KeLP Applications and Transitions

A number of applications use the KeLP library, or the underlying techniques.

1. Cell microphysiology. The MCell-K simulator enables large scale simulations of cell physiology at unprecedented scales and is a parallel variant of the successful MCell simulator. This is joint work with Terry Sejnowksi and Tom Bartol, Jr. who are with the Computational Neurobiology Laboratory at The Salk Institute. Further information is available on the MCell-K website.
2. Scalable free space Poisson solver. We have developed a fast 3-D Poisson solver based on a method of local corrections algorithm, called SCALLOP. This solver currently scales to 1000 IBM SP processors, and our ultimate target is to run on 4096 processors this year. This is joint work with Phil Colella in the Applied Numerical Algorithms Group at Lawrence Berkeley National Laboratory, and is part of an NPACI Alpha project involving an Immersed Boundary Method for modeling biological systems (with Kathy Yelick at the University of California, Berkeley). Further information is available on the SCALLOP website.
3. Structured adaptive mesh refinement. KeLP defines what an API should look like for handling communication in a way that encapsulates the details of low level message passing, e.g. MPI. As a proof of this concept the techniques embodied in KeLP have been transitioned into CHOMBO, and library for structured adaptive mesh refinement developed in the Applied Numerical Algorithms Group at Lawrence Berkeley National Laboratory. This is joint work with Phil Colella and Brian van Straalen, and is supported by NPACI.
4. Parallel reservoir simulation.   Mary Wheeler and co-workers at  the University of Texas at Austin  are using KeLP to implement UTPROJ3D, a single phase flow code for porous media. UTPROJ3D employs mortar spaces to provide clean numerical coupling between blocks, which generally are not aligned along the Manhattan directions. The problem is complicated by the fact that when we include chemical transport, different numerical computations may be carried out on different blocks which complicates load balancing. KeLP facilitates the development and enhancement of this application by managing the complexity of the underlying irregular representation. For more information, see the url http://king.ticam.utexas.edu/NPACI/IPARS_KELP_DAGH. This project is supported by NPACI as an Alpha project.
5. Turbulence simulation.  KDISTUF employs Direct Numerical Simulation to model turbulence and is based on a 10,000 line legacy fortran code called DISTUF. We are using KDISTUF modernized a 10,000  to simultaneously explore: new algorithms for handling chemical reaction,   high performance animation, and algorithmic formulation for realizing communication overlap. This is joint work with M.S. student Bill Kerney, and Dr. Keiko Nomura of the UCSD MAE Dept. along with Ph.D. students Tamara Grimmet and Peter Diamessis.  For an interesting visualization carried out by Nicole Borde see the URL http://vis.sdsc.edu/research/smallturublence.html. This effort is supported by the National Science Foundation.
6. Out-of-core KeLP library. KelpIO is a high-level C++ I/O library for application I/O,checkpointing, snapshoting, and out-of-core execution for programs written in the KeLP programming system. It was written by Bradley Broom and Robert Fowler with the Telescoping Compiler project headed by Ken Kennedy.  This library allows users of  KeLP to easily write and optimize I/O using the same high-level, abstract paradigm they use with KeLP. Moreover, it allows users to quickly convert in-core applications developed with KeLP into out-of-core applications, using a KelpIO out-of-core derivative of KeLP's XArrayX construct.   For more information see the URL  http://www.cs.rice.edu/~dsystem/kelpio/
7. Real space adaptive code for local spin density computations arising in first principles simulation of real materials. At 30,000 lines, this is the longest KeLP application we know of. The code was written by Scott Kohn with the Center for Applied Scientific Computing at Lawrence Livermore National Laboratory, and is part of a collaboration involving John Weare (Chemistry, UCSD), Beth Ong (CASC, LLNL), and Ryoichi Kawai (Physics, U. Alabama, Birmingham). For more information see the URL http://www.cse.ucsd.edu/groups/hpcl/scg/Research/first.html. This project was supported by the National Science Foundation.