[MUD-Dev] Net protocols for MUDing (was: Moore's Law sucks)

Chris Gray cg at ami-cg.GraySage.Edmonton.AB.CA
Mon Feb 23 21:31:07 CET 1998


[Chris L:]

:The key feature of TCP which I'd remove would be the error correction. 
:Given a predictive client its both unecessary and counter-productive.

Only to a certain extent. Sure, you can probably cruise over a bit of lost
data in graphics or audio output. Missing some text from the middle of
a paragraph could be a bit shocking to the average user, however. Also,
the channel back from the client to the server should not lose any of
the user's input. What you want is the ability to be sure that if a
message arrives, then it is the complete message that was sent, not some
fragment of one. You will also want to do your own checks on any header
of the message, since UDP doesn't even guarantee correctness. (It typically
is correct over ethernet, because of the way ethernet works, but it isn't
over SLIP, for example, which is one reason why you shouldn't run NFS
over a SLIP link.)

If the system gets a bit smarter, with some caching in the client, then
some sort of reliable protocol will be needed, both ways, to co-ordinate
that cache.

--
Chris Gray   cg at ami-cg.GraySage.Edmonton.AB.CA



More information about the mud-dev-archive mailing list