Elsevier

Computer Networks

Volume 115, 14 March 2017, Pages 29-41
Computer Networks

Auction-based resource allocation in OpenFlow multi-tenant networks

https://doi.org/10.1016/j.comnet.2017.01.010Get rights and content

Abstract

In this paper, we investigate the allocation of network resources (such as FlowTable entries and bandwidth) in multi-tenant Software-Defined Networks (SDNs) that are managed by a FlowVisor. This resource allocation problem is modeled as an auction where the FlowVisor acts as the auctioneer and the network Controllers act as the bidders. The problem is analyzed by means of non-cooperative game theory, and it is shown that the auction admits a unique Nash Equilibrium (NE) under suitable conditions. Furthermore, a novel distributed learning procedure is provided that allows each Controller to reach the game’s unique NE in a few iterations by exploiting only locally available information. An implementation in OpenFlow-compliant SDNs is also proposed in a way that exploits native procedures already offered by OpenFlow. Finally, simulation results show that the proposed auction-based resource management scheme leads to significant improvements in network performance (for instance, achieving gains of up to 5 × reduction in transmission delays).

Introduction

Multitenancy is a concept referring to the possibility for several customers to share certain resources such as physical network elements and links and use them as they were the sole users of those resources. In the last years, increased attention has been paid to the application of the multitenancy concept in the networking domain. Even if most efforts have been focused on the application of multitenancy concepts to the datacenter domain [1], [2], in many other scenarios such concepts can be exploited beneficially. Two relevant scenarios, for example, are that of virtual network operators in which several companies sell network access services using the network infrastructure owned by a third party [3], and that of virtualized network functions where hardware network elements, such as switches and firewalls, are substituted by software middleboxes that execute networking procedures on one or more virtual machines (VMs) deployed on the cloud [4].

In multi-tenant scenarios, the owner of the network infrastructure has two major needs:

  • Maximize the quality of service experienced by its customers, that is, render its customers satisfied;

  • Maximize its revenue.

In order to meet both of them, efficient resource management mechanisms should be considered. Recently, thanks to their capability to provide dynamic network management [5], [6], [7], [8], software-defined network (SDNs) have attracted much interest in the literature as a reliable framework to provide support for multitenancy [9], [10], [11], [12], [13] and energy-efficiency [14], [15]. In SDNs, control and data planes are decoupled. Network control and management are centralized and implemented in software, while the data/forwarding plane consists of an underlying physical network composed by several SDN-compliant switches and links. Although there are several ways to implement SDNs, in this paper we consider OpenFlow [16] as the most popular implementation thereof. In fact, as we show later, OpenFlow specifications already provide procedures to support dynamic resource allocation in multi-tenant networks.

In a SDN, multiple networks can coexist; thus, to properly manage their interactions, OpenFlow provides a FlowVisor [17], which is a high-level controller that is designed to act as a proxy between the physical network and multiple customers. By exploiting FlowVisor protocols, OpenFlow fully supports the multitenancy principle. In fact, FlowVisor and OpenFlow together allow the network owner to divide the network resources into slices and give full control of each slice to one customer that, to this purpose, runs a software program referred to as Controller. OpenFlow and FlowVisor ensure isolation between slices and therefore, each Controller can use its share of the network resources as if it were the sole controller doing so. In the following, we will identify the network owner with the FlowVisor and its customers with the corresponding Controllers.

In this paper we address the case where the FlowVisor reserves a portion of the network resources and divides it among the Controllers that compete with each other to obtain such resources. Our problem formulation is general and can be applied to several resource allocation problems in SDN scenarios. However, for illustration purposes we focus on two relevant resource allocation problems where each Controller competes to obtain either additional space in OpenFlow routing tables, i.e., Flow Tables, to store its routing policies, or bandwidth on a certain network link to improve its achievable throughput. It is worth noting that both Flow Tables and bandwidth are scarse resources in many OpenFlow applications. On the one hand, Flow Tables are implemented in finite capacity memories. On the other hand, bandwidth is well-known to be limited on the network links. Accordingly, efficient assignment of such resources is of extreme importance.

To this end, in line with a large body of literature on the design of distributed resource management techniques we consider auctions as the allocation instrument.

In this perspective, the FlowVisor acts as the auctioneer while the Controllers act as the bidders. Periodically the FlowVisor starts a new auction to sell a certain amount of network resources and each Controller makes a bid. Controllers determine their bids based on their interest in the network resources, i.e. the object of the auction. The FlowVisor then assigns each Controller a portion/share of the network resources which is proportional to the submitted bid.

