It is a question of the volatility of nodes compared to the number of requests. If you only have one node going down per ten requests or something, then I think it is fine to remove bad addresses when connecting to them fails.
If the volatility is higher then this (and there is of course a question of how high volatility the routing can hold up for) then something like you propose might be necessary to keep requests from being held up. I'm concerned about the large number of extra messages created though... On Sat, 20 May 2000, Ian Clarke wrote: > > The node will delete the reference addresses it tries to follow but fails to > > connect or send to, if it manages to send a message at all. I implemented > > this > > before 0.2 was released, and to my knowledge it works. > > > > I don't feel there is really any room for drastic measures beyond this. If > > anything, the reason that bad references are currently not being cleared > > fast > > enough is probably that request volume is so low. > > Hmmmm, surely since a node should be trying to answer requests as > quickly as possible, that is a bad time to do the testing as to the > validity of the references in the DataStore? > > How about a regular cleanup of the datastore (perhaps once an hour) > where all nodes which are referenced are handshaked, and those that fail > are removed? Obviously this cleanup should be conducted immediately > after the datastore is read or generated from inform.php. > > Recently I have been unable to get a single darn DataReply out of the > network (probably due to nodes going through their datastores trying to > find a proper node to talk to, but taking ages to do it), I feel we need > to take urgent measures to address this. > > Ian. > > _______________________________________________ > Freenet-dev mailing list > Freenet-dev at lists.sourceforge.net > http://lists.sourceforge.net/mailman/listinfo/freenet-dev -- Oskar Sandberg md98-osa at nada.kth.se _______________________________________________ Freenet-dev mailing list Freenet-dev at lists.sourceforge.net http://lists.sourceforge.net/mailman/listinfo/freenet-dev
