A novel software-defined networking approach for vertical … · 2016-07-05 · Wirel. Commun. Mob....

WIRELESS COMMUNICATIONS AND MOBILE COMPUTINGWirel. Commun. Mob. Comput. (2016)

Published online in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/wcm.2690

RESEARCH ARTICLE

A novel software-defined networking approach forvertical handoff in heterogeneous wireless networksLi Qiang1, Jie Li2*, Yusheng Ji3 and Changcheng Huang4

1Graduate School of Systems and Information Engineering, University of Tsukuba, Tsukuba, Japan2 Faculty of Engineering, Information and Systems, University of Tsukuba, Tsukuba, Japan3 Information Systems Architecture Research Division, National Institute of Informatics, Tokyo, Japan4 Department of Systems and Computer Engineering, Carleton University, Ottawa, ON, Canada

ABSTRACT

We propose a novel vertical handoff scheme with the support of the software-defined networking technique for hetero-geneous wireless networks. The proposed scheme solves two important issues in vertical handoff: network selection andhandoff timing. In this paper, the network selection is formulated as a 0-1 integer programming problem, which maximizesthe sum of channel capacities that handoff users can obtain from their new access points. After the network selection pro-cess is finished, a user will wait for a time period. Only if the new access point is consistently more appropriate than thecurrent access point during this time period, will the user transfer its inter-network connection to the new access point. Ourproposed scheme ensures that a user will transfer to the most appropriate access point at the most appropriate time. Com-prehensive simulation has been conducted. It is shown that the proposed scheme reduces the number of vertical handoffs,maximizes the total throughput, and user served ratio significantly. Copyright © 2016 John Wiley & Sons, Ltd.

KEYWORDS

vertical handoff; software-defined networking; integer programming problem; total throughput; user served ratio

*Correspondence

Jie Li, Faculty of Engineering, Information and Systems, University of Tsukuba, Japan.E-mail: [email protected]

1. INTRODUCTION

Heterogeneous wireless networks integrate a variety ofwireless techniques to provide ubiquitous services [1].In heterogeneous wireless networks, users may need totransfer their inter-network connections from one accesspoint to another. The transferring operation among dif-ferent kinds of access points is called vertical handoff[2]. There are two important issues [3] needed to besolved in vertical handoff: network selection and hand-off timing. The network selection issue is to select anaccess point, to which the inter-network connection shouldbe transferred. The handoff timing issue is to determinewhen the inter-network connection transferring shouldbe implemented. The emergence of software-defined net-working (SDN) technique [4] makes it possible to solvethese two issues of vertical handoff in a novel perspec-tive. SDN is a new networking paradigm, which pro-vides a global centralized control of access points. Inthis paper, we make use of this feature of SDN andstudy the vertical handoff problem for heterogeneouswireless networks.

Consider the scenario shown in Figure 1. There is ageneral heterogenous wireless network environment thatconsists of Wi-Fi, LTE, WiMAX and 3G. A user walksto the company from home. When the user is at home,his (or her) smart phone connects to the Wi-Fi in thehouse. After the user goes out of home, his (or her) smartphone may connect to LTE, or WiMAX of the publiclibrary, or 3G. There are three available networks. Thisuser needs to know which one should be selected (net-work selection issue) [5]. After a new access point isselected, this user also needs to know when the inter-network connection should be transferred to the new accesspoint (handoff timing issue) [6]. Solutions to the networkselection and handoff timing issues compose the verticalhandoff scheme [7].

Because of lack of the global view, most of existingvertical handoff schemes failed to be global optimal. Theemergence of SDN [8] technique provides a chance tobreak this limitation. An SDN controller has an abstractedcentralized control of access points [9]. We make useof this feature of SDN, and propose a novel verticalhandoff scheme named software-defined networking based

Copyright © 2016 John Wiley & Sons, Ltd.

A demonstration of the wireless communications and mobile computing class file L. Qiang et al.

Figure 1. Illustration of vertical handoff.

vertical handoff (S-DNVH). In S-DNVH, users are dividedinto two classes: non-handoff users and handoff users.Non-handoff users will stay in the connections with theircurrent access points. While handoff users will send theirhandoff requests to an SDN controller. The network selec-tion is formulated as a 0-1 integer programming problem[10] by the SDN controller, with the objective of maxi-mizing the sum of channel capacities that handoff userscan obtain from their new access points. After the networkselection process is finished, handoff users have to wait fora certain time period [11]. After the time period, the SDNcontroller evaluates the performances of current accesspoints and new access points for users. If the new accesspoints are consistently more appropriate than the currentaccess points during the time period, users will transfertheir inter-network connections to the new access points.The main contributions of this paper are summarizedas follows:

� To the best of our knowledge, we are the first toapply the SDN technique in the study of vertical hand-off problem for heterogeneous wireless networks. Weinvestigate the architecture of SDN, and design acompatible vertical handoff procedure.

� We formulate the network selection issue as a 0-1integer programming problem based on our previouswork [12] and propose a network selection algo-rithm to solve it. In the proposed network selectionalgorithm, an SDN controller will allocate the mostappropriate access point for each user.

� We propose a handoff timing algorithm to determinethe time when the network selection results shouldbe implemented. Based on limited information andsimple calculation, the proposed handoff timing algo-rithm can predict the movement directions of users.Only if users are certain to move away from their

current access points, will the network selectionresults be implemented.

� Through comprehensive experiments, we validate thatour proposed scheme significantly reduces the num-ber of vertical handoffs, maximizes the total through-put and user served ratio.

The rest of this paper is organized as follows. Section2 is the related work. Section 3 is the system descrip-tion and problem formulation. Sections 4 and 5 presentthe proposed scheme for network selection and handofftiming issues, respectively. Section 6 is the performanceevaluation. Section 7 concludes this paper.

2. RELATED WORK

Here, we introduce the most typical related works in thissection. Wang et al. [11] proposed an interesting policy-enabled vertical handoff scheme. The network selec-tion is made by users. Periodically, users collect currentnetwork conditions as the inputs of their policy mod-ules. The policy module makes trade-offs among networkconditions, cost, power consumptions, and other crite-ria, then determines the “best" networks. In order toavoid the handoff instability problem, a random hand-off timing mechanism is further proposed. After wait-ing for a stability period, users hand off to those net-works, which have sustained high-performance during thestability period.

A fuzzy logic-based network selection scheme is pro-posed by J. Hou et al. [13], in which three input fuzzyvariables that include the probability of a short interrup-tion, the failure probability of handover to radio, and thesize of unsent messages are considered. At first, these threefuzzy variables are fuzzified and converted into some inputfuzzy sets by a singleton fuzzifier. Then the input fuzzy setsare mapped into output fuzzy sets by an algebraic prod-

Wirel. Commun. Mob. Comput. (2016) © 2016 John Wiley & Sons, Ltd.DOI: 10.1002/wcm

L. Qiang et al. A demonstrationof the wireless communications and mobile computing class file

uct operation. Finally, the output fuzzy sets are defuzzifiedinto a crisp decision point, which indicates a networkselection result.

Lee et al. [14] carried out research from the aspectof optimization, and proposed an enhanced group hand-off scheme. In enhanced group handoff scheme, each userevaluates its available access points on the remaining band-width. The network selection is formulated as a convexoptimization problem. The objective of formulated prob-lem is to minimize the handoff blocking probability. Byusing the Karush–Kuhn–Tucker condition, the formulatedconvex optimization problem is solved and a new accesspoint can be determined. After the new access point isdetermined, a user transfers its inter-network connection tothe new access point after an adjusted delay.

An interesting scheme proposed by Ciubotaru et al. [15]is called smooth adaptive soft handoff algorithm (SASHA).In SASHA, a user obtains a weighted sum of various per-formance parameters, and calculates the quality of service(QoS) values of its available access points. The user allo-cates its traffic according to the QoS values. As the userleaves an access point and gets closer to another, the QoSvalue of the leaving access point gets lower, and the QoSvalue of the approaching access point gets higher. As aresult, traffic on the leaving access point is transferred tothe approaching access point gradually.

