Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummers-lowrise
|
Message boards :
Seventeen or Bust :
Bust 5.11 versus 6.05
| Author |
Message |
|---|
|
|
|
LS,
Can someone tell why 2 crunched SOB17's under 5.11 wille be granted credit and another crunched under 6.05 will not be credited and v.v.???
Berend | |
|
John Honorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
                 
|
Can someone tell why 2 crunched SOB17's under 5.11 wille be granted credit and another crunched under 6.05 will not be credited and v.v.???
App versions should not have any impact on results. It could be just coincidence as there are other WU examples of 6.05/5.11 both receiving credit. The 6.05 host may simply have returned an invalid result. However, I'll investigate further. Do you have a WU ID?
____________
| |
|
|
|
|
Hi John,
Sorry to react a little slow but I was outdoors.
Yes, I do have a WU: llr_sob_47466789
In this case 2 systems running sob v5.11 did get credit and mine, sob v6.05, didn't although it finished without an error-code. Little wast those 800.000 secs.
There is a second, still waiting for checking: llr_sob_47466438
This WU is now crunched by a sob v5.11 and because most systems use this program, probably also this WU won't get any credits.
One was granted credit in which case the system (http://www.primegrid.com/show_host_detail.php?hostid=131034) running a sob v5.11 didn't get any credit: llr_sob_47466406
Nevertheless, no hard feelings. When a new app is added, thing's like this can happen
Greetings, Berend and thanks for coming back. | |
|
 Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14037
ID: 53948
Credit: 477,382,143
RAC: 302,741
                               
|
Hi John,
Sorry to react a little slow but I was outdoors.
Yes, I do have a WU: llr_sob_47466789
In this case 2 systems running sob v5.11 did get credit and mine, sob v6.05, didn't although it finished without an error-code. Little wast those 800.000 secs.
There is a second, still waiting for checking: llr_sob_47466438
This WU is now crunched by a sob v5.11 and because most systems use this program, probably also this WU won't get any credits.
One was granted credit in which case the system (http://www.primegrid.com/show_host_detail.php?hostid=131034) running a sob v5.11 didn't get any credit: llr_sob_47466406
Nevertheless, no hard feelings. When a new app is added, thing's like this can happen
Greetings, Berend and thanks for coming back.
If you take a look at the WU's for the host you listed, out of the 18 total WUs currently assciated with that system, 14 have completed. Of those 14, one errored while computing, and three were marked as invalid.
All of those were. v5.11. So 5.11 is obviously faulty, right?
Not necessarily. If you look at those "marked as invalid" WUs, you'll see that in this one both of the valid results were from 5.11 (and there was an error from a computer running 6.03), in this one the valid results are from 5.11 and 6.03 and the errors/invalids are also from both 5.11 and 6.03, and only in the one you used as an example are the 6.03s valid and the 5.11s invalid.
This particular computer has had a high percentage of errors: 4 out of 14. It's an example of a computer that isn't running LLR very stably, and isn't necessarily indicative of a faulty app. LLR seems to be an application that stresses CPs to a degree that most software does not, and some computers that normally don't have issues do have trouble running LLR. (Why is a long, long explanation; if you search through the list of my posts for a really long posting you'll find it pretty quickly.)
I might be temped to say that LLR, regardless of version, is buggy, if otherwise perfectly running computers have problems running this one program. That's a very reasonable conclusion. But if you look at the results across many computers, a pattern emerges. Some computers generate a lot of errors with LLR and others rarely, if ever, have errors. But there's no pattern as to which kind of computer has errors. There's no correlation to operating system, manufacturer, CPU type, amount of memory, or anything else. There simply are computers that are having trouble running LLR, and others that do not.
In effect, LLR acts as a CPU stress test, causing the CPU to run hotter than other programs do, and revealing errors where other programs -- including other BOINC programs that also run 24/7 -- do not. Most of the time, the computers having trouble with LLR show no other signs of not operating correctly, and do everything else without problem.
EDIT: BTW, I'n VERY skeptical of anyone who says "nah, the software is fine, your hardware must be the problem." Example:
CPDN has been saying that for years. Their WUs are extremely long running, and many people have errors. CPDN claims that their software is fine, so everyone with an error must have faulty hardware. The problem, however, is that if you look at the computers with errors, you don't see a pattern where some computers have errors and others don't. It's just kind of random. That points to it being a software problem rather than being hardware. And it is -- but as it turns out, it's not CPDN's application that's at fault. So they're right when they say their application is not causing the errors. In their case, it's actually a problem with the BOINC software. Under Windows (I don't know about other systems), sometimes during system shutdown, BOINC doesn't close the clients correctly, and when they restart when the system reboots, the tasks error out. That can happen with any BOINC project, including here at PrimeGrid. If you see a task with a significant amount of processing time marked as "Error while computing", this is one possible cause of that error.
(For this reason, it's always a good idea to shut down the BOINC client prior to shutting down the computer.)
At PrimeGrid, the "marked as invalid" problem seems to be with specific computers, rather than a systemic problem affecting everyone.
Regards,
Mike
____________
My lucky number is 75898524288+1 | |
|
mackerelVolunteer tester
Send message
Joined: 2 Oct 08
Posts: 2652
ID: 29980
Credit: 570,442,335
RAC: 10,182
                  
