SuperBS: A Methodology for Resource Management in Heterogeneous Wireless Networks Ilias Tsompanidis Ahmed H. Zahran Cormac J. Sreenan Computer Science Department Electronics and Electrical Communications Department Computer Science Department University College Cork Cairo University University College Cork Cork, Ireland Cairo, Egypt Cork, Ireland Email: [email protected] Email: [email protected] Email: [email protected] Abstract—Resource management in heterogeneous wireless heuristics however, provide sub-optimal solutions that are not networks has been approached from various angles by the guaranteed to perfectly adhere to the policy. research community. The complexity of the network and the heterogeneity of clients make the conclusive comparison of the Deriving from our work in URM, this paper introduces a various resource allocations challenging. This work introduces process for comparing the performance of resource manage- superBS, an approach defining a theoretical optimal resource ment algorithms (RMA) for heterogeneous wireless networks. allocation that adheres to the required resource management The idea of superBS is presented, a single virtual Base Station policies and can be used as a reference f o r the performance (BS) that supersedes all available BSs in the real network. The of considered algorithms, mitigating the heterogeneity of the superBS theoretically serves all clients at once, considering system. T w o applications that leverage superBS are developed, an a number of parameters affecting the performance of the implementation of a heuristic resource management algorithm, and the enhancement of a popular fairness metric with support multi-BS network. It provides an upper limit to the total f o r clients of different classes and traffic demands. Simulations utility, ideally differentiates the allocated throughput of clients demonstrate the performance of superBS and the proposed that belong to different classes while maintaining Max-Min algorithms. fairness among clients of the same class. This allocation method assures that there is a single maximum throughput Keywords—Heterogeneous Wireless Network, Network Selec- in a class, achieved by all clients unless their demand is tion, Priority Class, Resource Management, Heuristic, Fairness lower. The superBS resource allocation vector can be used as a benchmark, i.e. a normalisation factor, for comparing and I. INTRODUCTION grading the performance of RMAs. Traditional cellular service providers are increasingly e x - This work makes two additional contributions. First, it panding their networks with new Radio Access Technologies presents Ideal-aware Resource Management (IRM), a novel (RAT), such as femtocells, LTE and WiFi hotspots, creating heuristic RMA based on superBS theoretical allocation. IRM partially overlapping Heterogeneous Wireless Networks. Most approaches the optimal resource allocation problem with linear modern client devices are also equipped with multiple wireless complexity, in contrast to URM’s analytic exponential com- interfaces, enabling them to connect to any RAT operated plexity, while adhering to the intra-class fairness and inter-class by the Wireless Service Provider (WSP). However, they are differentiation requirements. Second, it discusses the compli- configured to preferably attach to the faster RAT, e v e n if cations in defining fairness in a heterogeneous wireless net- this causes congestion, while leaving the slower ones under- work. The well-known fairness metrics have difficulty dealing utilised. In our previous work, we described Utility-based with different traffic demands and service differentiation. Jain Resource Management (URM), an approach that differentiates Fairness considering the superBS resource allocation vector as the offered service level between different classes of clients, optimal is shown to be able to cope with such client, traffic, and utilises all available wireless resources to achieve network- and network heterogeneity. wide provision of comparable level of service to users of the same class regardless of their attachment point, while offering II. RELATED WORK premium service level to users of higher classes [1]. Utility-based resource allocation in telecommunications In order to ensure that the network resources stay optimally networks was first presented by Kelly [3], where a Network utilised over time, URM has to be run in short intervals. Utility Maximisation (NUM) problem is formulated to e x - However, its computational complexity deems it unfit for press the source rates, link capacities and design goals of real-time operation and precipitates the need for heuristic the modeled network. NUM is used by many researchers algorithms that follow the same intra-class fairness, inter- to model a number of different resource allocation problems class differentiation policy. URM maximises the total per-client and network protocols. A survey paper [4] summarises the utility, a logarithmic function of the allocated throughput, theories, algorithms and applications that derive from NUM, different for every class and mathematically bound to follow including research on resource management packet level dy- the defined policy and provide Max-Min fairness [2]. The namics, mainly focusing on stohastic wireless network models. Bellavista et al. [9], in an extensive survey paper, provide a 978-1-4673-2480-9/13/$31.00 ©2013 IEEE 1290 TABLE I. INDEX OF TERMS number of partially overlapping heterogeneous BSs, nomadic Term Symbol users requesting connections at the beginning of their sessions, Client i 2 (1; ··· ; n) Base Station (BS) s 2 (1; ··· ;S) and that a number of parameters are known to the optimising Priority class of Client i Pi entity. The optimal allocation offers the same throughput to Priority class tuning parameter αi clients of the same class and differentiates between users of Bandwidth allocated for Client i B at the associated BS i different classes across all BSs controlled by the Wireless Throughput of Client i Ti Service Provider (WSP). In order to achieve this optimal List of visible BS to Client i Li allocation, the WSP uses the utility function to quantify the Bandwidth of BS s Cs value of the throughput offered to the clients of each class. Maximum achievable throughput estimation of Client i associated to BS s Tci;s We defined a single optimisation problem, that combines the Spectral Efficiency of Client i R BS selection and the resource allocation subproblems, solved associated to BS s i;s Maximum and minimum Demand by maximising the total per-client utility. As we reported in Dmax;Dmin of Client i (BS-independent) i i [1], the utility function used for URM is: Theoretical Bandwidth allocated for Client i B at superBS i;sBS e − 1 Theoretical Throughput of Client i at superBS Ti;sBS fPi (Ti) = αi ln Ti + 1 ; (1) αi where Ti is the throughput offered to client i and αi is a classification model for resource management and network se- class-specific tuning parameter to change the curvature of the lection algorithms. They identify three first-level classification utility function, different for each priority class Pi. Table I lists directions, namely management scope, evaluation process, and all the optimisation parameters and notation used in the rest continuity management. of this paper. The objective is to maximise the sum of user utility (eq. 2), and thus, optimise the intra-class fairness and Resource allocation in wireless networks is inextricably inter-class differentiation across all BSs. entwined with the notion of fairness, with many models defining different fairness types. The ones most commonly considered are max-min fairness, proportional fairness, utility- X Maximise U = fPi (Ti) (2) Ti based fairness, the popular Jain fairness index, the Gini index, i and other approaches based on the theory of majorization [3], [5]–[7]. It is common to extend these fairness metrics to better This work takes a closer look into the problem of band- fit specific resource allocation problems. For example, Dianati width management in a heterogeneous wireless network and et al. [8] demonstrate the need for an appropriate definition and more specifically examines the issues of optimality and fair- a clear methodology for quantization of fairness, and propose ness. URM showed that the identification of the optimal a fairness index suitable for a single-hop, single cell wireless resource allocation is computationally demanding, hindering network. The effect that different types of objective functions real-time operation. This paper considers the same problem by in NUM-based resource management have on fairness has been reducing its dimensions and contributes three computationally questioned in recent research. Collucia et al. [2] identify and feasible applications. It first defines superBS, a simplified view study a number of objective function families that can be used of the network that considers a theoretical single-BS scenario to achieve max-min fairness. and estimates the optimal allocation of the clients, used for Corci et al. [10] exploit the different mobility profiles of benchmarking. Second, it implements an efficient heuristic cellular clients. They demonstrate the benefits of simplifying resource management algorithm using the superBS outcome. the mobility management support for low-mobility clients. Third, it discusses the notion of fairness and augments Jain In- Heterogeneity in the context of competing WSPs is also a dex with superBS to support client and demand heterogeneity. popular research topic. Previous