mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Data

Reply
 
Thread Tools
Old 2022-08-30, 21:59   #1
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

2·5,399 Posts
Talking Thinking out loud about getting DC's "finished" by 2030

Ok people, we have had PRP with proofs for a while now. And PRP is what the server hands out for first time checks. With that the number of unverified first time tests has stopped increasing. The number of exponents that need a DC LL or PRP has been decreasing. I think that we should set a goal of, before this decade is out, completing the needed DC's in the range of 0M to 150M. As stands now we can consider everything below 62M effectively done. And the amount above 116M is minor. So, that is the real range that we really have to focus on. Right now below 150M we have 714853 exponents that need verification (yes there are some of those that are Certs, but those quickly handle themselves.)

The change over from adding more to the pile of DC's to be done, to fewer happened 2 years ago this month. After the initial wild swing in predicted date of DC being finished, the date dropped to 12/17/2029. It steadied a bit around the end of 2031/early 2032 for a while and now has started moving up toward August 2036. This is mainly a result of drop off in DC's being done. Unfortunately there are still PRP's being turned in that are without VDF's and a few people converting PRP's to LL.

To address this, I would like to see the following happen:
  1. The PrimeNet server stop handing out automatic FTC assignments to versions of Prime95 that can't do the VDF's.
  2. The PrimeNet server stop handing out automatic FTC's to machines that have been turning in PRP results without VDF's or have been turning in only LL results.
  3. Add a radio button for users requesting FTC's via manual assignments to the effect of "I understand that I should be doing PRP's with VDF's, I pledge to do so.", the default option would be "Please convert my FTC request to DC work".
  4. George adjust the default rate at which PrimeNet hands out DC's, to verify machine sanity, to a proportion of the number of FTC's turned in the last month. (If more machines do the odd DC once every 5 months or so, that would help, or once every XX exponents they turn in.
  5. Up the cut off point where machines are given DC's because they are too slow.
  6. Those that would like to help, start voluntarily doing one DC every 8 or 10 exponents that they do FTC's on.
  7. Those that want to help, assess their machines and move the slowest 10% of their machinescurrently doing FTCs to DC
  8. Those who want to do something fun, focus on the 100M to 110M range (each of those ranges is less than 1/4 of the normal for the ranges below them.)
  9. A few Mersennaries target the DC's from 120M to 150M to clear them out.
  10. GPU72 start adding a few DC-TF's to "what makes sense".
  11. Ask SRBase to do a single bit raising run through the range of 70M to 100M (or 110M).

Those are my thinking out loud ideas. I have been tracking the changes weekly +/- for the last year.
Attached Thumbnails
Click image for larger version

Name:	Time Gap.png
Views:	55
Size:	516.3 KB
ID:	27263  
Uncwilly is offline   Reply With Quote
Old 2022-08-30, 22:27   #2
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

22·2,713 Posts
Default

Quote:
Originally Posted by Uncwilly View Post
Those are my thinking out loud ideas. I have been tracking the changes weekly +/- for the last year.
It has been demonstrated that projects that have clear and obtainable goals often receive some attention.
chalsall is offline   Reply With Quote
Old 2022-08-30, 23:51   #3
tuckerkao
 
"Tucker Kao"
Jan 2020
Head Base M168202123

25·52 Posts
Default

Quote:
Originally Posted by Uncwilly View Post
GPU72 start adding a few DC-TF's to "what makes sense".
Ask SRBase to do a single bit raising run through the range of 70M to 100M (or 110M).[/LIST]
Those are my thinking out loud ideas. I have been tracking the changes weekly +/- for the last year.
Adding TFs of M120.2M, M120.9M, M122.0M, M122.2M, M122.7M, M123.6M, M123.7M, M123.9M, M124.2M, M125.6M, M125.7M and M125.9M to 278. Example: M120202123

Best if SRBase can run through M110M, will eliminate more potential candidates.

