Further Plans: 6th Drive, other ideas

[VBCurtis]

[The first part of this message, unrelated to the topic, can be found here, as post #71.]

...

Sheep's multi-k sieve is by agreement and prime sharing between he and I, and is for k's I have reserved plus k=5. 5 and 11 will be shared with RPS, possibly 45 also; the other 4 are for my personal testing. A 6th drive of new k's (probly most of the 4th drive k's, tested to 600k already) will begin sieving around Thanksgiving, with testing ready after 1/1. Sheep will likely do most of the sieving on this also, as he prefers to help larger groups. Testing ranges and number of k's are still up in the air- Carlos and I were considering 8 to 10 k's on a range from 600k to 2M. Still undecided is which of the available tested-to-600k k's to pick- highest weight, lowest, middle, mix of all? The 2M upper bound is also up for discussion. 1.5M would finish in about half the time, thus half the sieving time also or less.
-Curtis

[mdettweiler]

Originally Posted by VBCurtis

...A 6th drive of new k's (probly most of the 4th drive k's, tested to 600k already) will begin sieving around Thanksgiving, with testing ready after 1/1. Sheep will likely do most of the sieving on this also, as he prefers to help larger groups. Testing ranges and number of k's are still up in the air- Carlos and I were considering 8 to 10 k's on a range from 600k to 2M. Still undecided is which of the available tested-to-600k k's to pick- highest weight, lowest, middle, mix of all? The 2M upper bound is also up for discussion. 1.5M would finish in about half the time, thus half the sieving time also or less.
-Curtis

After thinking about this a bit, I remembered something: On the Prime Pages, in the top-20 project rankings by number of primes, RPS is in second place, 15 primes behind Prime Internet Eisenstein Search. From what I've gathered around here, it seems that the general opinion is that RPS should try to capture the top spot as soon as possible. Thus, with that in mind, I would think that we'd want to pick the highest weight k's possible (of the ones we're choosing from) for the 6th drive--so we can maximize the number of primes potentially to be found as quickly as possible. As for the upper bound of n, I guess it depends on whether we want the 6th drive to be done more quickly (i.e. to prevent people from losing interest--the 3rd drive is a prime example [no pun intended!] of one for which people appear to have have lost interest and let it languish), or whether we want it to go longer (but, of course, higher).

Should somebody set up a poll for this? Or would you rather this discussion be kept a little more confined?

[Kosmaj]

On the Prime Pages, in the top-20 project rankings by number of primes, RPS is in second place, 15 primes behind Prime Internet Eisenstein Search. From what I've gathered around here, it seems that the general opinion is that RPS should try to capture the top spot as soon as possible.

Hi Anon, I don't share your opinion that we should try to do it asap. We started RPS from zero two years ago and since we waited two years we can wait another two weeks or two months. Besides, sieving for any serious drive will take two weeks or more. I think we better use our resources to run more LLR tests during that time.

About the 3rd Drive: we have sieved numbers up to n=2M but large k's represented in the drive just take too long time to test, while so far no primes have been found. So the people who used to drive the 3rd Drive decided to move towards "greener pastures"...

[mdettweiler]

[The first part of this message can be found here as post #79.]

...

Originally Posted by Kosmaj

Hi Anon, I don't share your opinion that we should try to do it asap. We started RPS from zero two years ago and since we waited two years we can wait another two weeks or two months. Besides, sieving for any serious drive will take two weeks or more. I think we better use our resources to run more LLR tests during that time.

About the 3rd Drive: we have sieved numbers up to n=2M but large k's represented in the drive just take too long time to test, while so far no primes have been found. So the people who used to drive the 3rd Drive decided to move towards "greener pastures"...

I wasn't really saying that my personal opinion is that we should try to get in first place ASAP, just that the general opinion around here seems to be that everyone is in competition to do so, and I figured that thus RPS would probably be interested in doing high-weight k's for the 6th drive. However, IMO I don't really mind whether we do high-weight, low-weight, or in-between--they all have thier advantages and disadvantages. It wouldn't affect whether I would crunch for the drive.

As for the 3rd Drive: I had kind of figured that's what had happened, considering how long the number take to test, but I still think we should finish what we've started...maybe if one or two people with a bunch of resources could put one or two cores on it until it's done, then we could at least have it cleaned up and out of the way. I mean, it is a bit silly to still have the 3rd Drive going, when the 4th has already been completed, the 5th will probably be done within a few months, and the 6th is to begin sieving in about a week or two...so all I'm saying is that we should at least try to make some effort towards getting it finished.

