>This approach is reminiscent of an idea called "next generation routing" >that we tried with 0.5. Unfortunately we never got that to work too well, >although it is unclear whether the problem was with NGR, or with some other >aspect of 0.5's implementation. The basic principal of NGR was that each >node would collect data about the performance of other nodes (how long >they took to retrieve particular keys etc), and then when routing, use >this data to predict the node that will respond most quickly for a given >key - and route to it. > >You can find out more about NGR on this page: > > http://freenetproject.org/ngrouting.html > >I think, however, that we need to get the basic algorithm working before >we try to embellish it with something like this, and then when we do >introduce more sophisticated routing schemes, we need to do-so cautiously >lest we wind up with an algorithm that is clever, but far too complicated >to figure out why it isn't working. > >Ian.
hi Ian, I agree that this routing mechanism does look like NGR, although I also see huge differences. A node does not have to guess where keys are located, routing will always go to a specific target (a location) like the current routing. The reason I am posting this idea now, is not that I think it should be implemented right away. Still we could discuss this or other proposals and try to find the advantages and disadvantages. Perhaps other routing mechanisms would work better. Without rush, we should discuss about it. Last note, I think that the 'stochastic' routing is quiet close to the current routing and not too complicated. It's far less complicated as NGR imho. Ruud _________________________________________________________________ On the road to retirement? Check out MSN Life Events for advice on how to get there! http://lifeevents.msn.com/category.aspx?cid=Retirement
