Future direction of NPLB

gd_barnes · 2008-08-24, 20:20

We are taking suggestions on the future direction of NPLB. Here is what I would propose at this time. I think it should be used as a starting point for discussion and not set in stone:

1. How the drives will proceed or be divided up for the range of n=600K-1M. I would like to propose that we leave the k=300-400 range as is for a period of time but leave open the possibility of doing n-range searches like we've done for drives 1, 2, and 3 for k's that are unreserved once we have all k=300-1001 up to n=600K. For k=400-1001, I'd like to propose that we divide that up into smaller k-ranges as separate drives, as follows: (1) k=400-600 (2) k=600-800 (3) k=800-1001. My thinking is that this will make the project more interesting because each of the drives would likely be at different search limits. Once we hit an fftlen change at the higher k-values, that will make the lower k-values a little more appealing and they would be searched deeper. That would give more of a choice as to what n-value people choose to search at.

2. k<300 for unreserved k's only searched to n=600K. There are currently about 15 unreserved k's < 300 that are at n=600K and are not being worked on yet we are already nearing or at n=1M on several k=300-400. My thinking is that we should make this a separate drive and sieving effort before we start k>400 for n=600K-1M. Then we could open all k-ranges at once and have 5 drives running concurrently, all that will likely be at different n-ranges after a period of time. My thinking is that it doesn't make sense to search higher k's to n=1M while leaving lower k's at only n=600K.

3. k=1003-2000. This range is an extreme state of mess as a result of the fragmented Benson searches. Karsten and some folks at RPS have gotten the gaps filled in up to n=50K on all of them (nice work!) but there are plenty of gaps still to fill. I think there are many people that don't like the higher n-ranges that we are currently processing because it takes too long to find primes without large resources, hence a drop-off in popularity of NPLB in the last couple of months. I think that doing a team sieve and search would bring in a number of people to find smaller primes. Many people aren't concerned about top-5000 primes and those people could help us clean up this mess.

4. Allow for individual-k reservations for ANY k-value for n>600K. In the same vain as the what the Riesel/Sierp base 5 project does, people could reserve individual k's but they would not be removed from the main drives. It would be up to the people reserving them to stay ahead of the drives on their searching. In effect, the drives would 'double check' the individual-k's that are reserved. With nearly 350 k's available at that time, there should be little fear that the drives would be doing more than 2-5% double checking of individual-k reservations.

All thoughts and opinions are welcome.

Gary

mdettweiler · 2008-08-25, 04:45

Quote:

Originally Posted by gd_barnes

We are taking suggestions on the future direction of NPLB. Here is what I would propose at this time. I think it should be used as a starting point for discussion and not set in stone:

1. How the drives will proceed or be divided up for the range of n=600K-1M. I would like to propose that we leave the k=300-400 range as is for a period of time but leave open the possibility of doing n-range searches like we've done for drives 1, 2, and 3 for k's that are unreserved once we have all k=300-1001 up to n=600K. For k=400-1001, I'd like to propose that we divide that up into smaller k-ranges as separate drives, as follows: (1) k=400-600 (2) k=600-800 (3) k=800-1001. My thinking is that this will make the project more interesting because each of the drives would likely be at different search limits. Once we hit an fftlen change at the higher k-values, that will make the lower k-values a little more appealing and they would be searched deeper. That would give more of a choice as to what n-value people choose to search at.

Yes, I agree--having things split up into smaller k-ranges will sure be nice when each tests starts taking a half hour apiece.

Quote:

2. k<300 for unreserved k's only searched to n=600K. There are currently about 15 unreserved k's < 300 that are at n=600K and are not being worked on yet we are already nearing or at n=1M on several k=300-400. My thinking is that we should make this a separate drive and sieving effort before we start k>400 for n=600K-1M. Then we could open all k-ranges at once and have 5 drives running concurrently, all that will likely be at different n-ranges after a period of time. My thinking is that it doesn't make sense to search higher k's to n=1M while leaving lower k's at only n=600K.

Agreed. If you'd like, I could spearhead a public sieving effort for this--when should we get started on it?

