Date of Award
8-25-2011
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Computer Science
First Advisor
Kenji Yoshigoe
Abstract
The sophistication of the devices through which a man shapes his world are the tool-marks of his development upon history. One device currently shaping the lives of man is the High Performance Computer (HPC) system. HPC systems have been described as ``a game changing technology (King and BusinessWeek, 2010).'' The reason for HPC systems to be declared ``a game changing technology (King and BusinessWeek, 2010)'' comes from the fact that it is cheaper for a manufacturer to conduct thousands of tests in virtual space, than it is to try multiple experimental designs in the physical world. Through capabilities of HPC systems, the manufacturer can minimize material consumption, maximize efficiencies, material combinations, strengths and testing conditions in an effort to provide you with a better product at the cheapest costs available. HPC systems are ``changing the game'' by allowing manufacturers to supply superior goods and services to you at seriously reduced costs. However, HPC Systems are conglomerations of hardware components and software bundles, each having their own unique set of strengths, capabilities and speed of operation. If there is ever going to be an efficient use of available HPC system resources, a well-defined method for identifying the impacts of each unique set of hardware or software specifications to the performance of the overall HPC system. Such a method must be able to implement various hardware and software capability sets in every conceivable combination and configuration, in order to evaluate the impacts of each. This type of in-depth analysis can best be achieved through the development of an HPC simulator. In conformance to the goal of maximizing efficient utilization of all HPC system resource commodities, this paper describes the following evaluations through multiple ``live'' and ``simulated'' HPC system configurations: New method for generating jobs using High Performance Linpack New method of testing the performance of ``live'' HPC system New method for evaluating the performance of a ``live'' HPC system New comparison of the workload analysis from ``live'' HPC system in two different CPU configurations New method for modeling an HPC system from workload analysis New design for an HPC simulator New components of an HPC simulator New method of implementing workload analysis in the HPC simulator New Results of comparison ``live vs. simulated'' HPC systems
Recommended Citation
Hurst, William Bradley, "Modeling and Simulation of HPC System Through Job Scheduling Analysis" (2011). Theses and Dissertations. 287.
https://research.ualr.edu/etd/287