3. SYSTEM DESCRIPTION ANDPROBLEM FORMULATION

3.1. Network architecture

For the upcoming problem formulation, we first exploit thespecial network architecture of SDN. Traditionally, eachaccess point contains both a control plane and a data plane

[8]. The control plane decides whether a traffic flow isadmissible or not, and the route that the traffic flow shouldtraverse. The data plane forwards the traffic flow accord-ing to the decision made by the control plane. In the SDNarchitecture, an SDN controller separates control planesfrom data planes of access points and provides a central-ized control of these access points. The SDN controllercommunicates with access points via OpenFlow [16] andhas a global view of the network environment. This featureof SDN gives us an opportunity to design a global optimalvertical handoff scheme.

Normally, an SDN controller can manage thousands ofaccess points at the same time. When there are a large num-ber of access points in an area, several SDN controllerscan be deployed. Besides of this, in order to group man-age the access points, multiple SDN controllers are alsorequired so that each SDN controller can install a cus-tom policy in its own domain. Stallings [8] pointed outthe most common scenario is the numerous and nonover-lapping SDN domains scenario as shown in Figure 2. Aswe know, an SDN controller centrally controls the accesspoints in its domain [17]. In fact, logical centralized controlalso exists in the multiple SDN controllers scenario. TheInternet Engineering Task Force is currently working onthe SDNi protocol [18], which helps the SDN controllersto exchange information and gain the global information.Thus, if an optimal vertical handoff scheme is proposed forthe single SDN controller scenario, the proposed schemealso works for the multiple SDN controllers scenario. Theextension work is just to add an “information exchange"stage at the beginning of proposed scheme. In this paper,we will consider the scenario with single SDN controller.

Figure 2. A network architecture of software-defined networking.


A demonstrationof the wireless communications and mobile computing class file L. Qiang et al.

3.2. Problem formulation

In this subsection, we formulate the vertical handoff prob-lem for heterogenous wireless network environment byusing the SDN technique [19]. Specifically, we consider aheterogenous wireless network environment that consistsof m access points. Let A be the set of access points,A D fa1, a2, : : : , amg. These access points support differ-ent wireless technologies. With the support of the media-independent handover (MIH) standard [20], all of theseaccess points can be centrally controlled by a single SDNcontroller. Access point ai (ai 2 A, i D 1, 2, : : : , m) hasli channels, ai equally divides its frequency band amongthese channels. The bandwidth of each channel in accesspoint ai is denoted by bi in MHz. If the access point ai isconnected by a user, this user will occupy one channel ofai [14]. Hence, the access point ai can serve at most li userssimultaneously.

Consider that there are n users. Let U be the set of users,U D fu1, u2, : : : , ung. If user uj (uj 2 U , j D 1, 2, : : : , n)is inside the coverage area of an access point, this accesspoint is called an available access point of user uj. Inheterogenous wireless network environment, uj may haveseveral available access points. However, uj can connectto at most one of its available access points at anytime.The connected available access point is called the currentaccess point of user uj. An adjacency matrix ı.t/ is used toreflect the relationship between access points and users attime t as follows.

u1 u2 � � � un

ı.t/ D

a1a2...

am

26664ı11.t/ ı12.t/ : : : ı1n.t/ı21.t/ ı22.t/ : : : ı2n.t/...

. . ....

ım1.t/ ım2.t/ : : : ımn.t/

37775

where

ıij.t/ D

�1, ai is connected by uj at time t,0, otherwise

(1)

Because most of the applications (e.g., music, video,and game) require higher download and lower uploadrates, users are more concerned about the quality of down-link (from access point ai to user uj). All links refer todownlink unless otherwise specified from now on. For con-venience, the notations used in this paper are summarizedin Table I.

For each access point-user pair�ai, uj

�, assume that the

received signal power of user uj from available accesspoint ai at time t is sij.t/ in watts. Let dij.t/ denote theEuclidean distance between access point ai and user uj attime t. Let pi denote the transmission power of access pointai in watts. Then the value of sij.t/ can be calculated asfollows [21].

sij.t/ D pi � hij � dij.t/�˛ (2)

Table I. Notation summary.

m Number of access pointsA Set of access points faig, i D 1, 2, : : : , mA0j Set of available access points for user uj , A0j � AXj Set of access points which transmit signal by using the same frequency band as user uj , Xj � Ali Number of channels in access point ai

bi Bandwidth of each channel in access point ai in MHzn Number of usersU Set of users fujg, j D 1, 2, : : : , nı.t/ Adjacency matrix of A and U at time t, that is,

�ıij.t/

��j Basic bandwidth requirement of user uj in Mbpssij.t/ Received signal power of user uj from access point ai at time t in wattsdij.t/ Euclidean distance between access point ai and user uj at time tpi Transmission power of access point ai for each channel in watts

hij Channel fading gain of channel�ai , uj

�˛ Pass loss exponentnij.t/ Additive white Gaussian noise power of channel

�ai , uj

�at time t in watts

qij.t/ Channel capacity of channel�ai , uj

�at time t in Mbps

oa Control overhead of the authentication phase in Mbpsor Control overhead of the reassociation phase in MbpsV.t/ Identifier vector of user set U at time t, that is, .vj.t//, j D 1, 2, : : : , nRj.t/ Channel capacities provided for user uj at time t, that is,

�rij.t/

�, i D 1, 2, : : : , m

R.t/ Channel capacities provided for user set U , that is,h�Rj.t/

�Ti, j D 1, 2, : : : , n

Fj.t/ Network selection result of user uj at time t, that is,�fij.t/

�, i D 1, 2, : : : , m

F.t/ Network selection results of user set U , that is,h�Fj.t/

�Ti, j D 1, 2, : : : , n



where the channel fading gain hij follows an exponen-tial distribution with rate � (hij � exp.�/), and thepass loss exponent ˛ > 2 (varies depending on thechannel conditions).

With the help of some techniques such as orthogo-nal frequency-division multiplexing (commonly applied toLTE, WiMAX and IEEE 802.11 b/g supported devices)[22] and code division multiple access (commonly appliedto 3G devices), the interference among users that belongto the same access point can be controlled in a low level.In order to further weaken this kind of interference, weassume that the orthogonal codes are used in code divisionmultiple access.

For the interference from other access points, let gxj.t/in watts be the interference caused by access point ax

(ax 2 A, x ¤ i) to user uj at time t, where ax transmits sig-nal by using the same frequency band as user uj. Let dxj.t/denote the Euclidean distance between access point ax anduser uj at time t. Let px and hxj denote the transmissionpower of access point ax in watts and channel fading gain,respectively. Then the value of gxj.t/ can be calculatedas follows:

gxj.t/ D px � hxj � dxj.t/�˛ (3)

Because the transmission powers of users are relativesmall, and the locations of users erratically change all thetime, the interference from other users is usually negligi-ble [23]. If the access point ai is connected by a user,thisuser will occupy one channel of ai. Hence, the accesspoint ai can serve at most li users simultaneously. Notethat the bandwidth of each channel in access point ai

is bi MHz. Let nij.t/ denote the additive white Gaus-sian noise [24] power of the channel

�ai, uj

�at time t in

watts. Let Xj ( Xj � A ) be the set of access pointsthat transmit signal by using the same frequency band asthe user uj. According to the Shannon equation [25], thechannel capacity denoted by qij.t/ in Mbps is calculatedas follows.

qij.t/ D bi � log2

"1C

sij.t/Pax2Xj

gxj.t/C nij.t/

#(4)

