Range for k>300 that Peter Benson has searched

gd_barnes · 2009-01-10, 12:09

I PM'd Peter Benson a few days ago about the ranges for k>300 that he has searched. Here is a snippet of the response that I got from him yesterday:

Quote:

Just a note on your new search for k=1005-2000. I have already searched k=1003-1399 from n=300,000-479,470 & I am currently searching up to n=600,000. (I have completed some ks up to 600,000.)

I have followed up asking if he could provide results file(s) for the range. If so, we may change our k=1005-2000 drive to k=1400-2000 because he will clearly get to n=500K before we will. We'll just use the sieve file for that range to include in some future double-check effort. If not, we should probably include it in our current drive.

Opinions anyone?

I didn't ask about the ranges that he has searched for k=300-1001 and n>600K where he has 3 primes from 2006 because I wouldn't expect him to still have results on them. We'll just continue our 5th/6th/7th drives as we are...very little double-checking there.

Gary

Flatlander · 2009-01-10, 16:25

I see little point in testing k=1003-1399 even if P.B. can't produce the lresults. Why not just leave it for a future double check?

gd_barnes · 2009-01-11, 06:10

I'm beginning to feel like Chris is here. It goes against my "grain", so to speak, to not include efforts outside of NPLB in our ranges, but we obviously have already done quite a bit of double-checking on our 3rd drive and some of our 1st/2nd drives, not to mention our double-check drive, and this is a whole lot of work.

I'm becoming inclined to accept Benson's searched range as long as he will let us know when he is complete to n=500K on all of his k's, regardless of whether he can provide results files or not.

Peter, if you happen to see this, can you let us know if you have results files and when you are complete to n=500K for k=1003-1400?

Assuming that we accept his range as complete, here is what we have:
1. The 8th drive for k=1400-2000/n=350K-500K.
2. The 9th drive phase 1 for k=1005-2000/n=50K-200K. (no change)
3. The 9th drive phase 2 for k=1005-2000/n=200K-300K.
4. The 9th drive phase 3 for k=1400-2000/n=300K-350K.
5. All of k=1003-2000 for n=50K-500K will be in future double-check efforts.

This effectively has us doing:
1. k=1005-1400 for n=50K-300K.
2. k=1400-2000 for n=50K-500K.

It has us removing Peter's range of:
k=1005-1400 for n=300K-500K.

This will still do some double-checking in the top-5000 range (~20% now I think). I feel more confident in the accuracy of Peter's vs. other's ranges. That's because it is a contiguous k and n-range and is not sporadic k's like other's efforts in the area.

Any more feedback on this? I'll stick this thread temporarily since it is important to come to a conclusion before our sieving drive is done. Please don't think that part of our sieving has been wasted. It will definitely be used for future double-checking.

Gary

gd_barnes · 2009-01-12, 23:46

I got a response from Peter that his results files are spread across many machines and many of them may not be available.

With only one person responding here saying that we shouldn't search Benson's k=1000-1400 range, perhaps we should search all k=1005-2000 and do what now amounts to ~50% double-check on the 8th drive.

...gotta get some consensus folks.

Gary

mdettweiler · 2009-01-13, 00:52

Quote:

Originally Posted by gd_barnes

I got a response from Peter that his results files are spread across many machines and many of them may not be available.

With only one person responding here saying that we shouldn't search Benson's k=1000-1400 range, perhaps we should search all k=1005-2000 and do what now amounts to ~50% double-check on the 8th drive.

...gotta get some consensus folks.

Gary

I'm inclined to think that given this, we *should* search the whole k=1005-2000 range--it shouldn't be that terribly much more work, considering how fast we've been going on the 50K-200K range. Yes, these n-ranges are higher, but I'm thinking that it shouldn't be too bad. Making sure that we have residuals for everything is, IMO, a worthwhile trade-off for the extra work.

