factmsieve.py Questions

[EdH]

I'm using factmsieve.py in a scripted method with several machines to distribute sieving across them. They use the spairs.add feature to collect all the relations and have them incorporated into the .dat file on each check. The way I am doing this, is probably quite rough, but it is working seemingly (to me) well. (If anyone is interested, I will provide further details later.

For now, my first question is how I can modify factmsieve.py (or, something else) to take smaller steps for its qintsize. Is qstep (or, qintsize) calculated, or from a table? I'm expecting to work with ~120-140 digit composites.

My reason to want to decrease the size, is that with all the extra machines adding relations, I get some real oversieving sometimes. A recent c131 gave me a situation where factmsieve.py checked at ~91.5%, so it assigned the next set of tasks. But, by the time it finished those, enough relations had been added from the other machines to take the count over 150%.

My next question has to do with polynomial pair selection. Is there a method to aggregate polynomials from external source machines, similar to the spairs.add feature for relations gathering?

Alternately, can I just "cat" them to one of the files and would the entire file contents be considered when the best poly is chosen by msieve (or, factmsieve.py)?

Last question for now, if I can describe it correctly, is with my composites around 130 digits +/- 10, do I need to be sure the external (to the main) poly searching machines are deconflicted in their search regions? If so, what is my best method? Is there a program related way to determine where msieve has chosen (via factmsieve.py) to work?

Thanks for all your help...

[Brian Gladman]

The values used for quintsize (and hence qstep) are set by the def-par.txt file. The format of lines in this file are described in the file while individual lines set out the parameters used for various size factoring jobs. I believe that you should be able to do what you want by modifying appropriate lines to suit your needs (note however that factmsieve.py changes the values for very large jobs of 160 digits or more).

[EdH]

Originally Posted by Brian Gladman

The values used for quintsize (and hence qstep) are set by the def-par.txt file. The format of lines in this file are described in the file while individual lines set out the parameters used for various size factoring jobs. I believe that you should be able to do what you want by modifying appropriate lines to suit your needs (note however that factmsieve.py changes the values for very large jobs of 160 digits or more).

Thank you.

I thought perhaps I had done something a few years ago in that file, but I couldn't see a correlation with any of its lines and a recent run. Now that I know that is the place, I can work out my needs.

I had seen the section of code for 160+ digit composites. I almost altered it before noticing it was for that size and above.

Thanks for all.

[EdH]

Originally Posted by Brian Gladman

The values used for quintsize (and hence qstep) are set by the def-par.txt file. The format of lines in this file are described in the file while individual lines set out the parameters used for various size factoring jobs. I believe that you should be able to do what you want by modifying appropriate lines to suit your needs (note however that factmsieve.py changes the values for very large jobs of 160 digits or more).

Hey Brian,

I finally figured out (read correctly) the section of code I was trying to manage. I was missing that fact_pl['digs'] was equal to digits / 0.72 and therefore any composite over 115 digits was handled by the "if fact_p['digs' >= 160:" section, instead of def-par.txt. So, therefore factmsieve.py changes the values for composites over 115 digits.

Thanks for all,
Ed