The future of Msieve - Page 3

Batalov · 2008-10-29, 18:41

Jason,

May I contribute a general suggestion that if your programming time is currently limited, then it may be a perfect time for small robustness tweaks to the code. No new features, but just 99.44% purity at those that it does well. (The last trick you pulled with sqrt was quite in this vein.)

Robust, bullet-proof BL, as the first pain point, maybe? If orthogonality check fails - restart from an earlier checkpoint or stop and wait for master's decision (and all that stems from this - possibly 1. do not overwrite checkpoints, do at least two of them; 2. the .chk files should have passed a QC before writing... etc.)

http://www.mersenneforum.org/showpos...&postcount=406 and same thing happens every now and then to each of us, right? And then everyone gathers around, pats the guy on the back and sardonically welcomes to the club. (I've got trivial deps on my last week-long BL, too. Not the first such case for me, don't welcome me to the club.)

--Serge

jasonp · 2008-10-30, 02:23

Currently I'm focusing on implementing Kleinjung's new polynomial selection method, described in his CADO workshop slides. It's remarkably simple and can potentially produce better polynomials than pol5 can. Unfortunately, making pol51opt optimize polynomials with enormous skew requires major changes in the root sieve, so it's slow going. But this is more fun with the code than I've had in a long time.

I should ask Kazumaro Aoki about the Lanczos checks that his code implements; he's had to deal with everything that could possibly go wrong solving matrices.

2008-10-29, 18:41	#23
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 274C₁₆ Posts	orthogonality checks Jason, May I contribute a general suggestion that if your programming time is currently limited, then it may be a perfect time for small robustness tweaks to the code. No new features, but just 99.44% purity at those that it does well. (The last trick you pulled with sqrt was quite in this vein.) Robust, bullet-proof BL, as the first pain point, maybe? If orthogonality check fails - restart from an earlier checkpoint or stop and wait for master's decision (and all that stems from this - possibly 1. do not overwrite checkpoints, do at least two of them; 2. the .chk files should have passed a QC before writing... etc.) http://www.mersenneforum.org/showpos...&postcount=406 and same thing happens every now and then to each of us, right? And then everyone gathers around, pats the guy on the back and sardonically welcomes to the club. (I've got trivial deps on my last week-long BL, too. Not the first such case for me, don't welcome me to the club.) --Serge

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
What does net neutrality mean for the future?	jasong	jasong	1	2015-04-26 08:55
Future of Primes.	mfgoode	Lounge	3	2006-11-18 23:43
The future of NFSNET	JHansen	NFSNET Discussion	15	2004-06-01 19:58
GIMPS future	Complex33	Lounge	31	2003-12-05 09:08
15k Future?	PrimeFun	Lounge	21	2003-07-25 02:50

2008-10-30, 02:23	#24
jasonp Tribal Bullet Oct 2004 DE3₁₆ Posts	Currently I'm focusing on implementing Kleinjung's new polynomial selection method, described in his CADO workshop slides. It's remarkably simple and can potentially produce better polynomials than pol5 can. Unfortunately, making pol51opt optimize polynomials with enormous skew requires major changes in the root sieve, so it's slow going. But this is more fun with the code than I've had in a long time. I should ask Kazumaro Aoki about the Lanczos checks that his code implements; he's had to deal with everything that could possibly go wrong solving matrices.