Low n ECM Factoring Reservation - Page 4

Greenbank · 2006-04-26, 13:31

Doh, it's the stack allocation problem, I thought I had compiled GMP with the right options. Will be moving to GMP 4.2.0 soon so I'll make sure it's ok then. Looks like I'll have to avoid the higher B1 values until I get time to do this. Sorry for wasting people's time.

rogue · 2006-04-26, 15:16

Quote:

Originally Posted by Greenbank

Doh, it's the stack allocation problem, I thought I had compiled GMP with the right options. Will be moving to GMP 4.2.0 soon so I'll make sure it's ok then. Looks like I'll have to avoid the higher B1 values until I get time to do this. Sorry for wasting people's time.

I'm not familiar with that problem. Which shell do you use? You should be able to use ulimit (or limit) to up your stacksize.

Greenbank · 2006-04-26, 15:51

From the GMP FAQ:

"
Q3: I get a segfault in a program that uses GMP to calculate numbers with several hundred thousand digits. Why?

A3: Upgrade to GMP 4.2. (Previous version of GMP allocated huge blocks of temporary memory on the stack, causing problems for some system.)
"

It used to say something along the lines of "run configure with --alloca" etc but the problem has gone away in 4.2.

[EDIT] Even fiddling with ulimit -s doesn't help. 4.2 is the answer.

Anyway, another factor:-

6739318350856072010032659714103 | 98749*2^1210+1

Found on the 233rd curve of B1=250000.

I've almost got everything done up to the 30 digit level (just finishing off a couple now). And moving on to the 35 digit level.

Mystwalker · 2007-02-14, 22:14

258317*2^1479+1 Mystwalker [reserved]

ltd · 2007-02-15, 19:10

Hi Dennis,

as an Info for your k/n pair there are 52 Tests done on the B1=3M Level.
Can be seen here: http://www.psp-project.de:8194/

I will remove your k from the ecmnet server. within the next days.

Lars

Mystwalker · 2007-02-15, 20:07

Hi Lars,

Thanks for that hint. Do you know the B2 bound of those curves?
Were they the default for gmp-ecm6?

That would help me in the calculation of the required curves...

Cheers,
Dennis

ltd · 2007-02-15, 21:21

The B2 was default ECM 5.
Nearly all ecmnet curves were done when there was no ECM 6 available
or only beta.

Can you give me a list (or a link) of curves needed per B1 step with ecm 6 then i would like
to adjust the curves needed for the unstarted B1 values

Lars

Mystwalker · 2007-02-15, 22:01

From the README file:

Code:

       digits D  optimal B1   default B2           expected curves
                                                       N(B1,B2,D)
                                              -power 1         default poly
          20       11e3         1.9e6             74               74 [x^1]
          25        5e4         1.3e7            221              214 [x^2]
          30       25e4         1.3e8            453              430 [D(3)]
          35        1e6         1.0e9            984              904 [D(6)]
          40        3e6         5.7e9           2541             2350 [D(6)]
          45       11e6        3.5e10           4949             4480 [D(12)]
          50       43e6        2.4e11           8266             7553 [D(12)]
          55       11e7        7.8e11          20158            17769 [D(30)]
          60       26e7        3.2e12          47173            42017 [D(30)]
          65       85e7        1.6e13          77666            69408 [D(30)]

          Table 1: optimal B1 and expected number of curves to find a
          factor of D digits with GMP-ECM.

ltd · 2007-02-16, 17:49

I promise next time i will look into the readme before asking stupid questions.

Lars

VJS · 2007-02-22, 14:04

If I can make a suggestion...

It has been suggested in a couple of places that the B2 value should be a little larger than default.

Basically what you want is roughly equal time spent during Stage1 and Stage2.

Of course running out of memory or having to use the -k option will quickly increase the time for stage2 (this is sub-optimal of course).

One could also start right in at the 40 or 45 digit level the curves do not nessarily have to be done in order.

Mystwalker · 2007-02-22, 18:35

Quote:

Originally Posted by VJS

It has been suggested in a couple of places that the B2 value should be a little larger than default.

