"Roger L. Beeman" <[EMAIL PROTECTED]> writes: > On Wed, 10 Apr 2002, Hrvoje Niksic wrote: > >> Agreed wholeheartedly, but how would you *implement* a non-jittering >> ETA? Do you think it makes sense the way 1.8.1 does it, i.e. to >> calculate the ETA from the average speed? > > One common programming technique is the exponential decay model.
Sounds cool. Do you have a pseudocode or, failing that, a reference easy enough that even a programmer of Unix command-line utilities can follow it? :-) (I must admit that your email address adds a certain weight to whatever you have to say about measuring bandwidth.) > I believe that the method is chosen for its simplicity and that > justifications of its validity are completely after the fact. The > simplicity is that one keeps a previously calculated value and > averages that value with the current measurement and saves the > result for the next iteration, i.e. add and shift right. I thought about calculating the average between the "average" and the "current" speed, and use that for ETA, but it sounded too arbitrary and I didn't have time to gather empirical evidence that it was any better than just using average. Again, I'd be grateful if you could provide some code. > You must chose how to normalize the measurement based on > irregularity in the measurement interval, however. I'm afraid I can't parse this without understanding the algorithm.