Re: [tcpm] draft-eggert-tcpm-historicize-00

[<[e-mail]>]

On 9 Jun 2010, at 18:35, Hagen Paul Pfeifer wrote:
> 
> TCP Alternate Checksum Options _can eventually_ useful in the future for
> example Interplanetary TCP.

Unlikely.

TCP is usable (albeit slow) up until 1.5 light seconds, which just about
encompasses the Moon. Beyond that, interactions of the path delay with
TCP timers degrade performance considerably.

Long-distance interplanetary communication is more likely to be scheduled
and continuous, carefully managed, rather than just throwing a bunch of
competing flows down a pipe and out a dish and trusting TCP endpoints to
do anything useful with the bandwidth their algorithms prevent them from
using effectively. (The old 'loss due to channel corruption is not loss
due to congestion, so don't back off sending' problem.) Long-distance
interplanetary communication may use IP, but it won't use TCP.

Some details in
http://personal.ee.surrey.ac.uk/Personal/L.Wood/publications/index.html#protocol-radius
Lloyd Wood, Cathryn Peoples, Gerard Parr, Bryan Scotney, Adrian Moore,
'TCP's protocol radius: the distance where timers prevent communication,'
peer-reviewed conference paper, International Workshop on Space and
Satellite Communications (IWSSC '07), Salzburg, Austria, September 2007.

In any case, the TCP/UDP checksum often acts as a final demux check at
the endpoint thanks to the pseudo-header coverage; lower-layer frame
checksums detect transmission errors, while higher-layer
checksums or digests across e.g. files detect errors after reassembly
on final delivery.

> Maybe some military sites already employ RFC 1145.

Unlikely.

SCPS-TP (a 'restandardising' of TCP by other bodies aimed at the US
Department of Defense and others way back when) has found a niche
in PEP acceleration for satellite in US military networks - and that
only has the TCP checksum. (A rare example of stuff being left out in
restandardising, rather than being added in.)

If you want TCP behaviour but with a stronger checksum, use SCTP.
See RFC4960. (Won't help with large jumbo SCTP frames, though;
longer lengths need stronger checks, and CRC32c weakens after
about 14Kbytes of payload.)


Lloyd Wood
[e-mail]
http://sat-net.com/L.Wood