[IPOL discuss] [IPOL tech] Démo de la méthode K_SVD

[Nicolas Limare]

Oops, I was wrong.

> It's not the "demo system" (ie python code) that is stopping processes
> demanding too much memory, it is the operating system kernel. I think
> these interrupted processes result in error messages appearing in the
> logs, like segmentation faults, so if you did not see error messages I
> think it's not a memory problem.

For K-SVD (similar problem), it was a memory use problem.

> 1Gb is what we ask to the authors and the general rule in the
> softeware guidelines. But on the demo server, the limit is at 4Gb.
> (cf /etc/init.d/ipol_demo)

I just increased the hard limit to 8Gb on the demo server, but I think
the memory requirements of IPOL programs should be investigated,
because running with less memory is as good as running in less time.

One cause for large memory footprints can be the parallel code
sections, if a large array is declared as "private", ie replicated for
every execution thread (-> x32 on the demo server!). If this data is
only read or if it can be written without thread conflicts, it should
be declared as "public", and not replicated.
-> https://computing.llnl.gov/tutorials/openMP/#Clauses

-- 
Nicolas LIMARE - CMLA - ENS Cachan         http://limare.perso.math.cnrs.fr/
IPOL journal                                             http://www.ipol.im/
-> image processing, reproducible research, open science
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <https://tools.ipol.im/mailman/archive/discuss/attachments/20131024/0714ef0d/attachment.pgp>