Register FAQ Search Today's Posts Mark Forums Read

 2019-05-20, 02:59 #23 nomead     "Sam Laur" Dec 2018 Turku, Finland 4758 Posts About the manual assignment progress reports. Different programs have wildly different progress logs. Example from Mlucas: [May 20 03:38:41] M50346019 Iter# = 24280000 [48.23% complete] clocks = 00:26:42.587 [160.2587 msec/iter] Res64: 54912F2731A45A41. AvgMaxErr = 0.042192089. MaxErr = 0.062500000. Residue shift count = 12144693. But minimum common requirements should be current iteration count and Res64/shift so that interim residues could be stored, if needed. I propose that instead of having the manual results submission form decipher many different log formats, there should be a simple common format, and let the client submission script do the conversion to that.
2019-05-21, 02:20   #24
kriesel

"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

10111111110112 Posts

Quote:
 Originally Posted by nomead About the manual assignment progress reports. Different programs have wildly different progress logs. Example from Mlucas: [May 20 03:38:41] M50346019 Iter# = 24280000 [48.23% complete] clocks = 00:26:42.587 [160.2587 msec/iter] Res64: 54912F2731A45A41. AvgMaxErr = 0.042192089. MaxErr = 0.062500000. Residue shift count = 12144693. But minimum common requirements should be current iteration count and Res64/shift so that interim residues could be stored, if needed. I propose that instead of having the manual results submission form decipher many different log formats, there should be a simple common format, and let the client submission script do the conversion to that.
I'm not concerned about Mlucas format, because the word is Madpoo and EWMayer were working together on implementation of the Primenet API in Mlucas v19, in development now. Mlucas is a cpu application, which is what the Primenet API was designed for and implemented for on the server and in prime95 & mprime.

I requested only CUDALucas progress reporting support, initially.

But what client submission script? Some gpu applications lack any client submission script, so all work reservation and submission is manual. Manual progress updates will be done only as needed in that use case. CUDALucas, for example, and CUDAPm1, have no client submission scripts. (Also many earlier versions of gpuowl, but there are so many log styles in gpuowl versus version, and many different gpuowl versions are fastest depending on the exponent / fft length, so I think supporting all the older useful versions of gpuowl progress reporting by the manual submission page is too much to ask.)

2019-05-21, 04:08   #25

"Sam Laur"
Dec 2018
Turku, Finland

317 Posts

Quote:
 Originally Posted by kriesel I'm not concerned about Mlucas format, because the word is Madpoo and EWMayer were working together on implementation of the Primenet API in Mlucas v19, in development now.
And it currently uses a Python script adapted from mfloop.py, works well enough but of course doesn't handle progress reports.

Quote:
 But what client submission script?
For CUDALucas / clLucas, there's llloop.py included in primetools - haven't tested it though, since I put about 99% of GPU effort into factoring now. But it shouldn't be that hard to adapt to other GPU LL / PRP applications.

Quote:
 (Also many earlier versions of gpuowl, but there are so many log styles in gpuowl versus version, and many different gpuowl versions are fastest depending on the exponent / fft length, so I think supporting all the older useful versions of gpuowl progress reporting by the manual submission page is too much to ask.)
Pretty much what my point was, choose some simple common format and then stick to it. So if the client output format changes, adapt the script, not the submission page. Like worktodo.txt - easy to parse and it's been around long enough that pretty much everything should support it directly. mfaktc seems to be pretty flexible with its output even now, but I see that as less of a problem anyway. I mean, how many people do GPU trial factoring work that takes so long, that expiration without progress reports could be a problem?

2019-05-21, 17:17   #26
kriesel

"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

7×877 Posts

Quote:
 Originally Posted by nomead For CUDALucas / clLucas, there's llloop.py included in primetools - haven't tested it though, since I put about 99% of GPU effort into factoring now. But it shouldn't be that hard to adapt to other GPU LL / PRP applications.
I contacted the author some time ago; essentially no one used lloop.py and it was abandoned years ago. And it does not do progress reports.

Quote:
 how many people do GPU trial factoring work that takes so long, that expiration without progress reports could be a problem?
Agreed. And why I emphasized gpu primality test progress reporting, not P-1 or TF. Weeks-long TF or P-1 assignments are possible but not common.

A possible benefit of CUDALucas or cllucas progress reporting, that is not very relevant to the other gpu applications, is early detection of pathological runs that are producing all 0x0 or 0x02 or 0xfffffff800000000 interim res64, with potential savings of days or weeks of wasted run time. Some users don't recognize those as an issue, but if they could be persuaded to do interim progress reports, the server might educate them!
Mainly though, manual progress reporting is about preventing expiration of computations that are progressing, and reducing the chafing of other users who see no indicated progress at server reports and imagine wrongly that means no progress is occurring.

Last fiddled with by kriesel on 2019-05-21 at 17:19

2019-07-14, 15:11   #28
kriesel

"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

7·877 Posts

Quote:
 Originally Posted by Batalov Sorry, no, not better. Now this post looks like a tumor. Volume keeps growing, information content stays constant. To quote from Linda Richman, "A little basic background" is neither little nor basic. "Talk amongst yourselves. I am feeling varklempt" (...after reading just 1/3 of it). Sticking to the same question as before (and just that, for brevity): "current item #14" still says "For example, 26 - 1 = (22-1)2 23-1." Which neither explains the point, nor is (ahem) correct. (22-1)2 23-1 = 71, not 63.
I began writing this is an attempt to go from beginning definitions to a rudimentary basis for understanding or a bit more, with links forward, both for sincere newbies, and as a place to point nonsensical trolls and perhaps save time correcting or countering them. (Trolls can be initially amusing but quickly become tiresome.)

