mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   No Prime Left Behind (https://www.mersenneforum.org/forumdisplay.php?f=82)
-   -   Team drive #7 k=800-1001 n=600K-1M (https://www.mersenneforum.org/showthread.php?t=11163)

mdettweiler 2008-12-25 17:35

[quote=gd_barnes;155037]Why, is it too hard to copy the five n=200 files into one file? lol

[URL="http://gbarnes017.googlepages.com/NPLB-800-6000-6010.txt"]Here[/URL] is one file.

Max,

I think you are the only person who takes just one file at a time. Perhaps we should just start posting 2-3 n=1000 files at a time. If people want a smaller range, they can tell us. What do you think?

It's takes time to split the files up into smaller parts so I think it would make more sense to post bigger files.

For now, I'll post the bigger files in this drive. If you agree, then you/we can change the other drives to bigger files also.


Gary[/quote]
Okay, that sounds good--yeah, let's do that. 1K files it is. :smile: And, of course, as you were saying, if people want a smaller range, they can just say so, or pull it out of the larger file manually.

Max :smile:

MooooMoo 2008-12-27 04:52

I know this might sound greedy, but could I reserve 980K-1M? I want to beat my personal record for largest prime found, which is currently a bit over 275 thousand digits long.

I'm pretty sure that I'll finish the range before this NPLB drive gets there, but if I don't, I'll release the range and post all results found so far.

gd_barnes 2008-12-29 07:57

[quote=MooooMoo;155248]I know this might sound greedy, but could I reserve 980K-1M? I want to beat my personal record for largest prime found, which is currently a bit over 275 thousand digits long.

I'm pretty sure that I'll finish the range before this NPLB drive gets there, but if I don't, I'll release the range and post all results found so far.[/quote]


Sorry we didn't respond earlier...

Max and I had a PM discussion about this and I did a fair amount of thinking about it myself.

This would be far too big of a range for us to hand out to a single person at such a high n-range, even if that person was Beyond or Lennart who could likely process it in a few weeks. We would, of course, allow them to reserve a large range but it would have to be at or near the leading edge of our servers and they'd have to guarantee that they could finish it in a few weeks so it wasn't "left behind" for too long. :smile:

I'm not personally fond of huge gaps in testing k's. But I realize that everyone has different goals and tastes when it comes to DC and to prime searching so am willing to accomodate those differences. I am also keenly aware that neither NPLB nor anyone or entity owns any part of the math world, which includes the search for prime numbers.

Therefore, I'll offer something similar to something that Carlos and I had talked about a little bit previously and that Max brought up independently of that to me in our PM discussion:

Consider doing an n=980K-1M search on a small # of k's in the individual-k drive. Here is what I'd suggest: We're going to start a mini-drive on 5 k's of that drive sometime in Jan. They are the 5 lowest remaining unreserved k's that are divisible by 3, that is: 321, 327, 333, 339, and 345. Give those k's a shot to start with. Since all are divisible by 3, they are relatively heavy-weight.

You didn't say how many cores you could put on the effort. My perception is that 5 higher-weight k's at such a high n-range should keep a full quad busy for at least a week, perhaps even a month. I haven't done enough searching at that level to remember testing times nor have I done exact calculations on the # of candidates for those k's in those ranges.

If you like that idea, we'll only search n=600K-980K with our mini-drive. I like the idea because the gap in the k's will be filled in within a few months. If you want to start with that, please post your reservation in that drive and I'll send you the files. If you'd like to continue after you are done, you can do some more k's in that drive for the same n-range. We'd prefer if you only took pieces that you can complete in 3-6 months at a time so after those 5 k's, you could do 5 more there or even do the same k's for n=960K-980K if our mini-drive hasn't reached that point. That said: We don't want people to get too much in the habit of searching ranges backwards so please limit the "reverse" reservations as much as possible; likely just one reverse reservation per k like I described here.

Note to all: I don't want to set a bad prescident here. We're OK with this because it's a range right at the top of our drives. If people start wanting to reserve n=800K-850K or 700K-725K right in the middle of our drives, I/we won't be too fond of that until the drives get there. If people want the very top of our files to search, then let's stick with the individual-k drive and reserve ranges for n=950K or higher thru 1M.


