1991 Fiscal Year Final Research Report Summary
Development of Large - scale Linear Equation Solver for Parallel Computer
| Project/Area Number |
02555102
|
|---|
| Research Category |
Grant-in-Aid for Developmental Scientific Research (B)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
土木構造
|
|---|
| Research Institution | Okayama University |
|---|
Principal Investigator |
HIROSE Soh-ichi Okayama Univ., The Graduate School of Natural Science and Technology, Research Assistant, 工学部, 教授 (30026322)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
廣瀬 壮一 岡山大学, 大学院自然科学研究科, 助手 (00156712)
|
|---|
| Project Period (FY) |
1990 – 1991
|
| Keywords | Nested dissection / Parallel computer / Elimination method / Linear equation / Parallel computing / Sparse matrix |
|---|
| Research Abstract |
The aim of this investigation is the development of an efficient solver of large sparse set of linear algebraic equations for the parallel computer. Basi-C tool is the nested dissection method, and the method is modified so that it can be effective on a parallel computer.
(1)Theoretical study on the parallel nested dissection method
The nested dissection method is originally developed so that the amount of fill-in through the elimination process can be minimized. And, this process necessarily eliminates nodes which are not adjacent each other. This characteristic fits to the parallel computing.
(2)Development of the parallel nested dissection method
Two-CPUs were used for the development of the parallel nested dissection method. Developed code consists of 1)the master task, 2)the worker task, and 3)the configuration file. In order to give equal loading to two CPUS, the treated problems are divided into two, each of which is solved using one CPU. For the comparison of the efficiency of the proposed method, the Gauss elimination and also the conventional parallel elimination methods are used. The numerical tests show that the proposed method is the fastest among them.
(3)The generalization of the parallel nested dissection method
Proposed method may show good performance if the domain to be solved is subdivided into as many equal-sized subdomains as the number of CPUs being introduced.
|
