NPLB future direction

[gd_barnes]

That clinches it. We need to start sieving k=2000-10000 for n<3M.

Anyone game for that? lol

[Flatlander]

I'll sieve up to p=1000. Should be ready about August.

[kar_bon]

Originally Posted by Flatlander

I'll sieve up to p=1000. Should be ready about August.

don't do that immense work!

all k<10000 are searched upto n=20000 (but for ~ k>3000 only online to n=10000 so far)

[henryzz]

Originally Posted by gd_barnes

That clinches it. We need to start sieving k=2000-10000 for n<3M.

Anyone game for that? lol

i think that next time we start a sieve of new ks we should sieve a huge range of ks at once
certainly more than a range of 1000
maybe up to k=10000 n<1M would be a good idea
i suppose it depends on how much memory we would need to sieve such a humungous range

[Flatlander]

I think it would be good to search 2000 <k 3000 before top-5000 gets into n=500,000 territory.

[Flatlander]

All our reportable primes are now >620k. I miss the lower/easier ones.
(I'm Chris b.t.w. )

[gd_barnes]

Originally Posted by Flatlander

All our reportable primes are now >620k. I miss the lower/easier ones.
(I'm Chris b.t.w. )

Of course you have the mini-drive that you sent me your file for. :-)

That said, with Free-DC crunching away on the 10th drive, NPLB has gotten a bit stale and that is definitely something that we want to avoid. All server ranges with the exception of G8000 are now in the n=620K-700K range.

Therefore, I think it's time to open up sieving efforts for k=2000-3000 for n=20K-1M and k=300-1001 for n=1M-2M. These efforts have already been started on the side. Several users have expressed interest to me for both larger and smaller n-ranges. That was part of the intent of the 8th/10th drives but they got higher real quick. Not that I'm complaining; it's just that we need some more variety.

We could also sieve k=3000-4000 for n=20K-1M, catch it up to the k=2000-3000 range (which is still at a depth of P<1T) and then merge them together. We could also make a partial and manual "individual k" range out of some of these sieve ranges. Edit: I suspect we could get a k=2000-4000 range sieved to P=20T within < 2 months with a concerted effort, which should be sufficient for searching n<=600K. Chris, I'm sure we would be able to start on the range long before the 5000th place prime is at n=450K.

I'm open to ideas. I'll move this discussion to a new thread as responses come in.


Gary

[mdettweiler]

Originally Posted by gd_barnes

Edit: I suspect we could get a k=2000-4000 range sieved to P=20T within < 2 months with a concerted effort, which should be sufficient for searching n<=600K.

Hmm...interesting idea. It would be nice to get some fresh meat for the project once again.

One thing, though, that I think we should keep in mind. I believe that as we continue to expand this project further and further up the k-range spectrum, we need to start setting some hard limits for n-values for given k-ranges, so that we're not piling up drives upon drives that are all going nowhere because they've passed the point where they're "interesting".

How about this: for the range of k=1400-2000, let's set a "stopping point" of n=1M. That is, when it reaches 1M, we won't immediately move on to >1M. Instead, we'll spend those resources on the k=2000-4000 range, and get that up to whatever target we set--possibly 1M, or maybe 600K. Once that's done, we'd move on to the next range, possibly k=4000-6000 or something like that. Et cetera.

k=400-1001 could be left has the "no limit" range--i.e., that's the one that we continue to extend the n-range for whenever we finish the current range. That will fill out the high-n end of the project's selection of work.

Thoughts, flames, rotten tomatoes?

Max

[MyDogBuster]

How about this: for the range of k=1400-2000, let's set a "stopping point" of n=1M. That is, when it reaches 1M, we won't immediately move on to >1M. Instead, we'll spend those resources on the k=2000-4000 range, and get that up to whatever target we set--possibly 1M, or maybe 600K. Once that's done, we'd move on to the next range, possibly k=4000-6000 or something like that. Et cetera.

I'm interested in the higher n ranges, n>1M. I also like the n ranges 700K-1M. I actually don't particularly care for the n ranges n<700K.

I think you will find people who enjoy just about any range you can come up with. I do think most people enjoy the lower ranges, so they can do those ranges and people like me can pick up the rest.

Trying to keep up with Prime Grid does not interest me. My first thought about Lennarts' post was "Who Cares". I will lose exactly zero sleep trying to figure out how to beat Prime Grid. RPS was the goal. They are non-Boinc, so are we.

[Flatlander]

Originally Posted by gd_barnes

Of course you have the mini-drive that you sent me your file for. :-)
Gary

I'd forgotten about that!
And it looks so masterfully sieved too.
btw How is the sieve depth on this? I guessed 3 or 4T.

Originally Posted by gd_barnes

We could also sieve k=3000-4000 for n=20K-1M, catch it up to the k=2000-3000 range (which is still at a depth of P<1T) and then merge them together.
Gary

Let's do it!

edit:
Maybe this thread should be split?

[gd_barnes]

Originally Posted by Flatlander

I'd forgotten about that!
And it looks so masterfully sieved too.
btw How is the sieve depth on this? I guessed 3 or 4T.

Let's do it!

edit:
Maybe this thread should be split?

Why would you need to ask? You sieved it to P=4T. I just created the drive and posted the files. I didn't check optimum depth but my guess is for such a small k and n-range that is n<=500K, that is sufficient.

I'll split this thread off for further discussion.

If people agree, I could start sieving k=3000-4000 for n=20K-1M on Thursday and then coordinate with Karsten on merging it with his k=2000-3000. 300 k's move to fast (i.e. k=400-1001 & 1400-2000) and so would 500 k's (i.e. k=2000-3000) with all of the horsepower we now have. Even after the major Free-DC folks leave at the end of May, I'm thinking the project needs to have a drive that proceeds VERY slowly; that is one with 1000 k's in it! THEN we would have drives that stay much longer at different search depths. Later on after we get k=300-1001 for n=1M-2M sieved enough for the lower n-ranges of that range, my hope is that the project would have high top-5000 ranges, medium top-5000 ranges, and lower top-5000 ranges to search.

One further thing; like we did for the individual-k drive, I see no reason why we couldn't start searching n>1M for some k's, even if those k's are not yet contiguously searched to n=1M yet. But for Karsten's (and my) sanity, I think it's best to limit that to perhaps 50-100 k's at the most. The individual-k drive has worked well in that regard since it is only 50 k's. Since k=300-400 is our lowest k-range, that would be a good place to start searching n>1M. But one more thing: k=300-400 may be searched completely to n=1M by the end of the year, which may be about when sieving is complete for the lower n-ranges of n=1M-2M so the point may be moot.


Gary