Last fiddled with by tuckerkao on 2022-08-30 at 23:57
tuckerkao is offline   Reply With Quote
Old 2022-08-30, 23:58   #4
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

2·5,399 Posts
Default

Quote:
Originally Posted by tuckerkao View Post
Adding TFs of M120.2M, M120.9M, M122.0M, M122.2M, M122.7M, M123.6M, M123.7M, M123.9M, M124.2M, M125.6M, M125.7M and M125.9M to 278
That would add nothing to the DC completion. That is ahead of the FTC's.
Uncwilly is offline   Reply With Quote
Old 2022-08-31, 00:00   #5
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

22·43·47 Posts
Default

A noble goal.

One of GIMPS' founding principles "is do what you find fun". Thus, I'd rather not refuse to hand out PRP assignments to those that specifically request them. I'm happy to change the rules of clients that have selected "Do what makes the most sense" work type.

I just checked the server code for "what makes sense" rules. Pre-version 30.3b3 clients are getting assigned 100% DC. Version 30.3b3 and later are getting 15% DC (which means ~4% of its CPU time is devoted to DC). I can up this percentage to something suitable.

With Gerbicz error checking and PRP proofs, the server no longer hands out yearly DCs. First-time users do get a DC.

Large memory machines can help with P-1 on poorly P-1'ed exponents that need DC. Can mersenne.ca provide that list?
Prime95 is offline   Reply With Quote
Old 2022-08-31, 00:05   #6
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

2×5,399 Posts
Default

Quote:
Originally Posted by Prime95 View Post
With Gerbicz error checking and PRP proofs, the server no longer hands out yearly DCs. First-time users do get a DC.
When you made that change it may have changed the curve.

And again, I am just "thinking out loud".

How many machines in a 3 month period are asking for "What makes sense"? I'm guessing that 3 months is enough to get a good average from slow machines.
Uncwilly is offline   Reply With Quote
Old 2022-08-31, 00:22   #7
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

22·43·47 Posts
Default

Random thoughts:

I'm not sure FTC will be at 150 million by 2030. Perhaps a lower goal (and time frame) like 100 million? Or maybe subgoals of 100M, 125M, and 150M if the FTC wavefront gets there?

Lots of opportunities for subprojects. Clearing DC ranges, TF targets, P-1 targets, DC left to do milestones.
Prime95 is offline   Reply With Quote
Old 2022-08-31, 00:38   #8
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

33·263 Posts
Default

Quote:
Originally Posted by Prime95 View Post
I just checked the server code for "what makes sense" rules. Pre-version 30.3b3 clients are getting assigned 100% DC. Version 30.3b3 and later are getting 15% DC (which means ~4% of its CPU time is devoted to DC). I can up this percentage to something suitable.

With Gerbicz error checking and PRP proofs, the server no longer hands out yearly DCs. First-time users do get a DC.
I suggest that prime95/mprime clients >V30.3b3 that have temporary disk space set too low to produce a useful proof file, get at least 50+% DC so they are performing more DC than they are creating a need for. (By assignment count, or by GHD?)

Whatever rules are put in place, they need to allow for proof generation sometimes failing, despite the user's best intentions. (I've had a file share go away for too long and cause that. Sometimes there are MD5 errors. Etc.)

Quote:
Originally Posted by Uncwilly View Post
That would add nothing to the DC completion. That is ahead of the FTC's.
It's a ~third order & years-ahead effect at best. Slightly deeper TF means slightly more factors found, so slightly fewer primality tests needed when 120M is in the first-test wavefront, of which a small fraction might have needed DC, due to proof gen failures, or the occasional user that deliberately runs PRP without proof gen, or runs LL, because it's a small memory small storage system.

