Server is assigning already factored exponents in 21M range

GP2 · 2003-10-30, 03:36

So far, as of 30 Oct 2003 03:00 UTC, about 150 exponents have been assigned in the 21M range for first-time LL testing by Primenet.

And so far, 11 of these assignments have been of already-factored exponents!

Exponents assigned (LL testing) that already have a factor?

-------
factors
-------
21004351,148873110909256049
21005399,11649960263685090143
21005503,709024873310654959
21006287,61814481861109306697
21007229,4841612787038654993
21007463,206019461984496607
21008257,506391274569698201
21008293,44286284487345506393
21008749,2858517788830384319
21009533,1495804325892510527
21009587,66280845149824422151

-------
STATUS_L.TXT
-------
21004351, *,57,,0.1,26.9,86.9,,29-Oct-03 21:21,mmesser,Work_NT
21005399, *,63,,0.1,134.9,69.9,,29-Oct-03 22:20,rodspade,rods_pc
21005503, *,59,,0.1,18.9,78.9,,29-Oct-03 22:20,g8kwi,CFFF8B0D0
21006287, *,66,,0.1,12.9,72.9,,29-Oct-03 22:57,.,C7C862164
21007229, *,62,,0.0,15.0,75.0,,30-Oct-03 00:00,Team_Prime_Rib,sobo5
21007463, *,58,,0.1,21.9,81.9,,30-Oct-03 00:07,bmg9611,ppalmf58_1
21008257, *,59,,0.1,31.9,87.9,,30-Oct-03 00:38,S133906,C81BDFF35
21008293, *,65,,0.1,26.9,86.9,,30-Oct-03 00:42,S133907,CA53B16D3
21008749, *,61,,0.1,35.9,87.9,,30-Oct-03 01:09,S133913,C12746CB6
21009533, *,60,,0.0,54.0,61.0,,30-Oct-03 02:58,jeffwoods,XP-MULTI
21009587, *,66,,0.0,29.0,88.0,,30-Oct-03 02:59,S133925,C1AB008B1

I'm flushing these by manually re-checking-in these factors under GP2/dummy, but new ones get assigned every hour...

Curiously, all of these were assigned with a "*" indicating that further factoring is needed.

For instance, the first exponent 21004351 was assigned with an indication that it was trial-factored up to 57 bits and needs further factoring. But in fact it has a factor of 57 bits and does not need further factoring. Likewise, the next exponent 21005399 is assigned with an indication that it has been trial-factored up to 63 bits... but it has a factor of 63 bits.

One possible brute-force solution is to look for all factors in the factors database with exponents in the range 21.0M and factors of between 57 to 66 bits, and manually re-check-in all of them. I will probably go ahead and do this.

GP2 · 2003-10-30, 04:17

Well, I re-checked in 358 factored exponents in the 21.0M range.

The summary.txt changed from this:

Code:

  Exponent Range       Trial Factoring       Lucas-Lehmer      Double Checking
                        Avail   Out Factd   Avail   Out  Done   Avail   Out  Done
 --------------------   -----=-----=-----   -----=-----=-----   -----=-----=-----
21000000   21099999            39    10    2375   146

to this

Code:

21000000   21099999            39   337    2045   151

So about 325 out of 2375 (around 13%) of the available exponents in the 21.0M range were already-factored exponents.

I'm guessing the same problem exists in the 21.1M, 21.2M, etc. ranges.

GP2 · 2003-10-30, 04:25

Quote:

Originally posted by GP2
So far, as of 30 Oct 2003 03:00 UTC, about 150 exponents have been assigned in the 21M range for first-time LL testing by Primenet.

And so far, 11 of these assignments have been of already-factored exponents!

Actually, there were 13, not 11.

Two factors were already "re-found"

21006907,57, F,171229386539390033,29-Oct-03 23:55,rabbits77,rabbits77
21008287,59, F,448051365378175441,30-Oct-03 00:45,stevehs,cinergi

even though these factors were originally found last November:

21006907,57, F,171229386539390033,04-Nov-02 23:21,Northwestern,C20CE007A
21008287,59, F,448051365378175441,10-Nov-02 15:03,sdbardwick,SBSAPC_PP200

Fortunately, not much duplicated work since these are relatively small factors.

