Distributed Computing System Design and Operation Using Machine Learning

Complex adaptive systems include elements that function independently and, as a whole, to generate emergent behaviors. Distributed sensor and computer networks exhibit the properties of a complex adaptive system. When the number of networked electronic devices increases, the quantity of data to be processed and transferred likewise increases. Distributed data centers are required to satisfy the service delay requirements for time-sensitive applications such as real-time video processing. Although distributed computing reduces the cloud computing delay for such applications, a sub-optimal distributed computing infrastructure is unlikely to meet the stringent low-delay requirements. Existing cloud computing operations optimization methods are unsuitable for the optimization of fog infrastructures for new real-time applications because these methods require prior knowledge of network performance and end-user demands, which may not be available. Furthermore, numerous existing Task Allocation methods have been designed for predicable tasks with long execution times. However, these Task Allocation methods perform poorly in time-sensitive applications that offload tasks sporadically. This research proposes a method to optimize the fog node geo-distribution and computer resource placement, based on a series of operational tests involving Evolutionary Bayesian Optimization and a custom Cumulative Distance covariance kernel. In addition, to mitigate the limitations of the existing Task Allocation methods, this research proposes an asynchronous Task Allocation algorithm using Non-deterministic Deep Reinforcement Learning. EdgeCloudSim simulations performed using real police body-worn camera datasets show that the proposed methods outperform the state-of-the-art algorithms.

Author

• Aswin, Buddy

Language

• en

Date created

• 2021

Type of Work

• Dissertation

License

• All rights reserved

Rights statement

• In Copyright

GW Unit

• Systems Engineering

Degree

• Ph.D.

Advisor

• Etemadi, Amir

Committee Member(s)

• Holzer, Thomas
• Blackford, Joseph P
• Mazzuchi, Thomas
• Etemadi, Amir
• Sarkani, Shahram

Persistent URL

•  https://scholarspace.library.gwu.edu/etd/bv73c139h