On Fri, Jun 25, 2004 at 02:30:04PM +0200, antonio montagnani wrote: > Andrea Giuliano wrote/ha scritto, On/il 25/06/2004 13:40: > >On Fri, 2004-06-25 at 10:57, Fons van der Beek wrote: > > > > > >>perhaps (most probably) you should change your mtu settings of your eth0 > >>card common is 1490 or 1492 > >> > >I guess it is already. I'm almost sure I saw that value in the output of > >tcpdump. The MTU for ppp is 1500, instead, and I think that's right: > >1492+8, the eth0 MTU plus the 8 byte overhead of ppp. Changing/forcing the MTU is an optimization for most of us. If a packet is routed and the best MTU at one hop is smaller than the incoming packet then the packet must be fragmented. Since fragmentation happens only once no matter how many hops the fragments will be the smallest MTU that the standard permits. MTU discovery in most TCP/IP software stacks takes care of this. If you snoop you might see a handful of "do not fragment ICMP packets" pass as part of this MTU discovery process. You might also see incoming fragments as a hint but you cannot clearly know the reason for fragmenting the packet. With ppp there is encapsulation overhead that effectively reduces the maximum MTU to a number smaller than the ethernet MTU. If your environment blocks MTU discovery it is often nice to just set the local MTU so packets simply 'fit'. It turns out that a dropped fragment is hard to recover so if your packet of data are being fragmented to a minimum value (I think the number is 576) then you want to specify an MTU that passes end to end to optimize the link. This is mostly important when links drop packets. If you have local links that use 'large MTU packets' (FDDI, Gigabit+jumbo) by default and MTU discovery fails then pay attention. I happen to set my ethernet MTU to match what I know my NAT/firewall box must pass via PPP. It lightens the load on my inexpensive NAT box. For most of us today this is a don't care topic because MTU discovery just works (YMMV). -- T o m M i t c h e l l /dev/null the ultimate in secure storage.