Importing dolfin takes for ever when huge number of nodes are used on large clusters

[Johan Hake]

See this discussion for more details.

[Chris Richardson]

As discussed before, much of the slowness is probably due to the sheer number of files being imported.

from dolfin import * imports the following number of .py files (on my install).

5 six
20 instant
46 FIAT
70 dolfin
146 numpy
154 ffc
166 python
202 ufl
319 sympy

Much of this is out of our control, but it might be possible to push down the number of files in ufl - and maybe consider how to get AD without using sympy.

[Martin Sandve Alnæs]

Added this:

https://bitbucket.org/fenics-project/ufl/issue/52/reduce-the-number-of-python-files

Getting rid of the sympy dependency will of course be a larger contribution.

[Chris Richardson]

I'm not sure this is really a DOLFIN issue, anyway, but more about using python on HPC. Is it worthwhile keeping it here?

[Prof Garth Wells]

See https://bitbucket.org/fenics-project/fiat/issue/8

[Prof Garth Wells]

Closing this because it's fundamentally and Python + HPC problem. We might be able to make a dent in the issue by reducing the number of imported files, but what it really needs is a system solution.

See http://www.archer.ac.uk/documentation/white-papers/dynamic-import/ARCHER_wp_dynamic-import.pdf.

[Anders Logg (Chalmers)]

Removing milestone: 1.7 (automated comment)