Automated PRP discussions

ltd · 2006-07-18, 07:32

How about setting up llrnet server for the automatic distribution of k/n pairs.
Only problem with that is that you need two server instances on different ports to distribute "+1" and "-1" as each server ( and also client ) can only run one of the types at one time.

This is a setup similar to the PSP setup. I you are interested how such system can be build please contact me.
Only problem i am only sporadic online till the 27th. So don't expect answers before.

Lars

masser · 2006-07-18, 16:24

I think Lars has a really good idea here. It's something we should seriously consider once testing has reached n=100000 for each k value. I don't think we can perform a serious double check until our first pass testing is slightly more automated.

So far, double checking has uncovered 3 missed primes out of the 90 remaining Sierpinski k values, about 3%. We should probably expect a similar result for the Riesel k values. I have done negligible double checking for the Riesel values. Has anyone else made any progress here?

EDIT: I want to add that I believe the missed primes are probably the result of human error, and not the codes. I suspect that what happened is someone manually copied a list of primes found and skipped a line or two as they were scanning throught the list.

axn · 2006-07-18, 18:57

Quote:

Originally Posted by masser

I think Lars has a really good idea here. It's something we should seriously consider once testing has reached n=100000 for each k value. I don't think we can perform a serious double check until our first pass testing is slightly more automated.

Actually, this is something I've been thinking about quite some time. The manual reservation just won't scale up once we get more users (it's fine for sieving, but too much for PRP). The problem is that we need somebody to host this. Does anyone have a static IP by any chance? Anyone?

Of course, the logistics need to be worked out in terms of how to remove "found factors" from the PRP queue, what to do when a prime is found, etc. Also a unique problem will be to keep +1 and -1 queue at roughly the same level. And these are just the top of my head.

I am sure lots more things need to be looked at - but this is definitely something that we need to have in place, say, by the time Riesels reach 100k.

Quote:

Originally Posted by masser

So far, double checking has uncovered 3 missed primes out of the 90 remaining Sierpinski k values, about 3%. We should probably expect a similar result for the Riesel k values. I have done negligible double checking for the Riesel values. Has anyone else made any progress here?

I am doing a double check on this - around 15000 at the moment. Not putting too much resources on this, though. Let's see how that progresses.

Quote:

Originally Posted by masser

EDIT: I want to add that I believe the missed primes are probably the result of human error, and not the codes. I suspect that what happened is someone manually copied a list of primes found and skipped a line or two as they were scanning throught the list.

It is highly unlikely that this is anything _but_ human error

For numbers this small, a machine error is very unlikely. OTOH, I hope nobody is looking thru the lresults.txt for found primes - much better would be to look at the output file.

ltd · 2006-07-19, 07:00

Depending on the way the llrnet server is installed i think i can host them.
psp-project.de already runs two llrnet server so another two should be no problem.

And when we put the k/n pairs into a mysql DB even the feeding of the servers is no problem as i have scripts for automatic feeding in place.

Only thing would be the access to the DB for removing factors and marking primes. Either i would have to give you access to mysql or we would have to create some scripts that allow you the manipulations.

Method one has the problem that i did not setup the security on the server well enough to have several administrators in different areas.
Method two has the problem that somebody has to write the scripts.
I think i will not have enough time as i should have created some more things for PSP already.

But let us talk about it in detail when i am back from work.

Lars

Citrix · 2006-07-21, 15:13

I am not sure but as far as I know, LLRnet does not handle base 5. But this is something that can be easily fixed.

One thing, I would like to see in LLRnet is to send diffrent kind of numbers using only one server.
For example, if base 5, riesel, sierpinski both could be handled by the same server instead of setting up several servers and several clients. So you could set up a personal LLRnet and work on different projects, instead of sticking to one.

I think the best way to make LLRnet compatible for base 5 will be to contact the author or LLR.

ltd · 2006-07-21, 16:28

As far as i know the llrnet server hands out the "newpgen type" information from the pairs.txt. The output string can be configured within one of the configuration files. So the server does not know if it is "+1" or "-1" or which base is used. The server is only able to handle one type as there can only be one "newpgen type" information.

The client is limited in the same way. It stores the work to be done in one file with a "newpgen type" header line as it is done with the offlline versions also. So the problem with diferent types of requests (bases or "+/-") comes from the original design of llr/PRP and not from the llrnet client.

When i am back at home i will make some tests to check if the client server combination is really able to handle base 5.

Lars

geoff · 2006-07-22, 03:09

Quote:

Originally Posted by axn1

Also a unique problem will be to keep +1 and -1 queue at roughly the same level.