MyDogBuster · 2009-01-13, 03:34

I would tend to vote to do it all, but I have serious reservations that people won't participate because of the ~50% double check, thus leaving the bulk of the work to the same people to clean up.

gd_barnes · 2009-01-13, 03:45

Well, believe-it-or-not, I will abstain from having a set-in-stone opinion. For once, I do NOT have an opinion on something. :-)

I've already "waffled" on the topic with my more recent post meant more to draw people out to express their opinions.

I suppose we could split it up:

Do k=1005-1400 in one drive.
Do k=1400-2000 in another.

But that leaves one drive with potentially 100% double-check work and the other with about 20-30%. Will anyone work on k=1005-1400 in the foreseeable future?

One thing that I'll mention: In order to make quicker ground in the top-5000 "wars" with RPS and PrimeGrid, we'd have to stick with only doing k=1400-2000 and then do k=1005-4000 later. The problem with that is, will k=1000-1400 ever get done?

As you can see, I go both ways with it.

Any more opinions? We're seeing 2-1 in favor of doing the whole range right now.

Gary

IronBits · 2009-01-13, 05:10

The way I see it...
If there is any doubt, run it again.
There is no other option, all doubt must be removed, no matter how painful it might be or perceived.
How to go about this... one person could re-run it, or, put it on a server so everyone can toss at least one core on it...
Question, how long would it take 1 3GHz quad (all cores) to complete this?
I can do the math from there ;)

Also, I assume there was some creditable and salvageable work that will be accounted for and removed...
This would leave only the parts there is doubt about, so that we are only doing those sections we have doubt...

Go Team Go!

gd_barnes · 2009-01-13, 05:29

Quote:

Originally Posted by IronBits

The way I see it...
If there is any doubt, run it again.
There is no other option, all doubt must be removed, no matter how painful it might be or perceived.
How to go about this... one person could re-run it, or, put it on a server so everyone can toss at least one core on it...
Question, how long would it take 1 3GHz quad (all cores) to complete this?
I can do the math from there ;)

Also, I assume there was some creditable and salvageable work that will be accounted for and removed...
This would leave only the parts there is doubt about, so that we are only doing those sections we have doubt...

Go Team Go!

I'll get back with you on that. I'll have to run a few tests on my Intel quads to know for sure. I can say that it is several CPU years of work just for the k=1000-1400 range that Peter Benson did; although nothing near as much as our 5th-6th-7th drives will be.

It's 3-1 in favor of doing the entire range at once; Do I here some more votes?

Gary

gd_barnes · 2009-01-13, 09:18

Quote:

Originally Posted by MyDogBuster

I would tend to vote to do it all, but I have serious reservations that people won't participate because of the ~50% double check, thus leaving the bulk of the work to the same people to clean up.

Ian,

I reread this and realized I may have missed the intent. Do you find yourself undecided on the issue? If so, I'll put you in my camp...not voting at all one way or another.

Edit: I may take a poll on this because it's clear we're going to have a closely split decision on it.

Gary

MyDogBuster · 2009-01-13, 09:58

Quote:

Ian,
I reread this and realized I may have missed the intent. Do you find yourself undecided on the issue? If so, I'll put you in my camp...not voting at all one way or another.

I would vote for doing it all. I'm just having a hizzy fit about something else.

2009-01-12, 23:46	#4
gd_barnes May 2007 Kansas; USA 7×13×113 Posts	I got a response from Peter that his results files are spread across many machines and many of them may not be available. With only one person responding here saying that we shouldn't search Benson's k=1000-1400 range, perhaps we should search all k=1005-2000 and do what now amounts to ~50% double-check on the 8th drive. ...gotta get some consensus folks. Gary Last fiddled with by gd_barnes on 2009-01-12 at 23:49

2009-01-13, 05:10	#8
IronBits I ♥ BOINC! Oct 2002 Glendale, AZ. (USA) 1113₁₀ Posts	The way I see it... If there is any doubt, run it again. There is no other option, all doubt must be removed, no matter how painful it might be or perceived. How to go about this... one person could re-run it, or, put it on a server so everyone can toss at least one core on it... Question, how long would it take 1 3GHz quad (all cores) to complete this? I can do the math from there ;) Also, I assume there was some creditable and salvageable work that will be accounted for and removed... This would leave only the parts there is doubt about, so that we are only doing those sections we have doubt... Go Team Go! Last fiddled with by IronBits on 2009-01-13 at 05:14

