Stream

Join this community to post or comment
 
Fast continuous integration of Fenics applications in Travis CI using MiniConda.

The use of MiniConda allows me to use the container-based infrastructure of Travis that loads very quickly. Starts within seconds after git push and loads the Fenics environment within 2 minutes. See for details

https://travis-ci.org/mikaem/testtravis

I have set this up for fenicstools as well, and I think it could be very useful for other Fenics applications. The Fenics installation is very recent, less than a week old from master on bitbucket. The recipes for building Fenics with Conda suitable for Travis can be found on the conda-gcc branch on this fork from Juan Rodriguez' original fenics-recipes: https://github.com/mikaem/fenics-recipes/tree/conda-gcc


2
Mikael Mortensen's profile photoJan Blechta's profile photo
8 comments
 
Ok, thanks for the explanation. Anyway it is surprisingly fast;)
Add a comment...

Garth Wells
owner

Announcements  - 
 
FEniCS in Linux Containers poster is on Figshare (http://dx.doi.org/10.6084/m9.figshare.1472955)
1
Add a comment...

Garth Wells
owner

Announcements  - 
 
Check out the Docker images at https://registry.hub.docker.com/repos/fenicsproject/ for running FEniCS inside Linux containers. Some starter instructions were posted to the mailing list at https://www.mail-archive.com/fenics@fenicsproject.org/msg02615.html.
1
Add a comment...

Garth Wells
owner

Announcements  - 
 
FEniCS'15 Workshop will be held 29 June - 1 July at Imperial College London.
6
1
Petr Lukáš's profile photoAnders Logg's profile photo
 
I would like to ask whether it exists any URL with more details of this event...
Add a comment...
 
Dear computational mathematicians, scientists and engineers,

Hypothetically, suppose you had a web-based scientific computing service at your disposal. You could send it a numerical problem, which it would solve efficiently and stream the results back to you. Further imagine that the communication both ways is in some pleasant and standard representation, allowing e.g. interactive visualization on your tablet or phone.

Are you working on a problem that you feel will benefit from the existence of such technology? Do you know colleagues that might?
2
Add a comment...
 
I'm working on a demo of how to do global operators in FEniCS. This computes

  d/ds M(f+s*v)

with semi-automatic differentiation, where

  M = F(x,f)**2*dx

and

  F(y,f) = f(x)*dx(1),

with subdomain 1 depending on y. The differentiation is semi-automatic because a couple of relations here must be specified manually.

See the code for details:

https://bitbucket.org/fenics-project/dolfin/issue/413/add-a-demo-of-global-operators-with-semi

It would be nice to refine this into a high quality dolfin demo with a better example operator, as it shows off several advanced features.
4
Add a comment...
 
Solving Poisson on a Unit Cube is a fairly basic test, but even that gets challenging at large scale. Here are some results with 24576 cores on  ARCHER (12.5 billion dofs). Algebraic Multigrid is the way to go, and PETSc has two 64-bit interfaces, GAMG and HYPRE. GAMG converges in fewer iterations, but the setup time (not shown) is more, so in fact, there is little difference between them. Solve time is about 60s.
6
2
Peter Brune's profile photoChris Richardson's profile photoJan Blechta's profile photoWim Vanroose's profile photo
4 comments
 
My disclaimer: I'm no expert, so if anyone has some different parameters to suggest, I can try them out. I guess the simple geometry may favour different amg models, too. It is difficult to figure out the best combination to minimise pcsetup + num_its * cost_per_iteration...
Add a comment...

Anders Logg
owner

Discussion  - 
 
It works! Finally.
5
Anders Logg's profile photoAditya Kashi's profile photo
3 comments
 
Multi-mesh forms? Excellent! That's exactly what I need for my current project, but am currently having to find some CAD and mesh-level workarounds to do my job.
Add a comment...

About this community

The FEniCS Project is a collection of free software with an extensive list of features for automated, efficient solution of differential equations.

