mersenneforum.org  

Go Back   mersenneforum.org > Prime Search Projects > Riesel Prime Search

Reply
 
Thread Tools
Old 2007-10-20, 08:29   #34
BlisteringSheep
 
BlisteringSheep's Avatar
 
Oct 2006
On a Suzuki Boulevard C90

3668 Posts
Default

Quote:
Originally Posted by VBCurtis View Post
Sheep- I plan to be at 75T by 15 Nov. Start at 75T, and any work you get done will be a bonus. I know sr1 isn't the easiest for you to run, so perhaps just run on a small handful of cores? I'll send the appropriate sieve file for sr1 presently.

After 15 Nov, I'll sieve at about 6T/week until catching the big sieve at 150T.
-Curtis
I should be able to get something started today. Initial testing with sr1sieve-1.1.12 is reporting about 4.3-4.4 Mp/s per core. I'll be able to start with 12 cores, but should increase that number to around 60 cores before the weekend is over. Will that help any? (I'd start them now, but it's 0430 and past my bedtime )

Last fiddled with by BlisteringSheep on 2007-10-20 at 08:31
BlisteringSheep is offline   Reply With Quote
Old 2007-10-20, 08:34   #35
em99010pepe
 
em99010pepe's Avatar
 
Sep 2004

2×5×283 Posts
Default

BlisteringSheep,

Try the latest version of sr1sieve (1.2.0).

Carlos

Last fiddled with by em99010pepe on 2007-10-20 at 08:35
em99010pepe is offline   Reply With Quote
Old 2007-10-20, 20:55   #36
BlisteringSheep
 
BlisteringSheep's Avatar
 
Oct 2006
On a Suzuki Boulevard C90

2·3·41 Posts
Default

Quote:
Originally Posted by em99010pepe View Post
BlisteringSheep,

Try the latest version of sr1sieve (1.2.0).

Carlos
I didn't use it originally because it was notated as "experimental" on Geoff's page. I have tried it now; there's at most a 0.4% increase in performance.

Also, I have started the first 12 cores now. I'm going to have to write some support scripts to generate work files before I can do the larger roll-out, though. I'm open to suggestions, but my initial plan is to give each core their own copy of the dat file with a localized start pmin value, using the same dat file for input and output so the pmin value gets updated, and pass the pmax value via sr1sieve-command-line.txt.
BlisteringSheep is offline   Reply With Quote
Old 2007-10-20, 22:07   #37
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

76538 Posts
Default

Oh, boy.. Sheep is here to save the day!
Since your farm is back up earlier than expected, it's prudent to start at 60T instead of 75T. Gary and I will achieve 60T roughly 1 Nov. With 60 cores, you'll reach 1T/hr throughput, finishing 60-120T in roughly 3 days. Yeesh.

My Athlon needs something to do, so once you reach 100 or 120T, it's better to return to sr2 and the big sieve. I'll fill in the gap this fall to 150T while you reach unprecedented (for RPS) depths for n>1.2M.
-Curtis

edit: Sheep, I'd just give each core 1T to do. When it's done, move it on to something else. Alternately, just use 12 cores with wider ranges- sr1 just doesn't lend itself as well to a wide rollout like you have. 4T each on 12 cores is ~50T and 12-14 days.

Last fiddled with by VBCurtis on 2007-10-20 at 22:09
VBCurtis is offline   Reply With Quote
Old 2007-10-21, 05:20   #38
gd_barnes
 
gd_barnes's Avatar
 
May 2007
Kansas; USA

3×7×13×37 Posts
Default Off-the-wall thought...

Quote:
Originally Posted by VBCurtis View Post
Oh, boy.. Sheep is here to save the day!
Since your farm is back up earlier than expected, it's prudent to start at 60T instead of 75T. Gary and I will achieve 60T roughly 1 Nov. With 60 cores, you'll reach 1T/hr throughput, finishing 60-120T in roughly 3 days. Yeesh.

