Parallel algorithm to solve one million DOF eigenvalue problem by PCs
Seung Wan Kim, Chul Cho · 19th AIAA Applied Aerodynamics Conference · 2001
The eigenvalue analysis requires large computing time and memory. And everincreasing demand for more precise analysis of the complex systems like aerospace vehicles is requiring bigger size of eigen solutions. In this work, an efficient method to solve large eigenproblems is developed by common Internet PCs. To overcome the difficulties - slow communication speed, small memory and etc., appropriate algorithm and techniques needed to be developed. As an affordable way, the block Lanczos procedure and parallel multifrontal technique is adopted and improved for computing efficiency. A number of example problems, which are appearing in the vibration analysis of structures, are solved by PCs connected through Internet. Eventually, one million DOF eigenproblem in aerospace structure is to be solved by 32 PCs during idling time like off-working hours. This work promises an alternative affordable supercomputing since the PCs are brought and maintained by the individual user without additive expenses.