[quote=Kosmaj;118778]Starting a drive at n=1M doesn't look good to me, it will take us ages to complete it, there won't be many primes, and at the same time we'll leave many promising Ks at 600k.

Therefore I'd like to suggest that we keep on selecting Ks already tested to 600k from the table provided by Anon in a post a few days ago. If we go to 1.4M I think 12 Ks will be enough. If we want 16 or more Ks I think we stop at 1.1 or 1.2M.

Running 16 or 20 Ks up to 1.4M sounds very unrealistic to me, but if that's what people want then it's all right.

Let's see, me and Anon suggested k=199, 239, 143 and 229. Any other suggestions? BTW, if somebody wants to test some of suggested Ks [I]now[/I] we can remove them from the list.

As for k<100 we can consider adding k=71 and 95 from their respective tested limits at 736k and 700k (but neither of them looks very promising).[/quote]


For once, I agree with Kosmaj! :surprised :grin: (I bet you didn't think that was possible!)

The more I thought about the original suggestion of going to n=1.4M and then my later suggestion of n=1.5M, the more I thought that it would make for a very long drive.

We might want to consider just taking them from n=600K to 1M or 1.2M tops, complete the drive, and then perhaps start a 7th drive from that point on. (Hopefully increased computer speeds in a couple of years will make that a better option.)

Here's what I'm seeing with the 1.4 or 1.5M limit, regardless of whether we have 12 or 16 k's: Here we are late in the year 2009 kind of grinding along. We're up to 1.3M, and each file is taking forever to process and is producing no primes for very large intervals. I can just see the drive stalling out at that point or possibly taking another 3 years just to do the final n=200K. Especially with 16 k's.

I make this suggestion regardless of whether we have 12 k's or more. As you all know, the timings are intense above n=1M. k=5 has driven that point home to me.

Also, Kosmaj suggests k=71 and 95. Even though they don't look promising, they do appear to be very low weight and we never really know where the primes are so I think they are good choices. I think it is a good thing to consider k's other than just those tested to n=600K and k>100. Perhaps in doing that, we could eliminate a few for k>200. The timings will be somewhat less for the lower k's. I'm thinking that it won't be a problem if we aren't testing every k at every n-interval.


Gary

[quote=gd_barnes;118807]For once, I agree with Kosmaj! :surprised :grin: (I bet you didn't think that was possible!)

The more I thought about the original suggestion of going to n=1.4M and then my later suggestion of n=1.5M, the more I thought that it would make for a very long drive.

We might want to consider just taking them from n=600K to 1M or 1.2M tops, complete the drive, and then perhaps start a 7th drive from that point on. (Hopefully increased computer speeds in a couple of years will make that a better option.)

Here's what I'm seeing with the 1.4 or 1.5M limit, regardless of whether we have 12 or 16 k's: Here we are late in the year 2009 kind of grinding along. We're up to 1.3M, and each file is taking forever to process and is producing no primes for very large intervals. I can just see the drive stalling out at that point or possibly taking another 3 years just to do the final n=200K. Especially with 16 k's.

I make this suggestion regardless of whether we have 12 k's or more. As you all know, the timings are intense above n=1M. k=5 has driven that point home to me.

Also, Kosmaj suggests k=71 and 95. Even though they don't look promising, they do appear to be very low weight and we never really know where the primes are so I think they are good choices. I think it is a good thing to consider k's other than just those tested to n=600K and k>100. Perhaps in doing that, we could eliminate a few for k>200. The timings will be somewhat less for the lower k's. I'm thinking that it won't be a problem if we aren't testing every k at every n-interval.


Gary[/quote]
Ditto, I agree with both of you guys too. :smile:

You both have a very good point there about the drive stalling out after the testing times get too long--we don't want to have another 3rd Drive on our hands. :no:

So, I'd say 1.2M would probably be good, because even though times get longer past 1M, they're still not that terribly unbearable at first--so 1.2M would mean that we'd be able to take these k's up higher than just plain 1M, but without much of a chance of people getting bored. :yawn:

9 K's
 

If you are planning a 6th drive, has any consideration been given to the k values (all less than 300) listed here:

[url]http://www.mersenneforum.org/showthread.php?t=7064[/url]

Kosmaj has my old test files for these values - IIRC, they have been sieved up to 2^43, or almost 9T and cover the range 0.7 M < n < 1 M....

just a thought...

