On Jun 5, 2011, at 9:46 AM, Pascal Hambourg wrote:
Rick Thomas a écrit :
On Jun 3, 2011, at 10:46 AM, Jeffrey B. Green wrote:
The RFCs say that any conforming implementation MUST handle an MTU of
1280, and may not necessarily handle anything larger.
What is your point in mentionning this requirement? Do you mean that
the
server should not send packets bigger than 1280 bytes if it fails to
handle properly path MTU discovery ? If so, I fully agree.
My point is that by setting your MTU to 1280, you have done *your*
part. At least you can be assured that all your packets will get thru
without fragmentation, even if the host at the other end -- or some
intervening router -- is improperly configured.
If the host on the other end sets its MTU to something larger and an
intervening router doesn't do fragmentation, they (or the admins of
the router) need to fix that. An easy recommendation that you can
make in this case (if the server admin on the other end is clueless
but willing to help) is for them to set their MTU to 1280 as well.
That will fix the problem regardless of intervening routers.
Finding a (possible series of) mis-configured intermediate router(s)
and convincing the respective router-admin(s) to fix their
configuration is often difficult. It's easier if you have only one
person to talk to, the server admin on the other end.
So it makes
sense (if you're going to go to the trouble of setting the MTU in the
first place) to use that number.
Lowering the MTU at the client side does not fix the problem that
exists
at the server side. It is just a workaround that works for TCP because
it happens that TCP uses the *outgoing* MTU to calculate the
advertised
*incoming* MSS (how weird when you consider that internet routing is
likely to be asymmetric).
In the particular case in hand -- aptitude/apt-get talking to
schein.debian.org, that is: the ipv6 avitar of security.debian.org --
I have observed that setting your own MTU to 1280 is all that's
necessary.
Enjoy!
Rick
--
To UNSUBSCRIBE, email to debian-user-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Archive: http://lists.debian.org/eb0db985-c349-4108-a689-120c3c997...@pobox.com