Strategic Double Clicking

[ATH]

Originally Posted by storm5510

M59522479 remains "Unverified" so someone else will have to run it again.

Your result was correct: https://mersenne.org/M59522479

[storm5510]

Originally Posted by ATH

Your result was correct: https://mersenne.org/M59522479

I ran another since and it yielded a matching result, Thank you for your efforts!

[PBMcL]

Sorry in advance if this is the wrong thread for this. A couple of days ago the server assigned 40631933 to me for double check. Yesterday an anonymous user turned in a verification for this exponent. However, it is still in my local file as the next assignment. What is the proper procedure to clear this and get a new assignment? Forcing manual communication with the server had no effect.

[Mark Rose]

Originally Posted by PBMcL

Sorry in advance if this is the wrong thread for this. A couple of days ago the server assigned 40631933 to me for double check. Yesterday an anonymous user turned in a verification for this exponent. However, it is still in my local file as the next assignment. What is the proper procedure to clear this and get a new assignment? Forcing manual communication with the server had no effect.

Stop/quit Prime95/mprime. Remove it from your worktodo.txt and delete the temporary files for the assignment. Start Prime95/mprime.

[PBMcL]

Thanks, Mark!
--Phil McL.

[Madpoo]

Here's an updated list of potentially bad results to do DC's on. The "bad/good" values shown are just for the particular month for that CPU.

exponent	Bad	Good	worktodo
50895409	2	1	DoubleCheck=50895409,73,1
57805271	2	0	DoubleCheck=57805271,73,1
59133343	2	0	DoubleCheck=59133343,73,1
55232743	2	1	DoubleCheck=55232743,73,1
55570961	2	1	DoubleCheck=55570961,73,1
55858961	2	1	DoubleCheck=55858961,73,1
55859129	2	1	DoubleCheck=55859129,73,1
55864343	2	1	DoubleCheck=55864343,73,1
55868689	2	1	DoubleCheck=55868689,73,1
56084641	2	1	DoubleCheck=56084641,73,1
56084689	2	1	DoubleCheck=56084689,73,1
56535389	2	1	DoubleCheck=56535389,73,1
58314797	2	1	DoubleCheck=58314797,73,1
58317373	2	1	DoubleCheck=58317373,73,1
58317383	2	1	DoubleCheck=58317383,73,1
58317449	2	1	DoubleCheck=58317449,73,1
58317481	2	1	DoubleCheck=58317481,73,1
58317509	2	1	DoubleCheck=58317509,73,1
58317629	2	1	DoubleCheck=58317629,73,1
58317823	2	1	DoubleCheck=58317823,73,1
58317979	2	1	DoubleCheck=58317979,73,1
45933301	1	0	DoubleCheck=45933301,73,1
46055543	1	0	DoubleCheck=46055543,72,1
55360093	1	0	DoubleCheck=55360093,73,1
50543137	1	0	DoubleCheck=50543137,73,1
57380027	1	0	DoubleCheck=57380027,73,1
58243741	1	0	DoubleCheck=58243741,73,1
57405221	1	0	DoubleCheck=57405221,73,1
50892319	1	0	DoubleCheck=50892319,73,1
46451477	1	0	DoubleCheck=46451477,72,1
59932349	1	0	DoubleCheck=59932349,73,1
57338317	1	0	DoubleCheck=57338317,73,1
59150717	1	0	DoubleCheck=59150717,73,1
58845481	1	0	DoubleCheck=58845481,73,1
59574397	1	0	DoubleCheck=59574397,73,1
59342483	1	0	DoubleCheck=59342483,73,1
59342489	1	0	DoubleCheck=59342489,73,1
59343653	1	0	DoubleCheck=59343653,73,1
48733831	1	0	DoubleCheck=48733831,72,1
50113937	1	0	DoubleCheck=50113937,73,1
54364027	1	0	DoubleCheck=54364027,73,1
56069149	1	0	DoubleCheck=56069149,73,1
45769877	1	0	DoubleCheck=45769877,72,1
59693677	1	0	DoubleCheck=59693677,73,1
55132873	1	0	DoubleCheck=55132873,73,1
59870053	1	0	DoubleCheck=59870053,73,1
50397913	1	0	DoubleCheck=50397913,73,1
57436123	1	0	DoubleCheck=57436123,73,1
57441157	1	0	DoubleCheck=57441157,73,1
57966593	1	0	DoubleCheck=57966593,73,1
57966613	1	0	DoubleCheck=57966613,73,1
57966761	1	0	DoubleCheck=57966761,73,1
50885231	1	0	DoubleCheck=50885231,73,1
50890663	1	0	DoubleCheck=50890663,73,1
50891573	1	0	DoubleCheck=50891573,73,1
53454707	1	0	DoubleCheck=53454707,73,1
53461313	1	0	DoubleCheck=53461313,73,1
53552483	1	0	DoubleCheck=53552483,73,1
53591299	1	0	DoubleCheck=53591299,73,1
53611007	1	0	DoubleCheck=53611007,73,1
53653813	1	0	DoubleCheck=53653813,73,1
53661637	1	0	DoubleCheck=53661637,73,1
52890251	1	0	DoubleCheck=52890251,73,1
52898959	1	0	DoubleCheck=52898959,73,1
52907317	1	0	DoubleCheck=52907317,73,1
52939609	1	0	DoubleCheck=52939609,73,1
52940297	1	0	DoubleCheck=52940297,73,1
52975843	1	0	DoubleCheck=52975843,73,1
52977017	1	0	DoubleCheck=52977017,73,1
52982513	1	0	DoubleCheck=52982513,73,1
53488879	1	0	DoubleCheck=53488879,73,1
50178343	1	0	DoubleCheck=50178343,73,1
50548907	1	0	DoubleCheck=50548907,73,1
54357169	1	0	DoubleCheck=54357169,73,1
54710561	1	0	DoubleCheck=54710561,73,1
54750329	1	0	DoubleCheck=54750329,73,1
54884051	1	0	DoubleCheck=54884051,73,1
55919653	1	0	DoubleCheck=55919653,73,1
58998371	1	0	DoubleCheck=58998371,73,1
54481951	1	0	DoubleCheck=54481951,73,1
54481969	1	0	DoubleCheck=54481969,73,1
54483907	1	0	DoubleCheck=54483907,73,1
54524461	1	0	DoubleCheck=54524461,73,1
54537739	1	0	DoubleCheck=54537739,73,1
54538381	1	0	DoubleCheck=54538381,73,1
54626917	1	0	DoubleCheck=54626917,73,1
54631123	1	0	DoubleCheck=54631123,73,1
56047373	1	0	DoubleCheck=56047373,73,1
56069933	1	0	DoubleCheck=56069933,73,1
56074229	1	0	DoubleCheck=56074229,73,1
56077699	1	0	DoubleCheck=56077699,73,1
56078719	1	0	DoubleCheck=56078719,73,1
59690161	1	0	DoubleCheck=59690161,73,1
59708153	1	0	DoubleCheck=59708153,73,1
59711929	1	0	DoubleCheck=59711929,73,1
59712101	1	0	DoubleCheck=59712101,73,1

