Discussion related to cado-nfs

[Pierrick Gaudry <pierrick.gaudry@loria.fr>]

Dear Alessio,

When computing relations, there is a point where we have "survivors",
i.e. candidates that have a reasonable chance to become a relation. These
can be handled on-the-fly, and then `las` can do it, or in the case where
we prefer to accumulate a lot of them beforehand, we put them in files,
and then we use `finishbatch`.

As for why it is or it isn't better to accumulate (or batch) survivors is
a long story, and depends on multiple factors. Reading and understanding
all the ugly technical details of the paper is necessary.

Regards,
Pierrick


On Thu, Sep 12, 2024 at 01:52:56PM +0200, Alessio Proietti wrote:
> Dear all,
> 
> please let me introduce myself first.
> 
> I am Alessio Proietti, a student in applied mathematics  in Rome Italy and I
> am about to graduate with a thesis on constraint programming applied to
> analysis of symmetric primitives. My background is dominated by study of
> symmetric crypto but I had some exposure to computational number theory, ANT
> and asymmetric cryptography. Recently, I got interested also in analysis of
> asymmetric primitives and their underlying hardness assumption so that I
> found GNFS and ultimately cado.
> 
> I was reading/Comparing the Difficulty of Factorization and Discrete
> Logarithm: A 240-Digit Experiment /and linked resources at
> https://gitlab.inria.fr/cado-nfs/records.
> 
> I think I got most of it except for a passage in reproducibility of
> cost-estimation for sieving in DLP240.  For RSA240 and RSA250 is used
> `rsa250-sieve-batch.sh` while for DLP240 it seems to me as if `las` could
> completely replace `finishbatch` and do batch smoothness detection or in
> general do whatever `finishbatch` can do... Am I getting it wrong? Why does
> `finishbatch` exist at all? Could one have used a `XXXYYY-sieve-batch.sh`
> script using `finishbatch` for DLP too?  Thanks.
> 
> Best Regards,
> 
> Alessio
>