How often are triple checks released?

patrik · 2004-01-01, 13:36

In 2000 I tested a number (9666457) which had a non-matching double check returned Aug 06, 2003. This number has still not been released for triple checking. I am just concerned that regular triple checking has been forgotten when the discussion here has been about early double checking of error-prone machines.

Since we have been discussing that double checks preferably should take less than six months, perhaps we shouldn't let triple checks lie untested for five months.

Also, a few of my other machines have returned non-matching double checks, and I am very eager to know if it was my test or the original that had an error.

GP2 · 2004-01-01, 18:05

Well, the motivation behind early double-checking is to ensure that every exponent has at least one non-error-prone result, to reduce the chances of a hidden prime.

As long as there's one non-error-prone result, the final confirming check can be done routinely years later, as is the usual case.

Early triple checks get done, but only if both previous results were error-prone.

Regarding routine triple checks, I just assumed George has a system for releasing these at regular intervals. I'm certainly in favor of these getting released regularly, because every result in the BAD file only gets there as the result of a triple check, so regular triple checks are vital for improving statistics on which machines are error-prone.

BTW, some bad news regarding your exponent:

cleared.txt
9666457,65,D ,0xDF8FC03B96B1A9__,06-Aug-03 03:55,foxkid,DJ

HRF3.TXT
9666457,osytol,gamma,WV1,00000000
9666457,foxkid,DJ,WX1,00000A14

With the error code 00000A14, it's extremely likely that it was your result that was the bad one, not the other guy's.

Furthermore foxkid/DJ is an error-prone machine, and you also have a few more:

Code:

uv3h	uv3	uv2	bad	good
----	---	---	---	----
3       4       5       2       5       foxkid,DJ
0       0       0       4       0       foxkid,Eilif
0       0       0       1       5       foxkid,L-C
0       1       2       1       3       foxkid,RMikkelsen
0       0       0       1       9       foxkid,dj
0       0       0       7       1       foxkid,home-NT

good = verified good results
bad = verified bad results
uv2 = results needing a 2nd check
uv3 = results needing a 3rd or higher check
uv3h = uv3 results that also had a "harmful" error code

patrik · 2004-01-01, 18:59

The good news (for me) is that I am not foxkid. I'm patrik in cleared and status, but for some reason unknown to me, the name gets translated to osytol in hrf3 and lucas_v.

(Maybe it has something to do with running the v.17 of the client in 1999. That version of the client had a bug which made it return bad results. I think it was decided that the users could keep the credit despite the bad results (since it was a bug in the client). So maybe this was the way to handle that. Or it could be when I turned off and on "Use primenet" so my results were reported again, and emailed George, since I was worried they would count as double checks. (This was before the forums started.) BTW my v.17 bad results were removed completely from the stats files.)

GP2 · 2004-01-01, 20:24

Quote:

Originally posted by patrik
The good news (for me) is that I am not foxkid. I'm patrik in cleared and status, but for some reason unknown to me, the name gets translated to osytol in hrf3 and lucas_v.

This is actually very common, unfortunately (different usernames used by GIMPS data files and by PrimeNet). It makes certain kinds of datamining unreasonably difficult (for instance, BAD usually uses the PrimeNet usernames, but LUCAS_V.TXT and HRF3.TXT use the GIMPS usernames).

One more thing that will hopefully be resolved with the new server.

PS,
no osytol machines are error-prone.

patrik · 2004-01-29, 19:38

As I suspected, a few results of my ws001 computer are bad. I overclocked it to FSB 150 MHz and on the next exponent got an error almost at the beginning. So I lowered it to 149 MHz and started from scratch. In November I saw that three double-checks had non-matching residues, so I lowered the FSB again to 147 MHz. After that everything has verified.

However, the exponents I ran during the time it was too heavily overclocked maybe should be released for early triple-checking again (like the Nov. 2003 release). These are the results I got (with a few extra columns of my own):

Code:

                                           Credit for                         Total   FSB
                                    Status exponent                           credit  changed to
 prime   fact    Lucas-Lehmer residue or factor
exponent bits    [residues partially masked]        date returned   computer ID
-------- ---- -- --------------------------------  ---------------  ------------
19401049  67     0x18F254550847AE__ P              25-Aug-03 17:28  ws001     153.432 FSB150
19315313  67     0x9B3CA9248B456E__ P      2.021   03-Sep-03 04:32  ws001     159.938 FSB149
17381233  66     0x61FE7AE27C11EC__ P      1.625   09-Sep-03 19:09  ws001     168.671
16570577  66     0x74C1AF700A1E58__ P      1.549   16-Sep-03 06:36  ws001     172.792
16562111  66     0x6C48DB14F9FE0F__ P      1.548   21-Sep-03 18:30  ws001     181.815
16482581  66     0x690450F02D3105__ P      1.541   28-Sep-03 07:17  ws001     189.911
15521911  66  D  0x037076FFB68DEB__ T      1.451   04-Oct-03 07:05  ws001     196.272
20758553  67     0xCC73C63AF3D888__ P      2.809   15-Oct-03 19:28  ws001     212.787
10249501  65  D  0x27ED8080C667AC__ C      0.502   18-Oct-03 10:30  ws001     216.353
 9803429  65  D  0x7C0D6B0CC6B9B2__ C      0.480   20-Oct-03 18:51  ws001     219.808
 9804107  65  D  0x60E732F7D40E91__ C      0.480   22-Oct-03 19:27  ws001     222.257
10324879  65  D  0xF3F949DA179B18__ Bad    0.670   25-Oct-03 13:17  ws001     224.607
10339117  65  D  0xC582409ACEC489__ T      0.670   28-Oct-03 07:01  ws001     227.922
10370341  65  D  0x730D3231B6999B__ Bad    0.672   31-Oct-03 04:49  ws001     233.879
10378411  65  D  0xD63CD4AA26E2A6__ C      0.673   02-Nov-03 17:34  ws001     236.216

