mersenneforum.org  

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

Reply
 
Thread Tools
Old 2003-06-12, 03:42   #1
dotnet_dev
 
Jun 2003

5 Posts
Default Testing same number on multiple machines

Hi,

Can I test same number (Like a 10 million digits no.) on two machines? If so, how? :?

Thanks,

DD

dotnet_dev is offline   Reply With Quote
Old 2003-06-12, 06:15   #2
S80780
 
Jan 2003
far from M40

11111012 Posts
Default

Hi, DD!

If you mean setting one machine up to test a number and another one to DC it - yes, you can. But you'd have to email the DC - result to George.

If you mean splitting up the test into smaller tasks and perform these simultaneously on different machines - no, you can't. Because you need to know the residue of the last iteration to be able to perform the next one properly.
All you can do is let one machine perform the P-1 stages and the other performs the LL-Test. In this case, the P-1 machine should be the one with the biggest amount of available memory. Of cause, this could mean that you waste proceesing power of the machine perfroming the LL-Test as the P-1 Test may find a factor.

All in all it is far better to let each processor test its own exponent.

Benjamin
S80780 is offline   Reply With Quote
Old 2003-06-12, 23:26   #3
crash893
 
crash893's Avatar
 
Sep 2002

23×37 Posts
Default

you just make duplicate worktodo.ini's right

why woudl youw ant to do this out of curoisty
crash893 is offline   Reply With Quote
Old 2003-06-13, 12:24   #4
dotnet_dev
 
Jun 2003

5 Posts
Default Intent

>>
If you mean setting one machine up to test a number and another one to DC it - yes, you can. But you'd have to email the DC - result to George.
<<

Yes, my original intent is that regarding 10 million bit numbers, since it takes such a long time to do on any machine, be able to split parts of work on machines. As such, since I am a relative newbie to the GIMPS, can you kindly point me how to write the worktodo.ini for the two machines?

I would appreciate if you can point to all the stages which GIMPS would go through (i.e. Worktodo's for them) so I can decide on a manual schedule of my machines.

Other than writing this ini, is there anything else I need to modify/add/delete?

Thanks,

DD
dotnet_dev is offline   Reply With Quote
Old 2003-06-13, 15:24   #5
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

2·3·5·241 Posts
Default

You cannot split a single test across two machines.
Prime95 is offline   Reply With Quote
Old 2003-06-13, 15:53   #6
cheesehead
 
cheesehead's Avatar
 
"Richard B. Woods"
Aug 2002
Wisconsin USA

769210 Posts
Default Re: Intent

Further explanation:

The Lucas-Lehmer algorithm, used to determine whether Mersenne numbers are prime, is inherently serial. Each step requires use of the result of the immediately preceding step, and so cannot be performed before the immediately preceding step is completed. Thus, unlike many other algorithms, it cannot be parallelized so that the final result is computed faster on multiple computers than on one computer.

Now, the preceding statement applies to the L-L test on a single Mersenne number. But multiple computers can perform multiple L-L tests, each on one single Mersenne number, in parallel. This isn't running the algorithm in parallel; it's running multiple instances of the algorithm at the same time. So no individual result (i.e., whether or not a particular Mersenne number is prime) can be computed faster on multiple computers than on a single computer.

As S80780 pointed out earlier, one can run two L-L tests on the same Mersenne number at the same time on two computers. In that case, my "multiple computers can perform multiple L-L tests, each on one single Mersenne number, in parallel" statement still applies; it's just that the single Mersenne number on one computer is the same Mersenne number as the single Mersenne number on another computer.

Quote:
Originally Posted by dotnet_dev
my original intent is that regarding 10 million bit numbers, since it takes such a long time to do on any machine, be able to split parts of work on machines.
Yeah, (*sigh*) we've all wanted to be able to do that. :)
cheesehead is offline   Reply With Quote
Old 2003-06-14, 00:35   #7
clowns789
 
clowns789's Avatar
 
Jun 2003
The Computer

32×43 Posts
Default

I was thinking you could link two computers by 25-pin cable so the two computers can relay information.
clowns789 is offline   Reply With Quote
Old 2003-06-14, 09:15   #8
S80780
 
Jan 2003
far from M40

53 Posts
Default

Of cause, you can. But as LL - Testing is a serial type of work, you can't split up the LL - Test into several subtasks to work on. Besides, even if you could do so, it still would be better to let each processor work on its own exponent. This is due to the fact that the relay of information needs some cycles from each processor involved every time it is performed. So, even if LL - Testing could be parallelized, it would still be better for the Project as a whole to let each processor work on its own exponent.

Benjamin
S80780 is offline   Reply With Quote
Old 2003-06-14, 19:47   #9
smh
 
smh's Avatar
 
"Sander"
Oct 2002
52.345322,5.52471

29×41 Posts
Default

Quote:
Originally Posted by clowns789
I was thinking you could link two computers by 25-pin cable so the two computers can relay information.
Many computers are already connected by network cables to relay information.

One of the verification runs on 'M40' was done with a multithreaded version of glucas on a dual itanium processor. Although iterations have to be done in sequence, each iteration can be split among multiple cpu's.

This way you can speed up the test of a single number, but the throughput will be higer if you test two different numbers at the same time, each on one cpu.

I believe that the latency of even a gigabit network is way to large to speed up a test with two seperate pc's
smh is offline   Reply With Quote
Old 2003-06-15, 01:05   #10
chrow
 
Jun 2003

17 Posts
Default Re: Intent

I did not understand!!!


Quote:
Originally Posted by cheesehead
Now, the preceding statement applies to the L-L test on a single Mersenne number. But multiple computers can perform multiple L-L tests, each on one single Mersenne number, in parallel. This isn't running the algorithm in parallel; it's running multiple instances of the algorithm at the same time. So no individual result (i.e., whether or not a particular Mersenne number is prime) can be computed faster on multiple computers than on a single computer.

Does that mean that when it executes LL test in a mersenne are not sweeping all possible probabilities?
Because multiple machines can run multiple LL tests?

Moises Deangelo
chrow is offline   Reply With Quote
Old 2003-06-15, 08:43   #11
trif
 
trif's Avatar
 
Aug 2002

2×101 Posts
Default Re: Intent

Quote:
Originally Posted by chrow
I did not understand!!!
Does that mean that when it executes LL test in a mersenne are not sweeping all possible probabilities?
Because multiple machines can run multiple LL tests?

Moises Deangelo
All LL tests are double checked to ensure that they have the same result. The 64 bit residue of the last iteration is what is checked. If two separate runs on the same exopnent agree in this residue, then the result is deemed correct, and no further runs are needed. If they do not match, then at least one of the LL tests has had an error and the exponent will have to be tested again until two matching residues are found.
trif is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Error with large number on 32-bit machines EdH GMP-ECM 43 2013-03-01 14:27
Monitoring multiple client machines? dans Software 5 2010-01-03 04:30
Primality-testing program with multiple types of moduli (PFGW-related) Unregistered Information & Answers 4 2006-10-04 22:38
Putting prime 95 on a large number of machines moo Software 10 2004-12-15 13:25
Running on multiple machines Helfire Software 8 2004-01-14 00:09

All times are UTC. The time now is 12:20.

Sat Dec 5 12:20:01 UTC 2020 up 2 days, 8:31, 0 users, load averages: 1.42, 1.54, 1.68

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, 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.