Overbook in Advance, Trade in Future: Computing Resource Provisioning in Hybrid Device-Edge-Cloud Networks

TL;DR

This paper proposes a novel resource trading mechanism called OATF for hybrid device-edge-cloud networks, enabling overbooking and forward contracts to improve resource utilization and efficiency in IoT systems.

Contribution

It introduces OATF, a new overbooking-promoted trading mechanism that manages uncertainties and enhances resource utilization across device, edge, and cloud layers.

Findings

OATF improves resource utilization and time efficiency.

Mutually beneficial utilities for all parties achieved.

Effective overbooking rate design and risk management demonstrated.

Abstract

The big data processing in distributed Internet of Things (IoT) systems calls for innovative computing architectures and resource provisioning techniques to support real-time and cost-effective computing services. This article introduces a novel overbooking-promoted forward trading mechanism named Overbook in Advance, Trade in Future (OATF), where computing resources can be traded across three parties, i.e. end-users, an edge server and a remote cloud server, under a hybrid device-edge-cloud network with uncertainties (e.g., "no shows"). More importantly, OATF encourages a feasible overbooking rate that allows the edge server to overbook resources to multiple end-users (e.g., exceed the resource supply), while purchasing backup resources from the cloud server, by determining rights and obligations associated with forward contracts in advance via analyzing historical statistics (e.g., network, resource dynamics). Such a mechanism can greatly improve time efficiency and resource utilization thanks to overbooking and pre-signed forward contracts. Critical issues such as overbooking rate design and risk management are carefully investigated in this article, while an interesting case study is proposed with mathematical analysis. Comprehensive simulations demonstrate that OATF achieves mutually beneficial utilities for different parties (cloud, edge, and end-users), as well as substantial resource usage and commendable time efficiency, in comparison with conventional trading methods.