Quote:

3. k=1003-2000. This range is an extreme state of mess as a result of the fragmented Benson searches. Karsten and some folks at RPS have gotten the gaps filled in up to n=50K on all of them (nice work!) but there are plenty of gaps still to fill. I think there are many people that don't like the higher n-ranges that we are currently processing because it takes too long to find primes without large resources, hence a drop-off in popularity of NPLB in the last couple of months. I think that doing a team sieve and search would bring in a number of people to find smaller primes. Many people aren't concerned about top-5000 primes and those people could help us clean up this mess.

Also agreed. Because of the immense size of this k-range, I suggest we split it up into smaller, more manageable chunks--how about starting with 1005-1400? (1003 has already been searched to n=500K by yours truly.) We could do, say, n=50K-500K to start with, then do 1400<k<2000 next. Focusing on 1005-1400 at first is especially important since that's where most of the Benson mess is--1400+ is actually a lot "cleaner", though of course it's still a worthy range to do.

Quote:

4. Allow for individual-k reservations for ANY k-value for n>600K. In the same vain as the what the Riesel/Sierp base 5 project does, people could reserve individual k's but they would not be removed from the main drives. It would be up to the people reserving them to stay ahead of the drives on their searching. In effect, the drives would 'double check' the individual-k's that are reserved. With nearly 350 k's available at that time, there should be little fear that the drives would be doing more than 2-5% double checking of individual-k reservations.

Hmm...this might work. We'll have to be sure we hang onto both copies of the residuals for ranges searched in this manner separately--i.e., keep one "stash" of residuals from first-pass, and one "stash" for second-pass (which would include residuals from doublecheck drives once they get up that high.

Anon

Mini-Geek · 2008-08-25, 20:33

I agree with point 1 and point 2 and have nothing else to add there.

Quote:

Originally Posted by gd_barnes

3. k=1003-2000. This range is an extreme state of mess as a result of the fragmented Benson searches. Karsten and some folks at RPS have gotten the gaps filled in up to n=50K on all of them (nice work!) but there are plenty of gaps still to fill. I think there are many people that don't like the higher n-ranges that we are currently processing because it takes too long to find primes without large resources, hence a drop-off in popularity of NPLB in the last couple of months. I think that doing a team sieve and search would bring in a number of people to find smaller primes. Many people aren't concerned about top-5000 primes and those people could help us clean up this mess.

Quote:

Originally Posted by Anonymous

Also agreed. Because of the immense size of this k-range, I suggest we split it up into smaller, more manageable chunks--how about starting with 1005-1400? (1003 has already been searched to n=500K by yours truly.) We could do, say, n=50K-500K to start with, then do 1400<k<2000 next. Focusing on 1005-1400 at first is especially important since that's where most of the Benson mess is--1400+ is actually a lot "cleaner", though of course it's still a worthy range to do.

I like the idea, but to avoid just having to split it once it gets to the n-range where Drive 1 will be soon (i.e. the reason for point 1), I'd say to make it smaller k-ranges, similar to how point 1 suggests Team Drive 1 be split up. We could still follow the general plan of completing 1005-1400 earlier than 1400+, just as the actual separate drive/files make it smaller than k=~400 or k=600.

Quote:

Originally Posted by gd_barnes

4. Allow for individual-k reservations for ANY k-value for n>600K. In the same vain as the what the Riesel/Sierp base 5 project does, people could reserve individual k's but they would not be removed from the main drives. It would be up to the people reserving them to stay ahead of the drives on their searching. In effect, the drives would 'double check' the individual-k's that are reserved. With nearly 350 k's available at that time, there should be little fear that the drives would be doing more than 2-5% double checking of individual-k reservations.

I don't think we should have individual and team be doing the same work. One or the other (I'd say team-only, with individual-k work just how it is now). Either allow individual-k reservations in ranges that are currently being worked on the team drive and remove the k from the team drive, or just don't allow individual-k in the team drive's range.

gd_barnes · 2008-08-26, 01:59

Quote:

Originally Posted by Mini-Geek

I don't think we should have individual and team be doing the same work. One or the other (I'd say team-only, with individual-k work just how it is now). Either allow individual-k reservations in ranges that are currently being worked on the team drive and remove the k from the team drive, or just don't allow individual-k in the team drive's range.

Good point. Does anyone else have any input on allowing individual-k reservations for k>400? I think the point that Mini-Geek makes is that k=300-400 should be enough k's for individual-k reservations.

Of course if we don't have enough interest in individual-k reservations for k=300-400, we may need to do an n-range drive search for remaining unreserved k's like we have done for other drives.