Basically what you want is roughly equal time spent during Stage1 and Stage2.

As far as I know, gmp-ecm does quite good in choosing adequate default B2 values. Actually, one should lower the B2 value by ca. an order of magnitude when stage1 is done with prime95/mprime (for base2 numbers).

Quote:

Of course running out of memory or having to use the -k option will quickly increase the time for stage2 (this is sub-optimal of course).

Definitive.
Especially when using prime95/mprime for stage1, I'd advise to do some test curves with various B2 values and calculate with value is optimal.
And don't forget the -treefile option - it can be far better than increasing the k value in some situations.

2006-04-26, 15:51	#36
Greenbank Jul 2005 2·193 Posts	From the GMP FAQ: " Q3: I get a segfault in a program that uses GMP to calculate numbers with several hundred thousand digits. Why? A3: Upgrade to GMP 4.2. (Previous version of GMP allocated huge blocks of temporary memory on the stack, causing problems for some system.) " It used to say something along the lines of "run configure with --alloca" etc but the problem has gone away in 4.2. [EDIT] Even fiddling with ulimit -s doesn't help. 4.2 is the answer. Anyway, another factor:- 6739318350856072010032659714103 \| 987492^1210+1 Found on the 233rd curve of B1=250000. I've almost got everything done up to the 30 digit level (just finishing off a couple now). And moving on to the 35 digit level. Last fiddled with by Greenbank on 2006-04-26 at 15:55*

2007-02-15, 19:10	#38
ltd Apr 2003 2²·193 Posts	Hi Dennis, as an Info for your k/n pair there are 52 Tests done on the B1=3M Level. Can be seen here: http://www.psp-project.de:8194/ I will remove your k from the ecmnet server. within the next days. Lars Last fiddled with by ltd on 2007-02-15 at 19:12

2007-02-15, 21:21	#40
ltd Apr 2003 1404₈ Posts	The B2 was default ECM 5. Nearly all ecmnet curves were done when there was no ECM 6 available or only beta. Can you give me a list (or a link) of curves needed per B1 step with ecm 6 then i would like to adjust the curves needed for the unstarted B1 values Lars Last fiddled with by ltd on 2007-02-15 at 21:23

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
5*2^n-1 Reservation Thread	VBCurtis	Riesel Prime Search	661	2021-05-08 02:21
LLR reservation	hhh	Prime Cullen Prime	6	2007-09-17 04:27
Sieve reservation	hhh	Prime Cullen Prime	5	2007-08-03 04:37
P-1 reservation	hhh	Prime Cullen Prime	1	2007-05-16 20:12
15k reservation	gribozavr	15k Search	3	2005-03-11 14:04

2006-04-26, 13:31	#34
Greenbank Jul 2005 386₁₀ Posts	Doh, it's the stack allocation problem, I thought I had compiled GMP with the right options. Will be moving to GMP 4.2.0 soon so I'll make sure it's ok then. Looks like I'll have to avoid the higher B1 values until I get time to do this. Sorry for wasting people's time.

2007-02-14, 22:14	#37
Mystwalker Jul 2004 Potsdam, Germany 3·277 Posts	258317*2^1479+1 Mystwalker [reserved]

2007-02-15, 20:07	#39
Mystwalker Jul 2004 Potsdam, Germany 3·277 Posts	Hi Lars, Thanks for that hint. Do you know the B2 bound of those curves? Were they the default for gmp-ecm6? That would help me in the calculation of the required curves... Cheers, Dennis

2007-02-16, 17:49	#42
ltd Apr 2003 2²·193 Posts	I promise next time i will look into the readme before asking stupid questions. Lars

2007-02-22, 14:04	#43
VJS Dec 2004 299₁₀ Posts	If I can make a suggestion... It has been suggested in a couple of places that the B2 value should be a little larger than default. Basically what you want is roughly equal time spent during Stage1 and Stage2. Of course running out of memory or having to use the -k option will quickly increase the time for stage2 (this is sub-optimal of course). One could also start right in at the 40 or 45 digit level the curves do not nessarily have to be done in order.