mersenneforum.org  

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

Reply
 
Thread Tools
Old 2017-05-23, 03:37   #1
Mark Rose
 
Mark Rose's Avatar
 
"/X\(‘-‘)/X\"
Jan 2013

3·977 Posts
Default Best approach for P-1?

My computer cluster is loaded with memory that I'm not taking advantage of. I understand P-1 can use the memory, but I'm unsure where to start with configuring things optimally.

Each node has 4 cores and 32 GB of memory with good memory bandwidth. I'm currently using a single worker for LL/DC.

Suggestions?
Mark Rose is offline   Reply With Quote
Old 2017-05-23, 04:38   #2
S485122
 
S485122's Avatar
 
Sep 2006
Brussels, Belgium

2×5×167 Posts
Default

Use two workers : one for LL work and the other for P-1. On the machine where I do this, I assigned two cores to each worker. By running just one P-1 worker you avoid the problem of having two workers competing for high memory usage during stage 2.

Jacob
S485122 is offline   Reply With Quote
Old 2017-05-23, 12:32   #3
kladner
 
kladner's Avatar
 
"Kieren"
Jul 2011
In My Own Galaxy!

2×3×1,693 Posts
Default

IIRC, my experience was that Stage 1, P-1 makes use of 2 cores fairly well. Stage 2 seemed to use 1.5 to 1.7 cores out of 2. Perhaps this reflects lack of memory bandwidth on the machines I was using?
kladner is offline   Reply With Quote
Old 2017-05-23, 14:53   #4
Mark Rose
 
Mark Rose's Avatar
 
"/X\(‘-‘)/X\"
Jan 2013

B7316 Posts
Default

So maybe I should allocate 3 cores to an LL worker and 1 to a P-1 worker?

I'll have to do some benchmarking.
Mark Rose is offline   Reply With Quote
Old 2017-05-23, 17:27   #5
kladner
 
kladner's Avatar
 
"Kieren"
Jul 2011
In My Own Galaxy!

2·3·1,693 Posts
Default

Quote:
Originally Posted by Mark Rose View Post
So maybe I should allocate 3 cores to an LL worker and 1 to a P-1 worker?

I'll have to do some benchmarking.
I did a brief bit of P-1 recently. I did not check the behavior for 2+ cores in Stage 2. I did just what you said. 1 for P-1, and 3 on a single DCLL worker.

6700K @ 4.3 GHz, Kingston dual-rank 8 GB x 2, rated 2666, running @ 3200
A quick search of results.txt suggests that core completed an 81.5M assignment in 12-14 hours.
The 3 core DCLL took 57-60 hr, v 24-25 hr for all four cores.

Last fiddled with by kladner on 2017-05-23 at 17:43
kladner is offline   Reply With Quote
Old 2017-05-23, 19:15   #6
Mark Rose
 
Mark Rose's Avatar
 
"/X\(‘-‘)/X\"
Jan 2013

3×977 Posts
Default

Quote:
Originally Posted by kladner View Post
I did a brief bit of P-1 recently. I did not check the behavior for 2+ cores in Stage 2. I did just what you said. 1 for P-1, and 3 on a single DCLL worker.

6700K @ 4.3 GHz, Kingston dual-rank 8 GB x 2, rated 2666, running @ 3200
A quick search of results.txt suggests that core completed an 81.5M assignment in 12-14 hours.
The 3 core DCLL took 57-60 hr, v 24-25 hr for all four cores.
That's a brutal loss of throughput. Good data point.
Mark Rose is offline   Reply With Quote
Old 2017-05-23, 23:58   #7
kladner
 
kladner's Avatar
 
"Kieren"
Jul 2011
In My Own Galaxy!

2·3·1,693 Posts
Default

Quote:
Originally Posted by Mark Rose View Post
That's a brutal loss of throughput. Good data point.
I should probably verify that on additional exponents. With only two samples, both could have included down periods with no output.

Last fiddled with by kladner on 2017-05-23 at 23:58 Reason: ,
kladner is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Unorthodox approach to primes Erkan PrimeNet 7 2017-01-10 03:34
An Analytic Approach to Subexponential Factoring akruppa Math 2 2009-12-11 18:05
Best approach for 130 digits? boothby Factoring 10 2009-10-16 17:24
Factoring with the birthday problem approach ThiloHarich Factoring 0 2009-08-18 16:48
The binary multiply convolution approach Dresdenboy Math 4 2003-06-02 17:23

All times are UTC. The time now is 11:01.

Tue May 11 11:01:56 UTC 2021 up 33 days, 5:42, 1 user, load averages: 2.52, 2.33, 2.05

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.