New LLR

2003-12-25, 20:47

It is has now been one year, since LLR release into the mainstream!
Overall a great new program, but after the last problem found with it, maybe we should get to test the new LLR before its release?

SlashDude · 2003-12-27, 04:46

It looks like the Riesel Sieve project is getting close to having a networked version of PRP / LLR! Has anyone ever talked with b2 to see if their network version would be "Public?" (Does LLR.exe and it's code fall under the GPL license - if you modify it, the modifications become public

)

If it's not going to be public, does anyone want to work on our own network version?

2003-12-28, 03:24

How much would it cost to have it written without infringing?

Citrix · 2003-12-28, 05:25

The new LLR version will be available under the same liscence the current version is available under. As for the network abilities, we might have to add it into the source of LLR and then compile it.

In short, the LLR code will be available publically but not the network code. The network code may be available on request, which we can then compile.

Ill look more into this when the new LLR client is released.

Citrix

b2lee · 2003-12-28, 19:36

Hello all,

B2 here with www.rieselsieve.com. Currently Jean Penne is working on a new updated LLR client. He is quite busy with it and has only vaguely mentioned some very nice speed increases. I'm happily awaiting the time for beta testing it.

Now a little recent history. I've talked numerous times with Jean about getting LLR to be network capable. At first I thought he may not see the need for this. However, to my great surprise he was very open to the idea and does see it as the future direction in which almost all prime finding tools will need to go.

To put it a little more into focus, 'I' may have asked for this tool to be developed for Riesel Sieve, but the tool will be designed for EVERYONE. Hopefully it can be set up very quickly for any other project to use. We at Riesel Sieve are for the opensource of all utilities that help the prime community at large. Sometimes we as a project must give into the wishes of individual programmers but we promote an open environment that helps everyone.

So, yes a new client is currently in developement. Yes, network calls are going to be a part of it. And, hopefully....hopefully Jean will get some things working in the new client that he is most happy about and you will see one hell of a speed increase that will bring a renewed excitement back to finding primes.

Also, the mentioned P-1 client. I've had many thoughts of step one of LLR running a P-1 phase. If you could contact Jean Penne and ask him about the inclusion of this step I think he would be receptive to the idea. As n grows P-1 will become a valueable part of our project and I'm sure other projects as well.

Any questions or comments can be sent to admin@rieselsieve.com or you can simply jump on irc.freenode.net #rieselsieve and discuss it with us in real time.

Lee Stephens
B2
www.rieselsieve.com

SlashDude · 2003-12-29, 01:47

Hi b2lee!

Thanks for the information on the upcoming version of LLR! I wasn't sure if you were developing your own networked version of LLR, and I wasn't sure of it's licensing...

I know I could run more machines if we had a networked LLR client available

Congratulations on your projects latest prime! It's a biggie!

Thanks again for the information and I hope you find another prime soon!!

-Mark

smh · 2003-12-29, 08:25

I seem to remember there is a network versio of pfgw.

Haven't taken a look at it though.

Kosmaj · 2003-12-29, 15:42

My wishes for the new LLR version:

1) Built-in P-1 test for large n's.
2) Network version including support for LAN
3) Supprot for larger k's like those of Payam numbers when k > 10^20.

The following are minor but I'll appreciate their inclusion.
4) Fixed "Start on boot-up" option. Now it's broken (but LLR can be started using the Task Scheduler).
5) Better format of the output in the main window, so that one can check results quickly. Similar to the output written to lresults.txt, for example:

3611911875*2^171588-1 is not prime. Res64: 5DDCA44429DB04A6 Time : 192.457 sec.
3611911875*2^171598-1 is not prime. Res64: 91A3976838DCF85E Time : 192.522 sec.
3611911875*2^171599-1 is not prime. Res64: 8F1323E5347C6B68 Time : 192.797 sec.
3611911875*2^171600-1 is not prime. Res64: 50D5C0A620FB039B Time : 192.670 sec.
3611911875*2^171601-1 is not prime. Res64: 40DD1539B56A51C4 Time : 192.684 sec.

