In this paper we investigate the use of parallel computing to deal with the high computational cost of numerical algorithms for model reduction of large linear descriptor systems. The state-space truncation methods considered here are composed of iterative schemes which can be efficiently implemented on parallel architectures using existing parallel linear algebra libraries. Our experimental results on a cluster of Intel Pentium processors show the performance of the parallel algorithms.
We discuss a parallel library of efficient algorithms for model reduction of largescale
systems with state-space dimension up to O(104). We survey the numerical
algorithms underlying the implementation of the chosen model reduction methods.
The approach considered here is based on state-space truncation of the system
matrices and includes absolute and relative error methods for both stable and unstable
systems. In contrast to serial implementations of these methods, we employ
Newton-type iterative algorithms for the solution of the major computational tasks.
Experimental results report the numerical accuracy and the parallel performance of
our approach on a cluster of Intel Pentium II processors.
We describe a prototype web service for model reduction of very large-scale linear systems, with
dimension in the order of millions of states, that includes a user-friendly interface designed so that the computation
can be easily performed via the HTTP protocol. Access via a web browser isolates the user of the service from the
complexities of installing and using the parallel model reduction codes and the maintenance of the hardware. In case
the routines are found to be appropriate for the problem the user needs to solve, the library can be then downloaded
and installed on the user’s own computing resources.
This paper illustrates the major issues of the access procedure by means of graphical examples, and describes
the structure and implementation of the remote model reduction service. The service is offered in a cluster of Linux
machines.
We discuss an efficient algorithm for optimal Hankel norm approximation of large-scale
systems and an implementation which allows to reduce models of order up to O(104) using
parallel computing techniques. The major computational tasks in this approach are the
computation of a minimal balanced realization, involving the solution of two Lyapunov
equations, and the additive decomposition of a transfer function via block diagonalization.
We will illustrate that these computational tasks can all be performed using iterative
schemes for the matrix sign function. Numerical experiments on a cluster of Linux PCs
show the efficiency of our methods.