Dissertation Defense Announcement for Amna Shahid – 06/14/2023 at 12:00 PM

Dissertation Title:  

Resource Optimization of Edge Servers dealing with Priority-Based Workloads by utilizing Service Level Objective-Aware Virtual Rebalancing 

When: 06/14/2023  12:00 PM

Where: Simrall Hall Room 228 

Candidate: Amna Shahid

Degree: Master of Science in Electrical and Computer Engineering

Committee Members:

Dr. Samee Khan
(Major Professor)

Dr. Chaomin Luo
(Committee Member)

Dr. Yu Luo
(Committee Member)

 Abstract:

 The advent of the Internet of Things (IoT) provides a favorable platform for the profitable communication between devices that possess both sensors and actuators and the cloud. The limited accessibility of Edge data to Cloud analytics caused by network lag has posed a hindrance to the widespread adoption of real-time analytics. VRebalance, a virtual resource orchestrator aimed at ensuring end-to-end performance for priority-based workloads in concurrent stream processing at the Edge, presents a viable solution to address this issue. The prioritization of workloads is subjected to Bayesian Optimization (BO) within the context of VRebalance, in order to identify resource configurations that are close to optimal and enable efficient and adaptable resource management. In this study, the Apache Storm platform was employed as the stream processing engine in conjunction with the real-time Internet of Things (IoT) benchmark tool known as RIoTBench. These tools were utilized to assess the effectiveness of the VRebalance system. The outcomes of the study demonstrate that VRebalance exhibits superior efficacy compared to traditional methodologies, successfully fulfilling Service Level Objective (SLO) targets in the face of system variations. As compared to a hill climbing algorithm, VRebalance exhibited a substantial decrease in SLO violation rates, achieving a reduction of almost 30% for static priority-based workloads and 52.2% for dynamic priority-based workloads. Furthermore, it is noteworthy that the utilization of VRebalance resulted in a significant decrease of 66.1% in the rate of Service Level Objective (SLO) violations, as compared to the default resource allocation mechanism of Apache Storm.