> Thanks for the response. I agree that optimizing segment size
> will optimise throughput, particularly over satellite. However,
> for short file transmission, it is arguably not worth spending
> time doing the discovery, where this could be better spent sending
> the file. All I'm suggesting is that this get moved from
> stand-mech I-D to res-issue I-D unless someone has demonstrated
> its value for satellite links.
I do not think Path MTU discovery belongs in the research draft. I
am not sure what sort of research you would like to see. I can come
up with scenarios in which pMTU discovery takes a long time (see the
I-D) and I can come up scenarios where it isn't a big deal. I think
that given that we generally use fairly common packet sizes, pMTU
discovery will probably add nothing to the transfer time in general
(the "handshaking" you refer to only happens if you pick a packet
size that is too large). It seems to me that the value added by
larger segments outweighs the possible costs in a typical situation.
And, if a given host wishes, it need not attempt to send the largest
packet possible, in the hopes (or knowledge) that this will save
time.
allman
This archive was generated by hypermail 2b29 : Mon Feb 14 2000 - 16:14:37 EST