Refine
Year of publication
- 1997 (3) (remove)
Document Type
- ZIB-Report (3)
Language
- English (3)
Has Fulltext
- yes (3)
Is part of the Bibliography
- no (3)
Institute
- ZIB Allgemein (3)
We study the parallelization of linearly--implicit extrapolation codes for the solution of large scale PDE systems and differential algebraic equations on distributed memory machines. The main advantage of these algorithms is that they enable adapativity both in time and space. Additive Krylov--Schwarz methods yield high parallel perfomance for such extrapolation methods. Our approach combines a slightly overlapping domain decomposition together with a polynomial block Neumann preconditioner and a reduced system technique. Furthermore we get important advantages through the explicit computation of the matrix--products of the preconditioner and the matrix of the linear system. The parallel algorithms exhibit scalability up to 64 processors already for medium--sized test problems. We show that the codes are really efficient in large application systems for chemical engineering problems.
Benchmarking of ODE methods has a long tradition. Several sets of test problems have been developed and new problems are still collected. So, a whole variety of problems can be used to check the efficiency of a method under investigation. In general, efficiency is measured by the amount of work whi ch is necessary to get a reliable solution for a prescribed accura cy. In order to quantify the term ``amount of work'' usually not only the computing time is measured but also the number of calls of functional units. To quantify the term ``quality of a numerical solution'' usually the $l_2$--norm of the true error at the final point of the integration interval is used. In our contribution we first discuss some general aspects of benchmarking. Then we present a new test frame which allows to solve typical benchmark problems with some of the state of the art integrators within a unified framework. Finally we show some results of our benchmark tests. Part of the test frame can be used interactively in the World Wide Web.
We study the parallelization of linearly--implicit extrapolation methods for the solution of large scale systems of differential algebraic equations arising in a method of lines (MOL ) treatment of partial differential equations. In our approach we combine a slightly overlapping domain decomposi tion together with a polynomial block Neumann preconditioner. Through the explicit computation of the matrix products of the pre conditioner and the system matrix a significant gain in overall efficiency is achieved for medium--sized problems. The parallel algorithm exhibits a good scalability up to 32 proces sors on a Cray T3E. Preliminary results for computations on a workstation cluster are reported.