Edit: By the way, is the rps_3rd_drive yahoo address that we're supposed to send 3rd Drive results to still active? I know that Yahoo deactivates free accounts after 4 months of inactivity, and it would seem that might have happened (unless someone has logged in a couple times since the drive went inactive); if it's been deactivated, then I'll simply attach the lresults.txt file for my current 3rd Drive reservation to my "completed" post when I'm done, rather than send it in to an inactive email address.

[VBCurtis]

Originally Posted by Anonymous

After thinking about this a bit, I remembered something: On the Prime Pages, in the top-20 project rankings by number of primes, RPS is in second place, 15 primes behind Prime Internet Eisenstein Search. From what I've gathered around here, it seems that the general opinion is that RPS should try to capture the top spot as soon as possible. Thus, with that in mind, I would think that we'd want to pick the highest weight k's possible (of the ones we're choosing from) for the 6th drive--so we can maximize the number of primes potentially to be found as quickly as possible. As for the upper bound of n, I guess it depends on whether we want the 6th drive to be done more quickly (i.e. to prevent people from losing interest--the 3rd drive is a prime example [no pun intended!] of one for which people appear to have have lost interest and let it languish), or whether we want it to go longer (but, of course, higher).

Should somebody set up a poll for this? Or would you rather this discussion be kept a little more confined?

To find more primes, we should select more k's with a smaller range; 12 k's to 1.5M will be a shorter project than 8 k's to 2M, yet should contain more primes. I count 13 k's <200 with current n-max at 600k, 17 more in 200<k<300. How about 6 of the former, 6 of the latter? We'll choose the 6 with the greatest number of known primes as our weight judgement. Is this too much work? 5 from each category better?
Polls entail choosing from predetermined ideas- I'm looking for suggestions more than votes.
-Curtis

[mdettweiler]

Originally Posted by VBCurtis

To find more primes, we should select more k's with a smaller range; 12 k's to 1.5M will be a shorter project than 8 k's to 2M, yet should contain more primes. I count 13 k's <200 with current n-max at 600k, 17 more in 200<k<300. How about 6 of the former, 6 of the latter? We'll choose the 6 with the greatest number of known primes as our weight judgement. Is this too much work? 5 from each category better?
Polls entail choosing from predetermined ideas- I'm looking for suggestions more than votes.
-Curtis

Yeah, 12 k's to 1.5M sounds like a good idea; even with the whole number of primes thing aside, doing more k's, but to 1.5M instead of 2M, means that the drive would stay at lower n levels longer--thus, people don't lose interest as quickly as they might for higher n levels. I like that idea, of doing 12 k's to 1.5M.

As for which k's to do, I took a look at the k<300 stats page on 15k.org, and made a list of all the open k's that are at n=600,000, as well as the number of primes for each:

107 (32 primes)      203 (23 primes)
115 (30 primes)	     205 (32 primes)
133 (24 primes)	     215 (39 primes)
143 (38 primes)	     217 (17 primes)
145 (17 primes)	     227 (31 primes)
149 (13 primes)	     229 (45 primes)
167 (36 primes)	     235 (23 primes)
169 (25 primes)	     239 (6 primes) 
175 (36 primes)	     241 (22 primes)
179 (12 primes)	     257 (34 primes)
185 (32 primes)	     259 (31 primes)
187 (20 primes)	     265 (33 primes)
199 (56 primes)	     271 (20 primes)	             	               
 		     275 (27 primes)
		     287 (24 primes)
		     293 (19 primes)
		     295 (27 primes)

Based on that list, the top 6 k's (by number of primes), at n=600k, for k<200 and 200<k<300 are:
k<200: 107, 115, 143, 167, 175, 199
200<k<300: 215, 227, 229, 257, 259, 265
So, if we want to do the top 6 k's by number of primes in each of those ranges, the above would probably be the ones we want.

If we removed the limitation of having the same number of k's from the ranges k<200 and 200<k<300, then the balance could be weighed even more in favor of high-weight k's, since it seems that k<200 has more high-weight k's overall than 200<k<300. I guess it depends on whether we'd rather get k's with as high weight as possible, or spread around the k's more evenly throughout the k<300 range.

[gd_barnes]

I'm also very much in favor of having a 6th drive with as many k's as possible but have it's completion be a lower range of n. That said, sieving efficiency is much greater for higher ranges of n. Maybe we could sieve a lot of k's to a very high n with the drive only being a subset of that n-range.

Perhaps sieve 15-20 k's up to n=2.5M-3M but have a 6th drive go up to n=1.5M. Then people can individually sieve and test a k or a smaller group of k's for n>1.5M as they please after the drive is over.