Note that Mlucas users can not produce proof files, because a version supporting that feature has not yet been released.
Quote:
Originally Posted by Prime95 View Post
I'm not sure FTC will be at 150 million by 2030.
At the end of 2021 we were a year ahead in first test milestones of the extrapolation I posted in 2018 at
https://www.mersenneforum.org/showpo...5&postcount=11 and regularly update. It looks likely we'll get to the 111M milestone this year, 4M versus the extrapolated 6M. Ben Delo's large throughput reduced at least for a while. If we continue at ~4M/year (which requires increasing computing power as exponents rise), 2030 would be around 140M.

Last fiddled with by kriesel on 2022-08-31 at 00:56
kriesel is online now   Reply With Quote
Old 2022-08-31, 02:25   #9
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

2×5,399 Posts
Default

Quote:
Originally Posted by Prime95 View Post
I'm not sure FTC will be at 150 million by 2030. Perhaps a lower goal (and time frame) like 100 million?
I started tracking to 150M because that was an easy boundary to grab from the Work Distribution Map. And because it was far enough to not likely be exceeded soon. The main thing is cleaning up behind the FTC's. If those fall below 1000 or so, we have finished, no matter what remains above the FTC wave.
Uncwilly is offline   Reply With Quote
Old 2022-08-31, 02:32   #10
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

1079810 Posts
Default

Quote:
Originally Posted by kriesel View Post
I suggest that prime95/mprime clients >V30.3b3 that have temporary disk space set too low to produce a useful proof file, get at least 50+% DC so they are performing more DC than they are creating a need for.
Quote:
It's a ~third order & years-ahead effect at best.
Quote:
Note that Mlucas users can not produce proof files, because a version supporting that feature has not yet been released.
We shall have to withhold all funding to the developer until that issue is fixed.

Quote:
If we continue at ~4M/year (which requires increasing computing power as exponents rise), 2030 would be around 140M.
Thanks for all of those comments. I did not have the data for that last, nor did it occur to me about the first.
Uncwilly is offline   Reply With Quote
Old 2022-08-31, 03:11   #11
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

33·263 Posts
Default

Completing DC's on Uncwilly's figure 714853 exponents "by" (which I think means before) 2030 means averaging 714853/7.333 years or ~97484/year, 8124/month, 267./day. That average rate is near the high end of daily rates I observed below Mp#49* during several months of tracking progress in DC up to Mp#48.

Countdown to verifying all tests below M(74207281): 207 558
Countdown to verifying all tests below M(77232917): 262 781
Countdown to verifying all tests below M(82589933): 361 429
from https://www.mersenne.org/report_milestones/ last updated 2022-08-31 02:30:17 UTC,

In 60M-100M I calculate ~668264 needing Cert or DC.
In 100M-120M, 190176 verified, while 45698 needing verification, or over 19% of those tested at least once.
(Derived from https://www.mersenne.org/primenet/ PrimeNet Activity Summary 2022-08-30 23:00 UTC)

In 120-150M there are less than ~1200 total first or double checked.
There are only 3 cases of triple digits in a million-exponent-value bin in that span.

In 126M,
there are 219 PRP results. All the 210 unverified results are from the same user GAN with no verified results, submitted well after prp proof generation became available in gpuowl and mprime/prime95.
Also LL; 1 verified, 2 unverified by patgie.

In 131M,
14 verified PRP, 1 unverified 131500093 (by me before PRP proof gen was available).
Also a lot of LL; 1 verified exponent, 191 unverified (almost all by yphysics and all well after prp proof became available in both mprime/prime95 and gpuowl).

In 143M,
5 verified PRP; all the 248 unverified results are from the same user Franklin Webber with only 3 verified results; some of each are from after prp proof became available in gpuowl and prime95/mprime.
Also in LL 1 exponent verified, 1 unverified.

If those 3 users GAN, yphysics, and Franklin Webber would cease generating more unverified results requiring DC work, by generating proof files, it would help a lot, relatively speaking. (GAN and yphysics are each in the top 30 among first primality test producers in the past year.)
The utility of a first test that will require a full double check is low or potentially even negative.
There may be folks who don't know proof generation is an option, or thought they implemented it but haven't, or whatever.
Or they have what they feel are very good reasons for not doing so. I tried to contact them by PM to see what's up, but the PrimeNet usernames are not valid Mersenne forum usernames.