[storm5510]

@Madpoo. I was just looking at the current status of M59712101, (https://www.mersenne.org/report_expo...9712101&full=1), which appears in your current list. Other than never having a DC ran in over four years, I do not see anything unusual in the detail. Is there something else that made you feel it was suspect?

[S485122]

Originally Posted by storm5510

...
Other than never having a DC ran in over four years, I do not see anything unusual in the detail. Is there something else that made you feel it was suspect?

If you read the post where that exponent was cited :

Originally Posted by Madpoo

Here's an updated list of potentially bad results to do DC's on. The "bad/good" values shown are just for the particular month for that CPU.

you conclude that the reason that exponent was listed is that the same CPU turned in a bad result the same month it returned the result for the cited exponent.

Jacob

[storm5510]

Originally Posted by S485122

If you read the post where that exponent was cited :you conclude that the reason that exponent was listed is that the same CPU turned in a bad result the same month it returned the result for the cited exponent.

Jacob

I asked because said exponent has only one LL test. Now, if the same user has other results which were bad, then I see the reason. Otherwise, no.

[kriesel]

Originally Posted by storm5510

I asked because said exponent has only one LL test. Now, if the same user has other results which were bad, then I see the reason. Otherwise, no.

It may not be on a per-user basis, but on a per-device basis or per-instance basis of prime95, cudalucas, virtual core # in manual assignment... Madpoo wrote "for that CPU". Some users run multiple systems. Some systems have multiple cpus. Some (most now) cpus have multiple cores. Prime95 supports multiple worker threads using multiple or single cores each.

I assume Madpoo meant for that primenet computer name (which is prime95 computer name, or all manually reported results lumped under the name Manual test regardless of source for that user); not sure what misfit does, but it's not supported for LL testing; other scripting behavior unknown).

Tracking by system, bad results, can help identify systems less reliable so the issue may be resolved or a failing system removed from service.

What's the current primenet average percentage of bad LL test residues?

[Madpoo]

Originally Posted by storm5510

I asked because said exponent has only one LL test. Now, if the same user has other results which were bad, then I see the reason. Otherwise, no.

Yeah, that's the point of this little project, is finding the exponents only tested once and try to predict if that solo test was good or bad.

By looking at the CPU's history of bad/good results, we can make some educated guesses on these unknowns.

Without having to re-read all the previous posts, I'll summarize a few of my methods:

• Look at the overall history of that CPU (not per user as kriesel pointed out, I mainly look at the individual CPU) and figure the bad/good ratio
• Look at the history of the CPU broken down by month & year (some systems may be fine much of the time but had periods where they put out bad results)
• Systems where no double-check has ever been done on any of their past results, so they're a total unknown ... preference given when the *user* has a generally bad track record over all of their systems combined although that's not always predictive, but sometimes it's been useful.

Obviously, when the bad/good ratio for a CPU (whether for it's lifetime or on a monthly basis) is high, we've seen our strategic double-checks mismatch that first check in excess of 50% of the time. This is a weird project where we actually hope to get mismatches, because we're hoping to find the bad runs from the past.

When doing the more speculative testing, like doing at least one DC for any system that hasn't had one, even when the user has a generally bad track record I haven't really seen a mismatch rate any higher than the overall average. That said, it's still useful because when we do find those new mismatches, it puts that CPU on our radar and we can do further mining of it's other results.

One of my side projects, that is never-ending, is doing daily checks for any exponents that need a triple-check and getting it done. Besides the ones where I've already done one of the checks, I think there are only about 65 or so outstanding. I've got something like 28 of them queued up and the other 37 are assigned to someone else.

I've even poached some of those triple-check assignments that were abandoned or hadn't progressed at all since being assigned > 30 days ago (usually they're in the "churn zone" in the 45M range, picked up by anonymous users, probably doing stress testing and quitting).

Those triple-checks have helped pick the winner/loser in the previous mismatches. My queries tried to guess the winner in those cases, based on the record of each system involved, but it was only a guess and knowing the real winner/loser is much better.