View Single Post
Old 2017-02-02, 04:06   #9
VBCurtis
 
VBCurtis's Avatar
 
"Curtis"
Feb 2005
Riverside, CA

10010001111102 Posts
Default

I admit I don't grok the NFS algorithm as well as I'd like, but for your purposes it's quite useful to understand the stages and what progress looks like.

I'm running poly select on two GPUs this evening on your number; I'll let them each run a couple hours and post the best poly I find here (so you can compare what you find, which will surely be better).

In this subforum, you'll find a thread "best msieve poly scores". Post #7 notes a C170 has best score 3.91e-13. If you break that score, I'd end poly select and proceed right to the sieving step (and post the poly you found so I can note the new record!)

A C171 has score of 3.13e-13, so you should expect to beat that score with your search. Score is, roughly, a measure of the rate at which you can expect to find relations, so a larger score is better (and project length scales roughly with inverse of score- a 3e-13 project is *about* 10 times harder than a 3e-12 project).

Your search will produce quite a few okay polys, with occasional good ones. In some sense, we wait for a really good one to pop out of the search- but how long to wait, as well as what "good" is, isn't quite clear. That's why I track best scores- if you break someone else's best for a size, you've likely found a really good poly and can skip the rest of poly select. If you find a few around 3.2e-13, and then one batch has one or two like 10% better, you've probably found your poly. It's worth holding out a few days of poly select for a hit 8-10% better than the typical best results from a run (I check twice a day, and consider each half-day a "run"), because 8% shorter project length for an 8-week effort is ~5 days of saved sieving time.

The factmsieve script has been tested and refined for inputs up to about 150-155 digits; the automatically generated settings aren't the fastest for 150+ digits, simply because so few projects are done at those sizes the author lacked data to create a set of best parameters (so he guessed a little, and figured anyone serious enough to do C170 would already know enough to set parameters himself). The NFS@home BOINC project has recently done a bunch of factorizations in this size category, but they've not always optimized parameters for shortest project length; all the same, their parameters are useful to copy for this project absent better info.

If you decide you wish to do this factorization, I'll explain test-sieving in a future post, so you can test a couple possible parameters choices and hope to discover savings of a day or week in sieving via better-than-script-default choices.

Last fiddled with by VBCurtis on 2017-02-02 at 04:09
VBCurtis is offline   Reply With Quote