Assume that all of the transmitted data can be receivedcorrectly. Consider that a user may carry out multipletasks at the same time, there are infinite data needed tobe received by each user. Thus, we can make an assump-tion that the data receiving rates of users are equal tothe channel capacities. Furthermore, in order to guaranteethe quality of experience, user uj has the basic bandwidthrequirement denoted by �j in Mbps. Suppose that the cur-rent access point of user uj is ac. If the channel capacity ofcurrent access point cannot meet the corresponding basicbandwidth requirement (qcj.t/ < �j), user uj will per-form the vertical handoff. We call these users who needto perform handoff handoff users. If the channel capac-ity of current access point can satisfy the basic bandwidthrequirement (qcj.t/ � �j), user uj will stay in the connec-tion with its current access point ac. We call these users

who do not need to performed handoff non-handoff users.A vector V.t/ D .v1.t/, v2.t/, : : : , vn.t// is used to identifythe kinds of users at time t. The value of vj.t/ is given asfollows, where j D 1, 2, : : : , n.

vj.t/ D

�0, user uj is a handoff user at time t,1, user uj is a non-handoff user at time t

(5)

Let rij.t/ denote the channel capacity that access point ai

can provide for user uj at time t, if ai is selected as the newaccess point. The value of rij.t/ is given in Equation (6). Ifuj is a handoff user (vj.t/ D 0), rij.t/ is equal to qij.t/. If uj

is a non-handoff user (vj.t/ D 1), rij.t/ is 0.

rij.t/ D

�qij.t/, vj.t/ D 0,0, vj.t/ D 1

(6)

Let Rj.t/ D�r1j.t/, r2j.t/, : : : , rmj.t/

�. Based on Rj.t/,

the SDN controller selects a new access point for uj. Thenetwork selection result of uj at time t is denoted byFj.t/ D

�f1j.t/, f2j.t/, : : : , fmj.t/

�. The value of fij.t/ is

given as follows, where i D 1, 2, : : : , m.

fij.t/ D

�1, ai is selected by uj at time t,0, otherwise

(7)

The channel capacity that handoff user uj can obtain

from the new access point is Fj.t/ ��Rj.t/

�T in Mbps,

where�Rj.t/

�T is the transposition of Rj.t/.

Fj.t/ ��Rj.t/

�TD

mXiD1

�fij.t/ � rij.t/

�(8)

The SDN controller has to ensure that if an access pointai is selected as the new access point of user uj, the channelcapacity provided by ai should satisfy the basic bandwidthrequirement of uj. That is, Fj.t/ �

�Rj.t/

�T should subjectto the following constrain.

Fj.t/ ��Rj.t/

�T� �j (9)

If there is no access point can satisfy the basic bandwidthrequirement of user uj, the vertical handoff of uj is failed.Moreover, uj will be discarded by its current access point.

Note that if user uj is a non-handoff user (vj.t/ D 1),it does not have any new access point (

PmiD1 fij.t/ D 0).

Thus, we can get a constraint for non-handoff users thatisPm

iD1 fij.t/ C vj.t/ D 1. If user uj is a handoff user(vj.t/ D 0), when the vertical handoff of uj is failed, uj willnot have any new access point neither (

PmiD1 fij.t/ D 0). In

this case, we can obtain a relationship that isPm

iD1 fij.t/Cvj.t/ D 0. If user uj is a handoff user (vj.t/ D 0) and the



vertical handoff of uj is successful, the SDN controller willallocate a new access point to uj (

PmiD1 fij.t/ D 1). In this

case, we can obtain a relationship, that is,Pm

iD1 fij.t/ Cvj.t/ D 1. In summary, the network selection result of useruj should satisfy the following constraint.

mXiD1

fij.t/C vj.t/ � 1 (10)

Given a set of users U , their network selection results aredenoted by F.t/ D ŒŒF1.t/�

T , ŒF2.t/�T , : : : , ŒFn.t/�

T�.Note that access point ai has li channels, because eachuser will occupy one channel, ai can serve at most li userssimultaneously. There are

PnjD1

�vj.t/ � ıij.t/

�non-handoff

users are connecting to access point ai at time t. Therefore,the number of handoff users assigned to ai (

PnjD1 fij.t//

should satisfy the following constraint.

nXjD1


�C

nXjD1

fij.t/ � li (11)

We express T .F.t// as the sum of the channel capac-ities that handoff users can obtain from their new accesspoints. T .F.t// is then calculated as follows.

T .F.t// DnX

jD1

nFj.t/ �

�Rj.t/

�To (12)

The goal in the paper is to maximize the sum of chan-nel capacities that handoff users can obtain from their newaccess points T .F.t//. Non-handoff users will stay inthe connections with their current access points. The opti-mization problem of maximizing T .F.t// is theoreticallyformulated as follows.

Maximize T .F.t// DmX

iD1

nXjD1


�(13)

subject to

mXiD1

fij.t/C vj.t/ � 1, j D 1, 2, : : : , n, (14a)

mXiD1


�� j, j D 1, 2, : : : , n, (14b)

nXjD1


�C

nXjD1

fij.t/ � li, i D 1, 2, : : : , m (14c)

The first constraint Equation (14a) indicates that eachuser has at most one new access point. The second con-straint Equation (14b) guarantees that the channel capacityprovided by the new access point can satisfy the basicbandwidth requirement of a handoff user. The last con-straint Equation (14c) ensures that the number of users

connected to an access point is smaller than the number ofchannels in this access point.

The vertical handoff problem is formulated as a 0-1integer programming problem. Although the formulatedproblem is an NP-hard problem, the computation complex-ity is acceptable in its particular application context. Thelatest OPENFLOW version 1.4 supported access points pro-vide 40 GbE services [26], and the computing capability ofSDN controller is considered to be infinite [27]. Comparedwith the powerful SDN devices, most of vertical hand-off scenarios involve limited number of users. Thus, wethink that the network selection process is completed in avery short time interval which can be neglected. Moreover,if we eliminate non-handoff users from consideration, thecomputation complexity can be further reduced.

4. NETWORK SELECTIONALGORITHM

Based on previous formulations, we study the networkselection issue of the vertical handoff in this section. Weassume that users are selfish, they will select access pointsin the always-best-connected (ABC) way [28] if allowed.That is, users always choose the access points which havethe best performance as their new access points. As anexample shown in Figure 3, there are three network accesspoints (i.e., a1, a2, a3). These access points support dif-ferent wireless technologies. In the coverage area of threeaccess points, there are four users, that is, u1, u2, u3, u4).At first, user u1 was connecting to the access point a1.When the channel capacity of .a1, u1/ cannot meet its basicbandwidth requirement, u1 performed the vertical handoff.Suppose that the performance of a2 is better than a3, thusu1 will choose a2 as its new access point. At the same time,other users (u2, u3 and u4) may also choose a2 as their newaccess points. Because the resources are limited, a2 will beexhausted. While there is no user connects to a3. Systemresources are unreasonably utilized.

In order to efficiently utilize the system resources andconstruct a good access point associating strategy for eachuser, we propose a network selection algorithm for verticalhandoff. An SDN controller selects access points for usersin three phases: initialization, request matrix construction,

Figure 3. An example of the network selection issue.



Figure 4. The procedure of network selection.

and network selection. The network selection procedureis illustrated as Figure 4 shows in which we consider theOPENFLOW 1.3.3.

4.1. Algorithm details

Phase 1. InitializationAt the beginning of each time slot, users evaluate their

current access points and determine whether to performvertical handoffs or not. If a user needs vertical handoff(handoff user), it will send handoff request frames to itsavailable access points. The values of request frames areequal to the channel capacities. If a user does not needvertical handoff (non-handoff user), it also sends requestframes to its available access points. In this case, therequest frames are just like Hello messages, and theirvalues are 0. After receiving the handoff requests, accesspoints send EchoRequest frames to the SDN controller toinitialize the network selection. Then the SDN controllerwill reply the EchoResponse Frames to these access points.

Phase 2. Request matrix constructionAccess points put the features of users (i.e., the chan-

nel capacities) in the Data fields of Packet-In messages,and send the Packet-In messages to the SDN controller.These Packet-In messages will be converged in the SDNcontroller. According to these received messages, theSDN controller constructs a request matrix R.t/ DhŒR1.t/�

T , ŒR2.t/�T , : : : , ŒRn.t/�

Ti. At first, the request

matrix R.t/ is incomplete because users may not be ableto reach all access points. There are some elements, whosevalues are unknown. These elements are defined as theunassigned elements.

