[ Note: I composed this last week in response to Adrian Hook's
message, but for some reason did not send it. I'm sending it now,
but it may be a little out of date given what's been discussed on the
list since last week. -Tim ]
Adrian (and everyone else),
The goal of the WG in the IETF should not be to craft how TCP can be
optimized for communication that goes over a satellite. The goal
should be to craft what TCP should be for the Internet as a whole
(including satellites). The charter of the existing WG is to document
things as they are today. It was unfortunate that the WG was named
"TCP over Satellite" or just "TCPSAT", for that has caused much
confusion about what this group should be about.
"TCP over a more heterogenous Internet" would have been a better name
for the WG.
Whatever the WG recommends, it should be a recommendation that people
can take to all the large vendors of TCP implementations (you know who
they are) and tell them *these* features should be included in all
TCPs that you ship and should be turned on by default. (The computers
at the endpoints of a connection and those who manage them should not
be expected to have to know whether or not the packets between them
are going over a satellite link.) Asking users at the end points to
tweak knobs to tune their performance is not acceptable.
If your goal is to make use of the TCP/IP technology within a system
in which you control all the pieces, then you can do what you want and
hack your TCP however you want. But then exactly how that should be
done should be (IMHO) outside of the scope of the IETF (and this WG).
A TCP that knew how to cope with both congestion losses and corruption
losses gracefully that was safe (regarding congestion behavior) would
be ideal. But how to do that is a hard research problem, and gets at
the heart of the Internet technology we already have deployed today.
Perhaps in a future with RED gateways and everybody (routers and
endpoints) participaing in ECN, then perhaps some losses could be
understood to be error losses. But its hard for me to see how we will
ever get completely there, given where we are now.
(Also note that if IPSEC becomes popular, the TCP headers will be
hidden behind end-to-end encryption, and there will be no opportunity
to muck with them in the middle of the net.)
-Tim Shepard
BBN Technologies......GTE Internetworking
[email protected]
+1 617 873 2013
This archive was generated by hypermail 2b29 : Mon Feb 14 2000 - 16:14:38 EST