Gary

mdettweiler · 2008-08-26, 04:18

Quote:

Originally Posted by gd_barnes

Good point. Does anyone else have any input on allowing individual-k reservations for k>400? I think the point that Mini-Geek makes is that k=300-400 should be enough k's for individual-k reservations.

Of course if we don't have enough interest in individual-k reservations for k=300-400, we may need to do an n-range drive search for remaining unreserved k's like we have done for other drives.

Gary

Now that I re-think it a bit, I think I'm inclined to agree with Mini-Geek too. Having people be able to search their own k's in the same range that will be covered by a team drive later on is a recipe for high mess, and as you guys said, 300<k<400 should be plenty for individual k reservations--after all, there's still plenty of k's left to be had as of yet (and if the popularity of the individual k's takes a sudden spike, we can always open up more, possibly in a higher k-range if we'd like).

gd_barnes · 2008-08-26, 08:32

Quote:

Originally Posted by Anonymous

Now that I re-think it a bit, I think I'm inclined to agree with Mini-Geek too. Having people be able to search their own k's in the same range that will be covered by a team drive later on is a recipe for high mess, and as you guys said, 300<k<400 should be plenty for individual k reservations--after all, there's still plenty of k's left to be had as of yet (and if the popularity of the individual k's takes a sudden spike, we can always open up more, possibly in a higher k-range if we'd like).

OK, I'll concur with that. No individual-k reservations for k>400.

We could leave it open as a possibility if people really start reserving many individual k's for k=300-400 such that we're about to run out of them but that would be a long way off.

G

mdettweiler · 2008-08-26, 14:29

Quote:

Originally Posted by gd_barnes

OK, I'll concur with that. No individual-k reservations for k>400.

We could leave it open as a possibility if people really start reserving many individual k's for k=300-400 such that we're about to run out of them but that would be a long way off.

G

Yeah, that sounds good.

gd_barnes · 2008-09-07, 08:17

I have created this new thread to discuss the future direction of NPLB as was done previously in the News thread. I have moved all posts from that thread to here.

Gary

gd_barnes · 2008-09-07, 10:32

I am seeking opinions about how to process k=1003-2000 in the future.

k=1003-1400 is a mess because it had fragmented searches done by Benson and there have been several misc. efforts to fill in some k's, some of which were coordinated and some that weren't.

All k's have been searched up to n=50K by Karsten, Henry and me. Many k's contain holes for n=50K to n=?? that need to be filled.

My initial thinking is this:
Do some sort of team sieve/drive to get all of the k's searched from n=50K-200K.

Possible different options:
Higher or lower n-range.
Split up k's into smaller pieces.
Split up k=1003-1400 from k=1400-2000.

My main objective for 2009 is that I'd like to make the project more interesting than it has been with essentially only 3 different n-ranges that could be searched, 1 of which is ~50% double-check at its current n-level. I think that having the upcoming k<300 effort plus this smaller-prime effort in addition to the regular drives will do just that.

I know that several of you like filling in gaps and holes and otherwise searching for smaller primes. So regardless of whether you have been a regular contributor here, if the idea of sieving and searching for smaller primes and cleaning up ranges appeals to you, please offer your two cents here.

Gary

IronBits · 2008-09-07, 13:18

With my limited knowledge...
Put any range that is made up of many smaller ranges on a server so everyone can just put a core or three on it until it has been exhausted.

Flatlander · 2008-09-07, 13:21

I'm wondering if k=1003-1400 would get neglected if it was seperated from k=1400-2000? (Because primes are more likely to have already been found.)

