[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Setting MTU size for ppp0



Hi Sven

On Sun, 2005-02-27 at 02:23 +0100, Sven Nielsen wrote:

> PROBLEM IS: my attempts to limit MTU to 1412 are ignored.
> tried adding "mtu 1412" to
> /etc/ppp/peers/dsl-provider
> /etc/ppp/options
> 
> still, after reboot or after poff -> ifdown -> ifup -> pon dsl-provider, ifconfig always reports ppp0 with 
> MTU: 1492. Also tried #pppd noauth mtu 1412 while ppp0 is already active, no effect.

I found a similar thing happening. I did find that doing 

ifconfig ppp0 mtu xxx 

seemed to work. Set it to 576 or something really low when your testing
just to see :) Anyway, I'm not 100% sure this works as advertised (it
does change the MTU as reported by pppd), but it seemed to work for me. 

Digging a little deeper, I found that, by default, pppoeconf sets up the
kernel-mode pppoe driver, using the roaring penguin pppd plugin. Under
this setup it seems to resist listening to any pppd specified MRU/MTU
values. 

If, in /etc/ppp/peers/dsl-provider, you comment out the "plugin
rp-pppoe.os eth0" line like so:

#plugin rp-pppoe.so eth0

and uncomment

pty "/usr/sbin/pppoe -I eth0 -T 80 -m 1412"

which should be near the top of the file, pppd uses the user-space pppoe
driver instead. And then I find that setting the MRU/MTU in is honoured,
though I don't think there is a good reason to set it to anything other
than 1412, since the pppoe driver will be clamping it to 1412 anyway.

Actually, the config is very confusing. All the documentation I found
seemed to imply that the user space deamon will be used by default,
while it is definitely not the case!

> Is some other script overriding my attempts ?

> Anybody any clue why setting MTU is ignored ??

See above :)


> Thanks for the help !

Hope this helps!


> wunderful linux :-) . BTW, using Ubuntu AMD64 (in native 64bit-mode) with self-compiled 2.6.x kernel.

I'm jealous haha. I actually do computational stuff for a "living" (as a
PhD student), and multi-gig memory using simulations aren't all that
uncommon... 64 bits would be sweet.

Cheers
Neilen



Reply to: