| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: Nutritious Rice for the World
Thread: BOINC underestimating RICE WU's? |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 35
|
|
| Author |
|
|
tombell12
Advanced Cruncher Australia Joined: Oct 8, 2009 Post Count: 87 Status: Offline Project Badges: 
|
Each RICE simulation requires 120,000 seeds. If YOU've aborted a task due overcaching, YOU are responsible for the delay in getting there. Aborts/Errors get immediately replaced when that 120,000 target has not been reached. The No Reply waits for reissue to find if the other 18 get to the total (Actually statistically, knreed mentioned early in the project, that something like ~17.2 results are needed, but getting all 19 enhances resolution of the target, like the difference of a 1024 pixel LCD or a 1200 pixel LCD. May even be mentioned in a help item). In short, not getting the full 19 reduces maximum definition for the scientists! Feeling guilty? Yes, this is true. I am the one who foolishly decided to cache too far ahead and then find WU's were taking longer than was expected according to BOINC  I had always looked at the Min Quorum as the minimum that would be needed for that WU. But also I have seen many units with all 19 results returned and as you just mentioned this is the best and most desired result. The way you have described it to me makes much more sense now. I had really only looked at the general outline of the project and not the more intricate details such as this  |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
tombell12,
----------------------------------------To add (from that special result detail header text ref), the minimum quorum 10 for RICE and 15 for HPF2 was strictly for the earliest possible moment to allow validation as not to let volunteers wait too long for credit. Each subsequent result 11-19 resp. 16-19 is validated against the earlier returned minimum quorum results. These later returns then get a credit ratio per seed computed that has basically been determined by the first group of validated returns. Being in the first 10 or 15 makes it part of the credit per seed / folding sim. proration.
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges: 
|
Yes, this is true. I am the one who foolishly decided to cache too far ahead and then find WU's were taking longer than was expected according to BOINC It's a problem that DCF is per BOINC-project and not per application, since there can be large variations in DCF based on which WCG-project you've run before. But, even if you're running only Rice, and each Rice-wu takes 7 hours +- 1 minute or something, the estimates is still not correct, and this is due to WCG is constantly "adjusting" the estimates... Now this of course makes sence for HCMD2 and so on, there different batches does have variable requirements, and can't always accurately estimate this before starts distributing the work. In Rice on the other hand it makes little sence. Due to the constant re-adjustments of Rice, I've currently got these various estimated cpu-times on the same computer for unstarted Rice-wu's: 7:00:39 6:46:51 6:53:45 On another computer, the lowest estimate is 6:43:50, meaning if anyone is running with a 10-day-cache with this incorrect estimate they'll have 9.6 hours too much cached work than can be returned by the deadline... So, even if you're running only Rice, and variations in cpu-time between Rice-wu's is negligible, it's not recommended to use a 10-day cache-size, since some work will be too late. Better would be to use max 9, if you needs to cache lots of Rice-work to reach your badge-target before the end.  "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
The fpops in the RICE tasks headers have been off for a wee bit , and not just a little (understatement), thus running a HCMD2 combo with RICE gives a real real roller coaster feeling. We'll live through it with just a few weeks to go :D
----------------------------------------PS, the servers too adjust that fpops on a daily basis for all research, but that may not be the case for RICE since is run time correlated ... always the same :|
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Sekerob, sorry for this OT: Photobucket has removed your charts here
|
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Sekerob, sorry for this OT: Photobucket has removed your charts here Thanks, Ctril-F5 should show the fresh set. Sometimes the mass load script hangs, but this one is rather weird as the group that's missing is somewhere in the middle.[/OTT]
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
tombell12
Advanced Cruncher Australia Joined: Oct 8, 2009 Post Count: 87 Status: Offline Project Badges:
|
I feel the last few posts have really cleared the air for me. I will NOT be making this mistake again!
 For now I have kept the cache to 1 day, that should be a tad more manageable  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
The fpops in the RICE tasks headers have been off for a wee bit , and not just a little (understatement), thus running a HCMD2 combo with RICE gives a real real roller coaster feeling. We'll live through it with just a few weeks to go :D PS, the servers too adjust that fpops on a daily basis for all research, but that may not be the case for RICE since is run time correlated ... always the same :| Perhaps the above is relevant to my question. I have a few days of Rice cached and haven't downloaded any more in that time as I took Rice off my projects for now. The estimates on the unstarted wu's suddenly went from 7 hours and a bit, to almost 10 hours. I'd never seen that happen unless I returned an unusually length wu; not possible for Rice. The estimates are slowly dropping but probably not fast enough to get off rush mode before the cache completes. Still wondering why this would happen, if any of you more knowledgeable ones would answer, or speculate. Thanks |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
check your message tab, have you run benchmark recently?
|
||
|
|
Ingleside
Veteran Cruncher Norway Joined: Nov 19, 2005 Post Count: 974 Status: Offline Project Badges:
|
Perhaps the above is relevant to my question. I have a few days of Rice cached and haven't downloaded any more in that time as I took Rice off my projects for now. The estimates on the unstarted wu's suddenly went from 7 hours and a bit, to almost 10 hours. I'd never seen that happen unless I returned an unusually length wu; not possible for Rice. The estimates are slowly dropping but probably not fast enough to get off rush mode before the cache completes. Still wondering why this would happen, if any of you more knowledgeable ones would answer, or speculate. Thanks Well, 4 different possibilities are: If you're running any of the most resent BOINC-clients, like v6.6.xx or v6.10.xx, the manager using estimated Run-time, to be compatible with GPU-apps and multi-threaded apps. But, this means if you're been using the computer for something else, the run-time goes up, and this will be the same with the estimates. A re-run of benchmark is maybe a possibility, but this should re-adjust the DCF... But depending on client it's maybe a bug somethere, or if you're running a really old client the DCF was likely not adjusted after benchmark... Another possibility is you had one or more Rice-task with significantly less than 7 hours estimated time, while the rest had more "normal" close to 7 hour-estimate. If a too-small-estimate was finished, the DCF would be increased, and the "normal" tasks would get a correspondingly large increase... And one option at the end, you've finished another non-Rice-task, without remembering it.  "I make so many mistakes. But then just think of all the mistakes I don't make, although I might." [Edit 1 times, last edit by Ingleside at Mar 15, 2010 12:29:38 AM] |
||
|
|
|