[Cado-nfs-discuss] some questions about cado-nfs2.3.0

Emmanuel Thomé Emmanuel.Thome at inria.fr
Mon Dec 3 08:50:12 CET 2018


On Sun, Dec 02, 2018 at 08:47:54AM +0100, sara hendrix wrote:
> Hi
>  
> i have some questions about cado-nfs2.3.0.
>  
> 1- i want to use blocklonczos.c file in cado-nfs2.3.0 seperataly. how i can
> feed blocklonczos.c the relations after sieving and filtering procedures?

I assume you mean blocklanczos.

While there is some code for Block Lanczos, it is only here to make sure
that it compiles, and passes some basic tests. I have not plugged it with
the rest of the nfs machinery, so you're out of luck here.

> 2- i have run cado-nfs on multiple machines. When LinearAlgebra was started,
> the program was ran only on host not the clients. How i can solve this problem?
> the input number for factoring has 640 bits long.

The python script only distributes the polynomial selection and sieving
steps.

In order to run the block Wiedemann algorithm on several nodes,
then you'll have to do without the python script. You may have a look at
the sections "Using bwc with MPI" and "Using several sequences" in
linalg/bwc/README

> 3- If i have multiple machines which all of them have many logical cores, say
> 32 or higher, what is the best strategy for setting the values of adrange,
> qrange and number of slave on each client?

For the range values, this only indicates the granularity of the
computation. On first approximation you don't care.

As for the number of slaves, it mostly depends on the available RAM on
your nodes. If 32 indepent mono-threaded jobs fit, then it will probably
be the best option, but eight 4-threaded jobs might do just as well.

E.


More information about the Cado-nfs-discuss mailing list