mersenneforum.org  

Go Back   mersenneforum.org > New To GIMPS? Start Here! > Information & Answers

Reply
 
Thread Tools
Old 2021-05-31, 12:03   #1
drkirkby
 
"David Kirkby"
Jan 2021
Althorne, Essex, UK

23·72 Posts
Default What's the best way to do P-1 factoring, whilst doing a PRP test of the same exponent?

I manually requested an exponent to run a PRP test on, using a slow, FREE Amazon computer. I got assigned

https://www.mersenne.org/report_expo...8888137&full=1

A P-1 factoring test should have been done on that, but with only 1 GB RAM on the free account, and a 1.5% chance of a factor being found, I thought it was pointless doing it. So I started the PRP test, and skipped the P-1 factoring . As you can see, the PRP test is 0.1% complete, and is actually expected to take couple of hundred days. (If it gets to a category #1, then I will just move it to a faster computer and complete it there.)

However, it makes sense to run a P-1 test on it - the server gave me the assignment with the P-1 test. But I would rather run that on a much faster machine with a few hundred GB or RAM. If a factor is found, I will just stop the PRP test on the free Amazon account, with very little wasted time.

What's the best way to start that P-1 test using mprime? Obviously I will not get a valid assignment ID if I request one.

IMHO, it would be good if the server would issue a AID for a TF or P-1 test to someone, if they already have a valid AID for a PRP test on an exponent. They might feel it should be trial-factored a bit more before actually starting the PRP test.

Dave

Last fiddled with by drkirkby on 2021-05-31 at 12:04
drkirkby is offline   Reply With Quote
Old 2021-05-31, 12:34   #2
drkirkby
 
"David Kirkby"
Jan 2021
Althorne, Essex, UK

23·72 Posts
Default Suggestion: Allow someone to get an AID for TF or P-1 if they have an AID for PRP

Mod note: this post was moved into this related thread

If someone has an AID for a PRP test on an exponent, they may well feel it should be trial-factored a bit higher before starting the computationally expensive PRP test. If they have a fast GPU, they might want to do that. I think it would be good if that person could get an AID test for a TF on the exponent.

I feel the same should apply to P-1 factoring too. If one has a PRP test for an exponent that needs P-1 factoring, someone should be permitted to do that P-1 factoring on another computer if they want to. I currently have an AID for a PRP test on a machine with little RAM. It needs a P-1 factoring first, but I would rather do that on a machine with a few hundred GB of RAM, where it should be more effective.

Dave

Last fiddled with by Uncwilly on 2021-05-31 at 15:04 Reason: Explanation
drkirkby is offline   Reply With Quote
Old 2021-05-31, 13:09   #3
Viliam Furik
 
"Viliam Furík"
Jul 2018
Martin, Slovakia

22·3·72 Posts
Default

I have started a P-1 with B1 = 1,000,000 and B2 = 30,000,000. That should be reasonable for the exponent. It should finish in about two hours.

EDIT:
BTW, you don't need an AID to do the P-1. You can just do it.

" Pminus1=N/A,1,2,exponent,-1,bound1,bound2,tflevel " should do the work.

Last fiddled with by Viliam Furik on 2021-05-31 at 13:30
Viliam Furik is offline   Reply With Quote
Old 2021-05-31, 14:02   #4
drkirkby
 
"David Kirkby"
Jan 2021
Althorne, Essex, UK

6108 Posts
Default

Thank you Viliam Furík. Your two hours of P-1 factoring might save me 200 days on the slow, but free, Amazon account. Knowing how to do it was the main thing.
drkirkby is offline   Reply With Quote
Old 2021-05-31, 15:01   #5
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

978410 Posts
Default

Quote:
Originally Posted by drkirkby View Post
I manually requested an exponent to run a PRP test on, using a slow, FREE Amazon computer. I got assigned

108888137

and is actually expected to take couple of hundred days. (If it gets to a category #1, then I will just move it to a faster computer and complete it there.)
That is currently Cat 2. The rules say: Must be completed in as little as 180 days. You should not leave it on such a slow machine.

The idea behind the assignment scheming is:
Trial Factoring, then
P-1 factoring, then
First Time Primality Testing, then
Verification (DC or running the proof file).

