> On 07 May 2015, at 13:02, Michael Lange <[email protected]> wrote: > > Hi, > > I am trying to optimise the performance of the current file reader for binary > .msh files in PETSc by utilising the element grouping described in the binary > format (see http://www.geuz.org/gmsh/doc/texinfo/#MSH-binary-file-format). > However, using a recent version of Gmsh (version 2.9.3, with options: -3 > -bin), the generated .msh files contain a header + binary entry pair for each > element in the mesh, instead of two groups of same-type elements (volume and > surface). This unfortunately prevents me from using efficient bulk reads to > speed up file I/O and hinders performance when reading large meshes on > certain HPC architectures. > > Is there any way to enable this type of element grouping in the binary mesh > file generation? Am I missing an option by any chance, or does this require > significant code changes?
Hi Michael - Indeed, the .msh format is designed for better performance by dumping groups of elements, but the current implementation in Gmsh outputs elements one at a time. Have you tried the .med (HDF5-based) format? The current implementation is pretty fast: > time gmsh 6millionTets_binary.msh - Info : Running 'gmsh 6millionTets_binary.msh -' [Gmsh 2.9.4, 1 node, max. 1 thread] Info : Started on Thu May 7 22:58:24 2015 Info : Reading '6millionTets_binary.msh'... Info : Mesh is in binary format Info : 1030301 vertices Info : Vertex numbering is dense Info : 6121200 elements Info : Done reading '6millionTets_binary.msh' Info : Stopped on Thu May 7 22:58:29 2015 real 0m5.844s user 0m5.566s sys 0m0.272s > time gmsh 6millionTets_binary.med - Info : Running 'gmsh 6millionTets_binary.med -' [Gmsh 2.9.4, 1 node, max. 1 thread] Info : Started on Thu May 7 22:58:33 2015 Info : Reading '6millionTets_binary.med'... Info : Reading MED file V3.0.6 using MED library V3.0.6 Info : Reading 3-D unstructured mesh <<6millionTets_binary>> Info : Done reading '6millionTets_binary.med' Info : Stopped on Thu May 7 22:58:37 2015 real 0m3.474s user 0m3.170s sys 0m0.301s > > Thanks, > > Michael Lange > > _______________________________________________ > gmsh mailing list > [email protected] > http://www.geuz.org/mailman/listinfo/gmsh -- Prof. Christophe Geuzaine University of Liege, Electrical Engineering and Computer Science http://www.montefiore.ulg.ac.be/~geuzaine _______________________________________________ gmsh mailing list [email protected] http://www.geuz.org/mailman/listinfo/gmsh