In particular, removal of
"V1=21, Computing U0... Done Computing U0"
It's done in only 1-2 seconds, and IMO it's not that important. It can be written only for primes.

6) Play a sound or send e-mail when a prime is found.

Kosmaj · 2003-12-29, 16:04

> I seem to remember there is a network version of pfgw.

I did extensive tests with pfgw beta versions and the server in summer but it was not working. I notified Jim (who takes care of the software) but he said it's working on his site. I think there was a mismatch of protocol versions of the server and the clients. In the beta version of 20031027 there was a notice that the server had been improved but I didn't have time to test it again. I'm missing this feature a lot. Manual distribution of candidates and book-keeping is error-prone and time consuming.

Kosmaj · 2004-01-29, 03:57

While inspecting the results of LLR testing k=210885 I ran upon a strange line:

Cannot compute V1 to test 210885*2^390124-1

and LLR went on to test the next candidate. The test by pfgw shows it's not a prime. But I wonder why? Does it mean LLR doesn't work for some special combinations of (k, n)?

2004-03-05, 00:53

So emummm.. where's the new LLR?

Anyways, Chris Caldwell does not object to submiting non=15k as 15k primes.

I will be submitting gM, as 15k the way I had originally planned.
RMA software is delayed due to a packaging problem.(my software is out of date) but version 1.4 is polished and ready to rock!
Release date 2/11/04

SlashDude · 2004-03-05, 04:02

Welcome back TTn!!!

I'm guilty of submitting a prime as a 15k that really isn't one....but no one caught it

Will your RMA software work with 15k's? Do you want some space on the 15k.org web site?

The Riesel Sieve Project has some talk on their forum about a new version they are testing... Hopefully they can finish "Testing" and share with the rest of us

Dave - Congrats on your latest prime!!

Again, welcome back!!!!

2004-03-07, 02:31

Thanks for the welcome!

You can input a known 15k prime, as a starting point, but I dont think that is what you meant. RMA uses a fixed n, rather than fixed k. But I will see if I can adjust for that option, in the next couple of days. Good idea!

Is there any requests for particular options ?

Current options include Riesel/Proth mode, Odd/Even/EvenOdd k mode, Manual/Auto mode. Newpgen services are automated. File status.
Testing is one click away!

I am getting some forum/web space this week, so we'll have to link up.

Thanks again,

SlashDude · 2004-03-24, 15:42

New LLR - Go get it :)

http://www.mersenne.org/gimps/llr.zip

(I haven't tested it yet, but I'll post when I have some results)

Kosmaj · 2004-03-24, 18:28

I did a quick comparison but the speed-up is only 2-5% so for accurate results a better controlled test envirnment is required. For example:

3611911875*2^233410-1 is not prime. Time : 449.951 sec.
3611911875*2^233410-1 is not prime. Time : 439.596 sec.

On a dual Athlon-2000, both cpu's 100% busy, the 2nd cpu running pfgw.

No new feauters in the menus, by the way, one can only hope the "Start at bootup" is working now.

paulunderwood · 2004-03-24, 18:35

I have done any specific tests but I'm see a 400% increase for 3*2^n-1

Kosmaj · 2004-03-24, 18:43

What cpu are you using?

Is the new version for P-4/SSE-2 only?

Thomas11 · 2004-03-24, 18:59

The new LLR uses IBDWT only for small k's (see Jean Penne's post in the Yahoo groups below). For the k's you are testing at 15k Search you will not see any significant differences in speed. Since we at 321search have found some (minor) bugs of the new version, I would suggest that you should use the old version until the new one is stable enough.

Here is Jean Penne's comment:

Quote:

So, this program uses IBDWT only if k <= 511 in non SSE2 code, and (alas...=
) k <= 31 in
SSE2 code ; for larger k's, it works exactly like the previous LLR version,=
using Proth mode
and rational bases FFT. But for really small k's, it is a major improvement=
:

-- Thomas

Kosmaj · 2004-03-24, 19:09

Oh, I see, there is a speed up only for k<512 in k*2^n-1.
Good news for 3-2-1 search and small Riesel coefficients, nothing for us.
Well, a disappointment for us...

