A Methodology for Optimized VM Scheduling in Small and Medium Fog Environments

Abstract

In the evolving landscape of cloud computing, fog computing has emerged as a critical paradigm designed to bridge the gap between end-user devices and remote cloud datacentres. Driven by the increased usage of the Internet of Things (IoT) and other low-capability devices, there has been a notable surge in network traffic from end-user devices. This increase in traffic has degraded the quality of service (QoS) for edge users, mostly impacting real-time applications. Fog computing acts as an intermediary layer between edge devices and the cloud, aiming to enhance QoS by processing complex tasks closer to the users, thus reducing reliance on distant cloud servers. Within this paradigm, small fog datacentres face difficulties in managing new workloads due to their limited resources. These datacentres are managed by small fog service providers (SFSPs) with limited resources in terms of money and capacity. As a result, this limitation leads to the starvation of virtual machines (VM) and reduced competitiveness with larger cloud providers.

While numerous studies have explored resource limitations relevant to small fog datacentres, the challenges of horizontal scalability and VM starvation remain unresolved. Moreover, some proposed solutions suggest increasing small fog datacentres' capacity by relying on the cloud or other datacentres. While the former undermines the fundamental purpose of fog computing, the latter requires the formation of complicated agreements. Therefore, the current literature lacks a framework for horizontally scaling SFSPs in real time using resources within the fog itself.

To solve these pressing issues, this thesis presents an innovative approach for horizontally scaling the size of SFSPs by using volunteer computing to find the best placement for high-priority VMs. This solution leverages physical devices located in the vicinity of the SFSP, which act as supplementary nodes to accommodate the overflow from the SFSP's core resources. These external devices, offering their computing resources voluntarily, are integrated into the SFSP's datacentre through a recruitment process that ensures they meet specific performance and reliability standards. The selection of the volunteer nodes for VM hosting is governed by a multi-criteria decision-making (MCDM) process. This approach allows the SFSP to evaluate multiple factors, such as node performance, reliability, and cost, to make informed decisions about where to allocate VMs. Our evaluations demonstrate the validity of our proposed horizontal scaling approach and a significant decline in VM starvation by employing volunteer nodes in an SFSP.