> > There is always a trade-off between security and efficiency. Talking to
> > unknown nodes is a security risk. It is absolutely essential that you give
> 
> There is a difference between trade offs and breaking the system
> though. If something like this turns out breaking the system, then we
> simply can't have it.

I'm sure it wouldn't break the system. As sure as is possible without
simulation, that is.

The only difference between a private cluster and and any other part of
the network, as far as the network is concerned is that 1) the nodes
inside the cluster don't every send messages to nodes which they find out
about from DataSource fields, and 2) nodes inside the cluster (not
counting the periphery) never reset the DataSource field to themselves.

This means that path compression will happen around the cluster but not in
the cluster. This would be bad if the cluster was very wide as every
request routed to it would have to go all the way through it. No hopping
into or out of the middle (that's the point, after all, that the entrance
and exit is contained to the periphery).

However, private clusters will most likely be somewhat close to fully
connected graphs, so they will not be very wide. A private cluster should
act just like a node except that it will absorb 2 or 3 hops from the HTL
instead of 1.

It is arguable whether private clusters will be fully connected graphs
since it is theoretically possible to construct other topologies. However,
I see no reason that other topologies would develop. A private cluster is
based on a web of trust. Such things tend to be close-knit circles rather
than webs. That's also the most beneficial network topology since the
people you trust most likely share common interests so your private
cluster can act as a communal cache.


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

Reply via email to