You''re absolutely correct, Jon, which is precisely why my original suggestion was for allowing the project to be able to designate appropriate applications as "linear", so that the estimations for those apps could be corrected without having any impact at all on applications which are harder to predict.
I honestly don't understand why there's so much resistance to this. There's no downside. It will improve time estimates for a lot of applications (and by improve, I mean from "essentially useless" to "almost perfect"), have no effect at all on the other applications, and is fairly simple to implement. Mike On Mon, Feb 10, 2014 at 11:45 AM, McLeod, John <[email protected]> wrote: > Yes, but the fix for some applications makes it worse for others. There > is also no good way to tell which is which, and there is also the problem > of believing the project administrators as their estimates are often bad. > > > > > > *From:* Michael Goetz [mailto:[email protected]] > *Sent:* Monday, February 10, 2014 11:20 AM > *To:* McLeod, John > *Cc:* Richard Haselgrove; Jon Sonntag; BOINC Developers Mailing List @ > berkeley.edu > > *Subject:* Re: [boinc_dev] Estimated Time Remaining > > > > Jon, > > > > That's very true -- but that's no reason not to fix the estimates for > those apps where it can be fixed! > > > > There's an easy way to fix the estimates for a large number of > applications -- perhaps the majority of applications. We should not pass > on that opportunity because we can't fix the problem for all applications. > For some applications, after all, it's impossible to predict the run time > because they have multiple execution paths that differ greatly in execution > complexity. > > > > By definition it's impossible to create a fix that will work with every > application. Let's at least fix it for those estimates where it's possible > to do so. > > > > Mike > > > > On Mon, Feb 10, 2014 at 9:40 AM, McLeod, John <[email protected]> wrote: > > However, the % complete is often not measured well, and that is required > to be accurate for an estimate based on time spent and % complete. > _______________________________________________ boinc_dev mailing list [email protected] http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