My Athlon needs something to do, so once you reach 100 or 120T, it's better to return to sr2 and the big sieve. I'll fill in the gap this fall to 150T while you reach unprecedented (for RPS) depths for n>1.2M.
-Curtis

edit: Sheep, I'd just give each core 1T to do. When it's done, move it on to something else. Alternately, just use 12 cores with wider ranges- sr1 just doesn't lend itself as well to a wide rollout like you have. 4T each on 12 cores is ~50T and 12-14 days.

Holy cow. Both Sheep and Curtis, IMHO dedicating any more than 12 cores to one k for sieving is overkill. If you look back at previous team drives, I don't recall seeing one where we sieved so deeply for n<1M. So Sheep, can none of your machines do LLRing? Even if it's half the speed or less than the usual for the rest of us, having 48 cores doing LLRing...whew! Regardless, Kosmaj or Curtis, I'd suggest seeing what other 3-4 k's you think is better for him to work on that he could put about 12 cores each on.

Here's an off-the-wall thought...If Sheep can't LLR, if we want to do this optimally, we should see how many people we can get to drop their reservations for k < 300 and have Sheep sieve the entire ~145 k's at once. (Knocking out the few that are being done by other major projects.) That would be the ULTIMATE team drive! Then Kosmaj, Lousle, Curtis or whomever could start posting ranges for people to LLR, perhaps just 200-500 n at a time, which would contain plenty of work at any level of n for 145 k's! The thinking being that we're all reserving ranges of n instead of specific k's.

The 5th drive now is pretty cool with the 14 k's divisible by 3. Imagine expanding it to ~145 k's! As crazy as it sounds, in theory it's the most efficient way to do things, kind of like they do for RieselSieve and Seventeen-or-Bust. Sure, there would be logistical issues like starting the sieves and then removing ranges already searched but that would only take a small amount of time compared to the long-term benefit.

I know getting around the political issue of reservations would be a problem but even if we could have him sieving 50 k's at once would be outstanding!

With Carlos having several new cores and with newer folks on board like Anon and Patrick, I think we've got a good motivated group to do a new large team drive like this.

(At least my wild ideas aren't outside the scope of the project this time!)


Gary
gd_barnes is offline   Reply With Quote
Old 2007-10-21, 06:19   #39
BlisteringSheep
 
BlisteringSheep's Avatar
 
Oct 2006
On a Suzuki Boulevard C90

2×3×41 Posts
Default

Quote:
Originally Posted by gd_barnes View Post
So Sheep, can none of your machines do LLRing?

Gary
The only point that I'm really qualified to answer : No, they can't LLR at all; there is no client available for them.

I have two concerns about your suggestion that I take the entire range to sieve:
1. I think that some people genuinely prefer to sieve.
2. I can't guarantee machine time, especially right now. Most of my machines are currently down, and they've been heavily utilized recently. I'm anxious about committing to something I'm not sure I can fulfill.

That said, I do want to help out however and whenever I can.

Sheep
BlisteringSheep is offline   Reply With Quote
Old 2007-10-21, 07:33   #40
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

1111101010112 Posts
Default

Quote:
Originally Posted by gd_barnes View Post
Holy cow. Both Sheep and Curtis, IMHO dedicating any more than 12 cores to one k for sieving is overkill. If you look back at previous team drives, I don't recall seeing one where we sieved so deeply for n<1M.
Gary
There is no other k sieved so deeply below 1M, because no other k had a single sieve done to 4M. Efficiency increases greatly with a larger n-range; combined with empirical evidence that removing the small n's provide very little speed benefit to the overall sieve, it makes sense to sieve extremely deeply for large sieves. Sheep's farm(s) have sieved k=15, my group of k's (11,13,31,45,99,127, and now 5), rieselsieve, and some other stuff I'm forgetting. PowerPCs have no LLR client, alas.

