mersenneforum.org  

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

Reply
 
Thread Tools
Old 2021-10-08, 00:10   #1
piforbreakfast
 
Oct 2020
Terre Haute, IN

2×32×5 Posts
Default How to help pare down P95 backlog

What job categories need the most help right now? The replacement for my gaming computer that overheated and died will be here in a couple days and I think I'd like to have that one help whittle down the pile of neglected jobs. Which category is most in need of workers?

Also, is it possible to run a duplicate PRP or other test on exponents that appear to be reserved but which haven't shown any progress in several years? I've been going through a number of 100M-digit exponents on mersenne.ca and notice that several appear to have been in progress but no results have posted for at least a year or two.

Last fiddled with by piforbreakfast on 2021-10-08 at 00:49
piforbreakfast is offline   Reply With Quote
Old 2021-10-08, 02:33   #2
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

22·41·61 Posts
Default

Quote:
Originally Posted by piforbreakfast View Post
Which category is most in need of workers?
Doing DC's would be helpful.
Quote:
I've been going through a number of 100M-digit exponents on mersenne.ca
The 100M digit range is still decades away from routine testing. I would not advocate doing primality tests on them.
Uncwilly is offline   Reply With Quote
Old 2021-10-08, 03:35   #3
piforbreakfast
 
Oct 2020
Terre Haute, IN

2·32·5 Posts
Default

Quote:
Originally Posted by Uncwilly View Post
Doing DC's would be helpful.
Should I choose just double-checking or double-checking with proofs?

Last fiddled with by piforbreakfast on 2021-10-08 at 04:29
piforbreakfast is offline   Reply With Quote
Old 2021-10-08, 20:25   #4
kruoli
 
kruoli's Avatar
 
"Oliver"
Sep 2017
Porta Westfalica, DE

12158 Posts
Default

The difference in efficiency between them is small, lower than 3 % (estimated).

Doing a DC with a proof will give you an assingment for doing a PRP with proof on that number. This way, the DC is "done" in 99.9... % of cases. It is independet of the original run.

Doing "normal" DC might yield to a triple check etc. being performed if the original and your calculations mismatch. But in the vast majority of cases, they will match and thus tell the performer of the original calculation that their run was sane.

Personally, I am doing DC the "old fashioned" way. But: Do as you like.
kruoli is online now   Reply With Quote
Old 2021-10-09, 12:57   #5
piforbreakfast
 
Oct 2020
Terre Haute, IN

5A16 Posts
Default

For that matter, why are there still so many unverified LL tests in the system for exponents that have already been determined to not be prime? 57895237 was the last verified non-prime exponent listed on the front page as of a moment ago, but there are many exponents listed below that level, and not all of the unverified tests may be LL. Is P95 just ignoring these tests that are still in the system? Here's the link I was looking at:

https://www.mersenne.org/report_ll/?...dispdate=1&B1=
piforbreakfast is offline   Reply With Quote
Old 2021-10-09, 13:57   #6
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

22×41×61 Posts
Default

If you click through to see the actual status of those numbers, say for instance: 53551349
you will see that some that don't have a matching LL, do have a PRP with cert. Thus they are double checked and we should move on.

Last fiddled with by Uncwilly on 2021-10-09 at 13:58
Uncwilly is offline   Reply With Quote
Old 2021-10-09, 17:28   #7
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

578310 Posts
Default

Many exponents with a single LL test will never get a second matching LL test. And should never.

Compute effort of LL DC is 1 primality test + ~0.040816... TC, QC, etc for 2% / test error rate; ~1.0408 cost.

Compute effort of a PRP/proof as a DC substitute is ~1.005 primality test & only ~24 PPM error rate historically, so ~1.005024 cost; net savings ~3.56% by going with PRP DC/GEC/proof. (Overhead of Jacobi check for LL and GEC for PRP are comparable at ~0.2-0.3%, but GEC catches 99.99+% of errors while Jacobi catches 50.%.)

Most LL single tests are several years old, longer than typical hardware useful lifetime, so the benefit of identifying flaky hardware by LLDC is hypothetical.
kriesel is online now   Reply With Quote
Old 2021-10-09, 18:37   #8
sdbardwick
 
sdbardwick's Avatar
 
Aug 2002
North San Diego County

13×53 Posts
Default

Quote:
Originally Posted by kriesel View Post
Most LL single tests are several years old, longer than typical hardware useful lifetime, so the benefit of identifying flaky hardware by LLDC is hypothetical.
I run LL DC to test sanity of the DC hardware, not the first-time test hardware.
EDIT: In fact, one of your TC just let me know there is a machine that needs investigation.

Last fiddled with by sdbardwick on 2021-10-09 at 18:42
sdbardwick is online now   Reply With Quote
Old 2021-10-09, 20:00   #9
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

132278 Posts
Default

Quote:
Originally Posted by sdbardwick View Post
I run LL DC to test sanity of the DC hardware, not the first-time test hardware.
EDIT: In fact, one of your TC just let me know there is a machine that needs investigation.
PRP/GEC will give much quicker and more reliable feedback on system accuracy.
kriesel is online now   Reply With Quote
Old 2021-10-09, 20:10   #10
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

37·269 Posts
Default

Quote:
Originally Posted by kriesel View Post
PRP/GEC will give much quicker and more reliable feedback on system accuracy.
+1
chalsall is online now   Reply With Quote
Old 2021-10-09, 21:29   #11
sdbardwick
 
sdbardwick's Avatar
 
Aug 2002
North San Diego County

68910 Posts
Default

Without the need for frequent monitoring of the individual computers?
Currently, I just check the completed assignment report occasionally, and see if there are any mismatches. Takes 5 seconds 3 times a week.
PRP will complete successfully, albeit slowly, on computers with known issues.

Last fiddled with by sdbardwick on 2021-10-09 at 21:29
sdbardwick is online now   Reply With Quote
Reply

Thread Tools


All times are UTC. The time now is 18:12.


Wed Oct 20 18:12:50 UTC 2021 up 89 days, 12:41, 0 users, load averages: 1.37, 1.28, 1.24

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, 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.