In the past (15 years ago or more) some TF was done before handing out the assignment for primality testing. But, rarely was it all done before. So, the line in the worktodo tells m/Prime/95 what is left. But, with the advent of GPU's it made sense to do all of the TF work on those. So for years assignments handed out have had enough TF that m/Prime/95 is satisfied with the level. And for a while, we were doing the same with P-1. But, Ben has been so productive, that P-1 can't quite keep up.

Any assignment for a first time check will include the data if there is any TF or P-1 needed. So, there is no need for separate AID's on the same exponent. If you want to split that work up on your end, fine. But, since you have the AID, as long as the work is turned in by you (even if it is turned in on a different machine or without mentioning the AID) you will get credit for it. And until the primality test result is turned in (or it expires), the AID is still active.
Uncwilly is offline   Reply With Quote
Old 2021-05-31, 15:58   #6
drkirkby
 
"David Kirkby"
Jan 2021
Althorne, Essex, UK

23×72 Posts
Default

Quote:
Originally Posted by Uncwilly View Post
That is currently Cat 2. The rules say: Must be completed in as little as 180 days. You should not leave it on such a slow machine.

The idea behind the assignment scheming is:
Trial Factoring, then
P-1 factoring, then
First Time Primality Testing, then
Verification (DC or running the proof file).

In the past (15 years ago or more) some TF was done before handing out the assignment for primality testing. But, rarely was it all done before. So, the line in the worktodo tells m/Prime/95 what is left. But, with the advent of GPU's it made sense to do all of the TF work on those. So for years assignments handed out have had enough TF that m/Prime/95 is satisfied with the level. And for a while, we were doing the same with P-1. But, Ben has been so productive, that P-1 can't quite keep up.

Any assignment for a first time check will include the data if there is any TF or P-1 needed. So, there is no need for separate AID's on the same exponent. If you want to split that work up on your end, fine. But, since you have the AID, as long as the work is turned in by you (even if it is turned in on a different machine or without mentioning the AID) you will get credit for it. And until the primality test result is turned in (or it expires), the AID is still active.
I got the exponent by manual assignment yesterday - it was category 3. You can't get category 2 by manual assignment. As far as I can see from the category 3 rules

"Assignments are recycled if assignment is not started within 120 (unless manual testing) days or when the exponent moves midway into the first category and the assignment is more than 240 days old."

So does that not mean I have 240 days? According to the output from mprime/website, it will actually complete in 79 days, although it was saying 200 days yesterday. I don't trust the time estimates - a few days back I got category 0 assignment unassigned when the server (mprime?) concluded it would not finish until 2022. It was absolute rubbish.

I am not following you about the implications of P-1 factoring and Ben. Does it mean his throughput is making P-1 more or less necessary?

I notice that most of the category 0 and category 1 exponents I get don't need P-1. I get a few category 0 assignments, but not that many - most are category 1. When there are category 0 assignments available, what mechanism is used to determine who gets a category 0, and who gets a category 1?

Dave
drkirkby is offline   Reply With Quote
Old 2021-05-31, 16:30   #7
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

22·3·449 Posts
Default

Ben's high primality test throughput makes more P-1 factoring throughput necessary. Some of it by Ben.

Before Ben Delo joined, GIMPS P-1 factoring was keeping ahead of primality testing. Ben about doubled the total first-time primality testing rate, while P-1 factoring rate hardly increased. So the usual P-1 factoring ahead was no longer sufficient. I generally have a few cpu workers and a Radeon VII or more doing production first-test wavefront P-1 factoring continuously. Along with everyone else doing P-1 factoring, it is still not enough to keep ahead of Ben et al primality testing. It takes more Radeon VIIs doing P-1 factoring all day long (24/7) to temporarily supply Ben and others with enough ready-to-PRP exponents to keep up. At B1=650000, B2=24000000, a Radeon VII GPU can do a 104M P-1 factoring in ~36 minutes by Gpuowl V6.11-380, so ~40/day.

