Hi !
We have a 2.4.16 generic kernel and we run into troubles with arp. We
have an ethernet segment of 600+ machines. On one of these machines, I
have set up a static arp table by entering every mac-ip couple.
For several days, I have left the ARP learning (ifconfig eth0 arp) and
there was no problem. Every 5 minutes, I have checked if there was
learned address in the arp cache : there was none, so it didn't learn
any new address.
Today, I have switch to a real static arp table (with ifconfig eth0
-arp). Randomly, some machines were then unable to ping the
host.
After a "ifconfig eth0 arp ; ifconfig eth0 -arp", this some other
machine which were unable to ping the host where the first machines
are able to do it again. arp table is correct but randomly, some
machines are unable to get their echo reply, even if its entry is in
the arp table.
Are there known issues about large arp tables ?
On Mon, Feb 25, 2002 at 11:43:15AM +0100, Vincent Bernat wrote:
> Hi !
>
> We have a 2.4.16 generic kernel and we run into troubles with arp. We
> have an ethernet segment of 600+ machines. On one of these machines, I
> have set up a static arp table by entering every mac-ip couple.
>
> For several days, I have left the ARP learning (ifconfig eth0 arp) and
> there was no problem. Every 5 minutes, I have checked if there was
> learned address in the arp cache : there was none, so it didn't learn
> any new address.
>
> Today, I have switch to a real static arp table (with ifconfig eth0
> -arp). Randomly, some machines were then unable to ping the
> host.
>
> After a "ifconfig eth0 arp ; ifconfig eth0 -arp", this some other
> machine which were unable to ping the host where the first machines
> are able to do it again. arp table is correct but randomly, some
> machines are unable to get their echo reply, even if its entry is in
> the arp table.
>
> Are there known issues about large arp tables ?
You may want to try to increment the thresholds for garbage collection in
net/ipv4/arp.c or play with the userland arpd in case the kernel is having
problem allocating as many ARP entries as you are using.
--
____/| Ragnar H?jland Freedom - Linux - OpenGL | Brainbench MVP
\ o.O| PGP94C4B2F0D27DE025BE2302C104B78C56 B72F0822 | for Unix Programming
=(_)= "Thou shalt not follow the NULL pointer for | (http://www.brainbench.com)
U chaos and madness await thee at its end." [20 pend. Mar 10]
OoO Pendant le journal t?l?vis? du lundi 25 f?vrier 2002, vers 20:42,
Ragnar Hojland Espinosa <[email protected]> disait:
> You may want to try to increment the thresholds for garbage collection in
> net/ipv4/arp.c or play with the userland arpd in case the kernel is having
> problem allocating as many ARP entries as you are using.
Thanks, however, I am still under the 1024 limit and since there is no
learning, there is no need of garbage collection. However, I will try
to modify the thresholds (they can be modified via /proc/net/arp).