Sparse Scientific Applications: Improving Performance and Energy Characteristics on Advanced Architectures
Open Access
Author:
Lee, Ingyu
Graduate Program:
Computer Science and Engineering
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
December 04, 2006
Committee Members:
Padma Raghavan, Committee Chair/Co-Chair Mary Jane Irwin, Committee Member Mahmut Taylan Kandemir, Committee Member Zi Kui Liu, Committee Member
Keywords:
Performance Scientific Application Sparse Energy
Abstract:
In many scientific and engineering disciplines, computational modeling and simulation
is typically used to understand complex systems or physical phenomena, in addition to
the traditional approaches involving theory and experiment.
The models in many of these applications are described by partial differential equations (PDEs).
The numeric solution of such models using implicit or semi-implicit schemes depends on
efficient sparse linear system solution. In this thesis, we focus on enabling
faster and more reliable sparse linear system solution taking into account trends in
modern computer architecture.
Our contributions include the following:
(i) specialized sparse solution schemes for quantum nanomechanics,
and computational fluid dynamics simulations,
(ii) developing a new preconditioning scheme to accelerate the convergence of a
sparse iterative solver, and
(iii) characterizing the interactions between sparse solution schemes and
the features of advanced computer architectures to improve performance while
reducing system power and energy.