Definition 4.1 (Unassigned element). The unassignedelement is the element of a request matrix, whose value

is unknown becaue of the corresponding access point anduser cannot communicate directly.

Because the SDN controller has a global view, it cancomplete the request matrix R.t/ after a simple calculationin which the unassigned elements will be set to be 0.

Theorem 4.1. The calculation rule of unassigned ele-ment maintains the consistency of vertical handoff requestmatrix and will not affect the network selection results.

Proof. Each column of the request matrix R.t/ corre-sponds to a vector Rj.t/, where j D 1, 2, : : : , n. Fornon-handoff users, all elements in Rj.t/ are 0, and thevalue of corresponding column in R.t/ will be set to be0. During the network selection process, there is no newaccess point will be assigned to these users. For handoffusers, the unassigned elements in R.t/ that correspond totheir unavailable access points are set to be 0. During thenetwork selection process, some other access points withpositive evaluations will be selected.

Phase 3.Network selectionAfter constructing the vertical handoff request matrix

R.t/, the SDN controller selects new access points forhandoff users. The network selection is formulated as a 0-1programming problem (Equation (13)). The SDN con-troller calculates the network selection results by solvingthis 0-1 programing problem. There are many tools that canbe used by the SDN controller to solve the linear problems,such as LINGO and CVX on MATLAB. The networkselection results are presented as a m � n matrix F.t/.According to F.t/, the SDN controller sends out Packet-Out messages. If the element fij.t/ is 1, that means ai is thenew access point of user uj. Therefore, the SDN controllerputs this result in the data filed of Packet-Out message,



and sends the Packet-Out message [29] to access point ai.Then, access point ai sends a feedback frame to user uj tonotify this result.

4.2. An example

In this section, we will use an example to explain thenetwork selection algorithm in detail. Specifically, we con-sider a scenario shown in Figure 5. There are three accesspoints (i.e., a1, a2, a3). These access points support differ-ent wireless technologies. An SDN controller centralizedcontrols these access points. In the coverage area of threeaccess points, there are five users (i.e., u1, u2, u3, u4, u5).The basic bandwidth requirements of these five users are2 Mbps (i.e., �1 D 2 Mbps), 3 Mbps (i.e., �2 D 3 Mbps),4 Mbps (i.e., �3 D 4 Mbps), 4.5 Mbps (i.e., �4 D

4.5 Mbps) and 4 Mbps (i.e., �5 D 4 Mbps), respectively.Suppose that the channel capacity of .a1, u1/ is 4 Mbps,and u1 is a non-handoff user. Meanwhile, other four users(u2, u3, u4, and u5) are handoff users.

Phase 1. InitializationBecause u1 is a non-handoff user, it will send request

frames r11.t/ and r21.t/ to the available access pointsa1 and a2, respectively. The values of r11.t/ and r21.t/are 0. Meanwhile, other four users u2, u3, u4, and u5are handoff users; they also have to send out requestframes. Take u2, for instance, a1, a2 and a3 are avail-able to u2. Let r12.t/ denote the handoff request framesent from u2 to a1. In our example, r12.t/ is 7, whichmeans if u2 selects a1 as its new access point, the channelcapacity provided by a1 is 7 Mbps. Similarly, other userssend the vertical handoff request frames to their availableaccess points.

Phase 2. Request matrix construction

After receiving the Packet-In messages, the SDN con-troller constructs a request matrix R.t/ as shown inEquation (15). At first, there are four unassigned elements(i.e., r14.t/, r15.t/, r23.t/ and r31.t/) in the matrix R.t/.Take the element r14.t/, for instance, because the access

point a1 is unavailable to user u4, u4 will not send arequest frame to a1. Therefore, the SDN controller cannotdetermine the value of r14.t/ in the beginning.

u1 u2 u3 u4 u5

R.t/ Da1a2a3

24 0 7 5 r14.t/ r15.t/

0 1 r23.t/ 2 6r31.t/ 5 3 4 2

35 (15)

Based on this primary request matrix, the SDN con-troller calculates the values of unassigned elements. All ofthe unassigned elements are set to be 0, and the completedrequest matrix R.t/ is as follows.

u1 u2 u3 u4 u5

R.t/ Da1a2a3

24 0 7 5 0 0

0 1 0 2 60 5 3 4 2

35 (16)

Phase 3. Network selectionBased on the request matrix R.t/, the SDN controller

formulates the network selection as a 0-1 programmingproblem (Equation (13)). The solution of this 0-1 pro-gramming problem is the network selection result, whichis presented as a matrix F.t/. Assume that the numberof channels in access points a1, a2, and a3 are 2, 3, 4,respectively. After some calculations, the SDN controllercan obtain a 3�5 matrix F.t/ as follows.

u1 u2 u3 u4 u5

F.t/ Da1a2a3

24 f11.t/ f12.t/ f13.t/ f14.t/ f15.t/

f21.t/ f22.t/ f23.t/ f24.t/ f25.t/f31.t/ f32.t/ f33.t/ f34.t/ f35.t/

35

u1 u2 u3 u4 u5

D

a1a2a3

24 0 0 1 0 0

0 0 0 0 10 1 0 0 0

35

(17)As a non-handoff user, u1 will not obtain any new access

point. User u1 stays in the connect with its current access

Figure 5. An example of the proposed network selection algorithm.



point a1. Consequently, access point a1 can only serve onemore user. In order to maximize the sum of channel capac-ities that handoff users can obtain from their new accesspoints, the SDN controller allocates access point a1 to useru3 instead of user u2. For user u4, because the channelcapacities provided by its available access points a2 and a3

cannot satisfy the basic bandwidth requirements, the verti-cal handoff of u4 is failed. User u4 does not have any newaccess point and will be discarded by its current accesspoint a2. According to the network selection result F.t/,the SDN controller sends out Packet-Out messages. Thevalue of f13.t/ is 1 means the new access point of user u3

is a1. That is, u3 should transfer its inter-network connec-tion from the current access point a3 to a1. Therefore, a1

will send a feedback frame to u3 to notify this selectionresult. Similarly, a2 sends a feedback frame to u5. a3 sendsa feedback frame to u2.

5. HANDOFF TIMING ALGORITHM

Because users are always moving around in wireless envi-ronment, the network selection results should not be imple-mented immediately. For example as shown in Figure 6,user u1 is moving back and forth. When user u1 leavesthe access point a1 and closes to the access point a2, itsinter-network connection will be transferred from a1 toa2. When u1 leaves a2 and backs to a1, the inter-networkconnection will be transferred from a2 to a1 again. Theinter-network connection of user u1 is switched betweenaccess points a1 and a2 times and times again [30]. Thiscommon example reveals that inappropriate handoff timingwill incur numerous unnecessary handoffs.

Besides that, the ping-pong effect will also lead to someunnecessary handoffs. A user handoff from its currentaccess point to the new access point in order to obtainhigher channel capacities. However, the SDN controlleroptimizes the network selection results from the global per-spective which inevitably sacrifices the individual interests.If the performance gain between the current access pointand the new access points is less than the handoff overhead,the handoff is not a wise choice in this case.

Figure 6. An example of the handoff timing issue.

5.1. Discussions

In our proposed scheme, users will wait for a stabilityperiod � [11] after their network selection processes arefinished. During the stability period, if the new accesspoints are consistently more appropriate than their cur-rent access points, users will handoff to their new accesspoints. Before we start introducing our proposed hand-off timing algorithm, there are three things needed to beexplained specially.

The first thing is about the meaning of appropriate. Inthis paper, we use“appropriate" instead of “better". Thereason is even the new access point is better than the cur-rent one, if the current access point can satisfy the basicbandwidth requirement of a user, this user should notperform the vertical handoff.