In this context, Controllers behave selfishly: each of them aims at maximizing a utility function which is the difference between the benefit from the obtained resource and the effective cost (monetary and otherwise) for obtaining it. Obviously,

  • The Controller’s benefit increases with the amount of the obtained share;

  • The Controller’s cost, in general, increases commensurately with the placed bid.

Given that Controllers have conflicting interests, the above scenario can be modeled using non-cooperative game theory. In this paper we provide such a game-theoretic formulation and show that the resulting game admits a unique Nash equilibrium (NE) for a wide class of benefit and cost functions which are specific to the SDN scenarios. Such states represent a stable operation point of the system where no Controller has an incentive to deviate –and thus disrupting the system equilibrium. Also, we provide a novel exponential learning scheme which converges to the NE and we show that such convergence occurs within a few iterations of the auction. We also provide extensive numerical results that provide significant insight into the considered scenario.

Motivated by this analysis, we provide an implementation of the auction-based resource allocation mechanism which is based on procedures and commands that are already provided by OpenFlow, and we show that the proposed mechanism leads to significant improvement in network performance. Specifically, in the Flow Table auction which is specifically addressed in SDN scenarios because of the working mechanism at the Controller, our approach increases the probability to find a matching rule in the Flow Table by a constant term equal to 0.2 in many cases. Instead, in the bandwidth auction, the proposed auction-based resource allocation mechanism reduces transmission delays up to 5 times if compared to those experienced when static resource allocation policies are considered.

The rest of this paper is organized as follows. An overview of the related work is provided in Section 2. In Section 3 we introduce the proposed auction game in a multi-tenant OpenFlow scenario. The auction game is described and studied in Section 4, where a distributed learning scheme to reach the unique NE of the auction game is also presented. In Section 5 we illustrate numerical results that provide useful insights on both the auction game dynamics and the learning scheme. In Section 6 a discussion on implementation issues of the proposed auction-based resource allocation scheme in OpenFlow networks is presented; also, an implementation of the auction game is provided with extensive simulation results which show the effectiveness of the proposed approach. Finally, Section 7 concludes the paper. All relevant system parameters and variables are summarized in Table 1.

Section snippets

Related work

The problem of resource allocation and management is crucial in any networking environment and a large body of literature exists on the subject spanning all types of networks, e.g., [18], [19], [20]. Obviously, resource allocation is a major issue in software defined networks as well [21]. In this context, the possibility to support multi-tenancy raises a new set of issues regarding resource management [22], [23]. In fact, the necessity emerges to identify a tradeoff between the needs of

System model

Our model consists of a physical network of switches operating under an OpenFlow framework that enables software-defined networking. The behavior of OpenFlow switches is completely determined by the contents of a Flow Table (or several), whose entries specify flows and how packets of these flows must be treated. More specifically, each entry contains three sections:

  • Rules: this is used to match incoming packets to existing flows. In fact, this entry contains a set of header values against which

System stability and learning

In this section, we examine the existence and uniqueness of the Nash equilibrium and propose an exponential learning mechanism through which the bidders can converge to the equilibrium.

Theorem 1 NE existence and uniqueness

The gameG always admits a unique NE.

Proof

Note first that each player’s payoff function ui is concave in bi being it defined as the difference between a concave and a convex function; thus, existence of NE follows from the general theory of Rosen [47]. To prove the uniqueness of the NE, we define the affine benefit

Numerical examples

In this section, we provide some relevant numerical results that illustrate the dynamics of our auction-based resource management scheme under different scenarios and cost functions. We consider N=30 Controllers whose interest factors are uniformly distributed at random in the interval [0, 1] so as to consider the worst-case scenario where the entropy is maximized, and thus also the unpredictability. Also, unless explicitly stated otherwise, we set the maximum admissible bid of each Controller

The OpenFlow auction in action

In this section, we focus on a practical implementation of the auction framework in an OpenFlow-based SDN where the auctioned resource is assumed to be the available bandwidth on a given link. More in detail, Section 6.1 summarizes the main procedures and operations needed to properly implement the auction framework by exploiting OpenFlow features and mechanisms; in Section 6.2 we present extensive simulation results that show how auction-based bandwidth allocation can effectively improve the

Conclusions

In this paper, we have analysed an auction-based scheme for the allocation of network resources in multi-tenant Software Defined Networks. In particular, it is assumed that a FlowVisor manages the resource allocation mechanism by executing an auction of fractions of network resources. The bidders are the Controllers who compete to gain additional network resources at a cost which is related to their interest in getting a particular network resource. The resulting scenario is modeled and studied

Salvatore D’Oro received the B.S. degree in Computer Engineering and the M.S. degree in Telecommunications Engineering degree both at the University of Catania in 2011 and 2012, respectively. He received the PhD degree from the University of Catania in 2015. In 2013 and 2015, he was a Visiting Researcher at Université Paris-Sud 11, Paris, France and at Ohio State University, Ohio, USA. He is currently a postdoctoral research fellow at the University of Catania. In 2015, he organized the 1st

References (50)

  • W. Li et al.

