Oskar Sandberg <oskar at freenetproject.org> writes: > On Fri, Dec 14, 2001 at 11:14:31AM -0600, Edgar Friendly wrote: > > The user can give up and try again - but since the new request is likely > to get routed back to the same node (even if the node that ate the > request is just broken, not an attack) it will hardly make it better. > I disagree. Instead of having all the nodes along the way waiting for up to 30 minutes on a single request, they're freed up to process other requests. Also, with the CP stuff, the routing situation may get a lot better (or worse) between the two requests. The big thing is that it simplifies the node. The internet works grest for routing because the work is done at the endpoints of the network, so things scale better when you add more clients.
> < > > > > a) It cannot go on "forever" as the HTL is decremented every time the > > > timer is restarted, and will eventually reach zero causing a Timeout. > > > > > This is not true, as confirmed by scipient's examination of the code > > as well as my own personal experience of having a HTL=10 request > > return 11 Restarted messages over FCP (which correspond directly to > > "queryRestarted", right?) and then fail after a _30 minutes_ of > > waiting. > > I coded it that way - somebody else must have broken it. Since it's a > little subtle it's easy to miss, but that was definitely the intention. > I'd love it if someone took the time to fix this. > > > b) Freenet's structure is handles this form of attack well because Nodes > > > that don't respond correctly eventually loose references. > > > > > The attack would be to have a node that handled _lots_ of connections > > and then sent "restarted" messages for about a half-hour on each > > request, and then finally answered the query. The victim nodes will > > not penalize this at all. > > There are many such "slow down" attacks that are possible - sending the > trailing as a "byte trickle" for example would have much the same > effect. > Which is why I want much more than just a CP value for each node, but a full analysis of the node's performance (latency, bandwidth, request speed, % successful requests, etc.) to be kept for use in routing decisions. > > Limited branching can be good. > > I didn't say it couldn't - but radical changes right now is not a > good idea. > yes, we need to get the thing working first. Thelema _______________________________________________ Devl mailing list Devl at freenetproject.org http://lists.freenetproject.org/mailman/listinfo/devl
