Doctor of Philosophy
The Internet of Things (IoT) is a concept in which physical objects embedded with sensors, actuators, and network connectivity can communicate and react to their surroundings. IoT applications connect physical objects for the purpose of decision making by sensing and analysing generated data from the embedded sensors in physical objects. IoT applications are growing rapidly as sensors become less expensive. Sensors generate large amounts of data that may meaningless unless the data is used to derive knowledge with in a certain period of time. Stream processing paradigm is used by IoT to provide requirements of IoT applications. In a stream processing paradigm, unlike traditional data bases, data is not stored but rather processed as it is generated. To transfer generated data from distributed data sources to a processing center such as cloud may not allow for real-time processing due to the network delay. Another high-demand application is live streaming of video. The performance of live video stream systems is inferior when there is a sudden large demand in the number of users. This thesis addresses some of the limitations of current architectures for video streaming systems and IoT applications based on the use of nearby computing resources (e.g., cloudlet, fog).
First, we addressed the degrading performance in video stream systems when a flash crowd occurs. The performance of video streaming systems is affected by flash crowd and degrade the quality of service for subscribers to the content delivery system. A flash crowd happens when there is a sudden large increase in the number of users. Therefore, flash crowds increase network traffic for any particular server. The main challenge is to make sure that the video streaming system has sufficient capacity to handle the occurrence of flash crowds.
Second, we address the limitation of current architectures for running mobile applications by introducing a dynamic partitioning and offloading of a mobile application. Mobile devices have limited resources including short battery life, storage capacity and processor performance. This limits the applications that can run on it. Mobile applications can be partitioned so that some of the application runs on a cloud. This works well for applications with relatively little data to be transferred and that do not have a high level of interaction with the user. Challenges with applications that have large amounts of data to be transferred and have a high level interactiveness is the high latency incurred by the network and packet loss of the wireless network. A mobile application can be partitioned so that part of it runs on a nearby computing resource e.g., fog node or cloudlet. This thesis presents a framework that introduces fine-grained offloading approach and support for runtime and dynamic partitioning of an application.
Third, we present a solution for placement of stream operators over distributed fog nodes for live processing of data streams from geographically distributed data sources. This placement of stream operators takes place in such a way that it supports applications with a high volume of data that require real-time (or near real-time) analysis To this end, this thesis proposed a set of algorithms for placement of stream operators among fog nodes.
Bayat, Navid, "Partitioning and Offloading for IoT and Video Streaming Applications that Utilize Computing Resources at the Network Edge" (2018). Electronic Thesis and Dissertation Repository. 5937.