines
using the IP address allocated by DHCP from the main LAN, rather than
creating a subnet. This would also have the advantage that if the
secondary ethernet connection went down, the packets would be routed
through the "default" one.

What do I need to do in Linux to make this work, and where should I
start looking ?



Geronimo W. Christ Esq

Geronimo W. Christ Esq wrote:
>
> I've got a few Dell boxen running various permutations of Red Hat
> Enterprise Edition. These are basically used as a build farm; one is an
> NFS server, and it together with the others all do compile jobs in
> parallel using distcc.
>
> At the moment these communicate through a 100Mb ethernet switch onto the
> main LAN. Since the boxes have two ethernet ports, and since the
> ethernet ports are gigabit ports, I'd like to try to come up with a
> scheme such that packets destined for one of the machines in the build
> farm are routed through one of the ethernet ports, and packets destined
> for elsewhere on the LAN are routed through the currently used port. The
> tricky requirement is that ideally I'd like to refer to all the machines
> using the IP address allocated by DHCP from the main LAN, rather than
> creating a subnet. This would also have the advantage that if the
> secondary ethernet connection went down, the packets would be routed
> through the "default" one.
>
> What do I need to do in Linux to make this work, and where should I
> start looking ?
>

route command.

It should be possible to route single machines over a specific card. The
routing table does not work only with subnets.

BR
Hans-Juergen Lange

Hans-Juergen Lange wrote:

> route command.
>
> It should be possible to route single machines over a specific card. The
> routing table does not work only with subnets.

Thanks for that. It is indeed possible to use "route" to do what I want,
although I'm not sure how to get the machine to "fall back" to using the
normal ethernet interface if the gigabit one fails.

Geronimo W. Christ Esq wrote:
> Hans-Juergen Lange wrote:
>
>> route command.
>>
>> It should be possible to route single machines over a specific card.
>> The routing table does not work only with subnets.
>
>
> Thanks for that. It is indeed possible to use "route" to do what I want,
> although I'm not sure how to get the machine to "fall back" to using the
> normal ethernet interface if the gigabit one fails.

Hi,

use cron to ping your machine every Minute (with a script). With ping -c
you get the count of pings that reaches your machine. If it reached 0
check out the fallback line. If this works change your routing table.

BR
Hans-Juergen Lange