Skip to Content.
Sympa Menu

cado-nfs - Re: [Cado-nfs-discuss] Restarting linear algebra after running out of storage

Subject: Discussion related to cado-nfs

List archive

Re: [Cado-nfs-discuss] Restarting linear algebra after running out of storage


Chronological Thread 
  • From: Emmanuel Thomé <Emmanuel.Thome@inria.fr>
  • To: David Willmore <davidwillmore@gmail.com>
  • Cc: cado-nfs-discuss@lists.gforge.inria.fr
  • Subject: Re: [Cado-nfs-discuss] Restarting linear algebra after running out of storage
  • Date: Mon, 9 Dec 2019 22:01:42 +0100
  • List-archive: <http://lists.gforge.inria.fr/pipermail/cado-nfs-discuss/>
  • List-id: A discussion list for Cado-NFS <cado-nfs-discuss.lists.gforge.inria.fr>

Hi,

Glad that you were able to solve your problem.

For the record:

On Mon, Dec 09, 2019 at 02:09:33PM -0500, David Willmore wrote:
> Writing balancing data to c155.sparse.4x6.bin
> write(cached matrix file): No space left on device
> write(cached matrix file): No space left on device
> write(cached matrix file): No space left on device
> write(cached matrix file): No space left on device
> /home/willmore/factoring/cado-nfs-2.3.0/build/blade/linalg/bwc/prep:
> exited with status 1
> aborted on subprogram error at
> /home/willmore/factoring/cado-nfs-2.3.0/build/blade/linalg/bwc/bwc.pl
> line 512.
> ...propagated at
> /home/willmore/factoring/cado-nfs-2.3.0/build/blade/linalg/bwc/bwc.pl
> line 1406.

In this situation, the cached matrix files that are written in the
directory c155.bwc/c155.sparse.4x6/ are unusable and must be
redone. It would have been sufficient to just remove that directory, as
the cached matrix files would have been recomputed by whatever bwc binary
was left to call (prep).

And it's an obvious misfeature that cado-nfs does the write to the final
path of the cached matrix file, in fact. Instead, it should write to a
temp file, and attempt a filesystem reame (mv) operation, which is atomic
in most cases. However, in cases where the filesystem is full, we would
just be trading an error message for another one anyway.

E.

>
> > From Paul.Zimmerman@inria.fr
> > you can restart the linear algebra directly by running the bwc.pl command
> > at the top of c155.bwc.stdout.1.
> > ...
> > All files of the same type should have the same size. If one is short,
> > just remove it (or move it to another place) and restart the bwc.pl
> > command.
>
> Nice to hear from you again, Paul. It's been a long time.
>
> That seems to have done it. I've moved all of those files out as they
> were of several different lengths and restarted the command.
>
> It's running now. Will this complete the factorization or will I need
> to manually perform any other steps after this?
>
> Thank you both for your help! I'll share my results with the group if
> anyone is interested.
>
> Cheers,
> David Willmore
>
> On Mon, Dec 9, 2019 at 8:46 AM Emmanuel Thomé <Emmanuel.Thome@inria.fr>
> wrote:
> >
> > On Mon, Dec 09, 2019 at 06:27:52AM -0500, David Willmore wrote:
> > > Upon restarting the cado-nfs.py script with the location of the
> > > recovery parameter file, I was told that there was a short read from
> > > the .prep file. Moving that out of the way and restarting resulted in
> > > the same error.
> >
> > Can you give the exact error message please ? You can probably find it in
> > files called c155.bwc.std*
> >
> > You might want to try to remove the cache directories c155.bwc/c155.*x*/
> > ; those are likely to contain files that were written incompletely.
> >
> > E.




Archive powered by MHonArc 2.6.19+.

Top of Page