The second thing is about the handoff overhead. Whenwe are judging whether a new access point is appropri-ate or not, we need to take both the performance gain andthe handoff overhead into account. According to the IEEE802.11 standard handoff procedure [31], users will experi-ence the authentication phase and the reassociation phasebefore transferring to their selected new access points.Anshul et al. [32] studied various scenarios and observedthat the average authentication delay and the average reas-sociation delay are 1.3 ms and 2.3 ms, respectively. Duringthe authentication phase, two authentication frames will beexchanged between the handoff user and the new accesspoint. Normally, an Authentication frame contains 24 bytesheader, at least 6 bytes body and 4 bytes frame checksequence (FCS) [33]. Therefore, the control frames willcost oa Mbps bandwidth during the authentication phase asthe following equation shows.

oa D2 � .3.4e 5/ Mb

1.3e 3 s.D 0.0052 Mbps (18)

During the reassociation phase, the handoff user willsend out an reassociation-request frame and the newaccess point will reply an reassociation-response frame.An reassociation-request frame consists of 24 bytes header,at least 10 bytes body and 4 bytes FCS. While anreassociation-response frame has 24 bytes header, at least6 bytes body and 4 bytes FCS. Therefore, the controlframes will cost or Mbps bandwidth during the Reassocia-tion phase as the following equation shows.

or D.3.8e 5C 3.4e 5/ Mb

2.3e 3 s.D 0.0031 Mbps (19)

The last thing is about the length of a stability period.For a handoff user uj, its available access points set isdenoted by A0j, and the size of A0j is jA0jj. Let �x be thelength of stability period in the x th round, where x D1, 2, : : :. For an available access point ai (ai 2 A0j), its chan-nel capacity at time t is denoted by qij.t/. After waiting fora stability period �x, the channel capacity of ai becomesqij.tC�x/. Thus, the change rate of channel capacity during



the stability period �x can be represented by qij.tC�x/

qij.t/. We

make use of the average change rate of channel capacitiesfor all available access points to adjust the length of sta-bility period. If uj is suggested to wait for another stabilityperiod, the value of �xC1 can be calculated as follows.

�xC1 D

Pai2A0j

qij.tC �x/

qij.t/ˇA0jˇ �x (20)

Equation (20) indicates that if the average change rate ofchannel capacities during �x is bigger than 1 (i.e., �xC1 >

�x), that is the connection quality of user uj is getting better,uj will wait for a longer stability period in the next round.Otherwise, uj will wait for a shorter stability period in thenext round. H.J. Wang et al. [11] pointed out that the stabil-ity period is proportional to the handoff latency. Moreover,S. Sharma et al. [34] proved by experiment that the handofflatency is about 0.1 s. Thus, we accordingly set the initialvalue of stability period �1 to be 0.1 s in this paper.

5.2. Algorithm details

In this section, we take a handoff user ui for instance toexplain our proposed handoff timing algorithm. For theuser u1, suppose that its current access point is a1, and itsnew access point which has been determined by the SDNcontroller is a2 as shown in Figure 7. Following the previ-ous definitions, the channel capacities of a1 and a2 at time twere q11.t/ and q21.t/ respectively. After waiting for a sta-bility period � , the channel capacities of a1 and a2 becomeq11.tC�/ and q21.tC�/ respectively. Based on the channelcapacities, the Authentication overhead oa and the Reasso-ciation overhead or, the SDN controller makes a judgementand notifies ui whether it should transfer the inter-networkconnection from a1 to a2 or not.

The network performance can be regarded as stablewhen considering the user mobility, which is a widelyaccepted assumption [35], [36]. Therefore, we are workingunder the assumptions as shown in Figure 7. The coveragearea of a1 is the circular region insides c1. The closer toa1, the higher channel capacity that u1 can obtain from a1.c1, and c01 are concentric circles. If u1 moves along c01, thechannel capacity of .a1, u1/will not change [37]. Similarly,

the coverage area of a2 is the circular region insides c2. Ifu1 moves along c02, the channel capacity of .a2, u1/will notchange. There is a line L goes through the intersections ofc01 and c02. L is perpendicular to the line between a1 and a2.We consider the following two situations.

� If u1 moves to the left of L, which means u1 hasthe tendency of moving closer to a1. Because userwas moving back to its current access point dur-ing the stability period, it does not need verticalhandoff anymore, and the network selection result iscancelled.

� If u1 moves to the right of L, which means u1 hasthe tendency of moving closer to a2. Because userwas moving away from its current access point duringthe stability period, it has to transfer the inter-networkconnection to the new access point at once.

Because the movement trend of user is important forthe vertical handoff, some related work tried to predict themovement trend of a user. The existing work is based onlocation information [38], context [39] or historical record[40], and so on. Each of them requires a large amount ofstorage space. In this paper, we predict the movement trendof a user just based on the channel capacities of its cur-rent access point and new access point. Discussions areprovided for the following nine cases.

(1) q11.t/ < q11.tC�/ and q21.t/ < q21.tC�/. Becausethe channel capacity of a1 increases, u1 must beinside c01. For the same reason, u1 is also inside c02.That is to say, after a stability period, u1 locates atthe domain d1 shown in Figure 8(a). Because theline L passing through d1, we cannot determine themovement trend of u1. As a result, u1 should waitfor another stability period and then analyze thesituation again.

(2) q11.t/ < q11.tC�/ and q21.t/ D q21.tC�/. Becausethe channel capacity of a2 does not change, u1 mustlocate at c02. Furthermore, u1 is inside c01. That isto say, after a stability period, u1 locates at the linesegment l1 shown in Figure 8(b). l1 is on the left ofL, which means u1 moves back. As a result, u1 doesnot need vertical handoff anymore.

Figure 7. An example of the proposed handoff timing algorithm.



Figure 8. The movement direction of a user.

(3) q11.t/ < q11.tC�/ and q21.t/ > q21.tC�/. Becausethe channel capacity of a2 decreases, u1 must beoutside c02. Furthermore, u1 is inside c01. That is tosay, after a stability period, u1 locates at the domaind2 shown in Figure 8(a). d2 is on the left of L, whichmeans u1 moves back. As a result, u1 does not needvertical handoff anymore.

(4) q11.t/ D q11.tC�/ and q21.t/ < q21.tC�/. Becausethe channel capacity of a1 does not change, u1 mustlocate at c01. Furthermore, u1 is inside c02. That isto say, after a stability period, u1 locates at the linesegment l2 shown in Figure 8(b). l2 is on the rightof L, which means u1 moves away. For this case,if the handoff overhead is less than the performancegain (i.e., oa C or < q21.t C �/ q11.t C �/),u1 should handoff to the new access point a2 atonce. Otherwise, u1 will initialize another networkselection.

(5) q11.t/ D q11.tC�/ and q21.t/ D q21.tC�/. Becausethe channel capacities of a1 and a2 have no change,u1 still locates at the original point after a stabilityperiod. We cannot determine the movement trend ofu1. As a result, u1 should wait for another stabilityperiod, and then analyze the situation again.

(6) q11.t/ D q11.tC�/ and q21.t/ > q21.tC�/. Becausethe channel capacity of a2 decreases, u1 must beoutside c02. Furthermore, u1 locates at c01. That is tosay, after a stability period, u1 locates at the line seg-ment l3 shown in Figure 8(b). l3 is on the left of L,so u1 does not need vertical handoff anymore.

(7) q11.t/ > q11.tC�/ and q21.t/ < q21.tC�/. Becausethe channel capacity of a1 decreases, u1 must beoutside c01. Furthermore, the channel capacity ofa2 increases, u1 must be inside c02. That is to say,after a stability period, u1 locates at the domain d3shown in Figure 8(a). Similar to case 4, if the hand-off overhead is less than the performance gain (i.e.,oaC or < q21.tC �/ q11.tC �/), u1 should hand-off to the new access point a2 at once. Otherwise, u1will initialize another network selection.

