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)

gd_barnes 2010-09-29 22:21

[QUOTE=mdettweiler;231518]Ah, good catch. I changed it in the script's internal comments but not in the readme. That is fixed now and the corrected zip file is uploading right now. (I also added an entry in the changelog indicating that the version number was changed to match the Windows script, but that no actual changes to the do.pl script were made.)

Karsten, just to clarify: are you using V0.73 or V0.74 to denote the latest version of your script with LLR 3.8.2 in it? Either's fine with me, but if you're sticking with V0.73 then I should probably make sure that the Linux client matches that instead of being V0.74 (since, after all, the whole point of raising the version number is so that they match).[/QUOTE]

You do realize that you did not respond to my first point in the first para here. Correct?

gd_barnes 2010-09-29 22:37

Max,

The Linux client does not work with LLR 3.8.2. It keeps exiting immediately. Please test and correct it when you get a chance. Thanks.

Edit: A second attempt says that pairs are not available. The first attempt, I just plopped the LLR binary in my existing folder since there was no coding change to the script; only the comments changed. The second attempt, I used all of your files, just in case something else changed that I wasn't aware of. Both ways, no luck. I made sure I had the llrnetcongif configured correctly.


Gary

gd_barnes 2010-09-29 22:55

BTW, are you guys aware that the LLR 3.8.2 binary jumped to ~25 MB from LLR 3.8.1 that was only ~5 MB? Is this really something where we can just plop the binary into our existing folders on LLRnet and/or PRPnet? After my problems, I think we need to fully test the new clients.

rogue 2010-09-29 23:01

[QUOTE=gd_barnes;231983]BTW, are you guys aware that the LLR 3.8.2 binary jumped to ~25 MB from LLR 3.8.1 that was only ~5 MB? Is this really something where we can just plop the binary into our existing folders on LLRnet and/or PRPnet? After my problems, I think we need to fully test the new clients.[/QUOTE]

Yes, it is larger. George added a lot of code to support additional FFT sizes.

mdettweiler 2010-09-29 23:43

[QUOTE=gd_barnes;231980]You do realize that you did not respond to my first point in the first para here. Correct?[/QUOTE]
Ah, whoops. I was thinking that at this point, we may as well keep it at 0.74. But I can change it back if you'd like. Either way's fine with me.
[QUOTE=gd_barnes;231981]Max,

The Linux client does not work with LLR 3.8.2. It keeps exiting immediately. Please test and correct it when you get a chance. Thanks.

Edit: A second attempt says that pairs are not available. The first attempt, I just plopped the LLR binary in my existing folder since there was no coding change to the script; only the comments changed. The second attempt, I used all of your files, just in case something else changed that I wasn't aware of. Both ways, no luck. I made sure I had the llrnetcongif configured correctly.


Gary[/QUOTE]
Hmm, very strange. The new LLR should handle input and output files exactly the same--all that was changed was the gwnum version, which just changes some of the screen output regarding FFT sizes. It should not affect anything the do.pl script cares about.

I'll try testing it, though, and see what happens for myself. Clearly there is [i]some[/i] difference in how LLR 3.8.2 handles its input and ouput files and it's making the do.pl script go nuts.

Regarding the size of the llr binary: as Mark said, that's due to a lot more CPU-specialized code being packed into the latest gwnum. This is why LLR 3.8.2 is usually faster than 3.8.1.

gd_barnes 2010-09-30 00:03

Ah, heck yeah on the speed increase! Wee! When using PFGW 3.4.0 (Linux) and 3.4.1 (Windows), BIG speed up my CRUS base 28 stuff. 20% or more! I think Karsten is getting that increase or better on his n>1M base 2 stuff here. Very nice!

Yeah, I suspect there is some sort of input-output difference to/from the new LLR. It's always good to run a quick test before posting new software even if the change is what should be a small one. :smile:

kar_bon 2010-09-30 02:59

[QUOTE=gd_barnes;231991]
Yeah, I suspect there is some sort of input-output difference to/from the new LLR. It's always good to run a quick test before posting new software even if the change is what should be a small one. :smile:[/QUOTE]

As I mentioned, I've checked this and done as described for the WIN-script:
- stop client
- overwrite cLLR.exe with new one (Hint: use cLLR.exe not LLR.exe, the command version, not GUI-version!)
- start again client

I've done this with all of my 16 cores (XP, Vista, WIN7) and no problems occured.
Possible problem: Converting from lresults.txt to tosend.txt (check/test do.pl)

Please check again with UNIX.

The comments/screen outputs changed, because I've not patched again cLLR.exe, for example:
[code]
[2010-09-30 04:26:50]
Fetching WU #1/1: 327 1158098
Starting Lucas Lehmer Riesel prime test of 327*2^1158098-1
Using Core2 type-1 FFT length 72K, Pass1=96, Pass2=768
V1 = 3 ; Computing U0...done.
327*2^1158098-1 is not prime. LLR Res64: A4385FD1358EAB68 Time : 1607.416 sec.
[2010-09-30 04:53:40]

But this has no effect on the lresults.txt file.
[/code]

I've not yet updated the download-zip with the new cLLR version, because waiting for the next version would be better.

mdettweiler 2010-09-30 05:18

Okay, I found the problem. Turns out the llrnet and llr binaries didn't have +x permissions on them when they came out of the zip file. (Duh.) Running chmod +x on them got it working, so I replaced the binaries in the zip file with the +x'ed ones and re-uploaded it to the web site.

Gary, let me know if this one works. (It should work, since I really did test it this time. On a real Linux machine, too--one of yours, in fact. :razz:)

Edit: oh, and now that I've figured out what went wrong I can confirm that it should be OK to swap in the new LLR binary directly to an old do.pl client. If it doesn't work, run "chmod +x llr" and try again.

gd_barnes 2010-09-30 11:11

[QUOTE=kar_bon;231309]Jean Penné just released the new LLR V3.8.2 with gwnum-lib 26.2.

He mentioned a 10% faster testing with that version.

To use the new cLLR, download it from [URL="http://jpenne.free.fr/Development/cllr382dev.zip"]here[/URL], stop the client-script by hitting CTRL-C, overwrite the old cLLR.exe with the downloaded one and start the script again by calling 'do'.

I'll check if the speed is as assumed.

Later on I'll update the links in the first post with this new cLLR version.[/QUOTE]

The link does not work. I went directly to his page and there is no Windows or Linux 3.8.2 version. For my Linux machines, I used what Max posted in the 1st post here. I can't update my Windows machine now. Is 3.8.2 somewhere else?

kar_bon 2010-09-30 11:41

[QUOTE=gd_barnes;232038]I can't update my Windows machine now. Is 3.8.2 somewhere else?[/QUOTE]

All links work fine for me here.

I've uploaded the cLLR V3.8.2 WIN [url=www.rieselprime.de/dl/cllr382dev.zip]here[/url], too.

mdettweiler 2010-09-30 14:29

[QUOTE=gd_barnes;232038]The link does not work. I went directly to his page and there is no Windows or Linux 3.8.2 version. For my Linux machines, I used what Max posted in the 1st post here. I can't update my Windows machine now. Is 3.8.2 somewhere else?[/QUOTE]
In addition to the link Karsten posted, you can download all the LLR 3.8.2 versions at:
[URL]http://jpenne.free.fr/Development/[/URL]
It's not on Jean's main page since it's still a "development" version at this point. (So it's a kind of beta version, but should still be stable enough for use on production work.)


All times are UTC. The time now is 06:50.

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