| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Active Research
Forum: OpenPandemics - COVID-19 Project
Thread: Jumps in progress of some WUs |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 6
|
|
| Author |
|
|
[P3D]Magiceye04
Cruncher Joined: Jul 5, 2008 Post Count: 38 Status: Offline Project Badges: 
|
Hello,
i have observed some strange behaviour in the progress of some WUs. Most of them start at 0% and run nearly linear to 100%. But some jump in the first minutes or seconds to about 50% and then run to 100% the normal way. Others start at 0% and run much slower to about 50%, when all the others reach die 95...99% . But the last minutes the progress is getting faster and faster, nearly 1% per second at the end - and they reach the 100% at about the same time then the others. In may statistics all WUs have about the same run time. What is the reason of this? Are these WUs longer then usual but the project splits them to be final calculated on another PC when the estimated time is over? |
||
|
|
Falconet
Master Cruncher Portugal Joined: Mar 9, 2009 Post Count: 3296 Status: Offline Project Badges: 
|
That is normal behaviour.
----------------------------------------Those are workunits that have a short ligand and a long ligand. When the short ligand is being computed, the progress bar goes up to 50% very quickly. Once the short ligand is finished, the more compute intensive ligand begins which takes a lot longer. AMD Ryzen 5 1600AF 6C/12T 3.2 GHz - 85W AMD Ryzen 5 2500U 4C/8T 2.0 GHz - 28W AMD Ryzen 7 7730U 8C/16T 3.0 GHz |
||
|
|
Mike.Gibson
Ace Cruncher England Joined: Aug 23, 2007 Post Count: 12478 Status: Offline Project Badges:
|
That is correct.
Think of it this way. If it starts with a short ligand, it assumes that the second one will be just as quick, but then has to keep readjusting to allow for the long second one. Conversely, if it is a long ligand to start with, it takes longer to reach 50% and then catches up quicker. Pairing long ones with short ones is to keep the total time nearer the average. Mike |
||
|
|
alanb1951
Veteran Cruncher Joined: Jan 20, 2006 Post Count: 1000 Status: Offline Project Badges:
|
Often it's the size and complexity (number of branches) of the ligands being very different, but sometimes it's simply that one job within the work unit does a significantly different number of dockings than the other ones (and yes, I've seen work units with three or more different ligands in them).
I'm not sure whether there is anything non-trivial that could be done to get better overall progress predictions - unless the sub-jobs themselves know how long they are likely to take relative to the total time for the work unit and report their progress accordingly, I don't think there's a way. Cheers - Al.. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
AutoDock looks for the lowest possible 'best' docking energy, the lower the better. A threshold is set of how long the routine should try in the different positions and then abandon if nothing goes below the threshold. Hit the threshold soon, maybe a few more tries to see how low it can go then move on quickly. Of course, that's subject to getting falls positives, mentioned by scientists, and thus not my problem. Nothing much can anticipate that, so you get variable untimes.
----------------------------------------[Edit 1 times, last edit by Former Member at Sep 16, 2020 3:09:00 PM] |
||
|
|
[P3D]Magiceye04
Cruncher Joined: Jul 5, 2008 Post Count: 38 Status: Offline Project Badges:
|
Thanks for the explanation!
|
||
|
|
|