Thomas11 · 2004-03-24, 19:20

Quote:

Originally Posted by Kosmaj

I cannot beleive it! Is this all about speeding-up tests for k<512 ??
Who cares about such small numbers?

How about the 400% speed-up Paul Underwood is talking about?

Well, Paul (and a few others too) is testing numbers for k=3, so he's very happy about that speed-up ...

And I wouldn't call the primes 3*2^702038-1 and 3*2^727699-1 "small" numbers!

Nevertheless, we should expect another LLR version, capable to use IBDWT also for larger k's , very soon.

-- Thomas.

Kosmaj · 2004-03-24, 19:24

Sorry,
I mixed up k and n in k*2^n-1. I edited my post but it was too late.

Kosmaj · 2004-04-08, 19:14

Steven Harvey just reported yet another bug in the new LLR in his
message to the PrimeNumbers group:

25*2^130407-1 is not prime. Res64: 0000000000000000

while in fact the number is prime. So check your lresults.txt file for
Res64=0000000000000000.

While I'm thankful to Jean Penne for his work on LLR I must say that the latest version has too many bugs. It looks like he never did any serious tests on the new version.

ET_ · 2004-04-09, 11:43

Quote:

Originally Posted by Kosmaj

While I'm thankful to Jean Penne for his work on LLR I must say that the latest version has too many bugs. It looks like he never did any serious tests on the new version.

I'm having troubles with 3*2⁹⁹⁸⁹⁷⁵-1... It's the third time that the program crashes on this value

Luigi

2004-04-09, 11:46

Quote:

Originally Posted by Kosmaj

Steven Harvey just reported yet another bug in the new LLR in his
message to the PrimeNumbers group:

25*2^130407-1 is not prime. Res64: 0000000000000000

while in fact the number is prime. So check your lresults.txt file for
Res64=0000000000000000.

that is my output with the new llr. looks fine:

[Thu Apr 08 19:18:44 2004]
Using IBDWT : Mersenne fftlen = 6144, Proth fftlen = 14336, Used fftlen = 8192
fftlen seems to be too small, using next fftlen...
Using IBDWT : Mersenne fftlen = 6144, Proth fftlen = 14336, Used fftlen = 10240
25*2^130407-1 is prime! Time : 74.321 sec.

Quote:

Originally Posted by Kosmaj

While I'm thankful to Jean Penne for his work on LLR I must say that the latest version has too many bugs. It looks like he never did any serious tests on the new version.

I believe he tested as good as he could...

Kosmaj · 2004-04-09, 12:59

thommy, what cpu are you using?

2003-12-25, 20:47	#1
Val 3·269 Posts	New LLR It is has now been one year, since LLR release into the mainstream! Overall a great new program, but after the last problem found with it, maybe we should get to test the new LLR before its release?

2003-12-27, 04:46	#2
SlashDude Aug 2002 Minneapolis, MN 2²×3×19 Posts	It looks like the Riesel Sieve project is getting close to having a networked version of PRP / LLR! Has anyone ever talked with b2 to see if their network version would be "Public?" (Does LLR.exe and it's code fall under the GPL license - if you modify it, the modifications become public ) If it's not going to be public, does anyone want to work on our own network version? Last fiddled with by SlashDude on 2003-12-27 at 04:47

2004-03-05, 00:53	#11
TTn 3,739 Posts	LLR? So emummm.. where's the new LLR? Anyways, Chris Caldwell does not object to submiting non=15k as 15k primes. I will be submitting gM, as 15k the way I had originally planned. RMA software is delayed due to a packaging problem.(my software is out of date) but version 1.4 is polished and ready to rock! Release date 2/11/04

2004-03-24, 18:35	#16
paulunderwood Sep 2002 Database er0rr EB2₁₆ Posts	I have done any specific tests but I'm see a 400% increase for 32^n-1 Last fiddled with by paulunderwood on 2004-03-24 at 18:37*

2004-03-24, 19:09	#19
Kosmaj Nov 2003 3622₁₀ Posts	Oh, I see, there is a speed up only for k<512 in k2^n-1. Good news for 3-2-1 search and small Riesel coefficients, nothing for us. Well, a disappointment for us... Last fiddled with by Kosmaj on 2004-03-24 at 19:20*

2003-12-28, 03:24	#3
Val 420₁₀ Posts	How much would it cost to have it written without infringing?

