Discussion related to cado-nfs

[Emmanuel Thomé <Emmanuel.Thome@inria.fr>]

On Fri, Oct 14, 2016 at 07:32:06AM +0200, Pierrick Gaudry wrote:
> Hi,
> 
> Yes, there is definitely support in Cado-nfs for using many nodes.
> 
> Some time ago, I had written some kind of Howto that is accessible from:
>   
> https://gforge.inria.fr/tracker/index.php?func=detail&aid=16857&group_id=2065&atid=7445
> but this is at least partly obsolete.
> 
> Also in linalg/bwc/README, there is some more documentation about how the
> linear algebra step works. Again it starts with a warning that this is
> not up to date :-(

Well, it carries modification dates ! I did skim through it this summer,
and clean up some outdated stuff.

The section "Using bwc.pl with MPI" is probably where you want to start
from.

Next, you'll certainly want to have parameters such as linalg.bwc.mpi and
linalg.bwc.thr set to something reasonable.

> Look also at this thread:
>   
> https://lists.gforge.inria.fr/pipermail/cado-nfs-discuss/2015-June/000431.html
> 
> The thing is that the linear algebra has been deeply modified a few weeks
> ago, so some binaries, some files, some command-lines might have changed.

There used to be a binary called "split", which is now dead.

The command-lines of the "mksol" and "gather" binaries have been modified
a bit.


E.

> What I suggest is that you give it a try after reading the provided
> information, and then you don't hesitate to ask (maybe not on the list)
> if you have a problem. In the meantime, I (or Emmanuel who is the main
> author of this part of the code) will write a proper Howto for using the
> full linear algebra capabilities of CADO-NFS.
> 
> Regards,
> Pierrick
> 
> On Fri, Oct 14, 2016 at 06:31:40AM +0330, Amin Razavi wrote:
> > Hi,
> > We are trying to factor a 190 digits number in a 64 nodes cluster (48 
> > cores
> > for each node).
> > It takes 20 hours for polynomial selection, 3 days for sieving , 5 hours
> > for merging.
> > Now surprisingly after 3 days we have only 2% progress in linear algebra
> > (krylov) : 22000 of 1266000 iterations.
> > Size of all data on hard disk is 150GB.
> > We had some software crashes but resumed the process.
> > It is not what we had expected from linear algebra process.
> > Is there any way to distribute the "krylov" process to all the clusters
> > instead of only 1 node? How to break the sparse merged matrix to pieces 
> > and
> > run it separately on our mesh?
> > 
> > Thanks
> 
> > _______________________________________________
> > Cado-nfs-discuss mailing list
> > Cado-nfs-discuss@lists.gforge.inria.fr
> > http://lists.gforge.inria.fr/mailman/listinfo/cado-nfs-discuss
> 
> _______________________________________________
> Cado-nfs-discuss mailing list
> Cado-nfs-discuss@lists.gforge.inria.fr
> http://lists.gforge.inria.fr/mailman/listinfo/cado-nfs-discuss