Discussion related to cado-nfs

[narges mousavi <mousavi.nrgs@gmail.com>]

Dear Prof. Zimmermann

Thank you for your fast reply.

It was assumed that the reported results are the total spending time with 16 threads, not the sum up of all thread’s time; So, we made a mistake in comparison of our results with yours.

We plan to run the CADO-NFS 2.1 with 128 threads. It would be grateful if you can help us by answering this question:

We know that the sieving step can be run using almost unlimited number of threads, but we don’t have enough knowledge about the other steps. How many threads can we use in Polynomial selection, Linear Algebra and Square roots?  Is there any limitations? (Max or min…?)

Best Regards

Narges

On Mon, Jun 15, 2015 at 12:57 PM, paul zimmermann <Paul.Zimmermann@inria.fr> wrote:

       Dear Narges,

> Date: Mon, 15 Jun 2015 12:43:32 +0430
> From: narges mousavi <mousavi.nrgs@gmail.com>
>
> To whom it may concern,
>
> According to the results mentioned in the main page of CADO-NFS
> <http://cado-nfs.gforge.inria.fr/> (Features subsection), can we say
> CADO-NFS 2.1 is two times faster than CADO-NFS 1.1? If so, can we refer to
> this result for larger numbers or not?

as you can see in the table, the speedup of 2.1 wrt 1.1 is not monotonous:
2.3 for RSA-120, 1.3 for RSA-130, 1.4 for RSA-140 and 2.2 for RSA-155. Note
also that for another (say) 155-digit number you might get a different ratio.
Thus we cannot say that CADO-NFS 2.1 is two times faster than CADO-NFS 1.1.
However it shows that CADO-NFS 2.1 is globally much faster than 1.1. For
larger numbers we don't have real figures but I'm confident it is similar.

> As the second question, we had set of factorizations on large numbers with
> CADO-NFS 1.1 and overally got more better results than the reported results
> that are shown in the CADO-NFS's site (more than 6 times faster). We can
> not find out what's the main reason of this different performances. Our
> test bed features are as the following:
>
> - Intel  Xeon  X7560  2.27GHz  ,   4cpu*8*2 thr = 64 core (we used 60
> threads)
> - 25MB cache, 64GB RAM
>
> Best Regards,
> Narges

our table displays the total cpu time, i.e., with 16 threads we sum up the
time spend for all threads. It would be very surprising that you get a 6-times
speedup. Can you reproduce on the RSA-xxx numbers?

Best regards,
Paul Zimmermann