| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: BOINC Agent Support
Thread: BOINC error... |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 16
|
|
| Author |
|
|
KLiK
Master Cruncher Croatia Joined: Nov 13, 2006 Post Count: 3108 Status: Offline Project Badges: 
|
i noticed some strange thing on the dualCPU & dualCORE machines...the "time to finish" the calcs are smaller than actuall...and for some time i've been watching this, to figure what is wrong...
----------------------------------------what i noticed is that the "time to finish" is often 2x smaller than the actuall time needed for the calc to finish! so that got me thinking....the BOINC has the CPU benchmark, that gives score for both CPUs or both COREs...it uses that benchmark to calculate the finishing result, even though there are 2jobs on 2CPUs or 2COREs... in other words, it's a stupid error... so can you post that error forward, so that it can be fixed! |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Are you using an old version of BOINC?
The behaviour you describe would be a bug, but I believe that the current BOINC client doesn't behave like that. |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Don't think so KLiK. The benchmark (for 5.10) only tests core 0 and calculates the times to finish for each job in progress separately on basis of that test. If the estimated flops in the job header is too higher, you will see that BOINC will recompute the TTC based on actual progress so far and extrapolation there-off. It does not work proper if the job is heavily non-deterministic.... sometimes it takes allot more operations, sometimes allot less. At any rate, the TTC is highly volatile on non-linear calculations and only should be considered indicative. As the job progresses further, the TTC will get closer.
----------------------------------------
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 1 times, last edit by Sekerob at Feb 14, 2008 6:37:53 PM] |
||
|
|
KLiK
Master Cruncher Croatia Joined: Nov 13, 2006 Post Count: 3108 Status: Offline Project Badges:
|
i'm using the current 5.10.30 version...
----------------------------------------HCC finishes in 2x... FA@h finishes in 2x... HPF2 finishes in 2x... dddt finishes in 2x... is that the concidence? seen that on ABIT BP6 with 2x533Cels, than i use for internet browsing...on sister laptop that has Core2Cel on 2,0GHz with Vista...and on friend computer (that i was fixing) that has Core2Cel on 2,0GHz with WinXP Home... i'm running the 65% of CPU power on BOINC...other things are default! |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Remember that the times shown by BOINC are CPU times. The actual time will be longer. If you have the throttle set to 65%, then 35% of the time BOINC isn't working, and BOINC doesn't include that time in its estimates.
Is this what you are seeing? |
||
|
|
KLiK
Master Cruncher Croatia Joined: Nov 13, 2006 Post Count: 3108 Status: Offline Project Badges:
|
Remember that the times shown by BOINC are CPU times. The actual time will be longer. If you have the throttle set to 65%, then 35% of the time BOINC isn't working, and BOINC doesn't include that time in its estimates. Is this what you are seeing? firstly, if that is right, then you have explained 1/3 surplus time... given the 2/3 is 100% CPU time, then we can calculate that the job will be done in 3/2 actuall time...which is 150% of the given (writen) TTC! 150% is not 200% (2x)...so where does the other 50% goes? secondly, i conclude that the given TTC is allways wrong (by at leaste 50% plus time) and nobody can rely on it...and because of that all my HCC on one computer gives results slightly after TTC! so that means WCG has a lot of TTC overruns ('cause somebody else also has this problem in our million users)...which slowes down the computing network, 'cause of the resending the data! |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Normally BOINC is extremely conservative is TTC prediction to prevent excess buffering of work which could lead to running over the deadline. Remember that 60% throttle is only 60% of the spare cycles. If it's busy with other things the portion of available time reduces too. Overall system (in)efficiency further eats into that time.
----------------------------------------Edit: Inserted "(in)" to clarify
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 2 times, last edit by Sekerob at Feb 15, 2008 3:18:35 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
KLiK, BOINC doesn't work like that. Completion estimates are based on actual performance.
|
||
|
|
KLiK
Master Cruncher Croatia Joined: Nov 13, 2006 Post Count: 3108 Status: Offline Project Badges:
|
Normally BOINC is extremely conservative is TTC prediction to prevent excess buffering of work which could lead to running over the deadline. Remember that 60% throttle is only 60% of the spare cycles. If it's busy with other things the portion of available time reduces too. Overall system efficiency further eats into that time. that could be on my "p2" machine with BP6 mbo & 533Cels on it... but it also does that on Core2Cels 2,0GHz...so there must be some problem in the time calcs for the dualCPU/dualCore machines! i'll watch a little other single CPU/Core machines... maybe you should do the same with yours? 1h to TTC, mark time...in 1h return to see when it finished? |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Suggest you have a look in the client_state.xml or get hold of BOINCview to see your current average CPU efficiency. If it's e.g. 0.6 it means 1/0.6 = 1.67. My quad runs .95 and reports an average run time in the daily device statistics of about 22.5 hours, meaning 1.5 hours are taken up by other things. If there is lots of swapping from ram to disk due cram it only gets worse.
----------------------------------------The below sample is from a part time machine which thru throttling does not get all the available CPU time. <time_stats> <on_frac>0.992760</on_frac> <connected_frac>-1.000000</connected_frac> <active_frac>0.994774</active_frac> <cpu_efficiency>0.494336</cpu_efficiency> <last_update>1203088832.234375</last_update> </time_stats>
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
|