What is ffddm ?#
In the acronym
ff stands for FreeFem++ and
ddm for domain decomposition methods. The idea behind ffddm is to simplify the use of parallel solvers in FreeFem++: distributed direct methods and domain decomposition methods.
Parallelism is an important issue because, since about 2004, the clock speed of cores stagnates at 2-3 GHz. The increase in performance is almost entirely due to the increase in the number of cores per processor. All major processor vendors are producing multicore chips and now every machine is a parallel machine. Waiting for the next generation machine does not guarantee anymore a better performance of a software. To keep doubling performance parallelism must double. It implies a huge effort in algorithmic development.
ffddm, FreeFem++ users have access to high-level functionalities for specifying and solving their finite element problems in parallel. The first task handled by
ffddm is the data distribution among the processors. This is done via an overlapping domain decomposition and a related distributed linear algebra. Then, solving a linear system is possible either via an interface to the parallel MUMPS solver or by using domain decomposition methods as preconditioners to the GMRES Krylov method. The
ffddm framework makes it easy to use scalable Schwarz methods enhanced by a coarse space correction built either from a coarse mesh or a GenEO (Generalized Eigenvalue in the Overlap) coarse space, see also the book An Introduction to Domain Decomposition Methods: algorithms, theory, and parallel implementation. State-of-the-art three level methods are also implemented in
ffddm framework is entirely written in the FreeFem++ language and the '.idp' scripts can be found here. It makes it also a very good tool for learning and prototyping domain decomposition methods without compromising efficiency.
ffddm can also act as a wrapper for the HPDDM library. HPDDM is an efficient implementation of various domain decomposition methods and a variety of Krylov subspace algorithms, with advanced block and recycling methods for solving sequences of linear systems with multiple right-hand sides: GMRES and Block GMRES, CG, Block CG, and Breakdown-Free Block CG, GCRO-DR and Block GCRO-DR. For more details on how to use HPDDM within
ffddm, see here METTRE LE LIEN
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
macro dimension 2// EOM // 2D or 3D include "ffddm.idp" mesh Th = square(50,50); // global mesh // Step 1: Decompose the mesh ffddmbuildDmesh( P , Th , mpiCommWorld ) // Step 2: Define your finite element macro def(u) u // EOM macro init(u) u // EOM ffddmbuildDfespace( P , P , real , def , init , P2 ) // Step 3: Define your problem macro grad(u) [dx(u), dy(u)] // EOM macro Varf(varfName, meshName, VhName) varf varfName(u,v) = int2d(meshName)(grad(u)'* grad(v)) + int2d(meshName)(1*v) + on(1, u = 0); // EOM ffddmsetupOperator( P , P , Varf ) PVhi ui, bi; ffddmbuildrhs( P , Varf , bi ) // Step 4: Define the one level DD preconditioner ffddmsetupPrecond( P , Varf ) // Step 5: Define the two-level GenEO Coarse Space ffddmgeneosetup( P , Varf ) // Step 6: Solve the linear system with GMRES PVhi x0i = 0; ui = PfGMRES(x0i, bi, 1.e-6, 200, "right"); ffddmplot(P, ui, "u") Pwritesummary
This example solves a Laplace problem in 2D in parallel with a two-level GenEO domain decomposition method. To try this example, just copy and paste the script above in a file 'test.edp' and run it on 2 cores with
$ ff-mpirun -np 2 test.edp -glut ffglut
Explanations are given in:
|Quick introduction to domain decomposition methods and
|More about Domain Decomposition||here|
|Complete Documentation of ffddm||here|