Just out of curiosity:
What were the criteria you used to select the 1M range exponents you´ve been TCing lately?
At first sight, they don´t seem to fit in any of the groups elected to TCs.

[QUOTE=lycorn;404397]Just out of curiosity:
What were the criteria you used to select the 1M range exponents you´ve been TCing lately?
At first sight, they don´t seem to fit in any of the groups elected to TCs.[/QUOTE]

I have one machine where it's official capacity involves higher than normal CPU usage, so it wasn't really helping out the triple-check stuff too much (and it's one of the least powered machines I have... dual 4-core (Xeon E5630)).

A couple months back I did a side thing doing triple-checks of exponents below 1M that didn't already have triple-checks (many of them already had). At the time I thought it'd be fun to do triple-checks of exponents between 1M-2M. There's something like 10,900 or something like that... it'll keep that system busy for a couple months I think.

Yeah, no real reason except it was in a range that could somewhat easily be checked, and it's an interesting (to me) way of checking some old work with old software, just on the infinitesimally small chance there was a bug in old software (or hardware) that would come up with the same *wrong* residue twice. Extremely unlikely, to the point of "why bother" for most people, but this machine isn't useful for much else at the moment.

FYI, with 4 cores on each worker (and 2 workers on that system), it takes an average of 10 minutes to complete each test. The exponents range in size from 1.34M-1.99M. Curiously, all of the exponents below 1.34M had already been triple-checked in the past.

So I'm guessing it'll take that one system about 40-45 days to go through just under 11,000 checks.

I checked some in since I started it last night, but I'll probably wait and check in the rest when they're all done, in batches.

If it bugs anyone to see a ton of old stuff jamming up the latest result reports, I could temporarily tweak that to exclude these... unless any interesting results came through? Is that kind of a cheat though, excluding results I'm checking in that are boring triple-checks of small exponents?

And reminder, there is a daily XML file being generated now of *all* the work that checked in during that day.
e.g. [URL="www.mersenne.org/result_archive/2015/2015-06-18.xml.bz2"]www.mersenne.org/result_archive/2015/2015-06-18.xml.bz2[/URL]

Just in case I flood the hourly report with these, those XMLs would be the go-to for a full list of everything.

(And no, I have no plans to do the same from 2M - 3M ... I'll keep my crazy on a leash after these are done) :smile:

[QUOTE=Madpoo;404400](And no, I have no plans to do the same from 2M - 3M ... I'll keep my crazy on a leash after these are done) :smile:[/QUOTE]

Just wondering... Have you ever found a bad double check?

Not saying what you're is ill-advised -- I have sometimes spent hundreds of dollars double checking TF results for worker's who's statistics were "non-nominal".

But it would be interesting to know (and, frankly, furthers the integrity of GIMPS).

"Trust No One" -- Mulder et al.

[QUOTE=chalsall;404401]Just wondering... Have you ever found a bad double check?[/QUOTE]
He hasn't. If I be him and would, then I would boast everywhere, including indonesian television... :razz:

[QUOTE=Madpoo;404400]

Yeah, no real reason except it was in a range that could somewhat easily be checked, and it's an interesting (to me) way of checking some old work with old software, just on the infinitesimally small chance there was a bug in old software (or hardware) that would come up with the same *wrong* residue twice. Extremely unlikely, to the point of "why bother" for most people, but this machine isn't useful for much else at the moment.[/QUOTE]

Fair enough.

[QUOTE=Madpoo;404400]
If it bugs anyone to see a ton of old stuff jamming up the latest result reports
[/QUOTE]

As far as I´m concerned, it doesn´t bother me at all.
Also, after all you´ve done for this project, you are more than welcome to have any little excentricity you like...:wink: :beer:

[QUOTE=LaurV;404431]He hasn't. If I be him and would, then I would boast everywhere, including indonesian television... :razz:[/QUOTE]

As it turns out, I [I]might[/I] have found one. But of course my first step right now is to do a quadruple check.

I've been on vacation (or "holiday" as some of you would call it) for the past week. I came back and checked in the results that had built up and wouldn't you know...

[URL="http://www.mersenne.org/M39988591"]M39988591[/URL]

Now, I'm not claiming anything until the new check I just kicked off (on a different machine) finishes up in another 18 hours, but it does bear mentioning that the machine I ran that on hasn't had any bad results so far, very stable system.

