Tiny Credit for Gamma-ray pulsar binary search #1 on GPUs
13 May 2017 0:17:06 UTC
Topic 207592
(moderation:
Three GPU-based work units that took forever to receive a validation from a wingman only provided 2.19 credits to me, instead of the customary 3465 credits, which the wingman received. Hopefully these will not be purged before you can review them. Any idea why this happened?
https://einsteinathome.org/workunit/286377953
https://einsteinathome.org/workunit/287683699
https://einsteinathome.org/workunit/288576697
Prompted by your post, I
)
Prompted by your post, I examined the last pages of my Valid tasks list, and found about a dozen of these for me.
Common elements of mine:
1. application is Gamma-ray pulsar binary search #1 on GPUs v1.20 (FGRPopencl1K-nvidia) windows_x86_64
2. sent to me between March 24, 2017 and April 17, 2017
3. I returned them between March 24, 2017 and April 19, 2017
4. one or more quorum partners errored out or timed out
5. the validating quorum partner returned the work between May 3 and May 10
6. my validating quorum partner received 3,465 credit, and I received 2.19 credit
7. one or more of my quorum partners who failed to validate also received 2.19 credit (including error while computing, timed out, not started by deadline-cancelled)
Examples:
Example 1
Example 2
Example 3
Example 4
Prompted by both posts, I
)
Prompted by both posts, I started looking through the valid tasks for all my fleet. I didn't initially pick up on the significance of Peter's comment about "last pages" so I started going back from the latest validated tasks. After more than 20 pages without seeing anything unusual, the penny dropped and I decided to jump to the last page without knowing how many that would be (I had imagined perhaps a couple of hundred pages). I got quite a shock to see it was page 1424 and there they all were, showing the 2.19 credit awards just as documented above. There were approximately 3 full pages worth. Who knows how many more there might have been that have already been removed from the database.
There were examples for both AMD and NVIDIA GPUs as well as for CPU tasks and all of them got 2.19 credits. I didn't examine more than just a couple of quorums but the ones I did fitted in with Peter's thorough description of the characteristics. Hopefully a staff member might be able to work out what happened and fix it so it wont happen again. I don't care at all about the credit. It's just the peace of mind that bugs are fixed and that things work as they're supposed to.
EDIT: In case it's useful, this is a CPU task quorum that shows the problem.
Cheers,
Gary.
This is the result of a
)
This is the result of a hickup I had while manually granting credit for the aborted O1Spot1T testrun. I aborted the script as soon as I realized my error. I investigated the impact and there was no harm done.
Only old tasks were affected that already got the right amount of credit. So what the script did was granting 2.19 Credits again for tasks that were already validated or had errors. It did not subtract the previous credit as that was already granted before yesterday evening. In the end everyone affected got 2.19 credits extra per task.
Sorry for the confusion.
It's been quite a day today
)
It's been quite a day today :-). With GPU tasks going AWOL and strange credit awards ....
Thanks for fixing things for us. I'm sure there are quite a few people happy to see the work flowing again :-).
Cheers,
Gary.
Thanks for the explanation,
)
Thanks for the explanation, and for always being responsive.
Woohoo I got an extra 2.19
)
Woohoo I got an extra 2.19 points!