On Sun, 5 Oct 2014, Nicholas Malaya wrote:

> The old gridded vane cases were run before we had started moving to
> serial meshes, so they create many xda files (say, 512).
>
> I want to use the 'averager' you designed to average these fields, but
> this wants xda files from serial meshes.

Hmm... I think running meshavg (or anything else that reads our
restart format) with ParallelMesh should work iff it's run on the same
number of ranks as the restart was run on.  So you could run meshavg
with 'mpirun -np 512' to average the restarts (even locally! 512 ranks
doesn't have to imply 512 processors, and having n_ranks>>n_cores
isn't as bad with ParallelMesh), then meshplot with ParallelMesh to
turn the averaged file into a .exo for your code.

> So my first inclination was to start a run for 0 timesteps that
> points to the many xda files, run with serial mesh, and therefore
> create a single xda output. (I ran this for a simple test locally
> and it behaved as expected)

How many ranks do you use for that run?  512 ranks with SerialMesh run
locally would kill your RAM, yeah.

> However, I am getting a memory error with this. I'm guessing that
> since serial mesh requires the entire mesh to be held locally, this
> pushed us over the edge.

Depends on how many ranks were running per node... but that's also a
limitation of using grins for plain mesh processing.  Unlike the
meshavg, meshplot, etc tools, grins always generates sparse matrices,
which can suck up a ton of RAM themselves.

We could probably fix that, though.  If the meshavg+meshplot on
ParallelMesh doesn't work, let me know, and I'll modify grins to
detect "0 timestep transient" runs and avoid generating sparse
matrices in that case.

> Any idea how one could convert the many xda files to a single xda
> output without performing the entire run again from the start? Thanks,

Copying libmesh-users in case anyone else has better ideas or can
correct me if I'm wrong.
---
Roy

------------------------------------------------------------------------------
Meet PCI DSS 3.0 Compliance Requirements with EventLog Analyzer
Achieve PCI DSS 3.0 Compliant Status with Out-of-the-box PCI DSS Reports
Are you Audit-Ready for PCI DSS 3.0 Compliance? Download White paper
Comply to PCI DSS 3.0 Requirement 10 and 11.5 with EventLog Analyzer
http://pubads.g.doubleclick.net/gampad/clk?id=154622311&iu=/4140/ostg.clktrk
_______________________________________________
Libmesh-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-users

Reply via email to