Bogus networking
Marcelo A. Ferreira Gomes
mol-general@lists.maconlinux.org
Sat, 21 Dec 2002 14:18:07 -0200
At 11:16 -0800 20/12/2002, Rob Pfile wrote:
>I really dont think the MTU has anything to do with it. if you lower
>the MTU, the TCP/IP stack will never generate a packet larger than the
>MTU, unless of course the stack itself is buggy, or the MTU is so
>small that you can't even fit the IP and TCP headers into the packet,
>in which case it probably wont transmit anything.
That's exactly what I intend it to do: never generate a packet larger
than what will fit in that guy's physical medium. An MTU of 1450
bytes, for instance, is still large enough not to cause a performance
hit, yet probably small enough to pass through most tunneling
protocols around, including PPPoE, PPPoA, or some other
pseudo-transparent protocol that happens to be in use.
>this is a bug in MOL, if you look back thru the archives samuel has
>commented on it. it might even be fixed in the CVS version, i'm not
>sure anymore. i believe it only affects the 'sheep' networking; i
>think 'tun' works okay.
In my previous message, I suspected that there could be a bug in the
sheep or tun driver. You seem to reinforce my suspicion, since it
occurs only with sheep net. Looks like the sheep driver causes either
MoL or Mac OS to find out a wrong value when doing PMTU discovery,
thus generating packets larger than will fit in the "true" MTU for
that specific path.
I still don't know whether it's the sheep driver, MoL, or even some
other piece of software that is faulty, but I am pretty much
convinced that it all has to do with wrong values being guessed
during PMTU discovery.
--
<suporte [at] mac [dot] com>: n=E3o =E9 ".br", mas =E9 mais BR que os outros=
=2E
:-)