|
|
A computer that is unstable with LLR is plain unstable, even if it appears ok with 99% of everything else. A properly working computer should not fault over load periods. I know you can get into various external events that will make a fault free computer give bad data, but that is the rare exception.
On any new build (overclocked or not) I will at a minimum give some hours of ram and CPU stress testing for confidence it is functioning correctly before serious crunching. I have come across one case where I would get one error in memory stress testing on average every few hours. This might seem insignificant, but in normal use it was enough to give the odd random crash. If I was doing SoB in that state, that would be wasted effort. Once corrected it would work fine as long as I can be bothered to stress it.
The other thing is, you can't assume a good computer remains a good computer. Components do age. Dust build up and heat doesn't help. So it is worth keeping an eye on error rates and see if something might have changed.
Could I say that LLR implementations are all perfect? No. But they are relatively simple and if there is a real fault they likely would have been spotted by now.
If you want something possibly even more stressful than LLR/Prime95 type loads, look into the various linpack implementations, link LinX. I don't know exactly what they do, but they make the CPU run several degrees hotter than LLR!
To end, I can't answer the orignial question on SoB results. Personally I would first make sure the computer really is stable. In that case, the fault should lie elsewhere. | |
|
|
|
|
Actually I pity anyone that is running SoBs with Windows because I have seen a large number of failures with hosts running Windows (mine included). It is easy to suggest reasons like design and length of support for multiple cores, multiple users and x64 architecture but I know that some failures occur with other operating systems and that some Windows hosts are successful. Thus, I try to avoid SoBs on my Windows hosts.
| |
|
|
|
|
yeah i got to the point it was saying a year or two to finish.... and what i read here to. so o cancel it till they fix the bugs on it | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14037
ID: 53948
Credit: 477,382,143
RAC: 302,741
|
yeah i got to the point it was saying a year or two to finish.... and what i read here to. so o cancel it till they fix the bugs on it
There aren't any bugs in it, at least that I'm aware of. I've run enough of them now to get a silver badge, without any errors at all, so I've got no qualms about saying that the application is fine. It's the same LLR program that's used for half of the subprojects here, after all.
The only thing different about it is that it's a MUCH longer task because 'n' is in the 18,000,000 range. That's some 18 days worth of processing on a 2.4 GHz Core 2. Your computers are maybe 50% of that speed. That assumes crunching 24/7 at full power and 100% CPU share.
The run time estimation -- well, I could talk for hours and hours about that. Suffice it to say that there are times when the estimate will be accurate and times when it will be nonsense, but the estimate doesn't affect the actual run time. You can ignore it if it's not making sense.
____________
My lucky number is 75898524288+1 | |
|
|
|
|
Ho, ho,
I'm only trying to find out what the reason is why this happens.
The PC I'm running is an Asus K50IN laptop aged 11 mounths.
It's up most of the time for server purposes, running SuSE 11.3
I am not in for the points (900.000 in 4 years isn't that much).
The only reason I posted this message was to find out why and maybe help the project to resolve probable discrepancies in app versions.
I'll come back on the subject when the last wu, llr_sob_47466438 will be crunched (this one is now finished and waiting to be checked).
Thanks for the response this far, good crunching to all of you.
Berend | |
|
|
|
|
The SOB site has finally brought out faster clients. Any chance of rebuilding apps for BOINC?
____________
| |
|
Message boards :
Seventeen or Bust :
Bust 5.11 versus 6.05 |