Status: C=cleared, P=Pending double-check, T=triple-check needed, Bad=bad

The exponents to reassign would then be
19401049
19315313
17381233
16570577
16562111
16482581
15521911
20758553
since 10339117 is already assigned for triple testing.

dsouza123 · 2004-01-31, 22:19

Not to hijack the main point of the thread How often are triple checks released
but how do can/do you get different usernames ?

When running Prime95 it only asks for username and password.

What is the difference between GIMPS and Primenet ?

2004-01-01, 13:36	#1
patrik "Patrik Johansson" Aug 2002 Uppsala, Sweden 5²×17 Posts	How often are triple checks released? In 2000 I tested a number (9666457) which had a non-matching double check returned Aug 06, 2003. This number has still not been released for triple checking. I am just concerned that regular triple checking has been forgotten when the discussion here has been about early double checking of error-prone machines. Since we have been discussing that double checks preferably should take less than six months, perhaps we shouldn't let triple checks lie untested for five months. Also, a few of my other machines have returned non-matching double checks, and I am very eager to know if it was my test or the original that had an error.

2004-01-01, 18:05	#2
GP2 Sep 2003 5×11×47 Posts	Well, the motivation behind early double-checking is to ensure that every exponent has at least one non-error-prone result, to reduce the chances of a hidden prime. As long as there's one non-error-prone result, the final confirming check can be done routinely years later, as is the usual case. Early triple checks get done, but only if both previous results were error-prone. Regarding routine triple checks, I just assumed George has a system for releasing these at regular intervals. I'm certainly in favor of these getting released regularly, because every result in the BAD file only gets there as the result of a triple check, so regular triple checks are vital for improving statistics on which machines are error-prone. BTW, some bad news regarding your exponent: cleared.txt 9666457,65,D ,0xDF8FC03B96B1A9__,06-Aug-03 03:55,foxkid,DJ HRF3.TXT 9666457,osytol,gamma,WV1,00000000 9666457,foxkid,DJ,WX1,00000A14 With the error code 00000A14, it's extremely likely that it was your result that was the bad one, not the other guy's. Furthermore foxkid/DJ is an error-prone machine, and you also have a few more: Code: uv3h uv3 uv2 bad good ---- --- --- --- ---- 3 4 5 2 5 foxkid,DJ 0 0 0 4 0 foxkid,Eilif 0 0 0 1 5 foxkid,L-C 0 1 2 1 3 foxkid,RMikkelsen 0 0 0 1 9 foxkid,dj 0 0 0 7 1 foxkid,home-NT good = verified good results bad = verified bad results uv2 = results needing a 2nd check uv3 = results needing a 3rd or higher check uv3h = uv3 results that also had a "harmful" error code

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Triple Double	robert44444uk	Puzzles	7	2007-06-01 08:25
When are triple-checks put on the server?	eepiccolo	PrimeNet	5	2007-01-31 06:50
Double checks	Rastus	Data	1	2003-12-19 18:20
How do I get rid of the Triple Checks??	outlnder	Lounge	4	2003-04-07 18:06
Triple Checks- How are they disposed of??	outlnder	Lounge	6	2003-01-24 22:01

2004-01-01, 18:59	#3
patrik "Patrik Johansson" Aug 2002 Uppsala, Sweden 5²·17 Posts	The good news (for me) is that I am not foxkid. I'm patrik in cleared and status, but for some reason unknown to me, the name gets translated to osytol in hrf3 and lucas_v. (Maybe it has something to do with running the v.17 of the client in 1999. That version of the client had a bug which made it return bad results. I think it was decided that the users could keep the credit despite the bad results (since it was a bug in the client). So maybe this was the way to handle that. Or it could be when I turned off and on "Use primenet" so my results were reported again, and emailed George, since I was worried they would count as double checks. (This was before the forums started.) BTW my v.17 bad results were removed completely from the stats files.)

2004-01-31, 22:19	#6
dsouza123 Sep 2002 2·331 Posts	Not to hijack the main point of the thread How often are triple checks released but how do can/do you get different usernames ? When running Prime95 it only asks for username and password. What is the difference between GIMPS and Primenet ?