    Efficient allocation of resources in multiple heterogeneous wireless sensor networks

    Parallel Distrib. Comput. J.

    (2014)
  • B.P. Rimal et al.

    A taxonomy and survey of cloud computing systems

    Proceedings of the Fifth International Joint Conference on INC, IMS and IDC

    (2009)
  • J. Mudigonda et al.

    NetLord: A scalable multi-tenant network architecture for virtualized datacenters

    Proceedings of the ACM SIGCOMM 2011 Conference

    (2011)
  • C.S. Mower, M.A. Palmer, S.C. Mayhew, Networking as a service: delivering network services using remote appliances...
  • R. Yu et al.

    Network function virtualization in the multi-tenant cloud

    IEEE Netw.

    (2015)
  • S. Shenker et al.

    The future of networking, and the past of protocols

    Open Networking Summit

    (2011)
  • H. Jin et al.

    OpenFlow-based flow-level bandwidth provisioning for CICQ switches

    IEEE Trans. Comput.

    (2013)
  • N. Handigol et al.

    Plug-n-Serve: load-balancing web traffic using OpenFlow

    ACM SIGCOMM Demo

    (2009)
  • D. Tuncer et al.

    Adaptive resource management and control in software defined networks

    Netw. Service Manage. IEEE Trans. on

    (2015)
  • L. Sun et al.

    A network management solution based on OpenFlow towards new challenges of multitenant data center

    Proceedings of the 9th Asia-Pacific Symposium on Information and Telecommunication Technologies (APSITT)

    (2012)
  • A. TaheriMonfared et al.

    Multi-tenant network monitoring based on software defined networking

    On the Move to Meaningful Internet Systems: OTM 2013 Conferences

    (2013)
  • Y. Shin et al.

    The study on configuration of multi-tenant networks in SDN controller

    Proceedings of the 16th International Conference on Advanced Communication Technology (ICACT)

    (2014)
  • Z. Bozakov et al.

    AutoSlice: automated and scalable slicing for software-defined networks

    Proceedings of the 2012 ACM CoNEXT Student Workshop

    (2012)
  • M. Banikazemi et al.

    Meridian: an SDN platform for cloud network services

    IEEE Commun. Mag.

    (2013)
  • R. Bolla et al.

    Fine-grained energy-efficient consolidation in sdn networks and devices

    Netw. Service Manage. IEEE Trans. on

    (2015)
  • G. Faraci et al.

    An analytical model to design and manage a green sdn/nfv cpe node

    Netw. Service Manage. IEEE Trans. on

    (2015)
  • N. McKeown et al.

    Openflow: enabling innovation in campus networks

    ACM SIGCOMM Comput. Commun. Rev.

    (2008)
  • R. Sherwood et al.

    FlowVisor: A Network Virtualization Layer

    Technical Report

    (2009)
  • Z. Dziong

    ATM network resource management

    (1997)
  • I. Akyildiz et al.

    Mobility and resource management in next generation wireless systems

    Sel. Areas Commun. IEEE J.

    (2001)
  • I. Akyildiz et al.

    Research challenges for traffic engineering in software defined networks

    Network, IEEE

    (2016)
  • M. Casado et al.

    Virtualizing the network forwarding plane

    Proceedings of ACM PRESTO 2010

    (2010)
  • P. Skoldstrom et al.

    Network virtualization and resource allocation in openflow-based wide area networks

    Proceedings of IEEE ICC 2012

    (2012)
  • G. Schaffrath et al.

    Network virtualization architecture: proposal and initial prototype

    Proceedings of ACM VISA 2009

    (2009)
  • A. Fischer et al.

