Hi,

 we have a problem with the new kernel related to the IP routing policy.
it's a quite unusual setup I think, so some sort of optimization in the
2.2 kernel really annoyed us (since other new capabilities enjoy us, we
would like to upgrade).

 Basically, we're using Linux to test features and performance of
network
specialized equipment (such as router), to do so we use to connect this
black boxes to several ethernet interfaces, moreover we use ip aliasing
to simulate thousends of clients...

        |------------|
        |     Linux  |
a.b.c.d |            | x.y.z.t
   -----|eth0    eth1|------
   |    |------------|    |
   |                       |
   |  s.t.u.v              |
   ------|  thing to  |-----
         |    test    |

then we define routes so that the route to x.y.z.t go through a.b.c.d
(by defing s.t.u.v as a gateway).

With 2.0 kernel, it works, pinging x.y.z.t or establishing tcp/udp flow
go through the "thing", with 2.2 kernel, it seems to detect that both
ip are on local interface and go through lo (as route -C shows).

We tryed to find out a solution using iproute package, without any
success
by now.

Is there a way to make the kernel less smart ? If the code need to be
patched, where is the new route optization coded ?

If the code need to be changed, wouldn't it be a good idea to add a
compilation
option (with value=on by default), since Linux is especially a good OS
for testing/benchmarking network devices ?

Regards,

        JP. Messager -- Open Source Software Engineer
                           (http://www.alcove.fr)
-
To unsubscribe from this list: send the line "unsubscribe linux-net" in
the body of a message to [EMAIL PROTECTED]

Reply via email to