On 4 mrt 2008, at 17:14, Templin, Fred L wrote: >> But the part that really bothers me is that there will ALWAYS be >> fragmentation even when the source host would have been perfectly >> capable of reducing its packet size.
> Turning back to this, the current SEAL draft is silent > on behavior in the face of persistent fragmentation. From > "Fragmentation Considered Harmful", the pain point is gross > fragmentation in the presence of congestion, since the loss > of a single fragment will result in the whole packet being > retransmitted causing yet further congestion. Approaches to > handling this include 1) drop large packets, and tell sources > to start sending smaller ones, 2) treat the loss of fragments > as a congestion indication, 3) some combination of the two. If I were an implementer, what I'd probably look at is making a separate, shorter queue for packets that cause the encapsulated packet to be fragmented and send too bigs for those if there is a tail drop. _______________________________________________ Int-area mailing list [email protected] https://www.ietf.org/mailman/listinfo/int-area
