Article Preview
Top1. Introduction
In traditional cloud computing systems, computing and storage resources are often virtualized in a manner of virtual machine (VM), which enables users to deploy their non-trivial applications among multiple physical machines, or accommodate several application on a single power server (Lemay et al., 2012; Narmanlioglu and Zeydan, 2017). As to networking resources, they are still be managed through the conventional approaches in most of real-world cloud systems (Li et al., 2013; Claypool et al., 2014). Only in recent years, network virtualization technology has attracted sufficient attentions and is gradually becoming the ideal approach to overcoming the well-known problem in current Internet, that is, the current Internet has fierce resistance of the well-established architecture to accommodate the vigorous need for testing and deploying new network protocols, technologies and applications (Heisswolf et al., 2013; Caggiani-Luizelli et al., 2016).
Network virtualization technology allows multiple dedicated virtual networks (VNs) to efficiently share the underlying physical substrate network (SN) resources. In this way, a VN user can have a fully dedicated network, with full administrative control during its lifetime (Fukushima et al., 2013; Alaluna et al., 2019). Meanwhile, network virtualization ensures complete isolation and transparency among the co-existing VNs, which enable resource providers to achieve the desired security and management scalability while relieving the infrastructure provider from the detailed management responsibility (Su et al., 2013; Demirci and Ammar, 2014; Vijayakumar et al., 2018). For example, in a cloud platform, infrastructure providers can create a set of VNs that host various services to serve end-users. In this scenario, the IaaS providers (SPs) are responsible for allocating the appropriate substrate resources to construct the VNs, and the most challenging task facing an SP is to find an appropriate mapping between the nodes and links in the requested VN topology and the substrate nodes and links or paths, respectively. The objective of the SP is to find such a mapping while maximizing its net profit from selling the SN resources (Hammad et al., 2014; Zhang et al., 2014). Unfortunately, solving this problem is generally difficult if not impossible (Hsu and Shieh, 2013; Esposito et al., 2016).
In this study, we take efforts on the resource pricing mechanism when using network virtualization in cloud environments. To overcome the demerits of existing price mechanisms in terms of efficiency and fairness, we present a game-based pricing model, in which virtual resource configuration and provision among VNs and SNs are defined as a two-phrase gaming model. In this gaming model, a cooperative gaming model is applied to optimize the resource benefits, while a non-cooperative gaming model is used to balance the user’s costs and provider’s benefits. To model price and resource usage sensitivity of the VN requests, we present a new dynamic VN model that employs utility functions to quantify the users’ value of the VN resource demands.
The rest of this paper is organized as follows. Section 2 presents the related work. In section 3, we describe the framework of the proposed gaming model and their solutions; In section 4, experiments are conducted to investigate the effectiveness of the proposed model. Finally, Section 5 concludes the paper with a brief discussion of the future work.