Novel Cloud Load Distribution Management and Deployment Techniques

Adewojo, Adekunbi

Abstract

Cloud computing provides scalable, flexible, and cost-effective computing resources. Cloud adoption has, however, introduced resource utilisation and service unavailability issues during the process of designing, deploying, and hosting cloud-native applications. These challenges can occur due to large, sudden, yet legitimate influxes of user requests, also known as flash crowds, and resource failures. Interactive web applications that experience sudden surges in user activity are particularly susceptible to these challenges. In this research, novel cloud algorithms and techniques were developed to address these challenges. An experimental approach was used to evaluate these novel cloud algorithms and techniques.

The main contribution of this research is the creation of an experimentally characterised novel weight-assignment load balancing algorithm that combines five carefully selected server metrics to determine the server's capacity to efficiently distribute the workload of three-tier web applications among application servers. A novel decentralised multi-cloud architecture and algorithm were also developed to distribute the workload of three-tier web applications using geographical and improved load distribution techniques.

In this research, a private OpenStack cloud was configured followed by a bespoke cloud experimental testbed, and finally, a heterogeneous multi-cloud experimental testbed to evaluate the novel algorithms. The experimental evaluation validated the novel algorithms against the baseline load distribution algorithms and techniques. The algorithms were implemented as a software service and were tested using the workload of an open-source cloud-hosted E-Commerce application.

The experiments carefully measured response times, scalability, number of errors, and throughput during flash crowds and resource failure scenarios. Results showed that the novel load balancing algorithms and architecture are more resilient to fluctuating loads and resource failures than baseline algorithms. For example, the novel single cloud load balancing algorithm improved the average response times by 12.5% when compared to the baseline algorithm and by 22.3% when compared to the round-robin algorithm in the flash crowds' situation.

The main conclusion of this research is that cloud-native application developers can mitigate the adverse effects of flash crowds and resource failures by using dynamic load balancing algorithms that carefully combine selected server metrics that affect a specific class of applications. Furthermore, the success of implementing and coordinating heterogeneous cloud infrastructure demonstrates the usability of the experimental testbed to evaluate cloud algorithms.

Citation

Adewojo, A. (2023). Novel Cloud Load Distribution Management and Deployment Techniques. (Thesis). University of Salford

• Thesis Type: Thesis
• Deposit Date: Jul 21, 2023
• Publicly Available Date: Oct 30, 2023
• Award Date: Sep 29, 2023