Advanced search

Message boards : Number crunching : 'Long runs' as 'Short runs'?

Author Message
Profile Sutaru Tsureku
Send message
Joined: 1 May 09
Posts: 13
Credit: 3,655,193
RAC: 0
Level
Ala
Scientific publications
watwatwat
Message 45629 - Posted: 8 Dec 2016 | 19:50:44 UTC
Last modified: 8 Dec 2016 | 19:59:22 UTC

1wgq-SDOERR_OPMcharmm6P2-0-1-RND0123 - Short runs

My PC with NV GT730 got a Short runs WU... - but it lasts ~ 3 days like a Long runs WU.

A NV GTX980 (wingman) meeded 8 hrs and 44 mins for this above WU.

Is there something wrong?

Thanks.
____________

Profile caffeineyellow5
Avatar
Send message
Joined: 30 Jul 14
Posts: 225
Credit: 2,658,976,345
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwat
Message 45631 - Posted: 8 Dec 2016 | 21:09:43 UTC

It does seem like some of Stefan's tasks are being marked in the wrong category of short and long. Only his CASP22S_crystal_ss_contacts tasks seem to be in the correct areas because he knows how long they should last based on their complexity. Perhaps he does not know what category to put these others in until a few run and come back. I know his work specializes in finding new methods and tool development. So maybe these other tasks are using new methods or areas of the CPU/GPU/PC architecture that are just having unknown results until tested on a few different systems.

This is speculation, but I think it makes good sense based on his "function" in the research. The CASP22S_crystal_ss_contacts tasks had expected result times because they had set nanosecond variations. These others do not (maybe) and therefore could be having wildly different results than expected, thus in the wrong categories or even too long for any system to handle for some architectures or complexities of task to architecture usage.

Many of the tasks we get are from students who are using the 'tried and true' methods and tools the way they have been used in the past. Stefan's whole job is basically to see if there are different or better methods or tools that are going unused or could be better used. I expect that job to have results in the field that do not match what we are used to from other students and tasks. I would think that after some of these WUs come back to them as too long to run, aborted by users, application errors, too long for the long v short runs categories, etc, that they will be reassigned, shortened or cast away as proven bad methods or even that a new category might be created just for runs that are super long.
____________
1 Corinthians 9:16 "For though I preach the gospel, I have nothing to glory of: for necessity is laid upon me; yea, woe is unto me, if I preach not the gospel!"
Ephesians 6:18-20, please ;-)
http://tbc-pa.org

Profile caffeineyellow5
Avatar
Send message
Joined: 30 Jul 14
Posts: 225
Credit: 2,658,976,345
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwat
Message 45637 - Posted: 9 Dec 2016 | 0:36:19 UTC

I've got 2 of them on a system with 3 980TIs in it. Let's see how these do. They are running 2 per card, so the system is running 6 WUs at this time and these 2 are on separate cards, but with 'ride along' units of other types.

http://www.gpugrid.net/workunit.php?wuid=12185655
http://www.gpugrid.net/workunit.php?wuid=12185598

PappaLitto
Send message
Joined: 21 Mar 16
Posts: 511
Credit: 4,672,242,755
RAC: 0
Level
Arg
Scientific publications
watwatwatwatwatwatwatwat
Message 45640 - Posted: 9 Dec 2016 | 11:42:11 UTC - in response to Message 45629.
Last modified: 9 Dec 2016 | 11:42:27 UTC

My PC with NV GT730 got a Short runs WU... - but it lasts ~ 3 days like a Long runs WU.

A NV GTX980 (wingman) meeded 8 hrs and 44 mins for this above WU.

Is there something wrong?

Thanks.


If the short WU is too long for your GPU, just abort it and let someone else take it.

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 45642 - Posted: 9 Dec 2016 | 19:32:21 UTC - in response to Message 45631.

It does seem like some of Stefan's tasks are being marked in the wrong category of short and long.

Now have a Gerard that looks like it should have been in the long queue instead of short.
Wonder if the boys went out for a long lunch sometime this week. ;-)

Betting Slip
Send message
Joined: 5 Jan 09
Posts: 670
Credit: 2,498,095,550
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 45643 - Posted: 9 Dec 2016 | 20:07:03 UTC - in response to Message 45642.

It does seem like some of Stefan's tasks are being marked in the wrong category of short and long.

Now have a Gerard that looks like it should have been in the long queue instead of short.
Wonder if the boys went out for a long lunch sometime this week. ;-)


Educated people always always go out for looooong lunches, that's what makes them different to people who do work they don't like for a living as against people who do things for a living that they would otherwise pursue as a hobby.

Basically, people who do what they love and get paid for it. More power to them, just wish I could join them.

I also wish that the UK could retain FULL scientific relations with the EU while we also leave. Be assured we want to be part of the science, that coming from someone who voted leave is a big stetement.

