P-1 factoring anyone?

[James Heinrich]

Should there be any cause for concern from people whose P-1 factor-finding seems to deviate considerably from the expected rate? For example:
http://v5www.mersenne.org/report_top_500_P-1/
"Richard B. Woods" and "MCAS" are high on the top-500 and have checked a decent number of exponents but haven't found any factors (certain far from the expected ~5% rate). A couple others (e.g. "axn", "id_warrior") have only found 1 factor out of many attempts, again with far-below-expected success rate. Do these users have abnormally bad luck in being assigned exponents, or are the exponents being P-1 poorly in some way that prevents factors from being found that should be?

On the other end, there are also a few users who have a much-better-than-expected factor-finding rate (the king of which is George, at 85% success), but I'm less concerned about finding lots of factors than too few.

[MatWur-S530113]

Hello,

I don't think that you can compare George's finding-rate with someone else. I think often results (with factors) are reportet to him, which he added manually to the server.
But the king of factor finding is Will Edgington, have a look at the ECM's top-producer page: 1151 attempts, 1151 successes
*That* is effective

best regards,

Matthias

[Uncwilly]

Originally Posted by James Heinrich

Should there be any cause for concern from people whose P-1 factor-finding seems to deviate considerably from the expected rate? For example:
http://v5www.mersenne.org/report_top_500_P-1/
"Richard B. Woods" and "MCAS" are high on the top-500 and have checked a decent number of exponents but haven't found any factors (certain far from the expected ~5% rate).

Cheesehead (Richard)'s data are false. You can find the discussion in the forum someplace.

[petrw1]

Originally Posted by James Heinrich

Should there be any cause for concern from people whose P-1 factor-finding seems to deviate considerably from the expected rate?

2/20=10% (I should be at 7.58%) for me since I have had a dedicated core on P-1 working in the 48-50M range

[Kevin]

I'm 3 for 13 since I started consistently putting some CPUs on P-1 factoring at the beginning of January . Does this mean I'm a witch?

[James Heinrich]

Originally Posted by Kevin

I'm 3 for 13... . Does this mean I'm a witch?

lol, not in the least
I'm totally unconcerned about short-term deviations (in fact I'm 3/8 in the last week), and not really concerned (simply curious) about above-expected successes. I am concerned about hundreds or thousands of GHz-days being spent without finding any factors.

[Mr. P-1]

I suspect in George's case at least, he has been testing the P-1 code on exponents with known smooth factors.

[cheesehead]

Originally Posted by James Heinrich

"Richard B. Woods" and "MCAS" are high on the top-500 and have checked a decent number of exponents but haven't found any factors (certain far from the expected ~5% rate).

As Uncwilly noted, my ("Richard B. Woods") seemingly-high standing in the P-1 report is the result of a server bug that I semi-deliberately triggered to see what would happen. My number of P-1 attempts and amount of CPU time listed there is greatly overstated, as a result. George is well aware of the bug (and its relatively low priority) -- when he fixes it, I'll drop (deservedly) in the P-1 standings.

Furthermore, several of my reported P-1 attempts were conducted as extensions of P-1 runs at lower bounds, and thus naturally had a lower probability of success than those done with standard bounds computed by the client.

As if that weren't enough ...

Most of my past P-1 results were completed by the v24 client. When the v5 server gets a v24 report of a factor found during P-1, it does not recognize that the factor was found by P-1, but instead assumes it to be the result of ECM! So those P-1 successes of mine are not properly credited in the P-1 listing.

A couple others (e.g. "axn", "id_warrior") have only found 1 factor out of many attempts, again with far-below-expected success rate. Do these users have abnormally bad luck in being assigned exponents, or are the exponents being P-1 poorly in some way that prevents factors from being found that should be?

Perhaps their listed numbers are affected by the same factors as mine.

On the other end, there are also a few users who have a much-better-than-expected factor-finding rate (the king of which is George, at 85% success), but I'm less concerned about finding lots of factors than too few.

... but those two alternatives could be related.

If you look (way, way down) in the ECM standings, you'll see that I'm listed with 3 successes in 3 attempts! This 100% success ratio is even more remarkable when one considers that I've never reported an ECM result to the v5 server!! Those 3 factors should be listed among my P-1 attempts/successes rather than as ECM results.

Originally Posted by James Heinrich

I am concerned about hundreds or thousands of GHz-days being spent without finding any factors.

You can relax ... or you can redirect your concern to the fact that hundreds or thousands of GHz-days either are erroneously inflated in the list (as with the first server bug above), or are unfairly deflated and miscategorized (as with that second server bug). Sleep well. :-)

[Mr. P-1]

Originally Posted by cheeseheadWhen the v5 server gets a v24 report of a factor found during P-1, [I

it does not recognize that the factor was found by P-1, but instead assumes it to be the result of ECM![/I] So those P-1 successes of mine are not properly credited in the P-1 listing.

I have two factors in my results page attributed to v4_computers which are correctly classified as F-PM1. I have no spurious sucesses on the ECM page.

Strangely, according to the P-1 table, I have 9 successes, but there are only 6 factors in my results page.

[cheesehead]

Originally Posted by Mr. P-1

I have two factors in my results page attributed to v4_computers which are correctly classified as F-PM1. I have no spurious sucesses on the ECM page.

Hmmm ... I underspecified and overgeneralized. Thanks.

Here, I'll try to restate what I experienced without unwarranted generalization:

In the only 3 instances where I have used the v5 manual_reports page to report (by copying lines from results.txt) factors from P-1 runs that were performed by the v24 client, the v5 server replied with a message that it was accepting the result but could not determine which method had been used to find the factor, so would credit it as though the method had been ECM.

Note 1: Those factor reports were in the midst of other report lines about unsuccessful P-1 runs, which were correctly handled.

Note 2: In each of the 3 factor cases, the results.txt line reporting the factor was immediately preceded by a line stating, "P-1 found a factor in stage ...". Though I had copied all those lines to the manual_report page along with the others, there was no sign that the v5 server acknowledged the "P-1 found ..." lines or associated them with the immediately-following factor report lines.

[petrw1]

I had one V4 TF assignment from a v4 computer that completed and reported a factor to the v5 server which was subsequently classified as a F-PM1.