On Fri, 1 Aug 2003, Jeff Ludwig wrote: > Perhaps it would be beneficial instead of just routing a request to the > "best" node to instead score the nodes and sample from a weighted > distribution of scores. The proposed next-generating routing has the > ability to score based on routing time and successful completions - Why > not sample the 5 best scores then compute Boltzmann transition > probabilities (example below)? This would ensure most routing requests > go to the best node but there's a finite probability of selecting the > second, third, 4th or 5th best... this may allow the node to discover > new routing paths?? Or if the preferred path is slow maybe it would be > able to find an alternate path?
I'm not a routing guru either, but I can tell you that currently there's a randomness factor thrown in to accomplish much the same thing as you're talking about. It's been under debate, but the reasons given for keeping it there is just what you're talking about. It allows the node to discover new paths once in a while and kills/prevents "islands" of nodes. > The "temperature" is a user adjusted parameter which controls the width > of the distribution. At higher "temperatures" the probability of > picking a sub-optimal node increases (at 20K p1=37.754% p2=62.22%) -- > the distribution widens. At lower temperatures the system reduces to > the current routing scheme. First, what would be the purpose of having this be a user adjusted parameter, versus something the node handles itself? Second, why make it adjustable at all? What would be gained by twiddling with the setting? The best reason I can think of is to see if one setting gives better results than another, but it's hard to quantify that. What do you measure? In fact, that's why the random factor is being debated. It's so hard to figure out if it accomplishes anything. It's one person's thoughts vs. another's. However, having said all that, it'd be kind of neat to have that to play with. -todd _______________________________________________ Tech mailing list [EMAIL PROTECTED] http://hawk.freenetproject.org:8080/cgi-bin/mailman/listinfo/tech
