Faulty BRP task erroneously validated?

Ubik
Ubik
Joined: 30 Oct 09
Posts: 9
Credit: 6020737
RAC: 9716
Topic 198184

One of my Android devices recently froze. After restart it sent in one result as completed, even though the progress bar before the freeze reported 20%. Correspondingly, the reported CPU time is about 20% of the average time for this machine. Surprisingly, the task was nevertheless successfully validated (against an opencl-intel_gpu result).

I was wondering whether there could be a bug at work.
The relevant task is
p2030.20141113.G190.52-01.83.C.b5s0g0.00000_3280_1
from workunit 225002289

Holmis
Joined: 4 Jan 05
Posts: 1118
Credit: 1055935564
RAC: 0

Faulty BRP task erroneously validated?

Looking at the stderr for task#513670130 belonging to WU#225002289 it seems the task did finish normally but was restarted once after the science app had call boinc_finish. When restarted the output file for the completed tasks was found so no need to reprocess.

The progress report shown i Boinc was probably not accurate due to the crash/freeze and that goes for the reported run time as well. But everything else points to the task being completed normally as the stderr looks ok and the task was validated.
I'd say everything is ok and nothing to worry about even though the crash/freeze did somehow effect the reported run time.

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 713539007
RAC: 898105

We have some rather

We have some rather pedantic/paranoid safeguards in place to prevent validation of partially completed results.

First, the result file itself is only generated at the end of computation, intermediate results are stored in checkpoint files but those are different from the final result file so there is not even a theoretical possibility that BOINC could somehow upload something that is meaningful to the validator before the computation is actually finished.

Also, the result file itself contains a special 'end of file' marker in the last line to guard against truncation errors or files being sent before writing the file is finished.

And then of course there is the comparison to a second hosts's results (which might be affected by the same problem tho, so the measures described before are not entirely useless).

HB

Ubik
Ubik
Joined: 30 Oct 09
Posts: 9
Credit: 6020737
RAC: 9716

Thanks for checking!

Thanks for checking!

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.