2008-08-24, 20:20	#1
gd_barnes May 2007 Kansas; USA 13²×61 Posts	Future direction of NPLB We are taking suggestions on the future direction of NPLB. Here is what I would propose at this time. I think it should be used as a starting point for discussion and not set in stone: 1. How the drives will proceed or be divided up for the range of n=600K-1M. I would like to propose that we leave the k=300-400 range as is for a period of time but leave open the possibility of doing n-range searches like we've done for drives 1, 2, and 3 for k's that are unreserved once we have all k=300-1001 up to n=600K. For k=400-1001, I'd like to propose that we divide that up into smaller k-ranges as separate drives, as follows: (1) k=400-600 (2) k=600-800 (3) k=800-1001. My thinking is that this will make the project more interesting because each of the drives would likely be at different search limits. Once we hit an fftlen change at the higher k-values, that will make the lower k-values a little more appealing and they would be searched deeper. That would give more of a choice as to what n-value people choose to search at. 2. k<300 for unreserved k's only searched to n=600K. There are currently about 15 unreserved k's < 300 that are at n=600K and are not being worked on yet we are already nearing or at n=1M on several k=300-400. My thinking is that we should make this a separate drive and sieving effort before we start k>400 for n=600K-1M. Then we could open all k-ranges at once and have 5 drives running concurrently, all that will likely be at different n-ranges after a period of time. My thinking is that it doesn't make sense to search higher k's to n=1M while leaving lower k's at only n=600K. 3. k=1003-2000. This range is an extreme state of mess as a result of the fragmented Benson searches. Karsten and some folks at RPS have gotten the gaps filled in up to n=50K on all of them (nice work!) but there are plenty of gaps still to fill. I think there are many people that don't like the higher n-ranges that we are currently processing because it takes too long to find primes without large resources, hence a drop-off in popularity of NPLB in the last couple of months. I think that doing a team sieve and search would bring in a number of people to find smaller primes. Many people aren't concerned about top-5000 primes and those people could help us clean up this mess. 4. Allow for individual-k reservations for ANY k-value for n>600K. In the same vain as the what the Riesel/Sierp base 5 project does, people could reserve individual k's but they would not be removed from the main drives. It would be up to the people reserving them to stay ahead of the drives on their searching. In effect, the drives would 'double check' the individual-k's that are reserved. With nearly 350 k's available at that time, there should be little fear that the drives would be doing more than 2-5% double checking of individual-k reservations. All thoughts and opinions are welcome. Gary

2008-09-07, 08:17	#8
gd_barnes May 2007 Kansas; USA 10100001000101₂ Posts	Direction of NPLB I have created this new thread to discuss the future direction of NPLB as was done previously in the News thread. I have moved all posts from that thread to here. Gary Last fiddled with by gd_barnes on 2008-09-07 at 08:20

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Move 19 game direction	MooMoo2	Other Chess Games	8	2016-02-01 17:50
NPLB future direction	gd_barnes	No Prime Left Behind	16	2009-05-13 16:45
Thoughts on future NPLB searches	em99010pepe	No Prime Left Behind	38	2008-12-23 10:02
Future project direction and server needs synopsis	gd_barnes	No Prime Left Behind	6	2008-02-29 01:09
Poll on direction of conjectures effort	gd_barnes	Conjectures 'R Us	2	2007-12-19 18:15

2008-09-07, 10:32	#9
gd_barnes May 2007 Kansas; USA 13²×61 Posts	I am seeking opinions about how to process k=1003-2000 in the future. k=1003-1400 is a mess because it had fragmented searches done by Benson and there have been several misc. efforts to fill in some k's, some of which were coordinated and some that weren't. All k's have been searched up to n=50K by Karsten, Henry and me. Many k's contain holes for n=50K to n=?? that need to be filled. My initial thinking is this: Do some sort of team sieve/drive to get all of the k's searched from n=50K-200K. Possible different options: Higher or lower n-range. Split up k's into smaller pieces. Split up k=1003-1400 from k=1400-2000. My main objective for 2009 is that I'd like to make the project more interesting than it has been with essentially only 3 different n-ranges that could be searched, 1 of which is ~50% double-check at its current n-level. I think that having the upcoming k<300 effort plus this smaller-prime effort in addition to the regular drives will do just that. I know that several of you like filling in gaps and holes and otherwise searching for smaller primes. So regardless of whether you have been a regular contributor here, if the idea of sieving and searching for smaller primes and cleaning up ranges appeals to you, please offer your two cents here. Gary

2008-09-07, 13:18	#10
IronBits I ♥ BOINC! Oct 2002 Glendale, AZ. (USA) 3×7×53 Posts	With my limited knowledge... Put any range that is made up of many smaller ranges on a server so everyone can just put a core or three on it until it has been exhausted.

2008-09-07, 13:21	#11
Flatlander I quite division it "Chris" Feb 2005 England 31×67 Posts	I'm wondering if k=1003-1400 would get neglected if it was seperated from k=1400-2000? (Because primes are more likely to have already been found.)