On Mon, 26 Nov 2007, Benjamin Kirk wrote: > So I hear you two should be hammering the library now -- both the > Serial and Parallel Meshes... Right?
That's the plan. I'm swamped with other work, but I'm getting some big runs started in the lab tonight and I should be submitting jobs on Lonestar later this week. > I'm redoing the EquationSystems IO to be able to read/write restart files in > parallel on a distributed memory mesh without serializing it. This should > allow for a parallel simulation to dump restart files without ever > serializing, and then these restart files could be post-processed into > GMV/tecplot/whatever later. Sounds good. We'll eventually want parallel or chunked writes for the visualization formats, but for now "you need to have enough RAM for the serial mesh on the workstation where you postprocess" is a minor limitation compared to "you need to have enough RAM for the serial mesh and parallel matrix on every node where you run". > This is one major feature which I think is needed before 0.7.0, and of > course we need parallel Mesh IO as well. I agree. We need to be at the point where ParallelMesh is practical for a majority of users, not just a theoretical feature. > Depending on how the ES IO goes, I may or may not jump right in to the Mesh > IO. If not, what do you think about releasing 0.6.9 without it? There has > been *massive* code change in the library, and I'd like to get people using > it sooner rather than later. > > Thoughts? I'd say don't release 0.6.9 until I've had more time to try to shake out bugs with my application codes. That goes for the ES IO too once you're done; even if it's only useful on static meshes at first, I've got some uniform mesh runs that I can try it out with. After that, sure. I'm fine with making our users into guinea pigs once we're at least 99% certain there isn't a snake hiding in the new code. ;-) --- Roy ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2005. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ Libmesh-devel mailing list Libmesh-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libmesh-devel
