[ih] Significant milestones in the history of TCP/IP

Noel Chiappa jnc at mercury.lcs.mit.edu
Sat Sep 19 07:24:11 PDT 2015


    > From: Dave Crocker <dhc2 at dcrocker.net>

    > lots of smaller lines means a higher percentage of the bandwidth
    > devoted to headers and control chatter.

Sorry, I don't understand the mechanism there, for more/larger headers? Can
you elaborate?

(The control chatter, perhaps - I can definitely see more routing traffic,
not sure about flow control.)


    > From: Jacob Goense <dugo at xs4all.nl>

    >> I suspect the only way to say with any certainty how well a network
    >> built out of lots of slow lines, as opposed to a few fast ones, would
    >> have worked is a comprehensive simulation.

    > Well, there is an ARPAnet IMP in simh now.

Well, for the routing alone, it's a lot more complicated than that. The
routing is a distributed non-terminating computation, and the question is how
long it takes to stabilize after a perturbation. In the ARPANET, perturbations
weren't just link/node up-down events, but also link delay changes (based on
load), i.e. the inter-perturbation time was small. (Clearly, if Tstabilize is
> Tperturb, the routing will be unstable.)

To _accurately_ simulate it, one would have to accurately simulate i) line
_delays_ (i.e. node-node transmission times), ii) line _speeds_ (to correctly
model the rate at which updates could be sent), and iii) the computation
speed of the IMP itself (as to how fast the algorithm ran, in real time).

This latter is non-trivial - I do a fair amount of work with PDP-11
simulations, and the simulator I use (Ersatz-11) is about 20 times as fast as
the faster real -11 ever - and that's on a relatively elderly Athlon chip!

	Noel



More information about the Internet-history mailing list