Thanks for the hints. Just a question... It seems that you prefer the Nemesis format compared to the native xdr xda, is there any reason for that? I don't know if I will have time to do it in the near future because the end of my PhD is approaching very fast. Anyway I think that parallel IO for mesh and solution is crucial to run parallel (>100 processes) computations, and of course ghosted vector is crucial too (thanks Tim).
Lorenzo Il giorno 21/ott/2009, alle ore 17.47, Derek Gaston ha scritto: > 1) In certain circumstances.... but probably not huge. > > 2) Yes > > 3) You can read and write parallel xda/r files.... although I > haven't done it myself. BUT... it looks like that is only possible > when writing the entire equation system not just the mesh. I don't > know if there is a way around that. Maybe Ben will chime in. > > Although: > > 4) Currently libMesh does support _reading_ of Nemesis meshes. > These are Exodus meshes that have been _pre_ decomposed so that each > piece of the mesh resides in a different file and each processor > only reads it's part of the mesh. I can verify that that does work > with ParallelMesh (although a few of the more advanced things in the > Exodus reader aren't supported yet). > > However... the _writing_ of Nemesis meshes / solutions isn't > supported yet... so to get output you still have to do the old > serialization. I hope to change that in the not too distant > future... but it continues to slip down my TODO list. > > If you are interested in helping out with getting Nemesis writing > going you can get the documentation here: > http://sourceforge.net/projects/exodusii/files/ > . Once we have Nemesis read / write capability we will have a > complete end-to-end parallel capability. > > Derek > > On Mon, Oct 19, 2009 at 9:30 AM, Lorenzo Botti > <[email protected]> wrote: > I'm planning to perform a lot of parallel runs (without AMR) on a big > cluster and I would like to know a couple of details about the > parallel mesh > implementation. > 1) Is it faster than serial mesh? > 2) Does it use less memory per process? > > 3) Is it possible to read and/or write parallel .xdr/.xda data files? > I'm curious about the possibility of making a mesh partitioning and > write > the single partition files so that every process involved in a > parallel > computation loads just one portion of the mesh. Does this approach > gives > some benefits? > > 4) Have you planned to introduce some parallel solution output > format? I > think that having to serialize the solution is quite an expensive > operation > for unsteady computations. > > Thanks a lot. > Lorenzo > ------------------------------------------------------------------------------ > Come build with us! The BlackBerry(R) Developer Conference in SF, CA > is the only developer event you need to attend this year. Jumpstart > your > developing skills, take BlackBerry mobile applications to market and > stay > ahead of the curve. Join us from November 9 - 12, 2009. Register now! > http://p.sf.net/sfu/devconference > _______________________________________________ > Libmesh-users mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/libmesh-users > ------------------------------------------------------------------------------ Come build with us! The BlackBerry(R) Developer Conference in SF, CA is the only developer event you need to attend this year. Jumpstart your developing skills, take BlackBerry mobile applications to market and stay ahead of the curve. Join us from November 9 - 12, 2009. Register now! http://p.sf.net/sfu/devconference _______________________________________________ Libmesh-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-users