2003-12-28, 05:25	#4
Citrix Jun 2003 11000101110₂ Posts	The new LLR version will be available under the same liscence the current version is available under. As for the network abilities, we might have to add it into the source of LLR and then compile it. In short, the LLR code will be available publically but not the network code. The network code may be available on request, which we can then compile. Ill look more into this when the new LLR client is released. Citrix

2003-12-28, 19:36	#5
b2lee Oct 2003 Atlanta, GA USA 2 Posts	Hello all, B2 here with www.rieselsieve.com. Currently Jean Penne is working on a new updated LLR client. He is quite busy with it and has only vaguely mentioned some very nice speed increases. I'm happily awaiting the time for beta testing it. Now a little recent history. I've talked numerous times with Jean about getting LLR to be network capable. At first I thought he may not see the need for this. However, to my great surprise he was very open to the idea and does see it as the future direction in which almost all prime finding tools will need to go. To put it a little more into focus, 'I' may have asked for this tool to be developed for Riesel Sieve, but the tool will be designed for EVERYONE. Hopefully it can be set up very quickly for any other project to use. We at Riesel Sieve are for the opensource of all utilities that help the prime community at large. Sometimes we as a project must give into the wishes of individual programmers but we promote an open environment that helps everyone. So, yes a new client is currently in developement. Yes, network calls are going to be a part of it. And, hopefully....hopefully Jean will get some things working in the new client that he is most happy about and you will see one hell of a speed increase that will bring a renewed excitement back to finding primes. Also, the mentioned P-1 client. I've had many thoughts of step one of LLR running a P-1 phase. If you could contact Jean Penne and ask him about the inclusion of this step I think he would be receptive to the idea. As n grows P-1 will become a valueable part of our project and I'm sure other projects as well. Any questions or comments can be sent to admin@rieselsieve.com or you can simply jump on irc.freenode.net #rieselsieve and discuss it with us in real time. Lee Stephens B2 www.rieselsieve.com

2003-12-29, 01:47	#6
SlashDude Aug 2002 Minneapolis, MN E4₁₆ Posts	Hi b2lee! Thanks for the information on the upcoming version of LLR! I wasn't sure if you were developing your own networked version of LLR, and I wasn't sure of it's licensing... I know I could run more machines if we had a networked LLR client available Congratulations on your projects latest prime! It's a biggie! Thanks again for the information and I hope you find another prime soon!! -Mark

2003-12-29, 08:25	#7
smh "Sander" Oct 2002 52.345322,5.52471 29×41 Posts	I seem to remember there is a network versio of pfgw. Haven't taken a look at it though.

2003-12-29, 15:42	#8
Kosmaj Nov 2003 E26₁₆ Posts	My wishes for the new LLR version: 1) Built-in P-1 test for large n's. 2) Network version including support for LAN 3) Supprot for larger k's like those of Payam numbers when k > 10^20. The following are minor but I'll appreciate their inclusion. 4) Fixed "Start on boot-up" option. Now it's broken (but LLR can be started using the Task Scheduler). 5) Better format of the output in the main window, so that one can check results quickly. Similar to the output written to lresults.txt, for example: 36119118752^171588-1 is not prime. Res64: 5DDCA44429DB04A6 Time : 192.457 sec. 36119118752^171598-1 is not prime. Res64: 91A3976838DCF85E Time : 192.522 sec. 36119118752^171599-1 is not prime. Res64: 8F1323E5347C6B68 Time : 192.797 sec. 36119118752^171600-1 is not prime. Res64: 50D5C0A620FB039B Time : 192.670 sec. 3611911875*2^171601-1 is not prime. Res64: 40DD1539B56A51C4 Time : 192.684 sec. In particular, removal of "V1=21, Computing U0... Done Computing U0" It's done in only 1-2 seconds, and IMO it's not that important. It can be written only for primes. 6) Play a sound or send e-mail when a prime is found.

