New MB apps have false completion times

Message boards : Number crunching : New MB apps have false completion times
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
Profile RottenMutt
Avatar

Send message
Joined: 15 Mar 01
Posts: 1011
Credit: 230,314,058
RAC: 0
United States
Message 1008483 - Posted: 26 Jun 2010, 4:03:00 UTC

I'm getting a little work, unfortunately the time to completion is 60 time to large. I'm starting to take this personally!

should take 1.5 hours, says 80 hours...
ID: 1008483 · Report as offensive
Profile SciManStev Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Jun 99
Posts: 6651
Credit: 121,090,076
RAC: 0
United States
Message 1008484 - Posted: 26 Jun 2010, 4:07:28 UTC - in response to Message 1008483.  

About 2 hours ago, mine said about 8 hours. Now is says 45 minutes. In reality it is 7 minutes.

Steve
Warning, addicted to SETI crunching!
Crunching as a member of GPU Users Group.
GPUUG Website
ID: 1008484 · Report as offensive
Profile skildude
Avatar

Send message
Joined: 4 Oct 00
Posts: 9541
Credit: 50,759,529
RAC: 60
Yemen
Message 1008501 - Posted: 26 Jun 2010, 4:47:26 UTC - in response to Message 1008484.  

I had a whole batch of them that claimed to be 7 minutes on my CPU turns out they were 35 minutes. Then I turn around and get more but they say 80 hours. Gotta love the new calculations


In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope
ID: 1008501 · Report as offensive
Profile RottenMutt
Avatar

Send message
Joined: 15 Mar 01
Posts: 1011
Credit: 230,314,058
RAC: 0
United States
Message 1008528 - Posted: 26 Jun 2010, 6:13:06 UTC

well it's nice to know it isn't just me. my rig is full up with 80 hour work units and will not ask for more work. i will gladly leave if that is what they want.
ID: 1008528 · Report as offensive
Profile soft^spirit
Avatar

Send message
Joined: 18 May 99
Posts: 6497
Credit: 34,134,168
RAC: 0
United States
Message 1008533 - Posted: 26 Jun 2010, 6:37:40 UTC

one machine under estimating the time by 50% or so.. other is over estimating
the time by about 100%.

Has anyone else noticed the first few percent go really slow, then it speeds up for each unit?
Janice
ID: 1008533 · Report as offensive
Profile Pappa
Volunteer tester
Avatar

Send message
Joined: 9 Jan 00
Posts: 2562
Credit: 12,301,681
RAC: 0
United States
Message 1008535 - Posted: 26 Jun 2010, 6:54:12 UTC

In front of me I have one machine that has a ton of work and this one. This one as I looked DCF jumped from about 0.20000 to 3.6xxxx

With the current Servers changes it would appear that the FPOPS estimate has changed causing the local machine to adjust DCF.

Regards

Please consider a Donation to the Seti Project.

ID: 1008535 · Report as offensive
Profile soft^spirit
Avatar

Send message
Joined: 18 May 99
Posts: 6497
Credit: 34,134,168
RAC: 0
United States
Message 1008537 - Posted: 26 Jun 2010, 7:19:13 UTC

most of this falls under a very large category..

"stuff I do not understand".

Everything I take something off the list, a bunch more things get added.


Janice
ID: 1008537 · Report as offensive
Iona
Avatar

Send message
Joined: 12 Jul 07
Posts: 790
Credit: 22,438,118
RAC: 0
United Kingdom
Message 1008538 - Posted: 26 Jun 2010, 7:21:39 UTC

Hmmm. I've been getting some really strange estimated times on WUs, all week. I've had some that were as low as 3 mins and some, like the ones I'm currently going through, with 'times' of 17 hrs! The '17 hr' WUs are being crunched in about, 2 hrs.....



