Date of Award

4-2012

Document Type

Thesis

Department

Computer Science

First Advisor

Dr. Azad Azadmanesh

Second Advisor

Dr. Harvey Siy

Third Advisor

Dr. Robin Gandhi

Fourth Advisor

Dr. Andrew W. Swift

Abstract

In a Service Oriented Architecture (SOA), the hierarchical complexity of Web Services (WS) and their interactions with the underlying Application Server (AS) create new challenges in providing a realistic estimate of WS performance and reliability. The current approaches often treat the entire WS environment as a black-box. Thus, the sensitivity of the overall reliability and performance to the behavior of the underlying WS architectures and AS components are not well-understood. In other words, the current research on the architecture-based analysis of WSs is limited.

This dissertation presents a novel methodology for modeling the reliability and performance of web services. WSs are treated as atomic entities but the AS is broken down into layers. More specifically, interactions of WSs with the underlying layers of an AS are investigated. One important feature of the research is investigating the impact of dynamic parameters that exist at the layers, such as configuration parameters. These parameters may have negative impact on WSs performance if they are not configured properly. WSs are developed in house and the AS considered is JBoss AS. An experimental environment is setup so that controlled service requests can be generated and important performance metrics can be recorded under various configurations of the AS. On the other hand, a simulation model is developed from the source code and run-time behavior of the existing WS and AS implementations. The model mimics the logical behavior of the WSs based on their communication with the AS layers. The simulation results are compared to the experimental results to ensure the correctness of the model. The architecture of the simulation model, which is based on Stochastic Petri Nets (SPN), is modularized in accordance to the layers and their interactions. As the web services are often executed in a complex and distributed environment, the modularized approach enables a user or a designer to observe and investigate the performance of the entire system under various conditions. In contrast, most approaches to WSs analyses are monolithic in that the entire system is treated as a closed box.

The results show that 1) the simulation model can be a viable tool for measuring the performance and reliability of WSs under different loads and conditions that may be of great interest to WS designers and the professionals involved; 2) Configuration parameters have big impacts on the overall performance; 3) The simulation model can be tuned to account for various speeds in terms of communication, hardware, and software; 4) As the simulation model is modularized, it may be used as a foundation for aggregating the modules (layers), nullifying modules, or the model can be enhanced to include other aspects of the WS architecture such as network characteristics and the hardware/operating system on which the AS and WSs execute; and 5) The simulation model is beneficial to predict the performance of web services for those cases that are difficult to replicate in a field study.

Comments

A DISSERTATION Presented to the Faculty of The Graduate College at the University of Nebraska In Partial Fulfillment of Requirements For the Degree of Doctor of Philosophy. Copyright 2012 Cobra Rahmani.

Files over 3MB may be slow to open. For best results, right-click and select "save as..."

COinS