The 14 k's drive has quite a lot of life left for LLR before we need a new team drive; quite a few users prefer to do everything themselves (me included). I think building a new team drive with a dozen or so k's is a great idea, using Sheep to sieve; however, moving 100+ k's to a group effort puts global efficiency above individualism. A possible compromise is to build such a global sieve, but continue to allow users to reserve entire k's out of the sieve, rather than team LLR everything. Those who like to sieve (or have athlons or laptops that aren't well suited to LLR) can reserve ranges from this massive sieve. If we ever do this, we should still leave some k's untouched for those interested to process entirely on their own. Perhaps put 30 k's of various weights into a mass sieve... eventually.

Another way to look at this is max efficiency comes from the largest n-range, but we do not usually sieve very-large n-ranges. Why not? The same long-run "greater good" is achieved by sieving a range twice as high as sieving twice as many k-values. One chooses limits on both for practical, project-completion reasons. Only for very large n-ranges does one not need to finish sieving for max efficiency before starting LLR; if sieving 100 k's, that sieving step would be many CPU-years. And what about file size? The sieve needs to be practically emailed, which puts a 10MB or 20MB limit.

I tried 33 k's from 260k to 300k, in the 900-1000 range (see primesearch), as an experiment. It was NOT efficient. Many k's with a narrow n range is bad bad bad for sr1/sr2. Big fat n ranges are incredibly efficient. Thus, my decision to run this k=5 sieve to 4M, and the resulting ~100T sieve depth for n<1M.
-Curtis
VBCurtis is offline   Reply With Quote
Old 2007-10-21, 09:52   #41
em99010pepe
 
em99010pepe's Avatar
 
Sep 2004

1011000011102 Posts
Default

You guys are forgetting amphoria, he has a bunch of machines.

Can I have my K=5 files by the end of this week (25-26 Oct)?

Last fiddled with by em99010pepe on 2007-10-21 at 09:52
em99010pepe is offline   Reply With Quote
Old 2007-10-21, 16:45   #42
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

3·7·191 Posts
Default

Yes- sr1 1.2.0 has given us enough speed boost that I can get you the files Friday.
-Curtis
VBCurtis is offline   Reply With Quote
Old 2007-10-22, 02:09   #43
gd_barnes
 
gd_barnes's Avatar
 
May 2007
Kansas; USA

3·7·13·37 Posts
Default Great thoughts Curtis!

Quote:
Originally Posted by VBCurtis View Post
There is no other k sieved so deeply below 1M, because no other k had a single sieve done to 4M. Efficiency increases greatly with a larger n-range; combined with empirical evidence that removing the small n's provide very little speed benefit to the overall sieve, it makes sense to sieve extremely deeply for large sieves. Sheep's farm(s) have sieved k=15, my group of k's (11,13,31,45,99,127, and now 5), rieselsieve, and some other stuff I'm forgetting. PowerPCs have no LLR client, alas.

The 14 k's drive has quite a lot of life left for LLR before we need a new team drive; quite a few users prefer to do everything themselves (me included). I think building a new team drive with a dozen or so k's is a great idea, using Sheep to sieve; however, moving 100+ k's to a group effort puts global efficiency above individualism. A possible compromise is to build such a global sieve, but continue to allow users to reserve entire k's out of the sieve, rather than team LLR everything. Those who like to sieve (or have athlons or laptops that aren't well suited to LLR) can reserve ranges from this massive sieve. If we ever do this, we should still leave some k's untouched for those interested to process entirely on their own. Perhaps put 30 k's of various weights into a mass sieve... eventually.

Another way to look at this is max efficiency comes from the largest n-range, but we do not usually sieve very-large n-ranges. Why not? The same long-run "greater good" is achieved by sieving a range twice as high as sieving twice as many k-values. One chooses limits on both for practical, project-completion reasons. Only for very large n-ranges does one not need to finish sieving for max efficiency before starting LLR; if sieving 100 k's, that sieving step would be many CPU-years. And what about file size? The sieve needs to be practically emailed, which puts a 10MB or 20MB limit.