(8) q11.t/ > q11.tC�/ and q21.t/ D q21.tC�/. Becausethe channel capacity of a2 does not change, u1 mustlocate at c02. Furthermore, the channel capacity of

a1 decreases, u1 is outside c01. That is to say, aftera stability period, u1 locates at the line segment l4shown in Figure 8(b). l4 is on the right of L, so u1takes the same measures as case 4 and case 7.

(9) q11.t/ > q11.tC�/ and q21.t/ > q21.tC�/. Becausethe channel capacity of a2 decreases, u1 must beoutside c02. Furthermore, u1 is outside c01. That is tosay, after a stability period, u1 locates at the domaind4 shown in Figure 8(a). The line L passing throughd4, so we cannot determine the movement trend ofu1. As a result, u1 should wait for another stabilityperiod, then analyze the situation again.

As we have explained that after the new access point isselected, a user should transfer its inter-network connectionto the new access point at an appropriate time. For this pur-pose, we proposed a handoff timing algorithm. The SDNcontroller only needs to know the channel capacities of cur-rent access point and new access point for each handoffuser, and the handoff overhead. Based on limited informa-tion, the SDN controller can determine the time when ahandoff user should implement its network selection result.Following the aforementioned approach, we can see thatthe network selection result will be implemented only if theuser is certain to move away from its current access point.

6. PERFORMANCE EVALUATION

In this section, we provide the performance evaluation ofour S-DNVH scheme. We compare the proposed schemewith two typical existing schemes: the ABC scheme [28]and the SASHA [15] under various network conditions.The concerned performance metrics are the number ofhandoffs, total throughput and the user served ratio. As weknow, the SDN technique originated in and is commonlyused in the campus networks. Furthermore, most of accesspoints in the campus networks are the IEEE 802.11 stan-dard [41] supported devices. Therefore, we will refer to theIEEE 802.11 standard to set the parameters of access pointsduring the experiment. If other standards are required inthe practical applications, this experiment procedure canbe repeated by using the corresponding parameters. Simu-



Table II. Experimental parameters.

Parameter Value

Simulator Mininet, Floodlight, MATLABNumber of access points 9Coverage radii of access points 30 m, 50 m, 100 mTransmission powers of access points 0.002 watts, 0.005 watts, 0.02 wattsNumber of channels in access points 10, 20, 30Basic bandwidth requirements of users 2 MbpsBandwidths of each channels in access points 2 MHz, 1 MHz, 0.6 MHzInitial value of stability period �1 0.1 sTime slot 1 sChannel fading gain h h Ï exp.1/Noise power per channel n n Ï N.0, 1/ wattsMoving velocities of users 0 Ï 5 m/s

lation experiments are repeated one thousand times and theresults are presented with 95% confidence interval.

6.1. Experiment setup

Over a 500 m � 500 m rectangular flat space, there are nineaccess points and n users. These nine access points belongto three different types, and each type has three entities.Different types of access points have different coverageradii, number of channels, bandwidths and other differentattributes. Furthermore, different types of access points areassumed to use different frequency bands, and same type ofaccess points use the same frequency band. An access pointis available to a user when the distance between them issmaller than the coverage radius of this access point. Usersare moving around inside the considered area. If the currentlocation of a user is denoted by a two-dimensional coordi-nate .x, y/, this user will be inside .x˙ M t � ı, y˙ M t � ı/after a period of time M t, where ı is the maximal mov-ing velocity of the user. Following this rule, the movementmodel of users used in our experiments can be illustratedas Figure 9, where .x, y/ is the location of a user at thebeginning of a time slot,

�x0, y0

�is the location of the user

after a stability period, and�x00, y00

�is the location of the

user at the end of the time slot. We set the basic bandwidthrequirements of users to be 2 Mbps, which corresponds

Figure 9. Movement model of users.

to the video conference demanding. Based on references[42,43], we have the main experimental parameters, whichare listed in Table II.

In order to construct the SDN scenario and evaluate theperformance of proposed S-DNVH scheme, we simulatethe OpenFlow switches by using the Mininet VM 2.2.1.The SDN controller is implemented through the Flood-light V 1.2. The SDN controller and OPENFLOW switchesexchange information via OPENFLOW 1.3. In order toavoid too much change to the SDN controller, the cal-culation work is assigned to MATLAB. Hence, a logiccomplete SDN controller is composed by Floodlight andMATLAB. At first, we write a Python file that is able toinsert the features of users into the data fields of Packet-In messages. We make use of the Packet-In message tosend information from OPENFLOW switches to the SDNcontroller. Based on Eclipse, we encapsulate an I/O APIfor the SDN controller. This API is used for SDN con-troller to generate a file which contains the informationreceived from switches, and read a file which containsthe network selection results calculated by the CVX onMATLAB. Once the SDN controller obtain the networkselection results, it will insert the network selection resultsinto the data fields of Packet-Out messages, and send thePacket-Out messages to OPENFLOW switches. The archi-tecture of our simulation system is shown in Figure 10.The switches in Figure 10 represent OPENFLOW enabledaccess points, and hosts represent users in this paper.

6.2. Experiment results

6.2.1. Number of vertical handoffs.

We study the number of vertical handoffs when the num-ber of users varies under free space propagation (˛ D 2),flat-earth reflection (˛ D 3), and diffraction losses (˛ D 4)environment conditions in Figure 11. The general trend isthat there will be more vertical handoffs in each time slotas the number of users increases. For the same numberof users, larger pass loss exponent environments alwayshave more handoffs. From the longitudinal comparison,we observe that the proposed S-DNVH scheme can reducethe number of handoffs significantly. This advantage is



Figure 10. Structure of simulation system.

more remarkable in tough environments. For example inFigure 11(c), the number of handoffs in S-DNVH schemeis nearly 76% less than ABC scheme and 59.6% less thanSASHA in the worst case.

Another interesting phenomenon is that the number ofvertical handoffs will increase as more users join in. Afterthe number of users is bigger than around 120, the numberof vertical handoffs in the two contrast schemes becomesrelative stable. However, this phenomenon does not existin our proposed scheme. We also observe that the ABCscheme is particularly sensitive to environment condition.The number of handoffs in ABC scheme varies dramati-cally as the pass loss exponent changes. Compared withABC, situations in other two schemes are much morepeaceful. Furthermore, the number of handoffs in SASHAis getting closer to that in our proposed S-DNVH schemewhen the pass loss exponent increases.

6.2.2. Throughput.

We compare the total throughput under different net-work settings in Figure 12. The total throughput is definedas the sum of channel capacities that non-handoff users andhandoff users can obtain. Because each user is assumedto occupy at most one channel of access point, highertotal throughput will be achieved as more users join inat first. Then, the value of total throughput has slowergrowth when the number of users is larger than a cer-tain value. From Figure 12 we find that, this special valueis around 160. Note that, the experimental nine accesspoints theoretically can support maximum 180 users at thesame time. Our experimental result is very close to thistheoretical value.

From any sub-figure of Figure 12, we can find that S-DNVH scheme always has the highest total throughput.Figure 12 also reveals that an inverse proportion operatesbetween the total throughput and the pass loss expo-nent. Furthermore, from the crosswise comparison, wecan observe that the throughput in our proposed S-DNVHscheme is getting closer to that in SASHA when the passloss exponent increases. However, even in the worst case,the throughput in S-DNVH scheme is 60% more than inSASHA and nearly 100% more than in ABC scheme.

6.2.3. User served ratio.

The user served ratio is the ratio of users who havethe network service. Following the definitions and assump-tions made in Section 3, the user served ratio is equal

toPn

jD1 vj.t/CPm

iD1

PnjD1 fij.t/

n , where n is the number ofusers,

PnjD1 vj.t/ is the number of non-handoff users andPm

iD1Pn

jD1 fij.t/ is the number of handoff users whosevertical handoffs are successful. We compare the user

Figure 11. Number of users versus number of handoffs.

Figure 12. Number of users versus stotal throughput.



