Posts by John McLeod VII

1) Message boards : SZTAKI Desktop Grid : 2.06 does not correctly suspend computation. (Message 7178)
Posted 3649 days ago by John McLeod VII
I had a run on task. BOINC had requested 2.06 to stop processing, but it was running while BOINC after BOINC had started another task (a long time after, I would like to add).

BOINC client 5.10.45 installed as service under Windows XP.
2) Message boards : SZTAKI Desktop Grid : Ad@m\'s POD (page 7) (Message 3862)
Posted 4257 days ago by John McLeod VII
Not bad this wu37377
And an other candidate for a maximum CPU time exceeded, i\'m afraid

CPU time (sec) 523438.984375
stderr out <core_client_version>4.32</core_client_version>
<message>Maximum CPU time exceeded

6 days, 1 hour ...

Not bad but I can top that with wu39494 ;-)
CPU time (sec) 907928.667307

And all that time for nothing!?

Sigh, I have one (still running) that has used 30 days.
3) Message boards : SZTAKI Desktop Grid : Cancelling large WUs? (Message 3824)
Posted 4263 days ago by John McLeod VII
Just read that shorter WUs will be generated now, and longer may be cancelled...

Well, thats NOT REALLY NICE!

Got one of the larger ones, running for 20 hrs now (app. 25% done)

Minimum quorum is 3 out of 3!

One the first 3 copies was already erraneous, and isn\'t re-sent yet...

Have you guys even thought of how long it will take to get my credits granted if people do what you told them: Aborting them?

And NO, I wont abort it (and loose 200 credits up to now)

I have one running for 240+ hours and < 1% complete.
4) Message boards : Észrevételek, tapasztalatok : WU question (Message 3823)
Posted 4263 days ago by John McLeod VII
BOINC restarts them from the last checkpoint (if any).
5) Message boards : Észrevételek, tapasztalatok : Does SZTAKI have some very, very, very large WU\'s? (Message 3781)
Posted 4270 days ago by John McLeod VII
The BOINC estimations are always wrong with BOINC 5.* releases. Your own calculation is correct. Usually programs should be better and better with every new release, but BOINC 4.* estimated these remaining processing times correctly. I don\'t understand this error, as it is just a simple division. Good luck and a lot of patience!

Thanks ... I\'m not so sure BOINC\'s estimates are flat-out erroneous; I suspect they\'re trying to weigh in such factors as short-term progress, past performance with a given project\'s WUs, the host\'s DCF, &c. With the enormous variability in WU length here, I\'m not surprised the poor thing\'s getting confused ...

Now I see. So the algorithm is correct in itself, but cannot fit to Sztaki. Nevertheless I prefer the previous one. It produced much more reliable results (also in CPDN and LHC), except of course in situations where predictions are impossible, such as WUs with only 1 matrice.

The problem with the 4.x version was that it failed very badly if the progress was not linear, and there are several projects that are distinctly non linear. The 4.x version would also not deal well with projects that gave bad estimates (the 4.x version would fail just as badly here - on every result, not just the first one). The 5.x version does very badly if the initial estimate is very badly wrong. However, the DCF will (slowly if it is an over estimate, and very quickly if it is an underestimate) move toward a better answer after each work unit is completed. If a project has stable runtimes / estimates the DCF will converge to a reasonable (not perfect, I can\'t claim that) estimate, and then the progress estimate will be quite good. Note that a project can have wildly varying normal run times, but if the project can reasonably accurately estimate the run time then the DCF will converge nicely anyway.

Inputs into the run time estimate fpops_est from the project (estimate of the floating point operations), the current CPU time used, the current fraction done, and the current Duration Correction Factor (a number that is some history about results for the project). The worse the initial estimate and the fewer times the the DCF has been updated (results completed without error), the worse the estimate.
6) Message boards : SZTAKI Desktop Grid : Something strange... (Message 3725)
Posted 4277 days ago by John McLeod VII
I have one at 9 days CPU time and 2.00%, and another that is close to that on a different machine. One of these has the highest deadline of anything on the host, so I am just going to let it run and see what happens.
7) Message boards : SZTAKI Desktop Grid : Another very very long WU... (Message 3654)
Posted 4284 days ago by John McLeod VII
Szia ?á­?
Ez a munkacsomag is kb. 890-900 ó?? hossz?i 37,5 nap azaz t?mit 5 hé´Ž
Ennyi id?ak akkor, ha folyamatosan megy, de ez itthoni gé°?m, napi 4-5 ó??´ ha megy...
Való??Ž nem szá­­t a hatá?Šd?Szé° nyarat!

I believe that I have the same, or a very similar problem. Any chance of a re-post in English?
8) Message boards : SZTAKI Desktop Grid : Wus application 2.0 - Request from the Project (Message 3496)
Posted 4297 days ago by John McLeod VII
How do we tell if we hava a bad result?
9) Message boards : SZTAKI Desktop Grid : Workunits/deadline (Message 3495)
Posted 4297 days ago by John McLeod VII
I have an gigantic result as well. This project seriously needs a longer deadline and an increased fpops_est. I know that others are aborting these results.
10) Message boards : SZTAKI Desktop Grid : New applications 2.0 /2.03 online - Feedback (Message 3464)
Posted 4299 days ago by John McLeod VII
Could we please have longer deadlines?

I have one that is 3.4% complete at 22 hours. At this rate, there is no hope of getting done even close to on time, and there is other work on the system that will be returned late as well.

Next 10 posts

Home | My Account | Message Boards

Copyright © 2018 SZTAKI Desktop Grid