Reply With QuoteYes, very cool.
I'm volunteering as betatester and/or pre-launch siever (to reduce download size) if there is any need.
Sounds like a great new option.
Yes, very cool.
I'm volunteering as betatester and/or pre-launch siever (to reduce download size) if there is any need.
Excellent news Louie.
One question, I have been sieving k=4847 for n=3m to 20m for a couple days. Will this submission page take n values that high (if I read correctly, it will) and will it be able to handle a 1.5MB .del file produced by newpgen (or whatever equivalent form you script will accept it)?
Before I forget: Did all 12 k's from before up to 25M. If anybody is interested it's here (zipped, 3.05MB).
That's great news from Louie about the new sieve! My only reservation is that a sieving range of n from 3 million to 5 million seems a little conservative. We ran through all the exponents from 1 million to 2 million in about a month. At that rate, allowing for the fact that testing an exponent takes time approximately proportional to n^2, testing the range from 3 million to 5 million should take just over a year. However, since we also are hoping to attract new participants to the project, we have to assume that it may take less than that. Consider that the time it takes to sieve is proportional to the square-root of the sieving range. Therefore increasing the range by half to 3 million to 6 million only requires 22% more sieving time, but will generate twice as much work for the project. (The square-root of 3/2 is about 1.22.) Increasing the range to 3 million to 10 million would require 87% more sieving time, (since the square-root of 7/2 is about 1.87) but generate 10 times the work for the project. Of course we also can make future sieves more efficient by eliminating more k-values, but even eliminating 2 or 3 more k-values won't buy us the efficiency that increasing the sieving range does.
This is really good news.
With a good semi-automated sieving process we might finally start sieving to a good depth. While sieving is pulling out numbers at a rate which is greater than that which we would expect the SB client to process numbers in the same n range, then all this shows is that we're not sieving deep enough. The optimal point should be where sieve and expected SB rates are equal.
This means if you're running the sieve on a P4-1.7, and you're sieving for n around 3,000,000, if you're eliminating more than one number per day, then you're not sieving deep enough.
And from some of the posts I've seen on this thread, it would appear that we're not sieving deep enough.
Or maybe I'm missing something……
True, but don't forget that we are looking for primes, not for composites!! So testing can start as soon as a reasonable sieving depth is reached. You's never know that N=3000123 might produce a prime, and it would be a wast to sieve a whole range for weeks.Originally posted by MikeH
The optimal point should be where sieve and expected SB rates are equal.
No, not deep enough, but it takes a lot of cpu time to sieve deep enough. K=4847 has had a couple of 1Ghz weeks of sieving. But coordinating this by hand is really hard.Originally posted by MikeH
And from some of the posts I've seen on this thread, it would appear that we're not sieving deep enough.
I agree with Phil Moore, that 3-5M is a bit too small range for sieving. 7 to 10M seems more reasonable. We don't have to sieve the whole range extremely far before starting to test 3M numbers. The sieving can continue in the mean time.
Louie, Will you provide a way to reserve a range to sieve, or at least make it available on the site which ranges are sieved, to avoid duplication of work?
I also agree, opening the ranges a bit wider than 3 - 5M seems like a good idea.
I've been following this thread for a while, and since I have about 5 additional PCs which are on 24/7 but not able to get net access, I have thought about joining the sieving effort, but this "paper" co-ordination hasn't looked too good to me. So once Louie gets the web form and web co-ordination sorted, those PCs will be on the case.
I'm hopeful that there are other members of the project in similar cimcumstances. For sure we will then achieve deeper seiving, which can only be a good thing.
(But can we please make sure that a reserved range doesn't expire in 5 days!)
I'm sure it won't.Originally posted by MikeH
(But can we please make sure that a reserved range doesn't expire in 5 days!)
I guess you're right. The new sieving function might also attract new people who knew the project but didn't participate at full throttle (if at all) because of the net connection problem.
I think that 3 - 20 million makes more sense than 3 - 5 million since there is a small hit in speed by extending the range. If there's a chance we will test a range, we should start sieving it now basically. I'll start building the official sieve file to this size.
For now, the forum will be the coordination center to reserve ranges. There will not really be an expiration in this system.
Hopefully there can be a reservation system on the website and stats on which ranges have factors submitted already (and how many) but that won't be available right away.
