Date of Award

Fall 2005

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computational Analysis and Modeling

First Advisor

Chokchai Leangsuksun

Abstract

Cluster computing has been attracting more and more attention from both the industrial and the academic world for its enormous computing power, cost effective, and scalability. Beowulf type cluster, for example, is a typical High Performance Computing (HPC) cluster system. Availability, as a key attribute of the system, needs to be considered at the system design stage and monitored at mission time. Moreover, system monitoring is a must to help identify the defects and ensure the system's availability requirement.

In this study, novel solutions which provide availability modeling, model evaluation, and data analysis as a single framework have been investigated. Three key components in the investigation are availability modeling, model evaluation, and data analysis. The general availability concepts and modeling techniques are briefly reviewed. The system's availability model is divided into submodels based upon their functionalities. Furthermore, an object oriented Markov model specification to facilitate availability modeling and runtime configuration has been developed. Numerical solutions for Markov models are examined, especially on the uniformization method. Alternative implementations of the method are discussed; particularly on analyzing the cost of an alternative solution for small state space model, and different ways for solving large sparse Markov models. The dissertation also presents a monitoring and data analysis framework, which is responsible for failure analysis and availability reconfiguration. In addition, the event logs provided from the Lawrence Livermore National Laboratory have been studied and applied to validate the proposed techniques.

Download

Share

COinS