mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Completed Missions (https://www.mersenneforum.org/forumdisplay.php?f=35)
-   -   Two priority triple-check exponents (https://www.mersenneforum.org/showthread.php?t=1183)

GP2 2003-09-30 18:17

Two priority triple-check exponents
 
Two triple-check exponents:

DoubleCheck=13470473,65,1
DoubleCheck=13879783,65,1

These are for two machines that between them have 130 LL-tested exponents, of which 15 have a non-matching second LL test, and the rest don't have any second LL test.

I suspect these two machines may be error-prone, but no exponents have been verified good or bad yet.

If the above triple check confirms the bad results, then the remaining 128 exponents for these two machines should go to George for release as pseudo-first-time LL tests.


Post here to claim this set.

When you're finished, manually check in your results at [url]http://www.mersenne.org/ips/manualtests.html[/url]

garo 2003-09-30 18:24

You should also be able to let Prime95 report normally for Primenet credit. The only caveat is that these exponents may not be recognized as being assigned to you and Primenet may complain and/or kick them out of your worktodo. The solution to that is to turn on manual communication only and turn it off once the two exponents are done.

GP2 2003-09-30 20:09

[QUOTE][i]Originally posted by garo [/i]
[B]You should also be able to let Prime95 report normally for Primenet credit.[/B][/QUOTE]

Garo,

Can you post a detailed step-by-step explanation of how to get Primenet credit for manual tests? If so, that would be great, I think we'd get a lot more takers.

I've got dozens more triple-check exponents like the above. The chances of these producing a Mersenne prime are nil, since two separate LL tests produced non-zero (albeit differing) results, and to get a prime you'd have to have both the triple and quadruple checks returning zero residue. So it's unfair to automatically release these as pseudo-first-time checks to people who specifically requested only first-time checks with a bona fide chance of finding a prime. Nevertheless these triple checks are extremely valuable because they have an excellent chance of unmasking a previously-undetected error-prone machine, and then all of that machine's other, unverified exponents [i]can[/i] legitimately be released as pseudo-first-time tests.

The triple-check exponents I want to post here satisfy both a) a high chance of unmasking an error-prone machine, and b) each exponent that unmasks an error-prone machine triggers a fairly large number (at least 15+) of that machine's other, unverified exponents for pseudo-first-time LL testing.

garo 2003-09-30 22:14

You cna either create a new directory to run the tests or do them in your usual directory.

1) Go to the advanced menu (if it is inactive you can activate it by putting in the password 9876).
2) Choose Manual Communication and select do not contact Primenet Server automatically.
3) Stop Prime95.
4) Add the exponents to your worktodo and crunch.
5) When the inserted exponents are done - even if other older exponents that you go tfrom Primenet remain - go to the advanced/manual communication menu and unselect the do not contact Primenet automatically option.

That's it!

markr 2003-10-15 07:33

I'll do these two. They'll be on 1.8 P4s, should be about 10-12 days.

markr 2003-10-27 05:38

Both done now & reported via primenet. Let's see which results match!

UID: Rameau/ki, M13470473 is not prime. Res64: 33DF36051F03B177. WZ1: 1391C4B5,7007516,00000000
UID: Rameau/cc, M13879783 is not prime. Res64: F4AD48F79DCA2C29. WZ1: 2C022D09,2827643,00000000

markr 2003-11-02 06:00

We have two matches: from lucas_v.txt 31Oct2003,
13470473,sgrupp,sr,WW1,33DF36051F03B177,13112412,00000000
13470473,Rameau,ki,WZ1,33DF36051F03B177,7007516,00000000
...
13879783,ganser,CF107EFD7,WX5,F4AD48F79DCA2C29,6514129,00000000
13879783,Rameau,cc,WZ1,F4AD48F79DCA2C29,2827643,00000000

GP2 2003-11-02 07:39

The original two results are now confirmed bad:

13470473,bcoz,C26C3bcm3,WX1,7066EF67448F3CE3,10690,0700000E
13879783,bcoz,C26C3bcm4,WX1,D5C5A1C9ED5B213C,6894416,02000004

These machines will qualify as error-prone under new proposed criteria that I'll post about in the next few days.

outlnder 2003-11-02 20:06

GP2,

Will you be starting a thread naming all the error-prone machines??

I think this would be a good idea, so the owners can periodically check the thread to see which machines they need to work on.

GP2 2003-11-02 21:15

outlnder,

There are a few problems:

- The thread would be very long. A possibility is to put the information up on the Opteron server, but I haven't gotten around to this.

- The data files (BAD, LUCAS_V.TXT, etc) sometimes use different user names than Primenet does. For instance, user "PaJaSoft" appears in the data files as "S112790". So a users looking up their error rate might have to look under an alias (which they might not even be aware of).

- The main reason is because it would be of only limited use. It takes a triple-check to definitely determine that a result was erroneous, which means that the erroneous results were often returned years ago, and many of the machines now identified as error-prone are not even in service anymore. By contrast, there are a whole bunch of modern machines currently crunching first-time LL tests that we have little or no information about, because none of their exponents have ever had a double-check.

So, the current list of error-prone machines is mostly useful for identifying which exponents need an early double-check. It's not very useful for determining which modern machines are error-prone (unless those modern machines happen to be set to do double-checking rather than first-time testing).

markhl 2003-11-03 04:22

I can see the public benefit of a thread naming all the error-prone machines. However I'd be uneasy with this, a bit too much like publicly denouncing a list of people!

It'd be much nicer to just contact people privately, then they are more likely to be grateful and to work with you to sort out the issues.


All times are UTC. The time now is 06:13.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.