If it turns out my quad-check matches my previous one, and not the two other ones from user "Kali", then I'll ask someone else to do a verifying run and we can ask George for input on what to do.

I'm kind of hoping it was just a crazy result from my system but I feel confident enough in that result (it didn't report any problems along the way... no roundoff > 0.4, etc) that I'm 95% sure my result is correct.

The implications if mine turn out to be correct? Good question... I'd definitely want to investigate further (make sure those previous results really had different shifts, see what app version was used, see what other work that user has done, etc) and it makes me a little nervous to think about it, honestly. :smile:

updated list
 

Here's an updated list of the triple-checks to be done. 325 left.

I went ahead and bit off the last remaining ranges of work. I'm estimating being done with all of them in the next couple weeks, except M383838383 still has 22 days to go.

There'll be a few new ones sprouting up from time to time which I'll try to find and take care off here and there.

[QUOTE=Madpoo;404872]If it turns out my quad-check matches my previous one, and not the two other ones from user "Kali", then I'll ask someone else to do a verifying run and we can ask George for input on what to do.[/QUOTE]

Just for fun, I've thrown this on one of my "barbies". Should be done in about 24 hours.

[QUOTE=chalsall;404902]Just for fun, I've thrown this on one of my "barbies". Should be done in about 24 hours.[/QUOTE]

Thanks. Especially if my second run at this matches my first, I'd want someone to check my results independently.

Just 2.5 hours to go on mine. :smile:

[QUOTE=Madpoo;404872]As it turns out, I [I]might[/I] have found one. But of course my first step right now is to do a quadruple check.
...
The implications if mine turn out to be correct? Good question... I'd definitely want to investigate further (make sure those previous results really had different shifts, see what app version was used, see what other work that user has done, etc) and it makes me a little nervous to think about it, honestly. :smile:[/QUOTE]

Okay... hmmm... now what?
[URL="http://www.mersenne.org/M39988591"]M39988591[/URL]

Last night I went ahead and looked up a few things about that user "Kali".

That one is the only self-verified one by that user, so that's good. They've only done 28 LL tests overall.

The Prime95 version they used for the apparently false self verification in both runs was:
"Linux64,Prime95,v25 beta,(aka Wd8)"

No bad results when I looked, they were all either verified or unverified (and a couple where the exponent was later factored).

I picked a couple unverified results of theirs and ran my own test. One matched okay ([URL="http://www.mersenne.org/M34617743"]M34617743[/URL]) and one of them didn't match their result, but happened to match someone elses ([URL="http://www.mersenne.org/M35950301"]M35950301[/URL] ... it was due for a triple-check anyway).

I also picked one of theirs that was verified (not by them) and I'm still waiting on my triple-check of that, just to see . Hopefully this isn't a case of monkey business and just some weird data thing? I don't know though... both of the runs on that weird exponent had the same residue but different shift counts. I'd rhetorically ask "what are the odds?" but one of you would actually figure it out and tell me. LOL I know it's close enough to 1 in 2^64 if it were just random, which it's not, really. :smile:

When digging around, there's also a curious thing where exponent [URL="http://www.mersenne.org/M37124947"]M37124947[/URL] is still unverified, with a result by someone else. Yet in the old version 4 logs we've pulled in, you'll notice there's a result from that Kali user for it... so we have a log entry for it but apparently it didn't get processed and "cleared" ?

What that tells me is there might have been something going on with that result where the system rejected it for some reason? I don't have the ability to check the security checksum on it to see if that was the issue (if I'm reading it right, the checksum on it might be all zeros? I'm not sure which part of the raw message is the checksum)

At this point I'm tempted to just run my own LL tests on all of the other exponents that user did, just to figure out what's what. There are 11 that are verified, and now 13 unverified, so it's not a lot.

I've corrected the result state in the database. Unfortunately, we do not have the prime95 results.txt lines for the offending exponent. This indicates the results were likely submitted manually.

The unprocessed result from M37124947 is indeed missing its checksum. This indicates this was an executable compiled by the user from source code. The v4 primenet server would not have processed it, but I would have accepted it (and I don't know why I did not).

At this point we have a mystery. The user was capable of building his own executable. Perhaps he did so with M39988591 and something went awry. We may never know.


[edit] M37124947 Kali result now processed