| Popis: |
Summarization: In this paper, which extends and concludes our work in [3, 4], we deal with the problem of efficiently organizing the computation involved in the bi-conjugate gradient stabilized [5] iterative solution of large linear systems, in order to compute on a distributed memory parallel computer. These linear systems are associated with the discreti zation of elliptic boundary value problems using a collocation finite element method based on Hermite bi-cubic elements. The organization of the computation involved follows the methodology used in [3,4]. That is, having the collocation matrix in its normal 2-cyclic or red-black form for increased scalability, we map the computation involved on an appropriate virtual architecture which, in the sequel, is optimally repartitioned and mapped on to a fixed size architecture. The whole process leads to the development of an efficient and well-structured parallel algorithm. For demonstration purposes the algorithm was realized on a distributed memory Parsytee Cognitive computer CC-2. Παρουσιάστηκε στο: 4th European Conference on Numerical Mathematics and Advanced Applications |