I'd also like to point out that there may come a time where double checks of proth tests may be important. In other words, you shouldn't simply work to equal the rate of sieving to the rate of SB. A factor for a number is far more valuable than a proth test for the same number since it can be verified so easily. So I would say that it's more correct to sieve until it's at least 2x slower than just running SB. GIMPS uses the same methodology for the same reasons.
That combined with the fact that the new siever is faster than before means that sieving will probably never be done to an optimal level no matter how much CPU is thrown at it. THATS OK though. As smh reminds us above, we are looking for primes, not just the composites. A good % of SBs power should still go to the process that will actually find the primes since sieving a k deeper makes little sense once there's a prime. Also, you have to realize that sub-optimal sieve levels are only slightly sub-optimal. They don't rob the project of a more than a few % of it's total thoughput.
I am making the sieving public mainly for 2 reasons:
1) the project will no longer be dependant on my access to the U of M super computing clusters to sieve work and is not dependant on me manually creating new workunits.
2) by storing all the factors in the database, it will be an excellent consistency check to insure that we didn't somehow miss any numbers. by having the system more tightly coupled, it avoids possible human error on my part.
Minor reasons for making the sieve public include:
1) More processing power being put into sieving making sieve levels closer to optimal.
2) there is the possiblity that people will be able to factor the numbers in inventive new ways to seriously reduce the test space
3) I can spend my time I spend manually removing factors people send me doing something else. It's not a big deal every now and than, but when I get a few a day, it's kinda tedious and it's definately unnessisary.
OK. I'm writing a book on this. I'm done. Hope that helps answer some questions on how this will all work.
Summary: get ready to do lots of sieving.
-Louie
Since I lost my DSL connection, actually participating in the project with my machines is not very feasible for me.
I am another who would be able to contribute to the project by sieving who would not be able to contribute otherwise.
I would imagine that by time all is said and done, the resources involved with sieving will not end up much from the search for primes. Between machines that could not previously be used being put to work sieving and increased interest in the project as a result of the public sieving effort, we should break even or even gain computers in searching for primes. Only IMHO, of course.
1500-1600G done, 5 numbers removed.
Starting on 1340-1400G.
I just saw you sent me the removed numbers.
Could you send them directely to Louie? There's no need for me anymore to sit in between, and i'm not able to send the file until this weekend.
ok, sent to Louie.
Just a question -
as I understand, your new siever is able to sieve on *all* sierpinski numbers left to check (12 currently) at the same time, with a decent speedup, compared to sieving each separate.
there is a project (Riesel primes), similar to SB, searching for primes of form k*2^n-1 (sierpinski is +1), where some 113 numbers are left.
would it be possible to change the 'SB multi-k sieve' to check these numers as well (shouldn't be a problem IMO)
this might accelerate the other (riesel) project quite a bit
tom
Just a question -
as I understand, your new siever is able to sieve on *all* sierpinski numbers left to check (12 currently) at the same time, with a decent speedup, compared to sieving each separate.
That is correct.
there is a project (Riesel primes), similar to SB, searching for primes of form k*2^n-1 (sierpinski is +1), where some 113 numbers are left.
would it be possible to change the 'SB multi-k sieve' to check these numers as well (shouldn't be a problem IMO)
this might accelerate the other (riesel) project quite a bit
Indeed, it is very possible - in fact, just last night I created a version of the software to do exactly that, and sent it to Wilfrid Keller and Ray Ballinger. The only changes required - apart from cosmetic issues - were removing one minus sign and removing the characters "p-" in two places.
Regards,
Paul.
> in fact, just last night I created a version of the software to do exactly that, and sent it to Wilfrid Keller and Ray Ballinger
So I was a little late in suggesting it
Thanks
Where can I downloading the software ????
Cu
From Paul's recent postings, it would appear that things have moved forward. We now look forward to a "SoBSieve".Louie wrote
I know you're all excited to get this stuff up and running but it will probably be a few more days yet. I'm excited too but I want to give things a chance to test out and solidify before public factor submission begins. And we still need to write most of the sieve page.
Any update on how things are going, and as to when we meagre mortals (the public) can start sieving?
SoBSieve started.
For those who haven't noticed it yet, find it at Louie's thread here
Regards,
Orkun
Does "SobSieve" support SMP ? The other thread is for the coordination of reservation of p (sieve) ranges.
Last edited by jreyesiii; 01-19-2003 at 02:29 PM.
Not natively, but I imagine you can start 2 copies and set the affinity by hand.
Perhaps it's better that you ask in the other thread.
Posting Permissions
