mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   No Prime Left Behind (https://www.mersenneforum.org/forumdisplay.php?f=82)
-   -   LLRnet supports LLR V3.8! (LLRnet2010 V0.73L) (https://www.mersenneforum.org/showthread.php?t=13165)

kar_bon 2010-03-12 22:57

LLRnet supports LLR V3.8! (LLRnet2010 V0.73L)
 
1 Attachment(s)
[CENTER][SIZE=+2]LLRnet supports LLR V3.8 and higher! (LLRnet2010 V0.73L)[/SIZE][/CENTER]


LLRnet is a Client/Server program to search for primes by using LLR.

LLRnet was programmed in 2004-2005 by Vincent Penné in LUA. It uses internally LLR V3.5 programmed by Jean Penné.

The latest LLR-Version available is 3.8.1, which is 10%-20% faster than V3.7.1c, with more possibilities in testing different values and less issues in small n-values.

Vincent do not support a newer version of LLRnet, so the idea is to make both (LLRnet and LLR) working together with a script, using the feature for the client/server communication and the speed of the new LLR V3.8.1!

There are 2 versions of this script available:

[B]A Perl-script for Linux 32-bit [/B][URL="http://www.noprimeleftbehind.net/downloads/llrnet-script-perl-0.74-linux32.zip"]here[/URL] (V0.74 w/LLRnet 3.8.2).

[B]A DOS-script for Windows 32-bit[/B] [URL="http://www.rieselprime.de/dl/LLRnetV073.zip"]here[/URL] (V0.73 w/LLRnet 3.8.1).

Older versions:
[B]A Perl-script for Linux 32-bit [/B][URL="http://www.rieselprime.de/dl/llrnet-script-perl-0.71-linux32.zip"][COLOR=#000080]here[/COLOR][/URL] (V0.71 w/LLRnet 3.8.0).
[B]A DOS-script for Windows 32-bit[/B] [URL="http://www.rieselprime.de/dl/LLRnet_new.zip"]here[/URL] or [URL="http://noprimeleftbehind.net/downloads/llrnet-script-batch-0.72-win32.zip"]here[/URL] (V0.72 w/LLRnet 3.8.0).

Please read the README for more information about settings and handling.

[B]Many thanks to:[/B]
- Vincent Penné for the LLRnet version
- Jean Penné for the current LLR V3.8.0 (and further development).
- Max Dettweiler and Gary Barnes for the conversion of the script in Perl and testing
- Ian Gunn for testing.

This script was done during the last 5 weeks and was well tested in many cases.
If there're any issues or suggestions about this script, please post here.

[B]Notes:[/B]
- There were no changes on the Server-side from LLRnet, only the Client was changed.
Please uses all files included with the downloads for proper working!

- To use a newer Version of LLR, copy the latest cLLR (Win/Linux) in the client folder.

- To use a local LLRnet-server the Win-version contains a small example.


[B]Screenshot and batches:[/B]

Here is a screenshot of 4 clients running under WIN together with 2 batches to show information for:

- Pairs done all over
- Primes found all over
- Currently reserved pairs
- Setting of the work units cache
- Current line cLLR is working on

There's also a batch starting all 4 clients by calling the "do.bat" and entitle each DOS-box separately.

The batches can be downloaded [URL="http://www.rieselprime.de/dl/LLRnet2010_Batches.zip"]here[/URL].


Happy hunting with a new Dimension in Prime Searching with LLRnet!

K.Bonath

em99010pepe 2010-03-13 09:28

I'm getting CRC errors when unpacking the zip. Could you please check it up? I will try to install winrar to see if the issue is with IZArs.

EDIT: It was a download error, got it now. Sorry.

kar_bon 2010-03-13 09:31

i've downloaded all 3 zip's and everything is ok!


AND extracted them all without error!

PCZ 2010-03-13 21:27

I've given this a try out using the dos script.
Working OK so far.

Noticed when running bases other than 2 that the results file in llrnet server always reports base 2.
The hist files in the client directories report the base correctly.

Good Job

kar_bon 2010-03-13 21:30

[QUOTE=PCZ;208296]I've given this a try out using the dos script.
Working OK so far.

Noticed when running bases other than 2 that the results file in llrnet server always reports base 2.
The hist files in the client directories report the base correctly.

Good Job[/QUOTE]

edit the line
[code]
displayFormat = "%s*2^%s-1"
[/code]

in the 'llr-serverconfig.txt' (folder 'LLRnet_server') and substitute the '2' with the base you're testing!

mdettweiler 2010-03-13 21:31

[quote=kar_bon;208297]edit the line
[code]
displayFormat = "%s*2^%s-1"
[/code]

in the 'llr-serverconfig.txt' and substitute the '2' with the base you're testing![/quote]
Yes, this happens regardless of what client you're using if you have displayFormat set incorrectly.

