Pulsar
27 Aug 2012 0:28:02 UTC
Topic 196496
(moderation:
while this file was running i saw the white lines in the spectrum...google search suggested it was sign of pulsar find... but the file was not validated ... its marked invalid...?? any particular reason??? p2030.20111005.G64.25-00.28.N.b1s0g0.00000_224
Pulsar
)
I dont have access to the extended results (and neither the knowlodge to interpret them) so I cant say you specifically why it was marked as invalid, but, in general, all the work units are sent initially to 2 different users, when both results are returned the server compares them and if they are equal (or there are only minimal differences due to rounding errors) then both are marked as valid. If they are not enough equal, then the same work unit is sent to a third user and at the return of this third result all of them are compared, and depending on the result of that comparison it could happen that all of them get marked as valid or just 2 of them or even, that a fourth result is needed...
For example, if the first 2 usrs return lets say a "4" and a "6" and the maximun acceptable difference were 1, a third result is needed, and then if the third returns a "5" then all three will become valid, but if the third one returns "3" or "4" then the one that returned "6" will be marked as invalid, it the third result is a "2" or a "7" then a fourth result will be needed. (Of course results are not just one integer number)
AFAIK, (bu not sure how much of this is valid on Einstein) data with a lot of radio frequency interference is the most common reasson for highly noticeable patterns and also those noisy signals are the ones that have more chances to lead to different results on different hardware.
My application was :Binary
)
My application was :Binary Radio Pulsar Search (Arecibo) v1.22 (BRP4SSE)
while others were
Binary Radio Pulsar Search (Arecibo) v1.25 (BRP4cuda32nv301)
could it be case of simple mismatch of application?
http://einstein.phys.uwm.edu/
)
http://einsteinathome.org/workunit/129803839
Usually different apps for the same type of task (BRP4) running on different devices (CPU, GPU) have to give (are designed to give) similar results.
I can't say if yours or the 2 CUDA results are 'wrong', someone familiar with the apps have to look.
[pre] [/pre]
And who that might be?... any
)
And who that might be?... any idea?
RE: while this file was
)
What 'spectrum'? Are you referring to the output of the screensaver graphics? Please be aware that the graphics app produces images that have no relationship to the calculations being performed on the current workunit.
What google search? Can you give a link? At the time tasks are being crunched, it's not possible to know or see any indication of any sort of a 'find'. The data that is sent back to the project has to undergo extensive post-processing to see if anything of interest has been 'found'.
Unfortunately, not all successfully completed results are found to be valid. Most often, this is due to overstressed hardware in the machine doing the crunching - things like excessive heat, flaky memory, overclocking, non-clean power (ripple and noise), faulty capacitors, etc. Is your machine a laptop? Have you checked that the airflow is good - no blockages due to dust and/or lint? Do you check operating temperatures? Crunching puts a lot of stress on laptops.
A much smaller number of invalid results may be due to differences in the platforms on which the results were calculated - particularly where a 'CPU' result is being compared with a 'GPU' result. The validator tries to make allowances for these differences and is usually quite good at doing so. It's possible that your result was just not quite close enough to the other two.
The Admins do not have the time to manually investigate individual results when you consider the large numbers of these flooding in every day. If you continue to see invalid results and you are troubled by these and you are sure your hardware is OK, perhaps you could consider adjusting your preferences so as not to receive BRP4 tasks. These do have the highest invalid rates - take a look at the server status page to get a feeling for this. Currently in the online database there are over 3600 invalid BRP4 tasks but only 67 invalid S6LV1 tasks.
Thanks for your contribution and good luck with your future crunching.
Cheers,
Gary.