Why are certain people getting massive credits for failed tasks running on ARM processors where e.g., all tasks are "completed too late to validate"?
For example, Sgt.Sausage (ID 40904) gets all tasks to complete in under 2-3 seconds (please - on an ARM CPU!), then marked too late to validate, then gets full credit and for no real work done. I have overclocked, quad core, development ARM equipment and in 2-3 seconds there cannot ever be any meaningful work done.

Something's very, very wrong here.

Just look at the stderr output from his workunits, it's a badly broken ARM client in my opinion.
I don't see "massive" credits gained, most probably due to weird bugs his client does not report correct CPU and runtime, one of his "fastest" ARM has a performance level barely near a Celeron 300 running opt app.

Anyway I'll check the workunits and results just in case. Something is so wrong there that I wouldn't be surprised if it somehow breaks the validator logic. "Too late" results shouldn't be validated at all.
I might also consider banning the bugged client.

---

M4 Project homepage
M4 Project wiki

Thanks for that.
Very reassuring.

Clearly such faults if unreported would bring enigma@home into disrepute here and in the wider BOINC community.

I just checked the results as previously described ... it appears that the cleanup has started. Previously 90 tasks on one ARM device in one day, all failed as described elsewhere, 7,200+ credits awarded for no actual work.

it's a badly broken ARM client in my opinion

maybe ... or far worse - a badly and publicly visible failure of the project to spot obvious erm, tricks?

(FWIW I am old so I trust my gut feel:- I kinda know the impossible when I see it. Then I check it with equipment I can easily afford to buy & time I have to spare in abundance. Think, unpaid watchers. You're welcome BTW.)

The evidence - shown publicly for ARM device(s*) of Sgt.Sausage:-
Date ......... Credits
2012-11-23 5681.00
2012-11-22 8536.00
2012-11-21 12869.00
2012-11-09 12600.00

No amount of ARM devices (single/dual/quad core) has ever recorded anything near these figures. Sgt.Sausage shows many devices (*check?) with the same specification - 1512mhz, dual core. I have a quad core running at 1600mhz - nowhere near those figures.

Actually some of his results were valid and ARMs are not as slow as you think, go to the cpu/os stats at the main page to see for yourself.

Anyway I've just upgraded validator so now it doesn't tolerate a slighest corruption in the output data and/or templates, eventually those broken ARM clients will have their daily quota close to 0. Their output is so badly corrupted that it even affects other hosts crunching the same WU.

---

M4 Project homepage
M4 Project wiki