I'm open to constructive suggestions for the title, organization, and content.
Re https://www.mersenneforum.org/showpo...32&postcount=2, I saw M6 != M(6) and previously missed the missing parentheses, since added. (In the usual notation, M6=217-1; for other readers, https://www.mersenne.org/primes/)

 2019-07-14, 15:54 #29 Dylan14     "Dylan" Mar 2017 22·3·72 Posts A few suggestions: For 5. it should be that a single nontrivial factor of a natural number proves the number composite (meaning a number other than 1 or the number itself). For 20. link to the full list of programs (which is https://www.mersenneforum.org/showpo...91&postcount=2). And finally, perhaps after 28., we should note that some tests need to be done 3 times (or more) due to mismatches, and that there is a dedicated effort to handle these (https://www.mersenneforum.org/showthread.php?t=24148).
2019-07-14, 18:26   #30
kriesel

"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

7×877 Posts

Quote:
 Originally Posted by LaurV You do not need to write 2kp+1 with spaces. My joke related to the "operation of k" was to say that you need to be constant in notation, and do not use "x" for multiplication, but it was not my intention to say that writing 2kp with no spaces is wrong.
No worries. Your post led me to more generally reviewing conventions used, and state them, and getting rid of some ambiguities and inconsistencies along with making it I think more readable in spots. For example I had used f2 when meaning f2 not f times 2.
Thanks also to Dylan14.

2019-07-14, 20:56   #31
Batalov

"Serge"
Mar 2008
Phi(4,2^7658614+1)/2

25×3×101 Posts

Parentheses are important but that is only the first problem that comes up with that explanation.

Quote:
 Originally Posted by kriesel For example, 26 - 1 = 2(2*3) - 1 = (22-1)2 (23-1)
No. That's not algebra, that's like explaining that "this chemical reaction works because of... alchemistry!"
"Abracadabra, 63 = 3 * 3 * 7 = {and because of that} = (22-1)2 (23-1)"

No, the "3"s here are not (22-1)2. they are (2+1)(22-21+1)

Straightforward explanation:
26 - 1 = (23)2 - 1 = (23+1)(23-1), because this is a x2 - 1 and you learned this at school.
Or
26 - 1 = (22)3 - 1 = (22-1)(24+22+1), because this is a x3 - 1 and you learned this at school or you learned this now. Take a piece of paper, open the parentheses, and observe how everything cancels.
That is sufficient. If you want to get fancy, you can continue and say, "but there is more,
(23+1) = (2+1)(22-21+1) and that equals 3 *3 "

Quote:
 Originally Posted by kriesel ...and 215-1 = 2(3*5) - 1 = 32767 = 7 * 31 * 151 = (23 - 1) (25 - 1) 151.
Same thing. Because there is no explanation what 151 is, there is no coherent story just "this is so, because I said so".

215 - 1 = (25)3 - 1 = (25-1)(210+25+1)
Or
215 - 1 = (23)5 - 1 = (23-1)(212+29+26+23+1)
What's more for every 2(r*s) - 1 (with s>1), have a look at how the above factorization repeats every time: https://primes.utm.edu/notes/proofs/Theorem2.html

And finally, there is a much better way. Just say - "you will benefit from reading this ( https://www.mersenne.org/various/math.php ). Ask questions if you will have them."
P95's webpage is coherent, brief and convincing, and has links to more detailed explanations. Yours leaves the reader more confused that they were when they started reading it.

 2019-07-15, 12:57 #32 Dr Sardonicus     Feb 2017 Nowhere 14FE16 Posts Composite exponents As is well known, (xn - 1) /(x - 1) = xn-1 + xn-2 + ... + 1. Now let a > 1 and b > 1 be integers. We apply the above formula to xab - 1. If we substitute xa for x and b for n, we obtain (xab - 1)/(xa - 1) = xab - a + xab-2a + ... + 1. Similarly, substituting x^b for x and a for n, (xab - 1)/(xb - 1) = xab - b + xab-2b + ... + 1. The ne plus ultra of this is the "cyclotomic factorization" corresponding to the factorization into cyclotomic polynomials $x^{n}\; -\; 1\; =\; \prod_{d\mid n}\Phi_{d}(x)\text{.}$ Examples of its application may be found in this Forum, e.g. here For more on the application of cyclotomic polynomials to integer factorization I suggest this paper.
2019-07-15, 17:55   #33
hansl

Apr 2019

5·41 Posts

From new participant reference:
Quote:
 Originally Posted by kriesel Mersenneforum.org covers many different ways of aiding the GIMPS progress, through trial factoring, P-1 factoring, Lucas-Lehmer testing, probable-prime testing, double checking, etc. of Mersenne numbers, 2p-1.
Needs a superscript: 2p-1
Code:
2[sup]p[/sup]-1

 Similar Threads Thread Thread Starter Forum Replies Last Post kriesel kriesel 35 2021-12-07 15:54 Rincewind Five or Bust - The Dual Sierpinski Problem 57 2011-02-06 21:53 jasong Twin Prime Search 311 2010-10-22 18:41 philmoore Five or Bust - The Dual Sierpinski Problem 83 2010-09-25 10:20 clowns789 Soap Box 3 2006-03-09 04:05

All times are UTC. The time now is 23:25.

Fri Jan 28 23:25:25 UTC 2022 up 189 days, 17:54, 1 user, load averages: 2.45, 2.10, 1.97