The current version of femLego is exttended to runs on parallel computer with both distributed memory and share memory by using the Message Passing Interface (MPI) libary. This tool was successfully used to solve some problems in heat and mass transfer, materials science and free boundaries. A flowchart of femLego is shown in Figure 1. The mesh is partitioned using the ParMetis library. A Fortran core code takes care of the matrix assembling which is done in parallel. A matrix solution is obtained using the Aztec library. If adaptivity is switched on, the last computed results are used by an error criterion to indicate regions of high variation of variables, i.e., regions requiring finer mesh. A new mesh, adapted to the solution, will be generated for use at the next step. The new mesh is again partitioned by ParMetis and balanced using a smoothing function. And the process repeats until final time. To simplify implementation and coding for refinement/derefinement, STL (Standard Template Library) C++ is used.
http://www.mech.kth.se/~minh/femLegoPar/index.html
Dr. Minh Do-Quang
Phone: +46 8 790 7191
Fax: +46 8 796 9850
E-mail: minh@mech.kth.se
Mechanics, KTH
No comments:
Post a Comment