I'm glad that major routing bug was squashed. Well done! But I think there is yet one more big thinko somewhere...
It seems that the network is not learning who specializes in what areas of the keyspace from successful and failed requests. For a number of weeks now I have been running my node with a specially tuned datastore. I wrote a script to delete keys from the ds that are not in my randomly chosen area of specialization thus forcing a strong specialization. If the network is working properly other nodes will notice that I am unusually successful in fulfilling requests for a certain part of the keyspace and will route similar requests to me reinforcing this specialization. This is not happening. Here is a histogram of my datastore: Histogram of keys in in fred's data store These are the keys to the data in your node's local cache (DataStore) Oct 26, 2003 11:26:16 PM keys: 64291 scale factor: 0.008623012341558933 (This is used to keep lines < 64 characters) 0 |======================== 1 |======================= 2 |======================= 3 |======================= 4 |======================== 5 |======================== 6 |==================================== 7 |======================================= 8 |========================================= 9 |========================================== a |=============================================================== b |=========================================== c |======================================== d |======================================== e |==================================== f |======================== peaks (count/mean) 0 --> (0.6943429) 2 --> (0.6754289) 4 --> (0.7212207) a --> (1.8471014) This node specializes in "a". It used to be even more specialized but I haven't run my script in couple weeks and because the network isn't working it is diluting the specialization. But if other nodes are learning my histogram of requested keys should look similar. Here it is: Histogram of requested keys. This count has nothing to do with keys in your datastore Oct 26, 2003 11:23:13 PM keys: 98643 scale factor: 0.009727922268211842 (This is used to keep lines < 64 characters) 0 |============================================================ 1 |========================================================== 2 |========================================================== 3 |============================================================ 4 |================================================================ 5 |============================================================ 6 |============================================================ 7 |========================================================= 8 |=========================================================== 9 |=============================================================== a |=============================================================== b |========================================================= c |============================================================ d |========================================================= e |========================================================== f |========================================================= peaks (count/mean) 0 --> (1.0054845) 4 --> (1.0671208) a --> (1.0614438) c --> (1.0106748) e --> (0.978559) Absolutely no specialization. The little differences that exist are easily within statistical noise. My node has been running like this for many weeks and I'm running 6281 so the network has been given every possible opportunity to learn and it has not. I have been thinking about the possibility that the network is just too chaotic and unstable to ever converge but I have forced convergence on my node and other nodes still are not learning. To me this says bug. As we know, specialization is absolutely necessary for freenet to scale otherwise we are just a broadcast search network. It will also greatly increase the amount of data freenet can store if everyone specializes in a specific area and we do not need nearly as much redundancy. Right now with the totally random routing due to no specialization freenet can only store as much retrievable data as 25*n where n is the average size of the datastores on freenet and 25 is the current max htl. No bueno. -- Tracy Reed http://copilotconsulting.com
pgp00000.pgp
Description: PGP signature
_______________________________________________ Devl mailing list [EMAIL PROTECTED] http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl