> Christopher Fraser wrote this...
> >
> > in too much load on the remote machine. I think it also uses UDP
> > datagrams which are much more efficient than TCP sessions.
> they are? there is quite a bit of disagreement about this particulare issue
> floating about.. suffice it to say some people believe UDP is more
> efficient, and other TCP. let the holy wars begin...
It's the implementation and not an inherent characteristic of the
protocol. For "fast processing", see Jacobson's work (I believe he
presented a paper on the topic at a previous AARNet workshop) on
streamlining TCP involving "Header Prediction". He is able to process
the "average" packet and percolate it up the stack in about or less
than 100 instructions.
There are two clear limitations with FSP:
1. It is bandlimited to 1K per `Round Trip Time'.
2. It has no congestion avoidance or control measures, which make
it socially unacceptable in the particular sorts of networks
we are working with. People like Jacobson, Jain, Crowcroft/Wang,
and recently UArizona's X-kernel group have put considerable
effort into the examination and recification of problems in
this respect.
FSP isn't a particularly good protocol for the latter reason.
mg. (who's thesis involves the area of congestion control).
-- Matthew Gream (sw/hw engineer) <M.Gream@nospam.uts.edu.au> (02) 821-2043