High-availability bonding .. what's the problem?

Hi!

I have problems with a Linux 2.6.9 kernel and the bonding module. I
would like to run two ethernet interfaces in active backup mode and when
I set up my bonding device like it is described in bonding.txt, it seems
that everything is ok, because when I cat /proc/net/bonding/bond0 I
correctly get the state of the bonding group and if I remove the cable
from my first ethernet device, the /proc/net/bonding/bond0 file shows
that the active slave has changed and that the link for the first
ethernet device is now down. So I guess that MII monitoring is OK, but ...

I cannot get a ping through, even if both ethernet devices of the
bonding group show "Link state: up". Ping says "destination host
unreachable".

I tried this on a dumb hub and a $$ 3Com 3300 Superstack switch (where I
added both ports to the same "Aggregation Link".

With my old configuration (2.4.20 kernel) it works fine both on the
switch and on the hub, so to me it looks like a problem with the bonding
module...

If anyone has an idea, I really would like to hear - err - read it. If
you need further informations, please let me know. We are talking about
an embedded system, so getting screen dumps or something like that is
not that easy, but I can try if you need it.

Thank you in advance for your help,
kind regards,

Heiko




--
Meinberg radio clocks: 25 years of accurate time worldwide

Meinberg Radio Clocks
www.meinberg.de

Stand alone ntp time servers and radio clocks based on GPS, DCF77 and
IRIG. Rackmount and desktop versions and PCI slot cards.

Heiko Gerstung wrote:
> Hi!
>
> I have problems with a Linux 2.6.9 kernel and the bonding module. I

What distro? Self compiled kernel? Using busybox?

> would like to run two ethernet interfaces in active backup mode and
when
> I set up my bonding device like it is described in bonding.txt, it
seems
> that everything is ok, because when I cat /proc/net/bonding/bond0 I
> correctly get the state of the bonding group and if I remove the
cable
> from my first ethernet device, the /proc/net/bonding/bond0 file shows

> that the active slave has changed and that the link for the first
> ethernet device is now down. So I guess that MII monitoring is OK,
but ...
>
> I cannot get a ping through, even if both ethernet devices of the
> bonding group show "Link state: up". Ping says "destination host
> unreachable".

So ping is not simply timing out. You should be getting ping stats on
packets succeded/dropped and times. Are you pinging by IP or host
name?

> I tried this on a dumb hub and a $$ 3Com 3300 Superstack switch
(where I
> added both ports to the same "Aggregation Link".
>
> With my old configuration (2.4.20 kernel) it works fine both on the
> switch and on the hub, so to me it looks like a problem with the
bonding
> module...

Doubt that it has changed but have not checked. Presume you are
"duplicating" your old setup and now it don't work

> If anyone has an idea, I really would like to hear - err - read it.
If
> you need further informations, please let me know. We are talking
about
> an embedded system, so getting screen dumps or something like that is

> not that easy, but I can try if you need it.

As always, make sure you can ping in this order (where applicable):
localhost
gateway
dns server ?
neighbor on same subnet ?
host on different subnet

Did you try traceroute to confirm how far you are getting?

Double check that there is not a firewall rule boinking your packets.
Could be at the originating host(?), a router, destination host.

Confirm, and post output if possible, of:
ifconfig -a
route -n (or netstat -nr)

Is arping available? $ man arping for its use.

Is it reasonably possibe to ditch the bonding module and just use a
single nic in "normal" mode to test ping? At least then you would know
if it's a basic network problem or something about the bonding config.

Also double check that you are not defaulting to IPv6 instead of IPv4.
will check back later,
prg
email above disabled