Garth Wells
owner

Announcements  - 
 
More details on the FEniCS'16 workshop have been posted to the mailing list: https://groups.google.com/d/msg/fenics-support/m6HemwjCB28/FsuJ7Ps-CQAJ
Google Groups allows you to create and participate in online forums and email-based groups with a rich experience for community conversations.
1
1
Jan Blechta's profile photo
Add a comment...

Garth Wells
owner

Announcements  - 
 
FEniCS'16 Workshop will be held 18th-20th May 2016 at Centre for Biomedical Computing at Simula, Oslo, Norway. Keep an eye on http://www.fenicsproject.org for updates.
FEniCS'15 at Imperial College London. We are happy to announce that the 2015 FEniCS Workshop will be held on June 29 - July 1 2015 at Imperial College London, London, UK. More information can be found at the official FEniCS '15 webpage.
2
1
Jan Blechta's profile photo
Add a comment...

Garth Wells
owner

Announcements  - 
 
 
I have an opening for a two-year post-doctoral research associate position at University of Cambridge. The research area is HPC for solid mechanics. The project will use FEniCS. See http://www.jobs.cam.ac.uk/job/7120/ for more information on applying. Send me a message for informal enquiries.
Research Assistant/Associate in High Performance Computing for Problems in Solid Mechanics (Fixed Term) in the Department of Engineering at the University of Cambridge.
1
Add a comment...

Anders Logg
owner

Discussion  - 
 
 
Postdoc positions in mathematics at Chalmers. FEniCS skills (or equivalent) a plus! #fenicsnews
1
Add a comment...

Garth Wells
owner

Discussion  - 
 
 
Check out notes from FEniCS developer's meeting on the mailing list and take part in the discussion. #fenicsnews
1
Add a comment...
 
 
Some new weak scaling results from simulations of a triply periodic (unstable, Re=1600) Taylor-Green vortex, using the Oasis P2P1 Navier-Stokes solver on the Abel Supercomputer at UiO. Timings are performed every 10'th time step and the figure shows both the best, the worst and the standard deviation. Assemble is perfect:-), PETSc cg solver with hypre-amg is struggling. 1 million dofs per CPU.
2
Garth Wells's profile photoMikael Mortensen's profile photo
8 comments
 
+Garth Wells
Absolute value for pressure is 1, so 1e-8 should be ok. I have not tried with stricter tolerances, though, so don't know where the floor is. Note that solution is very much in agreement with what it should be, both for pressure and velocity.
Add a comment...
 
Over the last couple of weeks I've optimized UFL quite a bit, from the core representation to the main algorithms, and the time to compile more complicated forms is now shorter than ever. Many of the optimizations apply both to uflacs and vanilla ffc, but uflacs is still faster for some of the more nasty equations out there.

Do you have a form that takes forever to jit? I'd love it if you can time it, update to latest master version of all fenics components, time it again, and post the result here.
9
Add a comment...
 
The dolfin Poisson demo has been observed working with python3 on Johan Hake's computer. Coming soon to the official FEniCS repositories, although a patched version of SWIG will be required.
4
Aditya Kashi's profile photoAnders Logg's profile photo
2 comments
 
I like this formulation... According to rumours , FEniCS / Python3 has been observed to work on Johan Hakes computer.... But this is not surprising as that is typically where the magic happens. 
Add a comment...

Garth Wells
owner

Announcements  - 
 
Try out the work of Larry Lizao on running FEniCS portably using +Docker 

https://www.mail-archive.com/fenics@fenicsproject.org/msg01554.html
1
Add a comment...
 
Just improved the svg rendering of 1D and 2D dolfin meshes for ipython notebook (in dolfin next branch).
4
Garth Wells's profile photoMartin Sandve Alnæs's profile photo
2 comments
 
This is just a direct convertion of each edge and vertex of the mesh into svg primitives. Anything is possible but I am not planning to extend this further. 
Add a comment...