Hal says,
> Furthermore, the routing algorithm is the same for inserts as for
> requests, so overlaps occur for inserts as well.  If someone inserts under
> key "mp3" it goes onto 5-10 nodes.  Assuming we allow multiple docs per
> key, every such insert will overlap with each other on at least one of
> those 5-10 nodes.  (And once the two paths cross they will probably join
> and stay together for the remainder of the HTL.)  Therefore there will
> be nodes which will have at least 1/5 to 1/10 of all "mp3" keys stored
> in Freenet.

Mike replies:
> You are right here when you are talking about a network where each
> node knows about every other node (or at least is "near" every other
> node). However, I see this as being less and less true as freenet grows
> up.

The part I wrote above was the conclusion to an argument.  If you disagree
then you should say which premises or which element of reasoning you
disagree with.

My assumptions are: Freenet will "work", that is, data inserted will be
found.  Also, it will work for even large networks with relatively small
HTL values, 10 or less.

>From this it follows that most inserts must cross with requests for the
same key.  Otherwise, the first request for an inserted item won't work.
And if the first request doesn't work, no requests will work.  Hence
Freenet would not work.  Therefore inserts and requests for a given key
must cross with high probability.

But inserts and requests use the same routing algorithm.  Therefore inserts
of data under the same key must cross, as well.  Also, once requests cross
they are likely to stick to each other (unless the closeness database has
changed significantly in the interim).  Therefore almost all inserts of data
under the same key will lead to the same final node, and that node will
be expected to hold entries for every data item in Freenet that uses that
key.

QED.

This is fine for keys which are not expected to have duplicated data, OK
for keys which have only a little bit of duplicated data, but bad for keys
like "mp3" which would have millions of entries.

Please, if you disagree, attack the premises or the argument, not the
conclusion.

Hal

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

Reply via email to