G

[VBCurtis]

Anon- thanks for the footwork! I count 13 k's with more than 30 known primes (ignoring 115 at exactly 30). 6 are below 200, 7 above. How about these 13 for the 6th drive? Are there any suggestions for other k's to consider?

If we do settle on these 13, our discussion moves to the choice of n-max. 1.4M will complete in a reasonable time, 1.5 less reasonable, on up to 2M which would take quite a number of months to complete. Remember that 1.6M to 1.7M range will take as long for each k as 600k to 1000k!

The 5th drive has run for quite a long time-- do we want a project of similar length, a little longer, or much longer? Our sieving resources should also play some part- those of you interested in participating, what sieve resources can you contribute? Please rate in Ghz-months (and athlon or core2). I would be willing to use my dual-1.6ghz of athlon for 4 months to sieve this, so I have 12 ghz-months (athlon) to offer.

[mdettweiler]

Originally Posted by VBCurtis

Anon- thanks for the footwork! I count 13 k's with more than 30 known primes (ignoring 115 at exactly 30). 6 are below 200, 7 above. How about these 13 for the 6th drive? Are there any suggestions for other k's to consider?

If we do settle on these 13, our discussion moves to the choice of n-max. 1.4M will complete in a reasonable time, 1.5 less reasonable, on up to 2M which would take quite a number of months to complete. Remember that 1.6M to 1.7M range will take as long for each k as 600k to 1000k!

The 5th drive has run for quite a long time-- do we want a project of similar length, a little longer, or much longer? Our sieving resources should also play some part- those of you interested in participating, what sieve resources can you contribute? Please rate in Ghz-months (and athlon or core2). I would be willing to use my dual-1.6ghz of athlon for 4 months to sieve this, so I have 12 ghz-months (athlon) to offer.

The 13 k's you mentioned sound good, I think we should go with them.

As for n-max, you brought up an interesting point when you mentioned about the length of the 5th drive. It's been going since about a year ago, somewhere around now in November (according to the original posting date on the main status post at the beginning of the 5th drive thread), and that's not counting the sieving. I would personally think a shorter drive would be better, about 6 months might be good, to keep interest high. How long, roughly, do you think it would take, to run the 6th drive to 1.4M? How long for 1.5M? It would appear 2M is out of the question, so I won't bother asking about it.

As for sieving resources, unfortunately I'm just about a washout. All I would possibly be able to contribute is one hyperthread (with BOINC running on the other hyperthread) of a P4 3.2Ghz, running only about 10-12 hours a day. And that's all busy with 3rd drive work right now, and I was thinking of doing some k=105 LLR work right after that. I could possibly replace the k=105 LLR work with some sieving for this, though, if it's needed badly, and assuming that Carlos can do all the heavy lifting for getting k=105 to 1M by the end of the year like he says. Anyway, though, in the case that my CPU power is needed for sieving the 6th drive, it will probably become available around the beginning of December.

In the case that we don't have enough sieving resources here to do the job to optimal depth in a decent amount of time (I'm not saying that we don't, this is just an idea in case we end up in that situation), we could contact PrimeGrid and see if they could set up our sieving as a temporary BOINC subproject. The reason why I mention PrimeGrid is because they have already helped two Mersenneforum.org prime search projects (TPS and PSP) go BOINC through them (PSP only with sieving so far, LLR in the works), and they already have an sr2sieve wrapper for BOINC (developed for their PSP sieve application), so all they would have to do would be to simply add another app listing to their system, telling it to use the existing sr2sieve application files, and call it "RPS Sieve" or something like that...they could sieve the 6th drive, then remove the app from their system when it's done (since it would only be filling a temporary sieving need). Carlos, I know how you feel about BOINC, but it might be a good way to get some resources for sieving the 6th drive--not to mention possibly getting more people interested in RPS. I actually found RPS because of PrimeGrid's link to the prime search projects here (actually it was just to the TPS forum at the time). So, I guess it might not exactly be too bad to do it for a temporary thing to do this sieving. Unless, of course, we've got plenty of resources to do it ourselves.

[em99010pepe]

I passed my last three years sieving for PSP, RieselSieve and Riesel Base 5 so I am off sieving. For me only LLRing.
BOINC can be used to sieve and to LLRring but please leave the manual reservation

Carlos

[AES]

Originally Posted by VBCurtis

Please rate in Ghz-months (and athlon or core2). I would be willing to use my dual-1.6ghz of athlon for 4 months to sieve this, so I have 12 ghz-months (athlon) to offer.

I'll match that. 1 core of an AMD 6000+ @ ~3ghz for 4+ months

I don't use BOINC