Newer milestone thread

[Xyzzy]

Originally Posted by Madpoo

Maybe those 3 were used as tests for different code bases or something to verify the process.

Maybe they were part of a disk image that got pushed out? Or the media is read-only?

[Uncwilly]

七

[Primeinator]

Countdown to proving M(32582657) is the 44th Mersenne Prime: 7

Estimated completion is 11-17

[Madpoo]

Originally Posted by Xyzzy

Maybe they were part of a disk image that got pushed out? Or the media is read-only?

What's weird is that out of those 13 mismatched residues, 12 of those were all from one user, "chatmate".

I think that user must have been checking a machine that had a lot of problems and kept checking in bad results. The DB shows that poor user had error codes on just about every result. Even on one of the check-ins, the "error code" was zero but the residue was incorrect anyway.

He/she did finally get a good result in which was verified, only to have the whole thing get factored about a year ago by ECM.

Every exponent has it's own story I suppose. :)

[cuBerBruce]

M(32155297) now has three matching LL results.

[legendarymudkip]

Originally Posted by cuBerBruce

M(32155297) now has three matching LL results.

How ironic.

[Madpoo]

Originally Posted by Madpoo

What's weird is that out of those 13 mismatched residues, 12 of those were all from one user, "chatmate".

I think that user must have been checking a machine that had a lot of problems and kept checking in bad results. The DB shows that poor user had error codes on just about every result. Even on one of the check-ins, the "error code" was zero but the residue was incorrect anyway.

He/she did finally get a good result in which was verified, only to have the whole thing get factored about a year ago by ECM.

Every exponent has it's own story I suppose. :)

By the way, I've been working on and off updating the exponent report page... mostly some style changes, but a few other things. For instance, the LL section of each exponent will currently only show verified/unverified results (single/double check). If there are some suspect entries in there (mismatched residues) they only appear while that exponent is in the unverified status. Once a matching residue comes in, the exponent is categorized as verified, and any of those previously suspect results get marked as bad, because they were.

There's also a status for an exponent where it might be verified composite by double-checking, but then if someone does some deeper TF or ECM testing on it and finds a factor, those previous "verified" results get classed as "factored" meaning the LL was verified but it's moot since a factor was found. And I think some people might, for whatever reason, do an LL test on exponents that were already known composite because of a known factor... maybe for testing.

I think it's kind of interesting to see those previous bad results, or show the LL results even if the number was factored. So my test page is including that information right now.

What do you all think though? Is that kind of "inside baseball" stats useful? I mean, I can peek in the database and see those previous bad results, but for everyone else, if you were looking at http://www.mersenne.org/M32155297 right now you wouldn't see those other 2 bad results anymore, now that it's verified.

You can see one of those bad results in the history section of the details, but the other bad result pre-dated this version of the database (or the raw client messages only go back so far... same difference), so only the LL entry is available for that particular bad result from user "EspElement" pre-2008 or so.

Or if you were to look at the details for M2397103 , since it was eventually factored you'd never know the sordid tale of user "chatmate" and his numerous failed efforts at that one.

So... think that kind of warts and all history (including bad results in the LL details) is useful, or just a mere curiosity since it doesn't really affect the outcome of anything?

[ATH]

Originally Posted by Madpoo

So... think that kind of warts and all history (including bad results in the LL details) is useful, or just a mere curiosity since it doesn't really affect the outcome of anything?

I would love to be able to see all the history but not as default.

Maybe change the "Show full details (current assignment, history, LL residues)" to "Show more details" or "Show history" and then add another check mark called "Show complete details/history (including bad results)" or something like that which would show you all the things you can currently see in your version.

[Madpoo]

Originally Posted by ATH

I would love to be able to see all the history but not as default.

Maybe change the "Show full details (current assignment, history, LL residues)" to "Show more details" or "Show history" and then add another check mark called "Show complete details/history (including bad results)" or something like that which would show you all the things you can currently see in your version.

Well, on that note we're actually trying out some other tweaks. Right now if you show the full details, it will include every time someone checked in even just 1 curve of an ECM run. Those add up...

Right now on my test page I've got it setup to "roll up" all of the curve info from the history and present it as "14,376 curves run" or whatever. That's actually not a real indication of the ECM work since there have been plenty of curves run prior, they're just not part of what's in the history log going back to 2008'ish.

There's a metric regarding ECM work effort that I'm still wrapping my head around, which kind of relates to the # of curves run for each upper bound, and I think I'm on the verge of beginning to start to comprehend. I may show that info instead, perhaps as a percentage like "ECM work xx% done" or "ECM complete" if applicable.

George had me add in a tick box so that you *could* get the full "user xyz ran 15 curves with bounds a and b" but honestly, for some exponents like 1277 it's a LOT of stuff.

Trust me, including the occasional bad or suspect LL result won't be a lot of extra info... one or two here and there really. The vast majority of LL results come in clean as a whistle and they verify just fine.

[ATH]

Originally Posted by Madpoo

There's a metric regarding ECM work effort that I'm still wrapping my head around, which kind of relates to the # of curves run for each upper bound, and I think I'm on the verge of beginning to start to comprehend. I may show that info instead, perhaps as a percentage like "ECM work xx% done" or "ECM complete" if applicable.

The number of curves to run at each B1/B2 is the number required so there is 1/e ~ 37% chance (risk?) that a factor was missed at that size. I guess that is the most efficient time to move up to the next size level (where you can still find the smaller factors).

[Madpoo]

Originally Posted by ATH

The number of curves to run at each B1/B2 is the number required so there is 1/e ~ 37% chance (risk?) that a factor was missed at that size. I guess that is the most efficient time to move up to the next size level (where you can still find the smaller factors).

Okay... so maybe y'all can help me piece something together.

For an extreme example I've been looking at M1277 since there has been a lot of ECM activity on it.

There's a measurement in the database "total ECM effort" and right now for 1277 it's "42 656 688 085 105.5"

For a cheat sheet, I'm using the code behind the ECM report on the site, e.g.
http://www.mersenne.org/report_ecm/?...77&ecm_hi=1277

It tells me there are 53,321 curves tested for that 800e6 bound #1.

The calculations behind that are using a table of total curves for each of the bounds (same you see on that table), and for 800,000,000 it's 360,000 curves. The simple math it's doing is just 800e6 * 360e3 = 288e12. The total ECM work done, 42.66e12, is 14.81%... 14.81% of 360,000 curves = 53,320.86 curves.

So... that matches what the report is doing, and I'm okay with that; I understand how it's getting that 53,321 curves tested for that bound.

What I wonder though... is it more helpful to show that in the exponent report like "53,321 curves tested for Bound #1=800e6" or would it be more helpful to show something like "ECM progress = 14.81% complete" or some such? Or both?

The deal is I'm not familiar with ECM in it's particulars, so I apologize if this *should* be apparent what the best way to show that info is or if these seem like dumb questions.