On 18 mei 2009, at 22:51, Joe Touch wrote:
That is equivalent to a case where some packets have more TCP or IP
options than others. I would expect TCP would retransmit the lost
packets, and I would not expect a synchronization between the
smaller-packet payload frequency and the retransmission, so since
these
smaller packets are sufficiently infrequent, a modern TCP (with SACK,
e.g.) ought to handle it fine.
I.e., reporting the "large majority" MTU ought to be appropriate,
rather
than claming down on the smallest MTU on the path.
Hang on, are you saying to just drop the larger packets without
sending back a too big message?
If this large packet is needed to set up state in the decompressor, it
needs to go through the tunnel.
If it doesn't set up compression state, the retransmissions aren't
going to be smaller to the data will never get there.
Or are you assuming that the processing happens on the source host? I
was assuming somewhere in the middle of the network.
