Re: [cobalt-users] Raq3i: onboard-NIC dies under heavy load

[Gerald Waugh <gwaugh@xxxxxxxxxxxxxxxxxxxxxxx>]

On Mon, 23 Dec 2002, Gunther Stammwitz wrote:
>
> I'm using my cobalt RaQ 3i as a firewall. It is equipped with two onboard
> intel-nics and one additional tulip-based nic in the pci slot. The bios has
> been flashed and the box is running debian with a 2.4.x-kernel.
>
> The problem is, that the box is carrying about 30 to 40 megabits per seconds
> - which means routing from one nic to the other. After running fine for a
> period that changes between just 3 hours or even 3 days the box looses the
> network-connection on one the first onboard nic (eth0). The second
> onboard-nic has died since such an event. DMESG logs some kind of
> netdev-watch-message: NETDEV WATCHDOG: eth2: transmit timed out NETDEV
> WATCHDOG: eth2: transmit timed out NETDEV WATCHDOG: eth2: transmit timed out
> NETDEV WATCHDOG: eth2: transmit timed out NETDEV WATCHDOG: eth2: transmit
> timed out
>
> The link on the nic is still up, but I cannot transfer any data. After
> rebooting everything is fine again. Any ideas ?? Maybe - the nics are
> getting too hot ???
>

A few things I found
There is/was a kernel problem, I believe in 2.4.16 but was fixed in 2.4.17
Don't configure your kernel to use IO-APIC
Use 2.4.x-ac kernels, which have a workaround.

Gerald
--
http://frontstreetnetworks.com | http://raqware.com
Front Street Networks LLC  | Phone: +1 203-785-0699
229 Front Street, Ste. C, New Haven, CT. 06513-3203