Gary

MooooMoo 2008-12-29 22:05

[QUOTE=gd_barnes;155557]This would be far too big of a range for us to hand out to a single person at such a high n-range, even if that person was Beyond or Lennart who could likely process it in a few weeks

...

Consider doing an n=980K-1M search on a small # of k's in the individual-k drive.

...

321, 327, 333, 339, and 345. Give those k's a shot to start with. Since all are divisible by 3, they are relatively heavy-weight.

You didn't say how many cores you could put on the effort.

...

after those 5 k's, you could do 5 more there or even do the same k's for n=960K-980K if our mini-drive hasn't reached that point. [/QUOTE]
Searching those 5 k's is an interesting idea, but I'll have to refuse it since the chance of finding a prime isn't that great. Even if those high-weight k's were tested from 960k-1M, the chances of finding a prime there might be 50-60% at best. I wanted to test a range that almost certainly has a prime (which to me is a 95-99% chance), which is why I was asking to test this huge range.

Since you thought that it was too big of a range to be handed to a single person, could I take the smaller range of n=999K - 1M then? Once that's done, then I'll move to n=998K - 999K, and so on until I find a prime or until this drive gets there.

Anyway, I'll probably put only 2 cores on the effort. What little (compared to some people here) computing power I have is going to be stretched thin - one is on k=313, the other is on k=173, and I'll be putting my two remaining cores on this near n=1M effort.

gd_barnes 2008-12-30 01:51

[quote=MooooMoo;155674]Searching those 5 k's is an interesting idea, but I'll have to refuse it since the chance of finding a prime isn't that great. Even if those high-weight k's were tested from 960k-1M, the chances of finding a prime there might be 50-60% at best. I wanted to test a range that almost certainly has a prime (which to me is a 95-99% chance), which is why I was asking to test this huge range.

Since you thought that it was too big of a range to be handed to a single person, could I take the smaller range of n=999K - 1M then? Once that's done, then I'll move to n=998K - 999K, and so on until I find a prime or until this drive gets there.

Anyway, I'll probably put only 2 cores on the effort. What little (compared to some people here) computing power I have is going to be stretched thin - one is on k=313, the other is on k=173, and I'll be putting my two remaining cores on this near n=1M effort.[/quote]


I figured that was why you wanted the big range: To have a good chance at a prime. That's why I suggested doing the 5 individual-drive k's, then another 5 k's, etc.

No, I'd prefer not on these big drives. It's more than size of the reservation. It will simply take us too long to fill in the gaps. Can you imagine Karsten having to show n=999K-1M on 300 k's in addition to the current search limit? That's a big part of the issue. It's better to do fewer k's for larger n-ranges.

How about this: Do the 5 individual k's that I suggested, but do them for n=960K-1M instead on the 1st pass. You wouldn't have to do such a small piece as n=980K-1M. After you're done, whether you have a prime or not, you can do another 5 k's for n=960K-1M. I haven't calculated the odds but off the top of my head, I can say that would likely give you a 80-90%+ chance at a prime, especially if all of the k's are heavier-than-average weight. If you still don't have a prime, you can try another 5 k's if available or do n=920K-960K on the k's you've already done.

I can virtually guarantee there should still be plenty of k's available after you're done with the first 5 k's. But just in case, keep an eye on that drive.

Personally, I'm not all that fond of high-n searches and I'll likely keep 2-3 average to above-average weight k's there reserved at all times. I wouldn't have any problem letting you take n=960K-1M for the k's that I reserve in the future.

Doing it this way makes it far more easy for Karsten to maintain his pages, still gives you a virtual lock on finding the prime that you want, and leaves gaps in k's that it won't take us too long to fill in. Everyone wins! :smile:

One other benefit to doing k=300-400 vs. k=800-1001. The testing time's may be shorter because the k's are smaller. I haven't checked where the fftlen changes are but that possibility does exist here.


