Large-scale systems of nonlinear difference equations are well suited for implementations on a parallel vector processor. This report de scribes the fast solution methods encompassed by Project LINK, including implementation on an IBM 3090-200VF supercomputer. LINK comprises 79 macroeconometric models emulating the economies of countries or regions linked through international trade. The model, with more than 20,000 equations, is solved by nesting a Jacobi iteration procedure with a combined Gauss- Seidel and Newton-Raphson procedure. The so lution performance, in terms of elapsed time, can be improved almost proportionally to the number of parallel processors. One can take advantage of vector processing at several stages of the solution algorithm. By exploiting the dynamics of the model, it is possible to gain a factor of 2 or more in speed from vectorization. Furthermore, the availability of large computer memory has significantly reduced the time spent on disk input/output.
|Original language||English (US)|
|Number of pages||23|
|Journal||International Journal of High Performance Computing Applications|
|State||Published - Jan 1 1987|
All Science Journal Classification (ASJC) codes
- Theoretical Computer Science
- Hardware and Architecture