PCZ 2010-03-13 22:11

Thanks that fixed it.

gd_barnes 2010-03-14 09:22

I believe in giving credit where it is due: This was Karsten's original idea so it was up to him how to present it here. Max was instrumental in converting the Windows script to the Linux script in very quick order. I did extensive detailed testing on the Linux side and made several modifications to the Linux script.

Many thanks to both Karsten and Max for their extensive efforts on this and picking up the slack after I had to leave on a business trip a little over a week ago.

One note to the public: While this has been extensively "alpha" testing, it can still be considered in a public "beta" testing phase. By that I mean, to fully prove itself, it needs to have 100s of cores running it.

There is one thing that I wanted to bring up. I know that many people, including myself, don't always read the documentation. There is one thing that is somewhat important to understand with the new LLRnet. It tests "in batch", meaning it will not return any pairs to the server until it is completely done with its cache. For that reason, whenever you stop one of your cores, please be sure and run:

llrnet -c

What that will do is send already-tested results to the server and return the untested pairs at the same time. The nice improvement that we made is that you only have to do it once to return all pairs; both completed and uncompleted. (This was one of the more difficult things to make work correctly during testing.)

As an example, if you cached 5 pairs and decide you want to stop after completing testing on 3 of them. The above command will return the 3 completed results PLUS the 2 remaining untested pairs to the server.

This is important because if you don't do that, not only will you miss credit for your completed results, you could have a prime in those results and miss getting credit for a prime!

Sometime in late March or early April, we will plan to have our first rally in quite some time and of course we plan to use the new client. In the week prior to that, perhaps I can combine with another heavy hitter such as Ian or Lennart to put at least 100 cores on it in preparation for the upcoming rally.

Everyone have fun with it and if you find anything, please let us know. I know firsthand that it is a whole lot faster than the "old klunky" LLRnet and I can personally attest to the fact that it works on at least 35 cores running at once.


Gary

kar_bon 2010-03-14 14:11

I've just been aware of an issue in the Win-batch V0.70:

Scenario:
- the WUCachSize is set to a value greater than 1 (for example 5)
- stopping the script during LLR is testing and a prime was found (for example the 2nd pair was prime and stopped after 3 pairs done)
- cancelling undone jobs with calling 'do -c'

-> this prime was submitted to the server but not logged in the local file 'primes.txt'

I've updated the ZIP-file in the first post and attached the new script here.
Please rename the TXT-file to 'do.bat' and place it in the client-folder(s).

[b]Note on 2010-03-30: With the new version the attachment is obsolete and was deleted here! Please refer to the link in post #1! (kar_bon)[/b]

gd_barnes 2010-03-14 19:51

[quote=kar_bon;208363]I've just been aware of an issue in the Win-batch V0.70:

Scenario:
- the WUCachSize is set to a value greater than 1 (for example 5)
- stopping the script during LLR is testing and a prime was found (for example the 2nd pair was prime and stopped after 3 pairs done)
- cancelling undone jobs with calling 'do -c'

-> this prime was submitted to the server but not logged in the local file 'primes.txt'

I've updated the ZIP-file in the first post and attached the new script here.
Please rename the TXT-file to 'do.bat' and place it in the client-folder(s).[/quote]


Nice catch Karsten. That's what we have beta testing for. It's almost impossible to think of every situation that will come up in alpha testing. Other testing scenarios that you might check out: Do the same thing but where the 1st pair is a prime -or- where the most recently completed pair is a prime. For example in the above scenario having stopped after 3 pairs were processed out of 5, try having pair #1 be a prime -and- also try having pair #3 be a prime. Bugs frequently come out at the beginning or end of a process, especially when the process is interrupted by something.

One question: Did you update the client in post #1 here? If not, that would be helpful. The hotel I'm at on my trip has a fairly severe limitation on downloading anything of any size and is very slow in doing so. After a minute of attempting to look at your DOS link, I decided to stop it as it was only barely above 10%.

While this is a small issue that doesn't affect reporting to the server, all issues need to be ironed out. I will test the same scenario on the Linux client by Tuesday after I get back from my trip.

For just small 1 or 2-line changes like this, I'm going to suggest that we don't update the version # just yet until we get all known small issues ironed out. Once they are all fixed, then we can "officially" make it version 0.71 or something like that. If there is a big issue as a result of something in large-scale stress testing, then I think it would be good to have a fix for that be either version 0.71 or 0.8 immediately, depending on what you guys think.

Does that sound reasonable?


Gary

kar_bon 2010-03-14 20:02

[QUOTE=gd_barnes;208384]Does that sound reasonable?[/QUOTE]

it's ok for me!

i've also put the determination of the local prime-logfile (primes.txt) in the WIN-script at the start of it, so it's done only once for a batch-run. otherwise it would be done everytime a prime was found. not that big timing issue.


All times are UTC. The time now is 05:00.

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