On Tue, Jul 03, 2001 at 11:25:30PM +0200, Oskar Sandberg wrote:
> - If the central element does accept references from the new nodes, then
> there is a risk that we end up with a topological problem not unlike the
> current one (and as Tavin pointed out, the current network has _not_
> survived)
Really, why? I can retrieve most or all of what I try to request - have
you tried it lately?
>, where the nodes an announcee is introduced only to are those
> nodes which were announced only a short while ago (because the central
> node is loosing all other references fast). If the central node does not
> accept references from announcees, then we can once and for all reject
> the "this is just a default" argument, and also I'm not sure it makes
> things better - if the central node accepts the references then they are
> at least gaining one fairly trusted peer...
One strategy I was thinking of is to rely on a number of reliable but
publicly unknown nodes to harvest references and send them to the new
inform mechanism, which would parcel them out in a sensible fashion. It
would be very difficult to compromise a node when you don't know its IP
address.
> - The network learns about the announcee through the announcement
> protocol, which I agree does spread things out to some extent, but the
> announcee learns about the network by making normal data requests (which
> makes another point, the bandwidth load on the central node would be
> VERY large, since initial requests go through it - do you want your
> companies pipe saturated by that? Security is not the only reason we
> decentralize...), which do not spread the references very well.
The solution above addresses this issue.
Ian.
PGP signature
