On Nov 3, 2009, at 5:21 PM, Shang wrote:
I have the same problem. There are random white blocks in the
epsilon png output
when I run the simulation on the cluster with multi-node.
Any real solution for this problem? Is it solved in the latest meep
version?
Thanks!
The solution is, and always has been, to use the MPI version of HDF5.
With the non-MPI version of MPI, the only way we have to write out an
HDF5 file is to have one process at a time open the file, write to it,
and close it, but on a cluster the network filesystem semantics don't
seem strong enough for this to work. (I suppose we might try
inserting explicit fsync calls to try to force the filesystem to flush
to disk, but it always seemed like the right solution was to use the
parallel HDF5.)
Steven
_______________________________________________
meep-discuss mailing list
[email protected]
http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/meep-discuss