GP2 · 2003-10-30, 04:40

Quote:

Originally posted by GP2
I'm guessing the same problem exists in the 21.1M, 21.2M, etc. ranges.

OK, I think I flushed out all the remaining already-factored available-for-assignment exponents in the entire 21M range.

Old man PrimeNet · 2003-10-31, 00:25

It's good to have so many sharp eyes on GIMPS.

I reviewed the server's logs and assignment behavior, but did not see anything unusual happening. The exponents in question where recently explicitly reassigned for LL testing within a large range block.

Every hour of every participating machine is sincerely appreciated. Even if unintentional (George may have a reason), the cost of 'rediscovering' a factor is relatively low yet fully credited in PrimeNet CPU time when reported by the Prime95 program.

As such, no action should really be necessary. George might be reviewing the range for routine admin updates.

Prime95 · 2003-10-31, 00:30

For some unknown reason when I opened the 21 millions up for LL testing, the server also re-released these exponents for factoring. Strange.

In the future, I will use commands that make individual exponents available for testing rather than using the commands that operate on a range of exponents.

GP2 · 2003-10-31, 05:41

I use the fake machine name GP2/dummy for things like this. Results from this "machine" should not receive credit, now or in the future. In particular, I checked in about 3000 factors under this name using the manual reporting form.

Joe O · 2003-10-31, 16:54

Quote:

Originally posted by Old man PrimeNet

Every hour of every participating machine is sincerely appreciated.

Then reward us by rewriting the server code!

GP2 · 2003-10-31, 17:56

Quote:

Originally posted by Joe O
Then reward us by rewriting the server code!

That's a bit uncalled for, it's much easier said than done.

Rather than rewriting the server, maybe we should take a close look at the BOINC framework that will be used very soon in the next phase of SETI@Home.

The advantages would be:
1) someone else maintains the server code
2) users using BOINC can choose to split their CPU time among different projects.

Since SETI@Home has (literally) 100 times as many users as GIMPS, the BOINC code should end up fairly robust through beta-testing. And if even 1% of those SETI users choose to devote a percentage of their CPU time to GIMPS, we could get a large benefit.

And finally, teams like Ars Technica (Team Prime Rib) that work on multiple projects could more easily shift their resources around among BOINC projects to respond to challenges (like the recent surge by curtisc).

I downloaded and compiled the BOINC source code out of curiosity, but haven't done anything with it.

Joe O · 2003-11-01, 00:00

Quote:

Originally posted by GP2
That's a bit uncalled for, it's much easier said than done.

It needed to be said, because it needs to be done! Look at all the posts in this forum about replacing the server, especially those by George !

At least replace SQL Server with a real database server. That would go a long way to solving some of our problems. If SS were replaced with a Database Server that also runs on *nix, we would not be tied to run only on Windows NT/2000/2003. The effort would force a review of the server code that is long overdue.

It might not be easy, but very few things worth doing are easy.

ColdFury · 2003-11-01, 01:47

What does being SQL have to do with windows?

Most "real" database servers use SQL and most of them are on Unix. I don't remember off-hand what primenet runs.

