Time to Completion is counting UP
15 Aug 2008 17:12:24 UTC
Topic 193855
(moderation:
What is happening? The To Completion time on an Einstein task is counting up. The job Progress is moving in the right direction towards 100%.
I also run tasks for SETI and LHC (when available) on the same machine and they countdown correctly.
(This may have been the way it has been all along and I've just noticed it.)
Time to Completion is counting UP
)
A rate is derived = (work done so far)/(time so far). That is then used to estimate remaining time assuming if that rate held. Time always marches on, and if there is a downward fluctuation in the arrival rate of recently completed work units then this derived rate will fall. The estimate to complete then goes long. As a greater fraction of the work is done this will be much less noticeable.
It's a similiar situation/effect, though of opposite sense, to one you might see on file copying progress bars. There the first time estimate is a bit silly ( too long ) as it accounts for the setup delay time for the transaction without much actual data transfer happening. But as the pipeline then comes up to speed a truer sense of the actual data rate holds - as the setup time is a much smaller fraction now.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Mike, I think he's talking
)
Mike,
I think he's talking about an individual task, not the project time from the status page.
Tom,
When your computer first gets a task it assumes a length of time to complete it. This is based on an estimate provided from the project and a "duration correction factor" that your computer determines on it's own based on past work. For instance, if the project says your computer should take 10 hours to complete a task, but they usually take 11 hours, your computer will figure it's factor as 1.1.
If you get a task of a different type (as is happening right now as people transition from S5R3 to S5R4) then it still applies it's old correction factor to the new task. It doesn't know that it's going to work this task at a different rate.
A lot of computers had gotten to a fairly low factor towards the end of S5R3 (one of mine was as little as .21 so it thought tasks only took 21% of the time of the estimate). Then they get a task that is going to take pretty much 100% of the time of the estimate. Using my earlier example... Your computer gets a task that is estimated at 10 hours and applies it's correction. It now thinks it takes 2.1 hours (using my computer as an example) and starts computing. BOINC figures out pretty quickly when it's 1 hour into the task and only 10% done (not ~50% like it expected) that it needs to adjust the estimated time to completion.
The GOOD news is that you only have to work a few tasks and it will adjust it's correction factor to the new units.
In other words... Don't worry, the problem should work itself out shortly and you'll see better estimates of the time to completion from the moment a task is downloaded. (Though bear in mind that with E@H they all vary some)
RE: Mike, I think he's
)
Doh! He's definitely talking about an individual task! :-)
OK .... I'll file my answer as I feel it has some long term potential. It could be a good one in time. I reckon it just needs the right question to come along ..... :-)
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
RE: I'll file my answer as
)
I completely agree! You're just THAT far ahead of the curve here... answering questions that will be asked in the future! 8-O
(Actually, I'm surprised that more havn't asked about the status page. I've been watching it with a mild curiosity for where it will settle.)
RE: I completely agree!
)
It worked for Nostradamus! Speaking of tricky answers, we have just been watching ( on YouTube ) Abbott & Costello doing their Who's On First? skit. A classic! :-)
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Haha, on a lark I actually
)
Haha, on a lark I actually performed that briefly for the "duet" catagory in high school forensic speech competition. Normally folks limit the funny stuff to solo "Humorous Interpretation" and don't ever do anything that the general public would know well, but I'm a skinny guy and a rather rotund friend and I decided to pull it out for just a couple of events. Since that was nearly 25 years ago, I can't quite do it perfectly anymore (and of course I do Abbott much better), but I still joke with him about it once in a while.
Thanks for the laugh! :-D
RE: Mike, I think he's
)
Don't worry, Mike, I found it interesting anyway!
In fact, when the Factor (DCF) is too low it takes just one task to get it where it should be. (This assumes that the estimate for all tasks are exactly correct, but it will still be close if they're close). But if the DCF was too high, it would only adjust by 10% of the adjustment needed, so it would several tasks to get it right.
But, as Thunder said:
Cheers
Rod