Discussion related to cado-nfs

[Paul Zimmermann <Paul.Zimmermann@inria.fr>]

       Dear Elena,

> From: Elena Wilson <elena225318@gmail.com>
> Date: Mon, 22 Jun 2020 12:02:21 +0430
> 
> Hi everybody,
> 
> The noticeable differences between cado-nfs-2.3.0 and cado-nfs-master are as
> follows:
> - Using batch smoothness detection in sieving
> - Parallelizing the "merge" step in filtering
> - Parallelizing the "lingen" step in linear algebra 
> 
> However, after running the cado-nfs-master, sieving of the cado-nfs-master
> took a longer time than the sieving of the cado-nfs.2.3.0.
> The comparison is done under the same setup with c120.params included in the
> master version:
> 
> #####################################################################
> # Sieve
> #####################################################################
> tasks.lim0 = 3000000
> tasks.lim1 = 5500000
> tasks.lpb0 = 27
> tasks.lpb1 = 27
> tasks.sieve.mfb0 = 54
> tasks.sieve.mfb1 = 54
> tasks.sieve.ncurves0 = 14
> tasks.sieve.ncurves1 = 19
> tasks.I = 12
> tasks.qmin = 1000000
> tasks.sieve.qrange = 10000
> # with that value, most factorizations will need only one filtering attempt
> tasks.sieve.rels_wanted = 13419694
> #####################################################################
> 
> For example, here is the result of sieving at interval 4000000-4010000:
> cado-nfs-2.3.0:
> 32674 reports
> elapsed time = 549.45 s
> PeakMemusage = 2805 MB
> 
> cado-nfs-master:
> 32664 reports
> elapsed time = 355.6
> PeakMemusage = 261 MB
> 
> Can you please explain what is wrong with this comparison?
> 
> Thank you,
> Elena

there might indeed be a regression for "smaller" numbers in cado-nfs-master,
which was optimized for the latest records (rsa240, rsa250, dlp240).

However your timings (unless you mixed them) seem to indicate that
cado-nfs-master is faster.

Please can you post the detailed command lines (makefb and las) that you used?

Best regards,
Paul