Figure 13. Number of users versus user served ratio.

served ratio under different network settings in Figure 13.It shows that the proposed S-DNVH scheme always hasthe highest user served ratio in different scenarios. As thenumber of users increases, the user served ratio sightlyincreases then decreases. Moreover, user served ratio willbe less in higher pass loss exponent environments. Asthe pass loss exponent increases, the user served ratiosin SASHA and the proposed S-DNVH scheme get closerto each other. When comparing our proposed S-DNVHscheme with the SASHA, we can find that the difference oftheir user server ratios is between 4% and 29%.

Another interesting observation is that the turning pointsin the two contrast schemes are around 120. This specialvalue has also been found in Figure 11. Thus, we can inferthat the maximal number of served users in the two contrastschemes should be 120. If there are more than 120 usersare added to the scenario, the performance of two contrastschemes will degrade. However, our proposed scheme doesnot be affected by this limitation. Compared with these twoschemes, our proposed scheme S-DNVH scheme can makefuller use of the system resources.

7. CONCLUSION

In this paper, we proposed a novel vertical handoff schemewith the support of SDN technique. The proposed schemeensures that a user will transfer to the most appropriatenew access point at the most appropriate time. From thestandpoint of users, we choose the channel capacity as theperformance metric. When the channel capacities cannotmeet their basic bandwidth requirements, users will initial-ize vertical handoffs. We formulated the network selectionprocess as a 0-1 integer programming problem, with theobjective of maximizing the sum of channel capacities thathandoff users can obtain from their new access points.After the network selection process is finished, users haveto wait for a stability period and make a decision. Userswill not handoff to the new access points unless the newaccess points are consistently more appropriate than theircurrent access points during the stability period. We car-ried out comparison experiments under different networksettings. Comparison results demonstrate that even in theworst case, the number of handoffs in the proposed S-DNVH scheme is 59.6% less than in SASHA, and thethroughput in S-DNVH scheme is nearly 60% more than in

SASHA. Furthermore, their difference of user served ratiosis between 4% and 29%.

REFERENCES

1. Cao X, Wang X, Lu S. Function computation overheterogeneous wireless sensor networks. IEEE Transac-tions on Parallel and Distributed Systems 2013; 25(7):1756–1766.

2. Li J, Kameda H, Li K. Optimal dynamic mobility man-agement for pcs networks. IEEE/ACM Transactions onNetworking (TON) 2000; 8(3): 319–327.

3. Li R, Li J, Wu K, Xiao Y, Xie J. An enhancedfast handover with low latency for mobile ipv6. IEEETransactions on Wireless Communications 2008; 7 (1):334–342.

4. Namal S, Ahmad I, Gurtov A, Ylianttila M. SDNBased inter-technology load balancing leveraged byflow admission control. In Proceedings of the IEEE SDNfor Future Networks and Services (SDN4FNS), Nov. 11,2013; 1–5.

5. Song Q, Jamalipour A. A network selection mech-anism for next generation networks. IEEE Interna-tional Conference on Communications (ICC) 2005; 2:1418–1422.

6. Márquez-Barja J, Calafate CT, Cano JC, Manzoni P. Anoverview of vertical handover techniques: algorithms,protocols and tools. Computer Communications 2011;36(8): 985–997.

7. Yan X, Sekercioglu YA, Narayanan S. A survey of ver-tical handover decision algorithms in fourth generationheterogeneous wireless networks. Computer Networks2010; 54(11): 1848–1863.

8. Stallings W. Software-defined networks and openflow.The Internet Protocol 2013; 16(1): 2–14, Cisco.

9. Akhunzada A, Ahmed E, Gani A, Khan M, Imran M,Guizani S. Securing software defined networks: taxon-omy, requirements, and open issues. IEEE Communica-tions Magazine 2015; 53(4): 36–44.

10. Cormen TH, Leiserson CE, Rivest RL, Stein C. Intro-duction to algorithms: Chapter 34 NP-Completeness.MIT press: London, England, 2001 Sep.



11. Wang HJ, Katz RH, Giese J. Policy-enabled handoffsacross heterogeneous wireless networks. In Proceedingsof the 2nd IEEE Workshop on Mobile Computing Sys-tems and Applications, WMCSA’99, Feb. 25-26, 1999;51–60.

12. Qiang L, Li J, Huang C. A software-defined networkbased vertical handoff scheme for heterogeneous wire-less networks. In Proceedings of the IEEE Global Com-munications Conference (IEEE GLOBECOM), Dec.8,2014; 4671–4676.

13. Hou J, O’Brien DC. Vertical handover-decision-making algorithm using fuzzy logic for the integratedradio-and-OW system. IEEE Transactions on WirelessCommunications 2006; 5(1): 176–185.

14. Lee W, Cho D-H. Enhanced group handover scheme inmultiaccess networks. IEEE Transactions on VehicularTechnology 2011; 60(5): 2389–2395.

15. Ciubotaru B, Muntean G-M. SASHA – a quality-oriented handover algorithm for multimedia contentdelivery to mobile users. IEEE Transactions on Broad-casting 2009; 55(2): 437–450.

16. Agarwal S, Kodialam M, Lakshman TV. Traffic engi-neering in software defined networks. In Proceedings ofthe IEEE INFOCOM, Apr 14-19, 2013; 2211–2219.

17. Dabbagh M, Hamdaoui B, Guizani M, Rayes A.Software-defined networking security: pros and cons.IEEE Communications Magazine 2015; 53(6): 73–79.

18. Yin H, Xie H, Tsou T, Lopez D, Aranda P, Sidi R.SDNi: A message exchange protocol for softwaredefined networks (SDNs) across multiple domains. InIETF Draft, Work in Progress, Jun 27, 2012.

19. Jain S, Kumar A, Mandal S, Ong J, Poutievski L,Singh A, Venkata S, Wanderer J, Zhou J, Zhu M,Zolla J. B4: Experience with a globally-deployed soft-ware defined WAN. In Proceedings of ACM SIGCOMMComputer Communication Review, Aug. 12, 2013;3–14.

20. Taniuchi K, Ohba Y, Fajardo V, Das S, Tauil M, ChengY-H, Dutta A, Baker D, Yajnik M, Famolari D. IEEE802.21: Media independent handover: features, applica-bility, and realization. IEEE Communications Magazine2009; 47(1): 112–120.

21. Andrews JG, Baccelli F, Ganti RK. A tractableapproach to coverage and rate in cellular networks.IEEE Transactions on Communications 2011; 59 (11):3122–3134.

22. Novlan TD, Ganti RK, Ghosh A, Andrews JG. Analyti-cal evaluation of fractional frequency reuse for OFDMAcellular networks. IEEE Transactions on Wireless Com-munications 2011; 10(12): 4294–4305.

23. Yang K, Gondal I, Qiu B, Dooley LS. Combined SINRbased vertical handoff algorithm for next generation het-erogeneous wireless networks. IEEE Golbecom 2007:4483–4487.

24. Sato H. The capacity of the Gaussian interferencechannel under strong interference (corresp.) IEEETransactions on Information Theory 1981; 27(6):786–788.

25. Shannon CE. A mathematical theory of communica-tion. ACM SIGMOBILE Mobile Computing and Com-munications Review 2001; 5(1): 3–55.

26. Pre-loaded Switches. [Online] Available: http://www.pica8.com/products/pre-loaded-switches [AccessedJune 2rd, 2016].

27. Inter-Datacenter WAN with centralized TE usingSDN and OpenFlow. [Online] Available: https://www.opennetworking.org/images/stories/downloads/sdn-resources/customer-case-studies/cs-googlesdn.pdf[Accessed June 2rd, 2016].

28. Louta M, Bellavista P. Bringing always best con-nectivity vision a step closer: challenges and perspec-tives. IEEE Communications Magazine 2013; 51(2):158–166.

29. Nettle OpenFlow Messages. [Online] Available: http://hackage.haskell.org/package/nettle-openflow-0.2.0/docs/Nettle-OpenFlow-Messages.html#t:CSMessage[Accessed June 2rd, 2016].

