On May 30, 2013, at 3:47 PM, Manav Bhatia <bhatiama...@gmail.com> wrote:
> Hi, > > I am attempting to do analysis on a mesh with 8M cells. The mesh was > created with gridgen, and am using the ExodusII IO class to read the > mesh. I am using the ParallelMesh class with the library configured with > Parmetis (and also with Metis, but I am not sure if this would matter). > > The process increases the memory requirement to ~22GB irrespective of > number of processors, and I am wondering if the process of reading the mesh > and then partitioning is making each process read the entire mesh before > breaking it up. > > If so, is there a flag/setup that I am missing to enable reading the > mesh in pieces? ExodusII is a serial format, and the reader is serial too I think. ExodusII meshes can be pre-partitioned, however, and the pieces read in parallel. I helped put the capability in a few years ago but don't use it regularly - perhaps John or some of the MOOSE guys could expand on the preferred way to preprocess a mesh into pieces? A fallback may be to use a single large memory node (if you have one big enough!) to transform the Exodus file to an XDR file, which could be read in parallel, but I'd want to check that... -Ben ------------------------------------------------------------------------------ Introducing AppDynamics Lite, a free troubleshooting tool for Java/.NET Get 100% visibility into your production application - at no cost. Code-level diagnostics for performance bottlenecks with <2% overhead Download for free and get started troubleshooting in minutes. http://p.sf.net/sfu/appdyn_d2d_ap1 _______________________________________________ Libmesh-users mailing list Libmesh-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libmesh-users