Re doing other computation types while holding an AID, it's more general than while holding a primality-test assignment, particularly with manual assignments without any expiration occurring. One could obtain a low bit level TF assignment, modify the upper level to go deeper, do P-1 factoring and PRP sequentially or in parallel.
TF and P-1 in parallel is not much risk of lost time since the run times are lower than primality test, and the probability of finding a factor by one or by the other are lower than the probability by running both making a primality test in progress moot. If a factor is found in one, report the no-factor other result first; for example if factored in TF, report the P-1 NF factor first, then the TF factor found, to dodge the "result not needed' that would result if the factor found is reported first.
Any* outstanding assignment type prevents issuance of any* other suitable assignment type for the same exponent, so while holding a valid current AID for an exponent, all suitable computation types can be done. It's up to the user to generate the correct additional worktodo entries (conforming to formats documented in the reference info for months if not years), and coordinate the activities on his own kit. But nothing will protect it from someone poaching.


* edit: I think make that those types appropriately involved in progressing current wavefront exponents in prime searching: TF, P-1, PRP (or ugh LL, and ECM is probably not appropriate for current wavefront exponents; P+1 is known not to be effective use of resources for prime searching. Ok, not "any", a specifiable few.)

Last fiddled with by kriesel on 2021-05-31 at 17:19
kriesel is offline   Reply With Quote
Old 2021-05-31, 16:40   #8
slandrum
 
Jan 2021
California

3·72 Posts
Default

The expiration rules for assignments apply to the category that the assignment was in when it was assigned, not to the category that it's in now. In your own assignments page at https://www.mersenne.org/workload/ , you can see the category that each assignment was in when it was assigned to you.
slandrum is offline   Reply With Quote
Old 2021-05-31, 16:53   #9
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

23×1,223 Posts
Default

Quote:
Originally Posted by kriesel View Post
Any outstanding assignment type prevents issuance of any other assignment type for the same exponent, so while holding a valid current AID for an exponent, all computation types can be done.
ECM is not such a case.
Uncwilly is offline   Reply With Quote
Old 2021-05-31, 17:25   #10
drkirkby
 
"David Kirkby"
Jan 2021
Althorne, Essex, UK

23×72 Posts
Default

Quote:
Originally Posted by slandrum View Post
The expiration rules for assignments apply to the category that the assignment was in when it was assigned, not to the category that it's in now. In your own assignments page at https://www.mersenne.org/workload/ , you can see the category that each assignment was in when it was assigned to you.
I thought that was the case.

https://www.mersenne.org/report_expo...8888137&full=1
was in category 3 when assigned. I will probably get some idea how long it is going to take when it has been running a while, as I don't trust the estimates on the website one bit.

My computer "jackdaw" has been sending in an average of more than one wavefront PRP test each day for the last few months, but still the website thinks they are going to take several days. Last week I was really p****d off when it unassigned a category 0 PRP test it thought would take until next year. The estimates are way off, and don't seem to be improving, despite that machine is on almost 24/7.
Attached Thumbnails
Click image for larger version

Name:	zzz.PNG
Views:	14
Size:	184.3 KB
ID:	24989   Click image for larger version

Name:	XXXXXXXXXXXXXXXXXXX.PNG
Views:	14
Size:	245.0 KB
ID:	24990  
drkirkby is offline   Reply With Quote
Old 2021-05-31, 19:14   #11
slandrum
 
Jan 2021
California

14710 Posts
Default

Did you get the cat 0 through the normal process (was it requested), or did you just attempt to grab it by assigning it to yourself? I'm really surprised if the server actually assigned it to you in the first place if if thought you couldn't complete it in a timely fashion.

I wouldn't worry about cat 0 assignments anyway, they will get assigned very shortly after becoming available, there's no need to try to rush in and grab them.

How many workers do you have on the computer? Have you always had that many workers on it? If you've changed the number of cores assigned to each worker at some point, that could be confusing things.

Last fiddled with by slandrum on 2021-05-31 at 19:20
slandrum is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
i am new. How to manually test an exponent from 2^76 and up king Information & Answers 6 2018-02-08 14:28
Exponent fully factored whilst only 74% known mattmill30 Factoring 3 2016-08-14 18:09
What if you test an exponent on your own? LiquidNitrogen Information & Answers 2 2011-07-03 22:57
Is it possible to test an exponent > 79300000 ? Ding_Cong Software 4 2005-11-04 17:39
LL test of exponent in 41-42M range edorajh Lone Mersenne Hunters 2 2004-01-02 17:16

All times are UTC. The time now is 08:22.


Sat Jul 24 08:22:58 UTC 2021 up 1 day, 2:51, 1 user, load averages: 1.35, 1.64, 1.60

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.