As you know, if every variable in the formula was 90% accurate (which is extremely optimistic in our case), the accuracy of the formula would be at most 90% if only additions and substractions are used and much worse if there are other operations. The accuracy of each estimator in a live network is anywhere between 0 and 100%, and the older the last measurement is, the bigger the error. If the formula uses 3 or more variables from estimators, if a single of those measurements goes under 80% accuracy you can dump the result altogether.
To judge different formulas it is not enough to just plug them in and see which one produces least error; if that was the case I can tell you right now that the formula that uses the least # of variables and only additions and substractions will be the best ;). We need to be able to know which estimator tends to produce the biggest error and to have some known range for that error. What if estimator A consistently produced 10 times more error than estimator B? Wouln't you want to decrease the weight or totally get rid of pA in that case?
The only way to find the error of the various estimators is to know what the actual values are. And the only way to have that knowledge is to run the tests in controlled environment - either sim or watchme network.
P.s. personally I'm against using a formula alone to make the most important decision a node has to make, but if we're going to do it lets at least do it the right way.
pgp00000.pgp
Description: PGP signature
_______________________________________________ Devl mailing list [EMAIL PROTECTED] http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl
