-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 > As a matter of fact I intend to use my own finite differences when it > comes to solving the PDEs; I have already done that part and its working > on arbitrary geometries. I was looking for a good way to parallelize my > data structures so that I could use what I have done so far in a > parallel environment. From what you have indicated it seems to me that > using p4est through deal interface doesn't affect the performance of > p4est. Besides my own code is in c++ so I might stick with deal and use > the wrappers. > > Just one quick question though. I have noticed that deal has a good > tutorial/manual for its code but I couldn't find similar thing for > p4est. Do you have to know where I can get one?
Currently there is no tutorial for p4est. There are comments in the .h files, there is the example/ subdirectory in the code with some useful programs, and the preprint linked from www.p4est.org with in-depth information. Using p4est usually comes down to understanding a handful of high-level forest operations (new, refine, coarsen, balance, partition, ghost), and then extracting information from the ghost layer to learn about the parallel neighborhood. All floating-point math logic needs to be implemented on top of p4est, so if you are not using deal, that is a fair amount of work. Currently the ghost layer is only one element thick. You may need more layers for finite differences, or other applications. In this case p4est would need to be extended, which for this particular feature would not be very hard, and I'd be willing to help. Carsten > On Fri, Dec 3, 2010 at 12:57 AM, Timo Heister > <[email protected] <mailto:[email protected]>> > wrote: > > Hi Mohammad, > > > So I was wondering if anyone could help me deciding between deal > II or p4est > > for this task. > > P4est does only supply you with the mesh management in parallel - > either directly or using it through the distributed::Triangulation > class in deal.II. The scalability between both approaches should be > similar. I guess your answer depends on if you want to continue using > your finite differences or if you want to use finite elements from > deal instead. > But even if you decide to implement your own finite differences you > could use the wrappers in deal for the mesh and the linear algebra. It > depends on if you like the c++ or the c interface more. :-) Using > p4est directly has the advantage of more functionality... > > > (I guess HYPRE could also be called through PETSc > > interface within deal II?) > > Yes, in the subversion there is a class called > PETScWrappers::PreconditionBoomerAMG which is the AMG from the Hypre > package. > > -- > Timo Heister > http://num.math.uni-goettingen.de/~heister > > > > On Fri, Dec 3, 2010 at 3:41 AM, Mohammad Mirzadeh > <[email protected] <mailto:[email protected]>> wrote: > > Dear all, > > I am a graduate student at the Mechanical Eng. Dept. in UC Santa > Barbara and > > I am working with Cartesian grids built on octree data structures. > I already > > have a functional sequential FDM code for solving PNP equations as > in ion > > dynamics for arbitrary geometries. I have the intention of > parallelizing my > > code using PETSc and HYPRE to handle the linear solver part but I > need to > > use a package to handle the parallelization of my grid data structure. > > Following on the suggestions made in the PETSc mailing-list I > decided to > > consider using either deal II or p4est packages. > > So I was wondering if anyone could help me deciding between deal > II or p4est > > for this task. I am aware that it is possible to access both PETSc > and p4est > > through deal II interfaces (I guess HYPRE could also be called > through PETSc > > interface within deal II?) I also know that p4est has a very good > > scalability up to 200k CPUs (although I may hardly ever go beyond > 1024). > > That being said, I almost only need to use either packages to > parallelize my > > data structure since on each CPU I could still use my sequential > functions. > > I would greatly appreciate if you could help me decide the right > way for > > doing this. > > Thanks, > > Mohammad > > > > ================================================= > > Ph.D Candidate, > > CASL Group and Squires Group, > > Department of Mechanical Engineering, > > University of California Santa Barabra, > > Santa Barbara 93106-5070, CA > > ================================================= > > _______________________________________________ > > dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii > > > > > > > > > _______________________________________________ > dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.10 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iEYEARECAAYFAkz5ikoACgkQRClUBgYbo/nn0gCgj0pLePt2nFkeP8I8E3ojsZgJ tSsAoJ6E1t9E2PuolTz1SL6UaragWflU =j8q7 -----END PGP SIGNATURE----- _______________________________________________ dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
