All, In the last few days I have noticed two sites, quite by accident, to which MTU discovery does not work correctly. In one case the router which failed blackholed the traffic, and in the other case, the router returned an MTU value which is garbage (Friday I got 18K, today I get 63692!) Many sites depend on MTU discovery working properly, and the need for MTU discovery is not going away! I ask all of you provider types out there to please look over RFC1191 (it is short). If you don't have tools to diagnose MTU discovery problems, you can pull down ftp://ftp.psc.edu/pub/net_tools/traceroute.tar which includes a -M switch for MTU discovery. Thanks, --Jamshid
At least one of the MTU discovery failures that Jamshid observed is an interesting artifact of the IP-over-ATM paradigm. It seems that if the two ends of a VC don't agree on the MTU, MTU discovery can persistently fail in the following way: a large packet with DF set will arrive at an ATM reassembly buffer which is to small. Since the whole packet can not be delivered, the next hop can not generate the proper ICMP message to adjust the MTU. So, If you are using PVCs in the Internet, you must verify that the upstream end generates the proper ICMP message for the down stream end. --MM--
participants (2)
-
Jamshid Mahdavi -
Matt Mathis