I tried 33 k's from 260k to 300k, in the 900-1000 range (see primesearch), as an experiment. It was NOT efficient. Many k's with a narrow n range is bad bad bad for sr1/sr2. Big fat n ranges are incredibly efficient. Thus, my decision to run this k=5 sieve to 4M, and the resulting ~100T sieve depth for n<1M.
-Curtis
A lot of great points, Curtis. I agree somewhere closer to along the lines of what you suggested more than my own about the 'huge' sieve or 'team drive'...put 30 k's into some big mass sieve and sieve all of them up to about n=3M-4M or so. It could be called a 'drive' or whatever we want. The main point that is good about what you mentioned is to have some massive sieved files out there for a whole bunch of different k's and just let people reserve the k's and grab the ranges of the sieved files that they want. That would be cool! And if people only wanted to (or only could) sieve instead, they could just sieve on this massive file if they wanted to or add additional ranges of n at the high-end of the sieve file or something of that nature. Or if they wanted to do a k all by themselves, then leaving a number of them untouched and not sieved at all would fill the bill there.

Obviously the more k's sieved at once, the more efficient for the 'long-term good'. There's certainly nothing wrong with sieving 100's of k's at once but you're right, removing individualism would remove some of the fun, myself included. As I mentioned, you'd definitely want to sieve up to n=2M-3M. As you said, sieving a small range of n is highly inefficent and not worth doing.

Back on sieving k=5...I'm not debating your mathematics about when to split off the ranges to LLR. But I'm not so sure I agree with continuing to sieve so long on an effort like this for k=5 even though we've included the huge range of n=470K-4M in the sieve. The point here is about economics, that is good old supply and demand. We have a really big supply of LLRers ready to go.

My opinion...release every range that everyone has already reserved by 10/31. (Obviously earlier for Carlos.) (Also, my 630-660 will be done Wednesday and I could start on 740-760 Thursday if you release it.) It doesn't matter whether it's sieved to 58M or 60M or 120M or whatever, I'd still say just release the reserved ranges and keep sieving the rest. As people are ready for their next range, just send it to them regardless of how far sieved up to at least n=1M at this point. 60M is very sufficient for this effort to n=1M. That gives us the greatest chance to have everything LLR'd to n=1M by the end of 2007.

My reasoning has to do with 'elastic supply' in this case. That is, if you release things as fast as people are ready for them, other people may come into the picture. For instance, Kosmaj, Karsten, Lsoule, Amphoria, etc. may pick up a couple of ranges to LLR also. People love this fast LLRing low k.


Gary

Last fiddled with by gd_barnes on 2007-10-22 at 02:10
gd_barnes is offline   Reply With Quote
Old 2007-10-22, 05:45   #44
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

401110 Posts
Default

Gary- send me whatever you have sieved on Wednesday night. I'll remove the factors and release up to 780k then, then whatever range necessary (900k?) to satisfy demand on 11/1. We're now in a range where sieving is in the ballpark of enough, so I'm fine with your suggestion. When we were at 20T and "proper" depth was 50T+, I wasn't so keen on that when a week's delay meant 50% more sieving.

At this pace, 900k-1M will be released on 11/15.

Mass sieve: We already have masser's files, 8 k's from 700k-1M. I plan to perhaps extend these to 2M and sieve them a while. Cruelty has 12 k's to 2M also, and he plans to LLR each until he finds a prime, then release the file for public consumption. So perhaps such a plan as we seem to want already exists-- we should encourage people to submit untested sieves when they release numbers.
-Curtis
VBCurtis is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
15*2^n-1, n>1M Reservation Thread Kosmaj Riesel Prime Search 702 2018-04-11 07:36
LLR reservation thread (n=480K-500K) Oddball Twin Prime Search 33 2012-01-20 05:37
Octoproth Reservation Thread Greenbank Octoproth Search 2 2007-12-26 09:58
Dodecaproth Reservation Thread Greenbank Octoproth Search 30 2006-02-09 00:33
Hexadecaproth Reservation Thread (n=76) Greenbank Octoproth Search 0 2006-01-25 13:41

All times are UTC. The time now is 16:25.

Wed Apr 8 16:25:12 UTC 2020 up 14 days, 13:58, 3 users, load averages: 1.67, 1.75, 1.84

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

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.