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Peter Montgomery's Thesis	mickfrancis	Computer Science & Computational Number Theory	3	2015-06-25 14:32
Peter Montgomery (IMPORTANT)	R.D. Silverman	Factoring	8	2014-06-07 18:43
Types of primes searched for.	3.14159	Miscellaneous Math	2	2010-12-04 13:09
Benson's prime search	Kosmaj	Riesel Prime Search	67	2009-01-18 21:59
k's/n-ranges not searched for team drives	gd_barnes	No Prime Left Behind	20	2008-12-26 08:13

2009-01-10, 16:25	#2
Flatlander I quite division it "Chris" Feb 2005 England 31·67 Posts	I see little point in testing k=1003-1399 even if P.B. can't produce the lresults. Why not just leave it for a future double check?

2009-01-11, 06:10	#3
gd_barnes May 2007 Kansas; USA 7·13·113 Posts	I'm beginning to feel like Chris is here. It goes against my "grain", so to speak, to not include efforts outside of NPLB in our ranges, but we obviously have already done quite a bit of double-checking on our 3rd drive and some of our 1st/2nd drives, not to mention our double-check drive, and this is a whole lot of work. I'm becoming inclined to accept Benson's searched range as long as he will let us know when he is complete to n=500K on all of his k's, regardless of whether he can provide results files or not. Peter, if you happen to see this, can you let us know if you have results files and when you are complete to n=500K for k=1003-1400? Assuming that we accept his range as complete, here is what we have: 1. The 8th drive for k=1400-2000/n=350K-500K. 2. The 9th drive phase 1 for k=1005-2000/n=50K-200K. (no change) 3. The 9th drive phase 2 for k=1005-2000/n=200K-300K. 4. The 9th drive phase 3 for k=1400-2000/n=300K-350K. 5. All of k=1003-2000 for n=50K-500K will be in future double-check efforts. This effectively has us doing: 1. k=1005-1400 for n=50K-300K. 2. k=1400-2000 for n=50K-500K. It has us removing Peter's range of: k=1005-1400 for n=300K-500K. This will still do some double-checking in the top-5000 range (~20% now I think). I feel more confident in the accuracy of Peter's vs. other's ranges. That's because it is a contiguous k and n-range and is not sporadic k's like other's efforts in the area. Any more feedback on this? I'll stick this thread temporarily since it is important to come to a conclusion before our sieving drive is done. Please don't think that part of our sieving has been wasted. It will definitely be used for future double-checking. Gary

2009-01-13, 03:34	#6
MyDogBuster May 2008 Wilmington, DE 2²·23·31 Posts	I would tend to vote to do it all, but I have serious reservations that people won't participate because of the ~50% double check, thus leaving the bulk of the work to the same people to clean up.

2009-01-13, 03:45	#7
gd_barnes May 2007 Kansas; USA 7×13×113 Posts	Well, believe-it-or-not, I will abstain from having a set-in-stone opinion. For once, I do NOT have an opinion on something. :-) I've already "waffled" on the topic with my more recent post meant more to draw people out to express their opinions. I suppose we could split it up: Do k=1005-1400 in one drive. Do k=1400-2000 in another. But that leaves one drive with potentially 100% double-check work and the other with about 20-30%. Will anyone work on k=1005-1400 in the foreseeable future? One thing that I'll mention: In order to make quicker ground in the top-5000 "wars" with RPS and PrimeGrid, we'd have to stick with only doing k=1400-2000 and then do k=1005-4000 later. The problem with that is, will k=1000-1400 ever get done? As you can see, I go both ways with it. Any more opinions? We're seeing 2-1 in favor of doing the whole range right now. Gary