On Sat, Jul 07, 2001 at 02:10:16PM -0500, Timm Murray wrote:
> Currently, the probability that a request being sent back through the
> network will have it's DataSource field reset on any given node is 1/30.
> Unless there is some weird statistical thing going on that I'm not aware of,
> the total chances that the DataSource field will be reset when going through
> two nodes is 1/15 (2/30). Through three nodes it's 1/10 (3/30), and so on.
Your weird statistical thing is that the probability you want to calculate is
actually:
1 - (29/30)^n where n == number of nodes passed through
> Right now, people are reporting HTL values of 100 being necessary (I'm
> ignoring people who say they use HTL of 500, Fred will just put that down to
> 100 anyway). That means there is a 100/30 chance of reset. Yes, it is
> almost assured that the DataSource field will be reset 3 times in it's
> journey through the network. As simulations have shown, having the
> DataSource field reset too often tends to lengthen network connections.
1 - (29/30)^100 = 0.967
extremely high probability that the datasource is reset at least once.
the probability that it's reset twice or three times would be a little
harder to calculate and i'm sure someone who knows the formula off the
top of their head will jump in..
> My suggestion is to put the probability of reset much higher, perhaps 1/60.
> I realize that this makes it more assured that the node mentioned in the
> DataSource field really is the original source of the data, but we may view
> it as a temperary solution and can lower it if and when we see network
> improvement (or if we don't see any improvement at all over a given time).
> The value can gradualy be brought back near it's orginal setting as we see
> improvement/lack of improvement.
>
> I also realize that this may only be treating the symptoms of the 0.3
> network. Still, I think we should at least try.
--
: it's time we took labor saving away from machines :
: and gave it back to the people... :
_______________________________________________
Devl mailing list
[EMAIL PROTECTED]
http://lists.freenetproject.org/mailman/listinfo/devl
