From: Karen L. <kyl...@gm...> - 2010-03-09 23:05:14
|
I don't think I can get away without using MeshData. The reason I'm using libmesh is precisely that I have data on a given (huge) mesh that I am using as input... Without running in parallel then, do you think the problem is too large to converge within a reasonable time frame? Karen On Tue, Mar 9, 2010 at 6:01 PM, Roy Stogner <roy...@ic...>wrote: > > > On Tue, 9 Mar 2010, Karen Lee wrote: > > Thanks! Looks like I was just missing some entries of my MeshData since I >> was combining 2 node files... >> > > Oh, but one more suggestion: if you can avoid using MeshData, do so. > It doesn't work with more than one MPI rank, it hasn't been tested > with threads, and since it hasn't had an active developer in years I'd > be wary of it in serial as well. The trouble is that our developers > have gotten in the habit of (and in some cases express strong reasons > for preferring) using ExplicitSystem solutions to store function data > and using subdomain ids to index into tables of discrete data. You > seem to be running problems large enough that parallelizing them will > be beneficial, and it might be an obstacle if you find yourself having > to rewrite the MeshData class before you could run in parallel. > --- > Roy > |