    Virtual network embedding: a survey

    Commun. Surv. Tut. IEEE

    (2013)
  • Cited by (0)

    Salvatore D’Oro received the B.S. degree in Computer Engineering and the M.S. degree in Telecommunications Engineering degree both at the University of Catania in 2011 and 2012, respectively. He received the PhD degree from the University of Catania in 2015. In 2013 and 2015, he was a Visiting Researcher at Université Paris-Sud 11, Paris, France and at Ohio State University, Ohio, USA. He is currently a postdoctoral research fellow at the University of Catania. In 2015, he organized the 1st Workshop on COmpetitive and COoperative Approaches for 5G networks (COCOA), and served on the Technical Program Committee (TPC) of the CoCoNet8 workshop at IEEE ICC 2016. In 2013, he served on the Technical Program Committee (TPC) of the 20th European Wireless Conference (EW2014).

    Laura Galluccio received the Laurea Degree in electrical engineering in 2001 and the Ph.D. degree in electrical, computer, and telecommunications engineering in 2005 from the University of Catania, Italy. Since 2002 she was with the Italian National Consortium of Telecommunications (CNIT), working as a Research Fellow in the FIRB VICOM and NoE SATNEX projects. Since 2010, she has been an Assistant Professor with the University of Catania. In 2005 she was Visiting Scholar with the COMET Group, Columbia University, New York. Her research interests include unconventional communication networks, software defined networks, and network performance analysis. She serves in the editorial boards of Elsevier Ad Hoc Networks and Wiley Wireless Communications and Mobile Computing journals.

    Panayotis Mertikopoulos graduated valedictorian from the Physics Department of the University of Athens in 2003. He obtained the M.Sc. and M.Phil. degrees in Mathematics from Brown University, USA, in 2005 and 2006 respectively, and his Ph.D. degree from the University of Athens in November 2010. During 2010–2011, he held a post-doctoral fellowship at École Polytechnique, Paris, France. Since 2011, he has been with the French National Center for Scientific Research (CNRS) and the Laboratoire d’Informatique de Grenoble. PM received the best paper award at NETGCOOP 2012 and was an Embeirikeion Foundation Fellow between 2003 and 2007. He has also served as TPC co-chair for WiOpt ’14, general co-chair of AlgoGT ’13, publications chair for WiOpt ’13 and ValueTools ’12, and is on the TPC of numerous conferences on game theory and wireless networks. Since 2014, the PI is also a member of the steering committee of the Optimization and Decision Theory branch of the Société de Mathé matiques Appliquées et Industrielles (SMAI). His main interests lie in optimization, game theory, dynamical systems, and their applications to telecommunication networks.

    Giacomo Morabito received the laurea degree and the PhD in Electrical, Computer and Telecommunications Engineering from the Istituto di Informatica e Telecomunicazioni, University of Catania in 1996 and 2000, respectively. From November 1999 to April 2001, he was with the Broadband and Wireless Networking Laboratory of the Georgia Institute of Technology as a Research Engineer. Since April 2001 he is with the Dipartimento di Ingegneria Informatica e delle Telecomunicazioni of the University of Catania where he is currently Associate Professor. His research interests focus on analysis and solutions for wireless networks and Internet of Things.

    Sergio Palazzo received the degree in electrical engineering from the University of Catania, Catania, Italy, in 1977. Since 1987, he has been with the University of Catania, where is now a Professor of Telecommunications Networks. In 1994, he spent the summer at the International Computer Science Institute (ICSI), Berkeley, as a Senior Visitor. In 2003, he was at the University of Canterbury, Christchurch, New Zealand, as a recipient of the Visiting Erskine Fellowship. His current research interests are in modelling, optimization, and control of wireless networks, with applications to cognitive and cooperative networking, SDN, and sensor networks. Prof. Palazzo has been serving on the Technical Program Committee of INFOCOM, the IEEE Conference on Computer Communications, since 1992. He has been the General Chair of some ACM conferences (MobiHoc 2006, MobiOpp 2010), and currently is a member of the MobiHoc Steering Committee. He has also been the TPC Co-Chair of some other conferences, including IFIP Networking 2011, IWCMC 2013, and European Wireless 2014. He also served on the Editorial Board of several journals, including IEEE/ACM Transactions on Networking, IEEE Transactions on Mobile Computing, IEEE Wireless Communications Magazine, Computer Networks, Ad Hoc Networks, and Wireless Communications and Mobile Computing.

    This work has been supported by the European Commission in the framework of the FP7 Network of Excellence in Wireless COMmunications NEWCOM# (Grant agreement no. 318306), and by the French National Research Agency (ANR) grant NETLEARN (contract no. ANR-13-INFR-004).

    View full text