Very interesting, idea. It is the sort of thing we have to be
careful with not to upset any clustering effects, but since the
DataSource has a random setting anyways, letting the load of the node
play in _should_ be ok. 

My only concern is the cost of balancing out this sort of feature.
More and more I wish that we had kept a simulator working beside the
attempted real network node so that we wouldn't be so blind about
the effects of this sort of changes.

On Tue, 25 Apr 2000, Bill Trost wrote:

> Uh-oh, I was thinking again...
> 
> When a node N1 changes the DataSource of DataSend D1, it encourages
> other nodes receiving that DataSend to route future DataSends for keys
> close to D1's to node N1.  If your node has lots of free cache, it's
> probably a good idea to encourage this sort of thing -- but you want
> to back off quickly, because these advertisements are going to last a
> while, and you don't want the node to overfill.
> 
> Therefore, I propose that the chances of the DataSource in a DataSend
> being changed be something like
> 
>       1 in 4 + 30 * sqrt(currentDataStoreSize/totalDataStoreSize)
> 
> Under this regime, the chances of a change are as follows:
> 
>       size            probability
>        0%             1 in 4
>       10%             1 in 13
>       25%             1 in 19
>       50%             1 in 25
>       75%             1 in 30
>       99%             1 in 34
> 
> Thus, for the most part, nodes behave like they currently do, but when
> they are near empty, they are much more aggressive about changing the
> DataSource.
> 
> Does this sound reasonable?
> 
> _______________________________________________
> Freenet-dev mailing list
> Freenet-dev at lists.sourceforge.net
> http://lists.sourceforge.net/mailman/listinfo/freenet-dev
> 


_______________________________________________
Freenet-dev mailing list
Freenet-dev at lists.sourceforge.net
http://lists.sourceforge.net/mailman/listinfo/freenet-dev

Reply via email to