2003-12-29, 16:04	#9
Kosmaj Nov 2003 2×1,811 Posts	> I seem to remember there is a network version of pfgw. I did extensive tests with pfgw beta versions and the server in summer but it was not working. I notified Jim (who takes care of the software) but he said it's working on his site. I think there was a mismatch of protocol versions of the server and the clients. In the beta version of 20031027 there was a notice that the server had been improved but I didn't have time to test it again. I'm missing this feature a lot. Manual distribution of candidates and book-keeping is error-prone and time consuming.

2004-01-29, 03:57	#10
Kosmaj Nov 2003 2×1,811 Posts	While inspecting the results of LLR testing k=210885 I ran upon a strange line: Cannot compute V1 to test 210885*2^390124-1 and LLR went on to test the next candidate. The test by pfgw shows it's not a prime. But I wonder why? Does it mean LLR doesn't work for some special combinations of (k, n)?

2004-03-05, 04:02	#12
SlashDude Aug 2002 Minneapolis, MN 11100100₂ Posts	Welcome back TTn!!! I'm guilty of submitting a prime as a 15k that really isn't one....but no one caught it Will your RMA software work with 15k's? Do you want some space on the 15k.org web site? The Riesel Sieve Project has some talk on their forum about a new version they are testing... Hopefully they can finish "Testing" and share with the rest of us Dave - Congrats on your latest prime!! Again, welcome back!!!!

2004-03-07, 02:31	#13
TTn 100000001000₂ Posts	Thanks for the welcome! You can input a known 15k prime, as a starting point, but I dont think that is what you meant. RMA uses a fixed n, rather than fixed k. But I will see if I can adjust for that option, in the next couple of days. Good idea! Is there any requests for particular options ? Current options include Riesel/Proth mode, Odd/Even/EvenOdd k mode, Manual/Auto mode. Newpgen services are automated. File status. Testing is one click away! I am getting some forum/web space this week, so we'll have to link up. Thanks again,

2004-03-24, 15:42	#14
SlashDude Aug 2002 Minneapolis, MN 2²·3·19 Posts	New LLR - Go get it :) http://www.mersenne.org/gimps/llr.zip (I haven't tested it yet, but I'll post when I have some results)

2004-03-24, 18:28	#15
Kosmaj Nov 2003 3622₁₀ Posts	I did a quick comparison but the speed-up is only 2-5% so for accurate results a better controlled test envirnment is required. For example: 36119118752^233410-1 is not prime. Time : 449.951 sec. 36119118752^233410-1 is not prime. Time : 439.596 sec. On a dual Athlon-2000, both cpu's 100% busy, the 2nd cpu running pfgw. No new feauters in the menus, by the way, one can only hope the "Start at bootup" is working now.

2004-03-24, 18:43	#17
Kosmaj Nov 2003 2×1,811 Posts	What cpu are you using? Is the new version for P-4/SSE-2 only?

2004-03-24, 19:24	#21
Kosmaj Nov 2003 2×1,811 Posts	Sorry, I mixed up k and n in k*2^n-1. I edited my post but it was too late.

2004-04-08, 19:14	#22
Kosmaj Nov 2003 2·1,811 Posts	Steven Harvey just reported yet another bug in the new LLR in his message to the PrimeNumbers group: 25*2^130407-1 is not prime. Res64: 0000000000000000 while in fact the number is prime. So check your lresults.txt file for Res64=0000000000000000. While I'm thankful to Jean Penne for his work on LLR I must say that the latest version has too many bugs. It looks like he never did any serious tests on the new version.

2004-04-09, 12:59	#25
Kosmaj Nov 2003 E26₁₆ Posts	thommy, what cpu are you using?