[quote=masser;118815]If you are planning a 6th drive, has any consideration been given to the k values (all less than 300) listed here:

[URL]http://www.mersenneforum.org/showthread.php?t=7064[/URL]

Kosmaj has my old test files for these values - IIRC, they have been sieved up to 2^43, or almost 9T and cover the range 0.7 M < n < 1 M....

just a thought...[/quote]
It might be kind of interesting to include those in the drive...if they were sorted by n along with all the rest of the k's, then the test files from n=700k to n=1M would simply include those k's (and thus be a little heavier than the respective files outside that range--which, of course, wouldn't include any work from those k's). In that way, those k's could be seamlessly integrated into the drive.

Masser's files are being sieved presently, and will be made available when the sieve finishes for individual reservations of entire k's or as blocks of n's like a team drive. They are mostly rather high-weight k's, which is why they were not converted to a team drive originally.

As for n>1.2M testing: We are forecasting 18 months to complete the drive. Consider how fast CPUs will be in 18 months! My k=5 testing times at 1050k are under 1000 sec/test on a Core2-2840. Even with larger k's and an FFT jump or two, we're talking under 30 min per test on a fast current machine; in 18 months, a quad-3.2 will likely be fairly mainstream, capable of 6 or 7 tests per hour at 1.4M (all 4 cores).

It's not unreasonable to expect double the resources in 18 months- a drive to 1.4M won't be THAT long. Finally, if we run a bunch of k's (say 16 or 18) to 1M, we're left with a massive pile of k's at 1M, facing a new drive that should run 1M to 2M (sieving less than a factor of 2 wastes quite a lot of time). We're better off with this sort of bootstrap process we've had- 4th drive to 600k, 5th from 260 to 1M, 6th from 600 to 1.4M, 8thish from 1M to ??, etc. The end of a previous drive is near the midpoint of the new drive.

Again, we can declare the team-drive portion complete at 1.2M, allowing individuals to reserve k's for the 1.2-1.4/1.5 range at that point. With well-sieved candidates, this may appeal to some people come early 2009.
-Curtis

[quote=VBCurtis;118841]Masser's files are being sieved presently, and will be made available when the sieve finishes for individual reservations of entire k's or as blocks of n's like a team drive. They are mostly rather high-weight k's, which is why they were not converted to a team drive originally.

As for n>1.2M testing: We are forecasting 18 months to complete the drive. Consider how fast CPUs will be in 18 months! My k=5 testing times at 1050k are under 1000 sec/test on a Core2-2840. Even with larger k's and an FFT jump or two, we're talking under 30 min per test on a fast current machine; in 18 months, a quad-3.2 will likely be fairly mainstream, capable of 6 or 7 tests per hour at 1.4M (all 4 cores).

It's not unreasonable to expect double the resources in 18 months- a drive to 1.4M won't be THAT long. Finally, if we run a bunch of k's (say 16 or 18) to 1M, we're left with a massive pile of k's at 1M, facing a new drive that should run 1M to 2M (sieving less than a factor of 2 wastes quite a lot of time). We're better off with this sort of bootstrap process we've had- 4th drive to 600k, 5th from 260 to 1M, 6th from 600 to 1.4M, 8thish from 1M to ??, etc. The end of a previous drive is near the midpoint of the new drive.

Again, we can declare the team-drive portion complete at 1.2M, allowing individuals to reserve k's for the 1.2-1.4/1.5 range at that point. With well-sieved candidates, this may appeal to some people come early 2009.
-Curtis[/quote]
Hmm...that sounds like it might be a good idea. We can, say, sieve the k's to 1.4 or 1.5M, and set a goal of at least 1.2M for the drive. If interest is still high, go higher with the drive; if not, then open the k's up for individual reservation. I'm sure there's some people out there who'd be glad to be able to reserve a k that's already pre-sieved.

Curtis has all the math down pat as usual and I can't argue with it. Sieving to n=1.4-1.5M, team drive to n=1.2M, and having people reserve individual k's at 1.2M to take them up to n=1.4-1.5M sounds like a good idea to me.

