ECM users - version 30.9/30.10/30.11 (see post#195)

[cmpute]

Hi James, you can regard this as a special case. I was intentionally testing one of my computers which has a sign of bad mem stick. I intentionally didn't include the known factors in the assignment to test if the PM1 run can find all the factors. It turns out that it didn't. So currently I'm putting that computer to do ECMs and plan to abandon it someday this year.

I'm just too lazy to turn off the result uploading just for that one task.. I'm totally okay if the result is deleted from the database.. Sorry for making the confusion.. Normally this won't happen if the task is fetched from PrimeNet or mersenne.ca.

[cmpute]

Besides, I found the B2 calculation of PM1 in 30.11 a little confusing to me. I have an exponent with recommended B2 calculated in 30.10b4 with the following setting:

[Work thread Mar 20 20:39] P-1 on M59771 with B1=526208124, B2=TBD
[Work thread Mar 20 20:39] Using FMA3 FFT length 3K
[Work thread Mar 20 20:39] Inversion of stage 1 result complete. 5 transforms, 1 modular inverse. Time: 0.002 sec.
[Work thread Mar 20 20:39] Available memory is 57344MB.
[Work thread Mar 20 20:39] Switching to FMA3 FFT length 4K
[Work thread Mar 20 20:39] With trial factoring done to 2^66, optimal B2 is 20481*B1 = 10777268587644.
[Work thread Mar 20 20:39] If no prior P-1, chance of a new factor is 24.5%
[Work thread Mar 20 20:39] Estimated stage 2 vs. stage 1 runtime ratio: 2.617
[Work thread Mar 20 20:39] Using 57344MB of memory.  D: 64350, 7200x1799271 polynomial multiplication.
...
[Work thread Mar 20 20:39] Stage 2 init complete. 255457 transforms. Time: 2.582 sec.
...

Then I update mprime from 30.10b4 to 30.11b1, and continue with the stage 2. The stage 2 takes nearly an hour:

[Worker Mar 20 21:38] M59771 stage 2 at B2=10049888984850 [91.62%].  Time: 44.008 sec.
[Worker Mar 20 21:38] M59771 stage 2 at B2=10164744725850 [92.67%].  Time: 44.022 sec.
[Worker Mar 20 21:39] M59771 stage 2 at B2=10279600466850 [93.71%].  Time: 44.000 sec.
[Worker Mar 20 21:40] M59771 stage 2 at B2=10394456207850 [94.76%].  Time: 44.161 sec.
[Worker Mar 20 21:41] M59771 stage 2 at B2=10509311948850 [95.81%].  Time: 44.075 sec.
[Worker Mar 20 21:41] M59771 stage 2 at B2=10624167689850 [96.85%].  Time: 43.977 sec.
[Worker Mar 20 21:42] M59771 stage 2 at B2=10739023430850 [97.90%].  Time: 43.963 sec.
[Worker Mar 20 21:43] M59771 stage 2 at B2=10853879171850 [98.95%].  Time: 43.990 sec.
[Worker Mar 20 21:43] M59771 stage 2 complete. 1128089145 transforms. Total time: 3478.496 sec.
[Worker Mar 20 21:43] Stage 2 GCD complete. Time: 0.001 sec.
[Worker Mar 20 21:43] M59771 completed P-1, B1=526208124, B2=10968734912850, Wi8: FB449F91

Afterward, I run PM1 stage 2 on the next exponent with similar magnitude, this time the optimal B2 is more than 20 times bigger:

[Worker Mar 20 21:43] P-1 on M59887 with B1=525048872, B2=TBD
[Worker Mar 20 21:43] Using FMA3 FFT length 3K
[Worker Mar 20 21:43] Inversion of stage 1 result complete. 5 transforms, 1 modular inverse. Time: 0.001 sec.
[Worker Mar 20 21:43] Available memory is 57344MB.
[Worker Mar 20 21:43] Switching to FMA3 FFT length 4K
[Worker Mar 20 21:43] With trial factoring done to 2^66, optimal B2 is 563199*B1 = 295706999661528.
[Worker Mar 20 21:43] If no prior P-1, chance of a new factor is 28.6%
[Worker Mar 20 21:43] Estimated stage 2 vs. stage 1 runtime ratio: 1.786
[Worker Mar 20 21:43] Using 57344MB of memory.  D: 3573570, 322560x1527820 polynomial multiplication.
...
[Worker Mar 20 20:46] Stage 2 init complete. 255455 transforms. Time: 2.406 sec.

And this time it takes 400 secs less,

[Worker Mar 20 22:32] M59887 stage 2 at B2=277411096732770 [93.25%].  Time: 32.769 sec.
[Worker Mar 20 22:32] M59887 stage 2 at B2=280565486971770 [94.38%].  Time: 32.812 sec.
[Worker Mar 20 22:33] M59887 stage 2 at B2=283719877210770 [95.50%].  Time: 32.782 sec.
[Worker Mar 20 22:33] M59887 stage 2 at B2=286874267449770 [96.62%].  Time: 32.766 sec.
[Worker Mar 20 22:34] M59887 stage 2 at B2=290028657688770 [97.75%].  Time: 32.870 sec.
[Worker Mar 20 22:34] M59887 stage 2 at B2=293183047927770 [98.87%].  Time: 32.781 sec.
[Worker Mar 20 22:35] M59887 stage 2 complete. 631062905 transforms. Total time: 2923.268 sec.
[Worker Mar 20 22:35] Stage 2 GCD complete. Time: 0.009 sec.
[Worker Mar 20 22:35] M59887 completed P-1, B1=525048872, B2=296337438166770, Wi8: 8E92E947

How does this make sense? Much larger B2 but faster calculation (with all the other settings left the same)

[Andrew Usher]

You did find the factors expected in stage 1, but not the one in stage 2 - it's not surprising a memory problem would show in stage 2. I would expect to see 'MISSED FACTOR' on mersenne.ca, but I guess that's only generated when the factor is found later and not earlier (that's when I see it).

Special case or no, your test did show an issue with automatic PRPs - that they run regardless of the actual list of known factors, which is a nuisance at best and at worst (when two people factor an exponent close together) fails to accomplish the purpose of giving factor-finders the chance to get the PRP credit. James's proposal seems to only way to fix this - one can assume that the people benefitting from this are likely to be connected - and further the PRP should not run if no _new_ factor was found.

[James Heinrich]

Originally Posted by Andrew Usher

I would expect to see 'MISSED FACTOR' on mersenne.ca, but I guess that's only generated when the factor is found later and not earlier

If the NF run happens earlier than the factor discovery date, the factor is considered "missed". If the factor is known before the NF run it is assumed to be "known-and-skipped".

[Prime95]

Originally Posted by cmpute

I'm trying to running P+1 tasks, and sometimes it crashes before stage 2.

Will be fixed in 30.12.

I would not invest a lot of effort in P+1. If I get around to implementing polynomial stage 2 then P+1 would be more promising.

[cmpute]

Thanks! Appreciate it.