I don't think we should worry about this. If people want to work on +1 more than -1, or vice versa, then that is OK.

axn · 2006-07-28, 20:58

Quote:

Originally Posted by ltd

When i am back at home i will make some tests to check if the client server combination is really able to handle base 5.

Any progress on this?

Citrix · 2006-07-28, 21:01

Ltd isn't home yet. Give him a few more hours before he reaches home.

LLRnet works for base 5, I just tested it.

michaf · 2006-08-28, 23:05

Just wondering,

any progress made in this area while my computer was having
hiccups?

Something working already, or just thought yet?

axn · 2006-08-29, 18:27

Quote:

Originally Posted by michaf

Just wondering,

any progress made in this area while my computer was having
hiccups?

Something working already, or just thought yet?

Watch this space!

2006-07-18, 16:24	#2
masser Jul 2003 Behind BB 2×983 Posts	I think Lars has a really good idea here. It's something we should seriously consider once testing has reached n=100000 for each k value. I don't think we can perform a serious double check until our first pass testing is slightly more automated. So far, double checking has uncovered 3 missed primes out of the 90 remaining Sierpinski k values, about 3%. We should probably expect a similar result for the Riesel k values. I have done negligible double checking for the Riesel values. Has anyone else made any progress here? EDIT: I want to add that I believe the missed primes are probably the result of human error, and not the codes. I suspect that what happened is someone manually copied a list of primes found and skipped a line or two as they were scanning throught the list. Last fiddled with by masser on 2006-07-18 at 16:28

2006-07-28, 21:01	#9
Citrix Jun 2003 3×5×107 Posts	Ltd isn't home yet. Give him a few more hours before he reaches home. LLRnet works for base 5, I just tested it. Last fiddled with by Citrix on 2006-07-28 at 21:39

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Soapbox Discussions	only_human	Soap Box	41	2019-11-16 15:46
Our discussions here: how can we improve things?	Brian-E	Soap Box	105	2013-11-10 12:26
n=390000 discussions and old reservations	TimSorbet	Twin Prime Search	31	2010-05-22 23:13
Primegrid discussions	pacionet	Twin Prime Search	17	2007-01-20 11:22
Distributed sieving discussions	michaf	Sierpinski/Riesel Base 5	76	2006-09-12 18:37

2006-07-18, 07:32	#1
ltd Apr 2003 2²·193 Posts	How about setting up llrnet server for the automatic distribution of k/n pairs. Only problem with that is that you need two server instances on different ports to distribute "+1" and "-1" as each server ( and also client ) can only run one of the types at one time. This is a setup similar to the PSP setup. I you are interested how such system can be build please contact me. Only problem i am only sporadic online till the 27th. So don't expect answers before. Lars

2006-07-19, 07:00	#4
ltd Apr 2003 304₁₆ Posts	Depending on the way the llrnet server is installed i think i can host them. psp-project.de already runs two llrnet server so another two should be no problem. And when we put the k/n pairs into a mysql DB even the feeding of the servers is no problem as i have scripts for automatic feeding in place. Only thing would be the access to the DB for removing factors and marking primes. Either i would have to give you access to mysql or we would have to create some scripts that allow you the manipulations. Method one has the problem that i did not setup the security on the server well enough to have several administrators in different areas. Method two has the problem that somebody has to write the scripts. I think i will not have enough time as i should have created some more things for PSP already. But let us talk about it in detail when i am back from work. Lars

2006-07-21, 15:13	#5
Citrix Jun 2003 3×5×107 Posts	I am not sure but as far as I know, LLRnet does not handle base 5. But this is something that can be easily fixed. One thing, I would like to see in LLRnet is to send diffrent kind of numbers using only one server. For example, if base 5, riesel, sierpinski both could be handled by the same server instead of setting up several servers and several clients. So you could set up a personal LLRnet and work on different projects, instead of sticking to one. I think the best way to make LLRnet compatible for base 5 will be to contact the author or LLR.

2006-07-21, 16:28	#6
ltd Apr 2003 2²·193 Posts	As far as i know the llrnet server hands out the "newpgen type" information from the pairs.txt. The output string can be configured within one of the configuration files. So the server does not know if it is "+1" or "-1" or which base is used. The server is only able to handle one type as there can only be one "newpgen type" information. The client is limited in the same way. It stores the work to be done in one file with a "newpgen type" header line as it is done with the offlline versions also. So the problem with diferent types of requests (bases or "+/-") comes from the original design of llr/PRP and not from the llrnet client. When i am back at home i will make some tests to check if the client server combination is really able to handle base 5. Lars

2006-08-28, 23:05	#10
michaf Jan 2005 737₈ Posts	Just wondering, any progress made in this area while my computer was having hiccups? Something working already, or just thought yet?