Profile caffeineyellow5
Avatar
Send message
Joined: 30 Jul 14
Posts: 225
Credit: 2,658,976,345
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwat
Message 45652 - Posted: 10 Dec 2016 | 15:01:06 UTC - in response to Message 45637.

I've got 2 of them on a system with 3 980TIs in it. Let's see how these do. They are running 2 per card, so the system is running 6 WUs at this time and these 2 are on separate cards, but with 'ride along' units of other types.

http://www.gpugrid.net/workunit.php?wuid=12185655
http://www.gpugrid.net/workunit.php?wuid=12185598
Followup:
Second one finished in 15 hours 30 minutes approx.
First one took over 34 hours then error on validation!

Personal conclusion:
Surely should be long run AND unpredictable time to finish.
Possibly one of those tasks where if it will take a long time it will fail in some way, but if it finishes in a short time, it could be good.

One of the 2 guys above ^^^ might call them unstable, especially the longer it goes and easily damageable. Would probably blame power loss as a culprit.
I did find that PC frozen up and it dropped 2 tasks on reboot while this was running, but the other one of these plus 3 more did finish with no error after the freeze as well and none of them 'started over' from zero %.

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 45660 - Posted: 11 Dec 2016 | 17:07:31 UTC - in response to Message 45652.

One of the 2 guys above ^^^ might call them unstable, especially the longer it goes and easily damageable. Would probably blame power loss as a culprit.
I did find that PC frozen up and it dropped 2 tasks on reboot while this was running, but the other one of these plus 3 more did finish with no error after the freeze as well and none of them 'started over' from zero %.

Have you set disk write caching to off as Zoltan has suggested? I've also found that if the running WU resets to 0% on reboot and starts over, it will always fail validation even though it appears to have completed successfully.

Profile caffeineyellow5
Avatar
Send message
Joined: 30 Jul 14
Posts: 225
Credit: 2,658,976,345
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwat
Message 45672 - Posted: 12 Dec 2016 | 15:40:03 UTC - in response to Message 45660.

Have you set disk write caching to off as Zoltan has suggested? I've also found that if the running WU resets to 0% on reboot and starts over, it will always fail validation even though it appears to have completed successfully.

Yes, disk caching off on that machine.

I have seen the same thing on tasks that start over at zero. I was hoping to let it run to see the time it took based on this post.

ncoded.com
Send message
Joined: 16 Aug 16
Posts: 20
Credit: 528,183,913
RAC: 2,071,797
Level
Lys
Scientific publications
watwatwatwat
Message 45683 - Posted: 12 Dec 2016 | 20:10:52 UTC - in response to Message 45672.
Last modified: 12 Dec 2016 | 20:12:58 UTC

Hi Guys,

If a WU runs for N Hours over X normal hours of either a short or long run, will you be compensated for the extra time spent processing in terms of credits?

EG Would you get more credits than you would have on a 'normal' short or long WU in terms of credits. To be honest I not sure how WU credits are worked out on GPUGrid.

Thanks,

Chris

Betting Slip
Send message
Joined: 5 Jan 09
Posts: 670
Credit: 2,498,095,550
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 45684 - Posted: 12 Dec 2016 | 20:12:54 UTC - in response to Message 45683.

No, WU credits are fixed.

ncoded.com
Send message
Joined: 16 Aug 16
Posts: 20
Credit: 528,183,913
RAC: 2,071,797
Level
Lys
Scientific publications
watwatwatwat
Message 45685 - Posted: 12 Dec 2016 | 20:13:37 UTC - in response to Message 45684.
Last modified: 12 Dec 2016 | 20:16:33 UTC

OK Thanks Bettingslip.

Unfortunately electricity bills are not.

I wonder if this is something the developers could think about..

Sending someone a 'short' WU which states that it is going to take an hour (on fastest GPU), when infact it takes maybe 20 hours, and then assigning the same amount of credit as a short WU, seems a bit mean..

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 45689 - Posted: 12 Dec 2016 | 23:26:12 UTC - in response to Message 45683.

Would you get more credits than you would have on a 'normal' short or long WU in terms of credits. To be honest I not sure how WU credits are worked out on GPUGrid.

Hi Chris, WU credits are fixed for each WU type. Unfortunately a WU assigned to the short queue only receives about 1/2 the credits of the same length WU assigned to the long queue. So for instance a 7 hour WU in the in the long queue may receive 100,000 credits while the same WU assigned to the short queue only receives 50,000. The reason for this was explained long, long ago as a compensation for the many bad WUs that were being generated, and also to motivate people to run the long WUs. This isn't so much the case anymore and I'd personally encourage the admins to give the same credits/hour for all WU types. It seems only fair (and non-elitist). ;-)

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 45690 - Posted: 13 Dec 2016 | 0:09:27 UTC - in response to Message 45689.

