I've noticed problems with the estimates for WCG/BOINC as well...it seems to consistantly believe that WCG tasks will take 1.5 to 4+ times as long as they actually do...and I have the feeling that the time estimates go up greatly whenever I restart my computer, even if my computer was only off for a couple minutes.

I think it's just a buggy estimate program...it's designed to guess how much wall-clock time it'll take to complete a WU, but it seems like it responds to outages (like me restarting my computer, or your BOINC agent switching to another task) incorrectly. While I'm not 100% sure that's the reason for bad estimates, I'm sure there is one...one of my computers is currently estimating 22hrs to completion on a FAAH WU...it's never once taken that long to complete a FAAH unit (that comp averages somewhere around 6-9hrs for FAAH units), and it averages something like 20 hrs of BOINC/day, so actual outages don't explain the whacked-out estimate.

So...WCG/BOINC gives greatly exagerated times to completion. Just watch your % complete...if it seems to increase fast enough for your WU to finish on time, the WU will finish on time.

One other note on Rosetta WUs...I noticed when I was running Rosetta@Home WUs that the % complete would sometimes hold still for long periods of time, while the time to completion increased steadily. But, when the % complete finally did advance, the time estimate would drop drastically...Rosetta only gives accurate time estimates just after the % completed increases. From what I understand about how Rosetta works, this is less often the case for WCG WUs than for Rosetta@Home WUs, because WCG is primarily running Rosetta stage 1, which usually has more % progress increases per WU, while R@H (IIRC) is working on improving Rosetta stage 2 (which, IIUC, is what HPF2 will be using...so we can expect it to have fewer % completion increases, and much more trouble of this kind).

Thanks for the info guys.

I'll keep watching the % to complete rates rather than the time.

The first WU I mentioned had stopped at 40% for about 15mins, so that was when I killed it. Probably shouldn't have done that. Sorry to whoever.

Thanks again for the help.

teemac (Australia)

This thread could be subtitled "Member to Member Support".

The time estimates are notoriously inaccurate. Based on what the techs indicate, it is caused by the nature of the beast and will not likely be improved in the foreseeable future.

Great Job guys! Keep Crunching.

RT's large green star is a Boerni star, for 5,000,000 points. The regular star system gets cumbersome in a hurry if you or the team has lots of points.

Vulture Central II at grid.org




One of those also shows up below for Vulture Central III.

I am running one workunit for climateprediction. The hours to completion is not dropping as fast as I want, so I have had to dedicate more time to it.

May 30 105 hours, 3.578% 1667 hours to go, due 12/15/07
May 31 110 hours, 3.757% 1665 hours to go, due 12/15/07
June 1 115 hours, 3.953% 1664 hours to go, due 12/15/07
June 2 121 hours, 4.194% 1662 hours to go, due 12/15/07
June 3 129 hours, 4.505% 1659 hours to go, due 12/15/07

... and yet more time...up to 75% now. That should do it.

The climate projects do not seem to have as many checkpoints, and one machine on a different subproject started at the same point over a period of three days if shifting after 1.0 hours in the boinc profile. I changed to run a project for 3.0 hours before switching to another project. That fixed that problem and at least reached a new checkpoint each time.

That's mostly the result of the reissue deadline being recently stretched out to a week instead of 4 days. For example, one of your 2 initial "partners" for your 5 day old WU must have blown it off (or is really slow), and since it's still less than a week, it hasn't been reissued to anyone else yet.

This is just a temporary effect, and will go away once your pending validation queue is fully saturated again. The extended deadline pushes out your average verification times, so the oldest units will tend to be older than before, and the average queue length will be higher. You should be reaching the new average length and getting back to normal daily production soon.

I've also noticed a burst of longer than normal WUs from the HPF project over the last couple of days ("ex" series), which always take longer for others to validate, which again causes a temporary sag in points produced. Now that the next series (gu) has started, I'm seeing shorter than normal units. Looks almost like the WU length tends longer as the series progresses (going from "quickie overview" to "high detail" modes?).