If I may ask, can we include k=139 in the drive? And no, it's not 'officially' reserved by Nuggetprime, even though it shows as such. See Kosmaj's comment in another thread about him stating that he didn't have the resources to continue it in the near future and that one of us could reserve it if we wish.

I took it over and am planning on taking it to n=800K but if we include it in the team drive here, I'll stop at n=600K and we can put the rest in the team drive starting there like most of the other k's. I just started sieving n=400K-800K late this afternoon but will delete the 600K-800K candidates if people think it can be included in the team drive.

Kosmaj, is that OK, or do you think Nuggetprime would want to pick it up after I am done with it?


Gary

I ran almost all of the k's mentioned by Anon, except the ones already deemed excluded from consideration. Here's the results:
[CODE]sieves are from 600k to 1.5M, to 5B.
k # tests primes
115 51961 30
133 37707 24
145 28443 17
149 26003 13
169 45260 25
185 54820 32
187 28798 20
203 43969 23
205 37196 32
217 20936 17
227 65017 31
235 40948 23
241 42375 22
257 50048 34
259 40248 31
265 51003 33
271 39305 20
275 41706 27
287 33136 24
293 37097 19
295 30852 27[/CODE]

Proposed k's to select: 133,145,169,187,203,205,217,235,259,275,287,295
I omitted the k's with more than 50k candidates as too heavy, and those where # of candidates was near 2000x the number of primes as "bad" k's.

These 12 k's to 1.5M are my preference for the 6th drive. If there are no objections, I'll build an sr2 file for these this week. Once I get to 1T, I'll be willing to email it to whomever wishes to donate sieving. I'm fine with reducing it to 1.4M, but it makes more sense to me to reduce the number of k's to 12 rather than range to 1.4M. If we go with 1.4M, what k would you add? One or two? None?
-Curtis

I neglected a bit of forecasting work in the previous post:
The 12 suggested k's add up to 429,000 tests before sieving. I estimate after sieving, there will be 300-305k candidates left to test. Divide by the n-range of 900,000, and we have a 1k file containing an average of 340 tests. Compare to the 5th drive's average around 740. Each test in this drive may take twice as long as a 5th drive test, but there will be half as many tests per range.
If I were to pick a 13th k, I would choose 257, as it has the lowest weight of the remaining k's and also the highest number of known primes of the k's I tested.
-Curtis

[quote=gd_barnes;118855]Curtis has all the math down pat as usual and I can't argue with it. Sieving to n=1.4-1.5M, team drive to n=1.2M, and having people reserve individual k's at 1.2M to take them up to n=1.4-1.5M sounds like a good idea to me.

If I may ask, can we include k=139 in the drive? And no, it's not 'officially' reserved by Nuggetprime, even though it shows as such. See Kosmaj's comment in another thread about him stating that he didn't have the resources to continue it in the near future and that one of us could reserve it if we wish.

Kosmaj, is that OK, or do you think Nuggetprime would want to pick it up after I am done with it?


Gary[/quote]


Can I assume by the lack of response to my question here that we are not going to include k=139 starting at n=600K in the drive? I'm going to test it to n=600K myself and if not included in the drive, I'll go to n=800K.

k=139 has a weight of 2586 and k=257 has a weight of 1867 so not a huge difference.

k=139 is now shown as reserved by me.


G

k=139
 

I don't know about the drive. Regarding the limit to test I'm enclosing the direct mail I got from Nugget in June, and you decide by yourself (two unrelated parts removed). He hasn't mailed me residues he is referring to.

[QUOTE=nuggetprime]Hello Kosmaj!
I am sorry to tell you that I haven't enough computing power to participate to RPS at the moment but this will be much better at the end of the year because then I'll get a core2duo with 2.13 ghz then. So I want to "freeze" my current RPS reservations and continue them next year.... My RPS job status:

About k=139: tested to about 419500, i'll mail you residues.
...
If you think that "freezing" RPS jobs is unfair for the other researchers, then you should inform me and cancel my RPS reservations so others could test them.

Regards,
nuggetprime[/QUOTE]