Date of Award

2017

Degree Type

Thesis

Degree Name

Master of Science in Electrical Engineering (MSEE)

First Advisor

Kunal Mankodiya

Abstract

The Internet of Things (IoT) architecture currently being implemented in commercial applications does not fully realize the potential of IoT devices. The IoT devices themselves lack the computational power to perform significant and real-time feedback for the data being gathered at edge devices. The current architecture places most of the computational burden on various cloud servers. This creates a performance bottleneck due to unpredictability in network latency and reliability.

The amount of data flowing through the cloud-to-things continuum will only continue to grow, increasing stress on the cloud and network. This type of bottleneck creates difficulties for devices, such as those in Smart Communities, that require reliable connectivity with in depth real-time feedback. These communities host a level of context that is unique to each of them and can be used to help maximize the usefulness of the local network while sending more contextually relevant information back to the cloud for deeper learning.

The master thesis research was aimed at developing, implementing, and evaluating a Fog Computing based IoT system. The deployment of the framework on several IoT devices will be set up with example data to be processed. These devices will be set up in a mesh topology with fog gateway devices. In this research, we developed three testbeds to test the fog computing architecture and its performance in IoT applications where smart communities are targeted. Each of the three different testbeds will run the most appropriate OS for the device and will be capable of managing communication with the things via Bluetooth, as well as providing access to a cloud service.

The aims is to evaluate the timing, CPU load, and memory load, and network measurements throughout each stage of the cloud-to-things continuum during an experiment for determining features from a finger tapping exercise for Parkinson’s Disease patients. It will be shown that there are limitations to the proposed testbeds when trying to handle upwards of 35 clients simultaneously. These findings lead us to an appropriate distribution of processing the leaves the Intel NUC as the most appropriate fog device. While the Intel Edison and Raspberry Pi find a better footing at in the edge layer, bridging communication protocols and maintaining a self-healing mesh topology for “thing” devices in the personal area network.

Share

COinS