| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: The Clean Energy Project - Phase 2 Forum
Thread: WUs running much quicker now? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 14
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hi,
For some days now I've noticed a great reduction of needed time to complete WUs in CEP2. Almost a factor 10! It seems the boinc estimates are now finally tuning down, but they still typically predict about 3x the time it actually takes. It would be nice if someone could confirm that WUs are expected to run much faster now! In the latest project update it is hinted that some heuristics might now be used to give up early on certain jobs (most?) I'm crunching on an overclocked Linux/i7-5930K. |
||
|
|
jonnieb-uk
Ace Cruncher England Joined: Nov 30, 2011 Post Count: 6105 Status: Offline Project Badges: 
|
Welcome to the forum
---------------------------------------- Two things to note: For almost all projects WUs can vary in length. For CEP2 in particular WU runtime has been known to vary from 3-4 hours upto 18 hours. BOINC will eventually "learn" new WU lengths and the estimates will be closer to actual. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Ok. Thanks, things going perfectly normal then.
I assume the "Estimated computation size" that can be seen in Boinc / Task Properties is a somewhat unreliable estimate of the computation time needed in CEP2 because it seems to state about 52000 GFLOPS no matter how long time the task ultimately took. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
There's a runtime cycle with CEP2 [short and long ones], easy and gradually going to difficult, then plummeting for a new library. Visit http://bit.ly/WCGCE1 to see the history, recent and all time [at bottom].
----------------------------------------The 'estimated' runtime for a new task is based on what was actually needed for results on work the days before, i.e. they are adjusted with latency... those short at start of a new library run could still be using the longer times that it took at the end of the previous library. [Edit 1 times, last edit by Former Member at Feb 8, 2015 3:00:51 PM] |
||
|
|
svincent
Advanced Cruncher Joined: Jan 3, 2009 Post Count: 53 Status: Offline Project Badges: 
|
The name of the task contains the simplified chemical formula of the compound being investigated: it'll be something like C20H21N1. The number of carbons gives a rough indication of the size of the molecule and hence whether it'll be a short task or a long one: I'm seeing the number of carbons vary from about 12 to 40.
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
The name of the task contains the simplified chemical formula of the compound being investigated: it'll be something like C20H21N1. The number of carbons gives a rough indication of the size of the molecule and hence whether it'll be a short task or a long one: I'm seeing the number of carbons vary from about 12 to 40. That seems nicely to explain the times I see! I did of course expect WUs to vary in size, but I got a little suspicious when the differences got to be an order of magnitude... As a side note, the task progress percentage in the boinc manager is far off compared with the Elapsed/Remaining ratio (percentage way lower). I'm not sure why these should be inconsistent if they both come from boinc estimates based on previous jobs, so I guess one of them may be an estimate produced/enhanced by the job itself. Example (C18xxx) Progress: 11.6% Elapsed 2:16:00 Remaining: 1:27:00 |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Be careful. There are long running WUs still exist. Following is the sample I just aborted befor timeout.
E229113_ 438_ S.254.C31H33N5.YQXKJTZXOSOHJB-UHFFFAOYSA-N.20_ s1_ 14_ 1-- HOME0V User Aborted 15/03/06 22:19:09 15/03/09 10:44:47 6.94 / 7.07 282.0 / 0.0 At the timing, the percentage of progress was 38%. The rig is Core2Duo E3400/3.2GHz Win Vista 32Bit with 8GB RAM. This WU will never complete within 18H so I have canceled. This rig crunches CEP2 WUs about 5 hours in average. There still be the huge WUs for CEP2. Kiyo. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
The progress on CEP2 is -not- tuned to the possible 8 checkpoints written but to the 18 hours, hence the % can be -utterly- misleading... Doing all 8 in 3 hours but also in 17 hours on my piddly 2.1 ghz laptop, but not once seen a result not make it to the first checkpoint in 18 hours. In short, 6.94 hours is -NO- indication when a checkpoint will be written, it can be in the next second.
See Start Here FAQs how to verify, if and when the task does write these. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Just to make sure, even a single checkpoint turns a result valid, no matter if it hits the 18 hours limit! Just let it run.
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
All my CEP2 WUs have now run max 7-8 hrs for a while. I find this a pleasant or at least refreshing change, especially if permanent.
 |
||
|
|
|