Abstract

Our society is increasingly evolving to rely on computer mechanisms that perform a variety of tasks. From a self-driving car to a satellite in space relaying data from Mars rovers, we need these systems to perform optimally and without failure. One such point of failure these systems can encounter is tactic volatility of an adaptation tactic. Adaptation tactics are defined workflows that allow systems to navigate their environment. Tactic volatility is the variance in the behavior in the attribute of a tactic, such as cost and latency and/or the combination of the two. Current systems consider these tactic attributes to be static. Studies have shown that not accounting for tactic volatility can adversely affect a system's ability to operate effectively and resiliently. To support self-adaptive systems and address their limitations, this paper proposes a Tactic Volatility Aware solution that utilizes eRNN (TVA-E) and addresses the limitations of current self-adaptive systems. For this research, we used real-world data that has been made available for use by researchers and academics. This data contains real-world volatility and helps us demonstrate the positive impact TVA-E when used in self-adaptive systems. We also employ the use of uncertainty reduction tactics and how they can assist in accounting for tactic volatility. This work will serve as an evaluation and a comparison of using different machine learning methods to predict and account for tactic volatility. We will study different predictive mechanisms in this paper: Auto-Regressive Moving Average(ARIMA), Evolving Recurrent Neural Network(eRNN), Multi-Layer Perceptron(MLP), and Support Vector Regression(SVR). These methods will be studied with our TVA-E process and we will analyze how they can enhance a self-adaptive system’s performance when it accounts for tactic volatility.

Library of Congress Subject Headings

Self-adaptive software--Quality control; Adaptive control systems--Quality control; Uncertainty

Publication Date

12-2021

Document Type

Thesis

Student Type

Graduate

Degree Name

Software Engineering (MS)

Department, Program, or Center

Software Engineering (GCCIS)

Advisor

Daniel E. Krutz

Advisor/Committee Member

Qi Yu

Campus

RIT – Main Campus

Plan Codes

SOFTENG-MS

Share

COinS