8:30 AM-9:30 AM
Room: Texas Ballroom B
Chair: Robert Schreiber, Hewlett Packard Laboratories
With the increase in computer speed and capacity, industry can apply computer simulation not only for verification purposes, but also increasingly as an integrated step in the design and development cycle. Traditionally, industrial finite element codes have been quite problem specific and their algorithms for solving the linear equations have been tightly coupled to the rest of the code.
The use of state of art algorithms is increasingly more important with growing problem size and more complex computer architecture, but the traditional "in house" codes typically cannot afford the cost of frequent updates or even keeping up with the advances in (parallel) algorithms.
The PARASOL project, funded by The European Commission, has attempted to address a part of this problem by developing a library of algorithms for large sparse linear systems of equations. The library contains state of art methods, direct as well as iterative, and all algorithms should perform well on parallel computing platforms.
The speaker will describe overall library design, including the common interface to application codes. He will report on preliminary results from realistic, industry-generated test cases. He will discuss some of the design choices made as well as possible developments in the future.
Petter E. Bjørstad
University of Bergen, Norway