2003-10-30, 03:36	#1
GP2 Sep 2003 3·863 Posts	Server is assigning already factored exponents in 21M range So far, as of 30 Oct 2003 03:00 UTC, about 150 exponents have been assigned in the 21M range for first-time LL testing by Primenet. And so far, 11 of these assignments have been of already-factored exponents! Exponents assigned (LL testing) that already have a factor? ------- factors ------- 21004351,148873110909256049 21005399,11649960263685090143 21005503,709024873310654959 21006287,61814481861109306697 21007229,4841612787038654993 21007463,206019461984496607 21008257,506391274569698201 21008293,44286284487345506393 21008749,2858517788830384319 21009533,1495804325892510527 21009587,66280845149824422151 ------- STATUS_L.TXT ------- 21004351, ,57,,0.1,26.9,86.9,,29-Oct-03 21:21,mmesser,Work_NT 21005399, ,63,,0.1,134.9,69.9,,29-Oct-03 22:20,rodspade,rods_pc 21005503, ,59,,0.1,18.9,78.9,,29-Oct-03 22:20,g8kwi,CFFF8B0D0 21006287, ,66,,0.1,12.9,72.9,,29-Oct-03 22:57,.,C7C862164 21007229, ,62,,0.0,15.0,75.0,,30-Oct-03 00:00,Team_Prime_Rib,sobo5 21007463, ,58,,0.1,21.9,81.9,,30-Oct-03 00:07,bmg9611,ppalmf58_1 21008257, ,59,,0.1,31.9,87.9,,30-Oct-03 00:38,S133906,C81BDFF35 21008293, ,65,,0.1,26.9,86.9,,30-Oct-03 00:42,S133907,CA53B16D3 21008749, ,61,,0.1,35.9,87.9,,30-Oct-03 01:09,S133913,C12746CB6 21009533, ,60,,0.0,54.0,61.0,,30-Oct-03 02:58,jeffwoods,XP-MULTI 21009587, ,66,,0.0,29.0,88.0,,30-Oct-03 02:59,S133925,C1AB008B1 I'm flushing these by manually re-checking-in these factors under GP2/dummy, but new ones get assigned every hour... Curiously, all of these were assigned with a "" indicating that further factoring is needed. For instance, the first exponent 21004351 was assigned with an indication that it was trial-factored up to 57 bits and needs further factoring. But in fact it has a factor of 57 bits and does not need further factoring. Likewise, the next exponent 21005399 is assigned with an indication that it has been trial-factored up to 63 bits... but it has a factor of 63 bits. One possible brute-force solution is to look for all factors in the factors database with exponents in the range 21.0M and factors of between 57 to 66 bits, and manually re-check-in all of them. I will probably go ahead and do this. Last fiddled with by GP2 on 2003-10-30 at 03:37

2003-10-30, 04:17	#2
GP2 Sep 2003 101000011101₂ Posts	Well, I re-checked in 358 factored exponents in the 21.0M range. The summary.txt changed from this: Code: Exponent Range Trial Factoring Lucas-Lehmer Double Checking Avail Out Factd Avail Out Done Avail Out Done -------------------- -----=-----=----- -----=-----=----- -----=-----=----- 21000000 21099999 39 10 2375 146 to this Code: 21000000 21099999 39 337 2045 151 So about 325 out of 2375 (around 13%) of the available exponents in the 21.0M range were already-factored exponents. I'm guessing the same problem exists in the 21.1M, 21.2M, etc. ranges.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Odd Exponents Not Factored Deeply Enough	Gordon	PrimeNet	22	2015-08-12 09:30
M(10^9) to M(2^32) range factored to 2^64	James Heinrich	Data	40	2015-05-08 13:54
Full info for factored exponents	flashjh	PrimeNet	4	2012-02-25 07:21
Exponents Factored Vs Factors Found	CCol	PrimeNet	1	2008-05-21 13:32
factored exponents assigned	S485122	PrimeNet	2	2007-05-17 12:08

2003-10-31, 00:25	#5
Old man PrimeNet Jan 2003 Altitude>12,500 MSL 101 Posts	It's good to have so many sharp eyes on GIMPS. I reviewed the server's logs and assignment behavior, but did not see anything unusual happening. The exponents in question where recently explicitly reassigned for LL testing within a large range block. Every hour of every participating machine is sincerely appreciated. Even if unintentional (George may have a reason), the cost of 'rediscovering' a factor is relatively low yet fully credited in PrimeNet CPU time when reported by the Prime95 program. As such, no action should really be necessary. George might be reviewing the range for routine admin updates.

2003-10-31, 00:30	#6
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 41×199 Posts	For some unknown reason when I opened the 21 millions up for LL testing, the server also re-released these exponents for factoring. Strange. In the future, I will use commands that make individual exponents available for testing rather than using the commands that operate on a range of exponents.

2003-10-31, 05:41	#7
GP2 Sep 2003 2589₁₀ Posts	I use the fake machine name GP2/dummy for things like this. Results from this "machine" should not receive credit, now or in the future. In particular, I checked in about 3000 factors under this name using the manual reporting form.

2003-11-01, 01:47	#11
ColdFury Aug 2002 101000000₂ Posts	What does being SQL have to do with windows? Most "real" database servers use SQL and most of them are on Unix. I don't remember off-hand what primenet runs.