On Friday 01 September 2006 09:59, Dave Baker wrote: > On Friday 01 September 2006 12:45, Ed Tomlinson wrote: > > Hi, > > > > I wonder just how dark darknet is? For instance, I have not meet any of > > the > people running the nodes I trust. > > I suspect this is the norm and very probably will remain so. On the other > hand I would like a fair ammount > > of anonomity. I am not sure just how much anonomity open open will give. > Another point, 'fun' is being > > had getting darknet to perform. Introducing another algorythm into the mix > is not going to make this easier. > > > > Given that darknet is not all that dark and the opennet is going to make an > already tricky system more > > complex, I wonder why we do not just extend the darknet concept? We could > use the swap messages > > for this. I see the process working something like this. Nodes keeps > > stats > on their peers, watching > > for peers that are abnormal (eg never reply with data, drop too many > packets, are only up 50% of the > > time, do not update often etc.) When nodes agree to swap why not have them > swap this info too? > > I am not sugesting they just swap the data for thier peers (and hense tell > swap parteners who > > they are connected to), but rather that they swap all the data they have > found in previous swaps. We > > probably want to decay this data so older data istrusted less and we may > randomly want to omit > > some nodes data. > > > > The result of the above process is that a node will have an idea of nodes > that are trustworthy. This > > will allow a node to ask during the swap process, do you want to connect? > If both swap parteners > > have enough data to trust each other and they both need more peers they can > just exchange connect > > data. We may also want to take into account how close the locations are > > and > use this as part of > > the criteria to accept a connection (eg if all peers are close, we really > want new peers to be further > > away and vice versa). > > > > An algorythm like this would let a node build a connection to freenet with > just one reference. Its > > simple enough to put a message on the webpage telling a newbie that his > > node > is still integrating > > itself into the network and that slower performance should be expected. > > It > would also mean that > > existing node owners do not have to go ref fishing (irc freenet-ref etc) > every couple of weeks. I > > suspect it would end up producing a darknet that is considerabliy darker > than the current one. > > It would also reduce the urgency (and maybe the requirement) for opennet. > > Surely what you've just described *is* opennet? At least a form of it. It's > still vulnerable to harvesting. I also don't understand the correllation
You can harvest darknet too - just watch swap messages... > between nodes performing well and not being malicious. If anything, people > that want to attack the network can get a fast server which will always be > rated highly by its peers under this system, and thus will be offered a lot > of new connections. This, if anything, makes a sybil attack even easier. As it is now do you _really_ trust any of your peers? have you actually seen more and one or two? If no, then why do you trust them? At least with this proposal you would know they node is acting 'normal'... I think we are fooling ourselves if we think darknet as it stands now is really dark. Note if a node accepts a connection, takes the peer data and then drops the connection and never reconnects this is ground to distrust it. Trust data gets shared. So you would really have to have a large cluster what really contributed to freenet to harvest a large number of nodes. IIf one node has too many connections and hense is too slow or reject to many requests again it will not be trusted... > At the end of the day, you either have automatic connection establishing > (opennet), or you don't. If you do, you're vulnerable to harvesting and sybil > attacks. That much is straightforward. Exactly how opennet is implemented is > another matter, although I believe there are some excellent ideas. The point is that automatic connection does not have to mean using different algs. The swap method will work with just a little more tweaking. Ed