Further up the exponent scale:
In 160M
PRP: 489 exponents verified with proofs, 5 unverified (VERY GOOD!)
LL: 2 verified results for 1 exponent, 4 unverified results, 3 of which are by Borek after prp proof introduction.

In 332M:
PRP: 344 verified results for 341 exponents, 334 unverified for 333 exponents
LL: 62 verified results for 28 exponents; 715 unverified results; 10 known bad results

In 333M:
PRP: 69 verified results for 68 exponents; 56 unverified results
LL: 4 results verified for 2 exponents; 104 unverified results; 1 known bad result.

In 334M:
PRP: 3 verified results, 7 unverified (5 by GAN)
LL: 2 unverified results
Attached Files
File Type: pdf mp48verificationmilestoneestimate.pdf (31.6 KB, 20 views)
kriesel is online now   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
COMPLETE!!!! Thinking out loud about getting under 20M unfactored exponents petrw1 Data 1414 2022-08-31 09:50
Is "mung" or "munged" a negative word in a moral sense? Uncwilly Lounge 15 2020-04-14 18:35
Aouessare-El Haddouchi-Essaaidi "test": "if Mp has no factor, it is prime!" wildrabbitt Miscellaneous Math 11 2015-03-06 08:17
Loud thinking on irregular primes devarajkandadai Math 4 2007-07-25 03:01
Would Minimizing "iterations between results file" may reveal "is not prime" earlier? nitai1999 Software 7 2004-08-26 18:12

All times are UTC. The time now is 19:38.


Sat Nov 26 19:38:17 UTC 2022 up 100 days, 17:06, 1 user, load averages: 1.72, 1.32, 1.11

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2022, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.

≠ ± ∓ ÷ × · − √ ‰ ⊗ ⊕ ⊖ ⊘ ⊙ ≤ ≥ ≦ ≧ ≨ ≩ ≺ ≻ ≼ ≽ ⊏ ⊐ ⊑ ⊒ ² ³ °
∠ ∟ ° ≅ ~ ‖ ⟂ ⫛
≡ ≜ ≈ ∝ ∞ ≪ ≫ ⌊⌋ ⌈⌉ ∘ ∏ ∐ ∑ ∧ ∨ ∩ ∪ ⨀ ⊕ ⊗ 𝖕 𝖖 𝖗 ⊲ ⊳
∅ ∖ ∁ ↦ ↣ ∩ ∪ ⊆ ⊂ ⊄ ⊊ ⊇ ⊃ ⊅ ⊋ ⊖ ∈ ∉ ∋ ∌ ℕ ℤ ℚ ℝ ℂ ℵ ℶ ℷ ℸ 𝓟
¬ ∨ ∧ ⊕ → ← ⇒ ⇐ ⇔ ∀ ∃ ∄ ∴ ∵ ⊤ ⊥ ⊢ ⊨ ⫤ ⊣ … ⋯ ⋮ ⋰ ⋱
∫ ∬ ∭ ∮ ∯ ∰ ∇ ∆ δ ∂ ℱ ℒ ℓ
𝛢𝛼 𝛣𝛽 𝛤𝛾 𝛥𝛿 𝛦𝜀𝜖 𝛧𝜁 𝛨𝜂 𝛩𝜃𝜗 𝛪𝜄 𝛫𝜅 𝛬𝜆 𝛭𝜇 𝛮𝜈 𝛯𝜉 𝛰𝜊 𝛱𝜋 𝛲𝜌 𝛴𝜎𝜍 𝛵𝜏 𝛶𝜐 𝛷𝜙𝜑 𝛸𝜒 𝛹𝜓 𝛺𝜔