On Sat, May 26, 2001 at 10:47:32PM +0200, Oskar Sandberg wrote:
> On Sat, May 26, 2001 at 01:23:01PM -0700, Ian Clarke wrote:
> > On Sat, May 26, 2001 at 09:11:10PM +0100, Adam Langley wrote:
> > > Hopefully, under 0.4/0.5 with the announcment system, nodes will try
> > > to position themselves at their conter-of-keyspace (as found by the
> > > annoument protocol). The network should route requests to them because
> > > the initial links to them will be for the same center.
> > 
> > I don't think that nodes should try to artificially encourage a
> > particular epicenter.  The existing system should make nodes specialise
> > without artificially boosting this effect - which I suspect would have a
> > detrimental effect overall.
> 
> Absolutely agreeing here. We have to remember that we are trying to
> balance search efficiency with keeping the network none-static and hard to
> map - otherwise we could just use hypercube routing or alike.

I agree as well .. but I think it's trivially easy for a node to assume any
keyspace focus it likes, from the perspective of the rest of the network,
simply by deliberately failing on keys that aren't in the desired focus.

I've been musing whether there would be some way for the network to assign
a keyspace focus to a node and never allow it to deviate.  If the node
handles keys in that range poorly, it is simply a bad node -- we wouldn't
start asking it for keys in another range.

Suppose the network were literally organized into a graph where each node
(vertice) maintained a relationship with X neighbors (the edges), and
communication between nodes were strictly limited to occurring along these
edges.  Of course we don't want it to be wholly static, so the rearrangement
of these graph edges would have to become a network operation in itself.
The graph would need to be able to organize itself so as to minimize
the number of hops from any node to any piece of data.

Now, view the keyspace focus as a periodic function around a circle (not
necessarily with a single peak).  What each node is looking for in
maintaining its set of neighbors is for the sum of these focus functions
of the neighbor nodes to approximate uniformity.  Make the height of these
functions a measure of the node's capacity (bandwidth, latency, success rate),
and the node will try to replace or supplement weak neighbors in order to
achieve that uniform sum.

One major advantage of a system like this is the increased difficulty of
harvesting node IP addresses.  Of course this also leads to the challenge
of designing an algorithm for the graph to rearrange itself without
eliminating the protection against harvesting.

And of course, there is the question of how we would assign a keyspace
focus function to a node in such a way that the network would globally
respect it.

-- 

# tavin cole
#
# "Technology is a way of organizing the universe so that
# man doesn't have to experience it."
#
#        - Max Frisch


_______________________________________________
Devl mailing list
Devl at freenetproject.org
http://lists.freenetproject.org/mailman/listinfo/devl

Reply via email to