30. Li J, Chen H-H. Mobility support for ip-based net-works. IEEE Communication Magazine 2005; 43 (10):127–132.

31. Mishra A, Shin M, Arbaugh W. An empirical analysisof the IEEE 802.11 MAC layer handoff process. ACMSIGCOMM Computer Communication Review 2003;33(2): 93–102.

32. Jain A. Handoff delay for 802.11 b wireless lans.Project Report (2003). [Online]. Available: http://www.cs.columbia.edu/~hgs/papers/2003/AnshulJain80211.ppt [Accessed Jan. 10th, 2016].

33. Gast M. 802.11 wireless networks: the definitive guide.Chapter 4–802.11 Framing in Detail O’Reilly Media,Inc. Available: http://osnet.cs.nchu.edu.tw/powpoint/seminar/802.11/802.11Framing%20in%20Detail.pdf[Accessed June 2rd, 2016].

34. Sharma S, Zhu N, Chiueh T. Low-latency mobile IPhandoff for infrastructure-mode wireless LANs. IEEEJournal on Selected Areas in Communications 2004;22(4): 643–652.

35. Li J, Kubota A, Kameda H. Location management forpcs networks with consideration of mobility patterns. InProceedings of the IEEE International Conference onComputer Communications (IEEE INFOCOM), March13-17, 2005; 1207–1216.

36. Ma D, Ma M. A QoS oriented vertical handoff schemefor WiMAX/WLAN overlay networks. IEEE Transac-tions on Parallel and Distributed Systems 2012; 23(4):598–606.

37. Aristomenopoulos G et al. An autonomic QoS-centricarchitecture for integrated heterogeneous wireless net-


http://www.pica8.com/products/pre-loaded-switches

http://www.pica8.com/products/pre-loaded-switches

https://www.opennetworking.org/images/stories/downloads/sdn-resources/customer-case-studies/cs-googlesdn.pdf



http://hackage.haskell.org/package/nettle-openflow-0.2.0/docs/Nettle-OpenFlow-Messages.html#t:CSMessage



http://www.cs.columbia.edu/~hgs/papers/2003/AnshulJain80211.ppt



http://osnet.cs.nchu.edu.tw/powpoint/seminar/802.11/802.11Framing%20in%20Detail.pdf

http://osnet.cs.nchu.edu.tw/powpoint/seminar/802.11/802.11Framing%20in%20Detail.pdf


works. Mobile Networks and Applications 2011; 16(4):490–504.

38. Mumtaz S, Monteiro V, Rodriguez J, Politis C. Posi-tioning for performance enhancement in shared LTEand HSPA networks. In Proceedings of the Interna-tional Conference on Telecommunications and Multime-dia (TEMU), Jul 30, 2012; 169–174.

39. Ahmed T, Kyamakya K, Ludwig M, Anne KR,Schroeder J, Galler S, Kyamakya K, Jobmann K.A context-aware vertical handover decision algorithmfor multimode mobile terminals and its performance.In Proceedings of the Euro American Conference onTelematics and Information Systems (EATIS), Feb. 7-10,2006; 19–28.

40. Kim T, Han S, Han Y. A QoS-aware vertical handoffalgorithm based on service history information. IEEECommunications Letters 2010; 14(6): 527–529.

41. Hiertz GR, Denteneer D, Stibor PL, Zang Y, Costa XP,Walke B. The IEEE 802.11 universe. IEEE Communica-tions Magazine 2010; 48(1): 62–70.

42. Effect of Transmit Power Changes on AP Cell Sizing.Available: http://wifinigel.blogspot.jp/2014/11/effect-of-transmit-power-changes-on-ap.html [Accessed June2rd, 2016].

43. Best Practices for High Density Wireless Net-work Design in Education and Small/Medium Busi-nesses. Available: http://www.netgear.com/images/pdf/High_Density_Best_Practices.pdf [Accessed June 2rd,2016].

AUTHORS’ BIOGRAPHIES

Li Qiang is a PhD candidate atthe Graduate School of Systems andInformation Engineering, University ofTsukuba, Japan. She received the BEdegree from Anhui University, China,in 2009, and ME Degree from Uni-versity of Science and Technologyof China, China, in 2012. Her main

research interests include mobility management, hetero-geneous wireless networks, 5G, and software-defined net-working (SDN).

Jie Li received the BE degree inComputer Science from Zhejiang Uni-versity, Hangzhou, China, and the MEdegree in Electronic Engineering andCommunication Systems from ChinaAcademy of Posts and Telecommuni-cations, Beijing, China. He receivedthe Dr. Eng. degree from the University

of Electro-communications, Tokyo, Japan. He is with theUniversity of Tsukuba, Japan, where he is a professor.He has been a visiting professor in Yale University, USA,and in Inria, France. His current research interests are in

mobile-distributed computing and networking, big dataand cloud computing, IoT, OS, and modeling and perfor-mance evaluation of information systems. He is a seniormember of IEEE and ACM and a member of IPSJ (Infor-mation Processing Society of Japan). He is the chair ofTechnical Sub-committee on Big Data (TSCBD), IEEECommunications Society. He has served as a secretary forStudy Group on System Evaluation of IPSJ and on sev-eral editorial boards for the international journals, and onSteering Committees of the SIG of System EVAluation(EVA) of IPSJ, the SIG of DataBase System (DBS) of IPSJ,and the SIG of MoBiLe computing and ubiquitous com-munications of IPSJ. He has also served on the programcommittees for several international conferences such asIEEE INFOCOM, IEEE GLOBECOM, and IEEE MASS.

Yusheng Ji received BE, ME, and DEdegrees in Electrical Engineering fromthe University of Tokyo. She joinedthe National Center for Science Infor-mation Systems, Japan (NACSIS) in1990. Currently, she is a professor at theNational Institute of Informatics, Japan(NII), and the Graduate University for

Advanced Studies (SOKENDAI). Her research interestsinclude network architecture, resource management, andquality of service provisioning in wired and wireless com-munication networks. She has been a Board Member ofTrustees of IEICE and is an Expert Member of IEICETechnical Committees on Internet Architecture, and Com-munication Quality, and a Committee Member of QualityAware Internet (QAI) SIG and Internet and OperationTechnologies (IOT) SIG of IPSJ. She is an editor of IEEETransactions of Vehicular Technology and has been anassociate editor of IEICE Transactions and IPSJ Journal,an editor of KSII Transactions on Internet and Informa-tion Systems, a guest editor-in-chief, guest editor, guestassociate editor of Special Sections of IEICE Transactions,guest associate editor of Special Issues of IPSJ Journal,and so on. She has also served as a TPC member of manyconferences, including IEEE INFOCOM, ICC, GLOBE-COM, VTC, and so on, and symposium co-chair of IEEEGLOBECOM 2012 and 2014, track co-chair of IEEE VTC2016 Fall.

Changcheng Huang received his BEng. in 1985 and M Eng. in 1988both in Electronic Engineering fromTsinghua University, Beijing, China.He received a PhD degree in ElectricalEngineering from Carleton University,Ottawa, Canada in 1997. From 1996 to1998, he worked for Nortel Networks,

Ottawa, Canada where he was a systems engineering spe-cialist. He was a systems engineer and network architectin the Optical Networking Group of Tellabs, Illinois, USAduring the period of 1998–2000. Since July 2000, he hasbeen with the Department of Systems and Computer Engi-neering at Carleton University, Ottawa, Canada where he


http://www.netgear.com/images/pdf/High_Density_Best_Practices.pdf

http://www.netgear.com/images/pdf/High_Density_Best_Practices.pdf


is currently a full professor. Dr. Huang won the CFInew opportunity award for building an optical networklaboratory in 2001. He is an associate editor of Springer

Photonic Network Communications. Dr. Huang is a seniormember of IEEE.


A novel software-defined networking approach for vertical … · 2016-07-05 · Wirel. Commun. Mob....

Documents

Transcript of A novel software-defined networking approach for vertical … · 2016-07-05 · Wirel. Commun. Mob....