View Single Post
Old 2011-11-04, 20:11   #6
Andi47
 
Andi47's Avatar
 
Oct 2004
Austria

2·17·73 Posts
Default

Quote:
Originally Posted by jrk View Post
Yes that is a good poly. My expected murphy score for a number of this size is about 2.71e-12 and your poly is 2.81e-12, so that is good. You should use ggnfs siever 14e and sieve on the -a side starting from specialq 15M until you have gathered 56M raw relations. Sieving will take about 3 to 4 weeks on your two cores, and the matrix about 2 days using msieve.
I am not sure whether LaurV can finish this one as an one-man job.

One of my biggest "one man" factorizations so far was a c153 cofactor of 2,1766M, and I used a similar computer for it (core 2, 2 gigs RAM).

After finishing sieving I attempted to do the matrix, and what happened then, can be seen in that thread beginning from post #157: Matrix generation failed with a "failed to reallocate 1.63 GB" error. On a 32 bit windows system it was not possible to do this linalg job, not even when I moved to a P4 with 3 Gigs RAM (also with 32-bit windows), because 32-bit windows is not able to handle the address space (reallocating huge chunks of data) which is needed for this matrix. When I switched to linux, it happily worked out.

So if LaurV's computer is running on a 64 bit system and with a 64-bit binary of msieve, 156 digit GNFS should be possible (but close to the edge of available RAM) (will take approx. 2 GB RAM).
For 32-bit systems it could end up in a balancing act between extreme oversiving for squeezing the matrix in smaller amount of RAM an failing when it's way too oversieved and runs out of cliques in the filtering step.

Last fiddled with by Andi47 on 2011-11-04 at 20:34 Reason: typo, clarification, ...
Andi47 is offline   Reply With Quote