[CALUG] Gigabit Ethernet

Dave Dodge dododge at dododge.net
Thu Oct 20 00:26:14 CDT 2005


On Wed, Oct 19, 2005 at 06:31:51PM -0400, Ray Lischner wrote:
> On Wednesday 19 October 2005 05:54 pm, Tim Thomas wrote:
> > The problem is that non-jumbo equipment will see a jumbo frame
> > as a "jabber" and will effectively lop off anything above the
> > standard data size of 1500 bytes.
> 
> So why doesn't PMTU discovery work in this case? Why can't the jumbo 
> frame device discover the path MTU is only 1500 (or whatever) and send 
> frames accordingly when it needs to talk to that device?

PMTU relies on the routers along the path between the sender and
recipient.  The sender marks the packet with a "don't fragment" flag.
If a router gets the packet and notices that it is too large for the
MTU of the next-hop network, it sees that flag replies to the sender
with an ICMP message instead of fragmenting and forwarding the packet.
The sender then retries with a smaller packet size.

If the machines are on the same LAN, they will try to send packets
directly to each other without involving any router.  In this case
there's nobody in the middle to broker the different MTUs and send the
ICMP messages.

Another way to look at it is that PMTU ICMP messages take place at the
IP layer of the network stack.  If you send a jumbo frame directly to
a non-jumbo device, the Ethernet layer will reject the packet before
it ever even reaches the recipient's IP layer.  Senders and recipients
on the same LAN have to agree on an MTU ahead of time.

                                                  -Dave Dodge


More information about the lug mailing list