Would you get more credits than you would have on a 'normal' short or long WU in terms of credits. To be honest I not sure how WU credits are worked out on GPUGrid.

Hi Chris, WU credits are fixed for each WU type. Unfortunately a WU assigned to the short queue only receives about 1/2 the credits of the same length WU assigned to the long queue. So for instance a 7 hour WU in the in the long queue may receive 100,000 credits while the same WU assigned to the short queue only receives 50,000. The reason for this was explained long, long ago as a compensation for the many bad WUs that were being generated, and also to motivate people to run the long WUs. This isn't so much the case anymore and I'd personally encourage the admins to give the same credits/hour for all WU types. It seems only fair (and non-elitist). ;-)

Speaking of WUs in the wrong queue, it looks like a bunch of GERARD_AGGRE WUs just got dumped into the short queue when they should be in the long queue. How come I never get to have long lunches? ;-)

ncoded.com
Send message
Joined: 16 Aug 16
Posts: 20
Credit: 528,183,913
RAC: 2,071,797
Level
Lys
Scientific publications
watwatwatwat
Message 45691 - Posted: 13 Dec 2016 | 1:00:55 UTC - in response to Message 45690.
Last modified: 13 Dec 2016 | 1:04:04 UTC

Thanks Beyond, I totally agree.. On both points!

Its strange about the WUs because for the last 8 months short WU's have been short, and long WU's have been long - at constant timelines.

--

In our opinion the whole credit system is a complete mess. You have Collatz that you can earn 100K per hour off a GTX970 doing non-life saving math WU's, and yet a project as important as GPUGrid assigns 200K credits-ish from a 9 hour WU.

And we won't talk about Bitcoin Utopia running off ASC devices which can earn Millions of credits an hour!

What I do not understand is why GPUGrid, World Community Grid, Rosetta, and ClimatePreditication etc eg the Really important projects, do not just match Collatz on credits. If they did they would get massive amounts of extra crunchers which would equate to massively more processing power, and its not like the credits cost anything to give out more.

PappaLitto
Send message
Joined: 21 Mar 16
Posts: 511
Credit: 4,672,242,755
RAC: 0
Level
Arg
Scientific publications
watwatwatwatwatwatwatwat
Message 45692 - Posted: 13 Dec 2016 | 11:34:47 UTC - in response to Message 45691.

In our opinion the whole credit system is a complete mess. You have Collatz that you can earn 100K per hour off a GTX970 doing non-life saving math WU's, and yet a project as important as GPUGrid assigns 200K credits-ish from a 9 hour WU.

And we won't talk about Bitcoin Utopia running off ASC devices which can earn Millions of credits an hour!


Credits are not equal among projects, you cannot directly compare credits from one project to another at all. If you are worried about gridcoin, they take this into consideration and divide the average credits of all users on that project and then place you higher or lower than that average depending on where you are. This determines your magnitude, which determines how many coins you get. Do not worry about collatz's crazy credit, or how low SETI's credit is. It's all in relation to the project itself.

Erich56
Send message
Joined: 1 Jan 15
Posts: 1120
Credit: 8,875,620,176
RAC: 33,427,509
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwat
Message 45693 - Posted: 13 Dec 2016 | 11:52:56 UTC - in response to Message 45690.


Speaking of WUs in the wrong queue, it looks like a bunch of GERARD_AGGRE WUs just got dumped into the short queue when they should be in the long queue.

you are definitely right :-(
My GTX970 just crunched two of them, duration 12 hours each, and only 26.700 credits ea. (normally, for a "long runs" with 12 hours on the GTX970, credit is minimum 160.000 or more!)

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 45711 - Posted: 13 Dec 2016 | 20:10:45 UTC - in response to Message 45693.

Speaking of WUs in the wrong queue, it looks like a bunch of GERARD_AGGRE WUs just got dumped into the short queue when they should be in the long queue.

you are definitely right :-(
My GTX970 just crunched two of them, duration 12 hours each, and only 26.700 credits ea. (normally, for a "long runs" with 12 hours on the GTX970, credit is minimum 160.000 or more!)

I've completed 5 of them so far for 26,700 each. Hoping it was just a mistake after having a bit too much vino at lunch. :-)

Erich56
Send message
Joined: 1 Jan 15
Posts: 1120
Credit: 8,875,620,176
RAC: 33,427,509
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwat
Message 45751 - Posted: 15 Dec 2016 | 20:18:15 UTC - in response to Message 45711.
Last modified: 15 Dec 2016 | 20:20:53 UTC

... after having a bit too much vino at lunch. :-)

sometimes life is just fun :-)

Post to thread

Message boards : Number crunching : 'Long runs' as 'Short runs'?

//