On November 22, 2003 09:12 am, Edward J. Huff wrote:
> On Fri, 2003-11-21 at 04:23, Ian Clarke wrote:
> > Edward J. Huff wrote:
> > > It seems to me that NGR can't possibly do certain things.
> > > When the standard deviation exceeds the mean, you can't
> > > even predict the sign of your random variable.
> >
> > The standard deviation exceeding the mean, if it does occur, isn't the
> > fatal problem you think it is.  NGR's data is likely to be extremely
> > noisy, but provided that it is able to extract some generalization it
> > should work, irrespective of the std dev.
>
> The question at issue is whether NGR can be used to distinguish
> between nodes based on bandwidth, while the bandwidth available
> to a transfer is allowed to vary randomly depending on requests
> from other nodes.  The issue is complicated:  any node can
> have its marginal available bandwidth fall near zero, and there
> is a positive feedback loop through the estimators leading to
> oscillation and inefficient use of resources.

This is a very very important point.  The first NGR implementation
did not use transfer rates in its estimators - maybe, exploitable as
it was, it made more sense.  It really depends on if transfer rate is
predictable for a given node.   Given that outbound bandwidth is
more often than not the limiting constraint on a node I suspect 
that its not all that predictable...

Ed

> I think this is a valid and important question.  Your response
> does not address it.
>
> -- Ed Huff
>
>
> _______________________________________________
> Devl mailing list
> [EMAIL PROTECTED]
> http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl
_______________________________________________
Devl mailing list
[EMAIL PROTECTED]
http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl

Reply via email to