Summary: | 碩士 === 國立成功大學 === 機械工程研究所 === 84 === Due to the dramatic improvement in the workstation computation
speed,solving a large problem in workstation cluster is an
economical way.How to solve a large eigenvalue problem in
workstation cluster is a big challenge.Existing sequential
algorithms need to be modified or replaced by algorithms which
take advantage of the power flexibility of the new computer
systems. Since Ethernet , which is shared by many computers,
are used in the campus network, the data transfer rate between
the work stations are much lower than the computation speed in
the work stations. The extra data communication between
workstations must be considered when distributed finite element
calculations are used. PVM(Parallel Virtual Machine), a public
domain subprogram, will be used to transfer data between
workstations and to request the other workstations to do finite
element calculations. The ratio between the communication rate
and the computation speed will be measured first, and used as a
reference to distribute computation work. The eigenvalue
problem associated with structural vibrations is a major,
computationally intensive activity in large -scale finite
element calculations. The forward lanczos method is considered
as one powerful method to extract small number of eigenvalues
in a large eigenvalue problem. This a algorithm will be
implemented in a single workstation. Distributed computation on
workstations and the speed up will be studied too. Due to the
complexity of the geometry of the machine tool, to model the
dynamic characteristic of the machine tool needs a lot of
elements and huge computations. A machine tool will be built
and workstation cluster will be used to calculate its natural
frequency and natural mode.
|