Gary

MyDogBuster 2008-12-30 02:04

[QUOTE]Personally, I'm not all that fond of high-n searches and I'll likely keep 2-3 average to above-average weight k's there reserved at all times. I wouldn't have any problem letting you take n=960K-1M for the k's that I reserve in the future.
[/QUOTE]

Part of the reason I take a whole k is the 900's. I don't particular like doing them but it's part of the k and it gives me a chance at a big prime. If the top 40N are missing, I probably won't take that k. Please keep some k's intact.

gd_barnes 2008-12-30 02:14

[quote=MyDogBuster;155699]Part of the reason I take a whole k is the 900's. I don't particular like doing them but it's part of the k and it gives me a chance at a big prime. If the top 40N are missing, I probably won't take that k. Please keep some k's intact.[/quote]


OK, shall do. That's why I was suggesting he do them 5 k's at a time. I think there will be enough high-n ranges to go around for everyone.

MooooMoo 2008-12-30 03:22

[QUOTE=gd_barnes;155698]No, I'd prefer not on these big drives. It's more than size of the reservation. It will simply take us too long to fill in the gaps. Can you imagine Karsten having to show n=999K-1M on 300 k's in addition to the current search limit? That's a big part of the issue. It's better to do fewer k's for larger n-ranges.

How about this: Do the 5 individual k's that I suggested, but do them for n=960K-1M instead on the 1st pass. You wouldn't have to do such a small piece as n=980K-1M. After you're done, whether you have a prime or not, you can do another 5 k's for n=960K-1M. I haven't calculated the odds but off the top of my head, I can say that would likely give you a 80-90%+ chance at a prime, especially if all of the k's are heavier-than-average weight. If you still don't have a prime, you can try another 5 k's if available or do n=920K-960K on the k's you've already done.

I can virtually guarantee there should still be plenty of k's available after you're done with the first 5 k's. But just in case, keep an eye on that drive.

...

One other benefit to doing k=300-400 vs. k=800-1001. The testing time's may be shorter because the k's are smaller. I haven't checked where the fftlen changes are but that possibility does exist here.
[/QUOTE]

[QUOTE=MyDogBuster;155699]Part of the reason I take a whole k is the 900's. I don't particular like doing them but it's part of the k and it gives me a chance at a big prime. If the top 40N are missing, I probably won't take that k. Please keep some k's intact.[/QUOTE]

Is it OK if I test those 5 k's from 920K-1M then? If I don't find a prime in that range, then I'll extend my k=313 reservation past n=1M and hope for a prime there.

gd_barnes 2008-12-30 06:12

[quote=MooooMoo;155704]Is it OK if I test those 5 k's from 920K-1M then? If I don't find a prime in that range, then I'll extend my k=313 reservation past n=1M and hope for a prime there.[/quote]


OK. That works. Can you give me an ETA? Use this for estimating: Look at the # of candidates in each of the 5 k's in the individual-k drive and multiply each by 1/5th (since 80K/400K = 20%). Then run a test on your machine at n=975K and use that as an average test time. The higher-than average n-range test accounts for a possible fftlen change.

6 months or less is preferred for any reservation as suggested in that drive. If it's a little over, it's no big deal. We'll likely look to complete the mini-drive in about that time frame, perhaps a little longer.

I think the bottom line here is: I'm pretty confident that you can find an n=920K-1M prime in < 6 months on 2 cores (assuming >= 2 Ghz) so I'm guessing that you'll be in good shape with that reservation.

After you calculate an estimate, I'll send you the files.

Ian, the 5 k's that I'm recommending are what we're going to put in a mini-drive so they won't take from other individual k's that are still available. I think this is a good solution for all involved. Also, it will shorten the mini-drive quite a bit, which won't bother me at all. lol


Gary

MooooMoo 2008-12-30 06:35

[QUOTE=gd_barnes;155719]OK. That works. Can you give me an ETA?

...

6 months or less is preferred for any reservation as suggested in that drive
[/QUOTE]
After estimating, I've decided to change the reservation to 950K-1M. The ETA for that range is sometime in early June.

gd_barnes 2008-12-30 13:21

Better load 'er up Max. :smile:

With Ian on here full-time, n=10K is fine.


All times are UTC. The time now is 14:52.

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