Don't take life too seriously, as you'll never come out of it alive!
ID: 1008538 · Report as offensive
Profile James Sotherden
Avatar

Send message
Joined: 16 May 99
Posts: 10436
Credit: 110,373,059
RAC: 54
United States
Message 1008543 - Posted: 26 Jun 2010, 7:41:27 UTC
Last modified: 26 Jun 2010, 7:42:24 UTC

I have an AP on my old P4 that started out as 320 hours to completion.Well it will finsh in about 53 hours wich is about normal with Opp apps. MY i7 GPU has times of 32 hours but being done in 30 Minutes with Opp Apps. Didnt we have this problem a while ago? Im not worried it will sort its self out,just throws you for a loop a bit.
[/quote]

Old James
ID: 1008543 · Report as offensive
Iona
Avatar

Send message
Joined: 12 Jul 07
Posts: 790
Credit: 22,438,118
RAC: 0
United Kingdom
Message 1008545 - Posted: 26 Jun 2010, 7:56:30 UTC - in response to Message 1008543.  

Normally, the estimated times are pretty close, but most of the WUs I've had this week, have been a 'country mile', out. The other snag is, these same times are used for the WU cache....thats a bit of a problem, if we're going to have longer 'outages'.



Don't take life too seriously, as you'll never come out of it alive!
ID: 1008545 · Report as offensive
Profile James Sotherden
Avatar

Send message
Joined: 16 May 99
Posts: 10436
Credit: 110,373,059
RAC: 54
United States
Message 1008547 - Posted: 26 Jun 2010, 8:04:35 UTC

I just looked at my GPU times and they have come down from 30 hours to 15 hours.
So things should work out in the end.
[/quote]

Old James
ID: 1008547 · Report as offensive
Profile Fred J. Verster
Volunteer tester
Avatar

Send message
Joined: 21 Apr 04
Posts: 3252
Credit: 31,903,643
RAC: 0
Netherlands
Message 1008550 - Posted: 26 Jun 2010, 8:28:47 UTC - in response to Message 1008547.  
Last modified: 26 Jun 2010, 8:31:36 UTC

Only on my Slowest CUDA rig, I have 15 MB WU's estimated 2 hours.
3 other are without SETI MB.
Ehh, jusst got 1 AP WU of 20 hours.

Getting hot, in the Lowlands :), gives my host some cooling down time....

Let them run for a hours, no WU's, then I just shut them down, saves a lot of electricity and don't have to use the AC.

And thus saves money, for other things........;-)
ID: 1008550 · Report as offensive
FiveHamlet
Avatar

Send message
Joined: 5 Oct 99
Posts: 783
Credit: 32,638,578
RAC: 0
United Kingdom
Message 1008552 - Posted: 26 Jun 2010, 8:34:48 UTC

I have GPU tasks on Rig on a Bench estimated at 60 hours infact they take 10 mins, also a couple of AP tasks esimated at 400 hours completed in 12 hours.
Where's the logic in that, as a consequence the rig is not calling for new jobs and will run out before it gets any.

Dave
ID: 1008552 · Report as offensive
Iona
Avatar

Send message
Joined: 12 Jul 07
Posts: 790
Credit: 22,438,118
RAC: 0
United Kingdom
Message 1008555 - Posted: 26 Jun 2010, 8:41:21 UTC - in response to Message 1008552.  
Last modified: 26 Jun 2010, 8:43:06 UTC

I have GPU tasks on Rig on a Bench estimated at 60 hours infact they take 10 mins, also a couple of AP tasks esimated at 400 hours completed in 12 hours.
Where's the logic in that, as a consequence the rig is not calling for new jobs and will run out before it gets any.

Dave



Indeed. Thats the problem I was trying to emphasise...then add in the factor of longer 'outages' and things could get 'ugly', with people increasing their cache sizes to compensate for those longer 'outages'. If push comes to shove, that big green button on the front of the PCs is going to be used a bit more.
Don't take life too seriously, as you'll never come out of it alive!
ID: 1008555 · Report as offensive
Profile Skywalker66 @ Berlin

Send message
Joined: 31 Jan 01
Posts: 78
Credit: 27,692,349
RAC: 0
Germany
Message 1008597 - Posted: 26 Jun 2010, 13:21:49 UTC

I have get Cuda´s with 9:30h now. That is very bad, before this coming big outtage. My Cache of my most Crunchers is very emty. So i get not enough work, to stay with work on the outtage.

So i must corrupted my flops to get enough. Every Day new things, so i must travel across my town to change in the configuration on my PC´s :-(
ID: 1008597 · Report as offensive
Profile perryjay
Volunteer tester
Avatar

Send message
Joined: 20 Aug 02
Posts: 3377
Credit: 20,676,751
RAC: 0
United States
Message 1008601 - Posted: 26 Jun 2010, 13:29:55 UTC

Be careful of messing with your cache size until this thing calms down and we can figure out where it is going. As someone said they are getting times on one machine that is way high and on another that are way low. It doesn't seem to matter what you are running and a bunch of shorties might trigger a response of the server thinking all your work can finish that fast. In other words, you set a 10 day cache while your estimates are way high then all of a sudden it goes the other way, you wind up with way more work than you can handle.(or is that the other way around? Whatever, you get my point!)


PROUD MEMBER OF Team Starfire World BOINC
ID: 1008601 · Report as offensive
Profile RottenMutt
Avatar

Send message
Joined: 15 Mar 01
Posts: 1011
Credit: 230,314,058
RAC: 0
United States
Message 1008669 - Posted: 26 Jun 2010, 17:40:09 UTC - in response to Message 1008601.  

I only have less then 10% of what i should have in my queue, I'm way under fed. It looks like it just isn't worth it any longer!
ID: 1008669 · Report as offensive
Josef W. Segur
Volunteer developer
Volunteer tester

Send message
Joined: 30 Oct 99
Posts: 4504
Credit: 1,414,761
RAC: 0
United States
Message 1008807 - Posted: 27 Jun 2010, 1:54:26 UTC - in response to Message 1008533.  

soft^spirit wrote:
...
Has anyone else noticed the first few percent go really slow, then it speeds up for each unit?

CUDA work definitely has that, the progress updates for Gaussian, Pulse, and Triplet searches were missing in the code from which those apps were built. Near the end there's a safety feature which forces it to catch up so it'll finish at 100%. The missing stuff has been added to the current code base, just hasn't been used for release builds yet.

Another thing you might be seeing is just a matter of not having accurate progress during startup and initialization. IIRC there's a rough guess for the baseline smoothing, but the optimum function testing is different for hosts with different capabilities so it would be more trouble than it's worth to try to even estimate what fraction of the run time it represents.

On CPU, there's also an effect due to using FFT lengths 65536 and 131072 only during the first part of a task (roughly up to 50% progress). The FFTs themselves are typically less efficient on CPUs than shorter lengths. The only searching done at those FFT lengths is the quick spike finding (and triplet finding for VLARs), so the FFT speed difference affects the progress more.
                                                                  Joe
ID: 1008807 · Report as offensive
Profile zoom3+1=4
Volunteer tester
Avatar

Send message
Joined: 30 Nov 03
Posts: 65709
Credit: 55,293,173
RAC: 49
United States
Message 1008808 - Posted: 27 Jun 2010, 2:19:56 UTC

I put the flops count tag back in My app_info file, Things look better now right now work looks like 1:14:36(HH:MM:SS) on one and that's dropping, It would be nice to be notified when the project has turned the flops calculator back on of course.
The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's
ID: 1008808 · Report as offensive
KB7RZF
Volunteer tester
Avatar

Send message
Joined: 15 Aug 99
Posts: 9549
Credit: 3,308,926
RAC: 2
United States
Message 1008815 - Posted: 27 Jun 2010, 3:01:51 UTC

I just got a bunch of work, and all of mine are as normal as they should be. Roughly 4 hours per unit, but running the optimized apps, so it'll be a little lower. But no weird times here. Running BOINC version 6.10.18, don't know if that has anything to do with it. Never got any weird messages either about the app_info.xml. Maybe I'm special??? LMAO!
ID: 1008815 · Report as offensive
1 · 2 · Next

Message boards : Number crunching : New MB apps have false completion times


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.