Sie sind auf Seite 1von 65

IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 11, NO.

1, JANUARY 2012 5 Design and Analysis of the Gateway Relocation and Admission Control Algorithm in Mobile WiMAX Networks Zong-Hua Liu, Student Member, IEEE, and Jyh-Cheng Chen, Senior Member, IEEE Abstract The WiMAX Forum has defined a two-tiered mobility management to minimize handover delay and packet loss. However, it leads to another problem: When to perform ASN GW relocation? The standards only define the ASN GW relocation procedures without specifying when the ASN GW relocation should be performed. It is left fo r vendors and operators to develop their own proprietary solutions. In this paper, we propose an algorithm, which incorporate s traditional Admission Control (AC) and Wiener Process (WP)-based prediction algorithms to determine when to carry out ASN GW r elocation. We further develop an analytical model to analyze the proposed algorithm. Simulations are also conducted to evaluate th e performance of the proposed algorithm. The results show that the proposed algorithm can improve the performance significantly in te rms of blocking probability, dropping probability, average serving rate, and average signaling overhead. Index Terms Mobility management, resource management, admission control, WiMAX net works, statistics and stochastic process, and wireless networks. 1 INTRODUCTION HE IEEE 802.16-series standards [1], [2] are exp ected to

ASN Anchored Mobility refers to the procedures T provide broadband wireless access for a va riety of

associated with the MS s movement between BSs, multimedia services. Like other IEEE 802-series standards,

which may belong to the same or different ASN IEEE 802.16 working group standardizes physical (PHY)

GWs. In ASN Anchored Mobility, the context of the layer and Medium Access Control (MAC) layer only . To

designated MS is transferred from the previous BS to build a complete system, higher layers are still necessary.

the new BS. Without performing CSN Anchored One of the major objectives of WiMAX Forum [3], thus, is to

Mobility, ASN Anchored Mobility can minimize develop and standardize the WiMAX Forum Ne twork

handover delay and packet loss. For example, an MS Architecture [4], [5], [6], [7], which is evolvi ng into Internet

dover

may perform intra-ASN han (e.g., changing Protocol (IP)-based wireless network. The archit

ecture is

from Flow (2) in Fig.

(1) to Flow 1) while still

depicted in Fig. 1. In Fig. 1, the Access Servic

e Network (ASN)

attaching to the same ASN GW. In addition, an MS provides wireless radio access for WiMAX subscri bers. It

may perform inter-ASN handover (e.g., changing from consists of one ASN Gateway (ASN GW) and many ba se

Flow (2) to Flow (3) in Fig. 1) where the ASN GW A is stations (BSs). Each ASN is connected to Connect ivity Service

the traffic anchor point and responsible for ASNNetwork (CSN), which provides IP connectivity se rvices. To

CSN tunneling. That is, traffic is still sent to ASN support IP mobility, M obile IP (MIP)1 is adopted. The

GW A, which then further tunnels traffic to ASN Home Agent (HA) of a Mobile Station (MS) is loca ted in the

GW B. In Flow (1) and Flow (2), the MS is called

CSN of the MS s Home Network Service Provider (H-N SP).

Serving MS of ASN GW A. In Flow (3), the MS is ASN GW supports the Foreign Agent (FA) functiona lity.

called Anchored MS of ASN GW A and handover MS The WiMAX Forum has defined a two-tiered mobilit y

of ASN GW B. In such case, the ASN GW A and management: ASN Anchored Mobility and CSN Anchored ASN GW B are called anchored ASN GW and Serving Mobility: ASN GW, respectively. CSN Anchored Mobility refers to the process of 1. Without loss of generality, we only discuss MIP in this paper. changing the traffic anchor point and is independent Although Client MIP (CMIP) and Proxy MIP (PMIP) are discussed in the WiMAX standards, they are variants of MIP. The technical problems and of the MS s link layer hand over [4]. It is also called solutions discussed in this paper apply to both CMIP and PMIP as well. ASN GW relocation. For ex ample, if CSN Anchored Mobility is not performed , when the MS roams from ASN GW B to ASN GW C in F ig. 1, ASN GW A will . Z.-H. Liu is with the Department of Computer Science, National Tsing Hua University, Hsinchu 30010, Taiwan. E-mail: horselui@gmail.com. tunnel traffic to ASN GW C. The MS is still served . J.-C. Chen is with the Department of Computer Science, National Chiao by two ASN GWs (ASN GW A and ASN GW C). As

Tung University, Hsinchu 300, Taiwan. E-mail: jcc@cs.nctu.edu.tw. aforementioned discussio n, the MS is called AnManuscript received 21 Apr. 2010; revised 11 Sept. 2010; accepted 25 Oct. chored MS of ASN GW A. La ter on, the ASN GW A 2010; published online 20 Dec. 2010. may request the MS to carry out CSN Anchored For information on obtaining reprints of this article, please send e-mail to: Mobility, i.e., ASN GW relocation. This may happen tmc@computer.org, and reference IEEECS Log Number TMC-2010-04-0185. Digital Object Identifier no. 10.1109/TMC.2010.265. due to the heavy load of the ASN GW A [8], to 1536-1233/12/$31.00 2012 IEEE Published by the IEEE CS, CASS, ComSoc, IES, & S PS

6 CTIONS ON MOBILE COMPUTING, VOL. 11, NO. 1, JANUARY 2012 Anchored Mobility and CSN Anchored Mobility in WiMAX networks. to-end latency, or for resource optimizaere is no resource for the newly arrived MS, es [4], [5]. After performing ASN GW GRAC will request an Anchored MS to the traffic anchor point is changed to relocation if there are Anchored MSs in he MS then is not served by ASN reover, because handover MSs are sensitive is shown in Fig. 1 after changing from ng and handover latency, we also propose a gorithm based on Wiener Process Flow (5). [18] to

IEEE TRANSA Fig. 1. ASN reduce endarrives and th tion purpos the proposed relocation, perform ASN GW ASN GW C. T the system. Mo GW A. This to call droppi prediction al Flow (4) to request Ancho

red MSs to perform ASN GW relocation Although the two-tiered mobility management defined early. Thus, handover MSs are not dropped when the in WiMAX potentially can minimize handover delay and system load is full. In addition, handover MSs do not need packet loss, it leads to another problem: When to perform

to wait for the completion of ASN GW relocation so ASN GW relocation? The WiMAX standards, however, only handover laten cy can be reduced. Furthermore, we develop define the procedures for ASN Anchored Mobility and CSN an analytical model to investigate the performance of the Anchored Mobility. The standards do not address when the proposed GRAC . The model analyzes the performance Anchored MSs should perform ASN GW relocation to bounds of the system. Extensive simulations are conducted relocate the traffic anchor point from the anchored ASN to validate th e analysis. The results show that the proposed GW to the serving ASN GW. The problem is left for vendors GRAC can effec tively reduce the blocking probability of and operators to develop their own proprietary solutions. new MSs and th e dropping probability of handover MSs. Besides, the problem is closely related to Admission Control The average signaling overhead is also reduced. The (AC), which is widely used in wireless networks to ensure average servin g rate is increased. service quality and reduce network congestion by limiting The contributi ons of this paper include: 1) The proposed the number of MSs served in the network. However, GRAC provides a systematic way to solve the problem traditional AC algorithms [9], [10], [11], [12], [13], [14], [15], effectively. 2 ) The proposed GRAC is [16], t be used directly dards, and can be used mobility management is fully compatible with when the two-tiered with other AC deployed in WiMAX. As afore[17] canno the WiMAX stan algorithms. 3) We derive the performance bounds mathementioned discussion, some MSs may be served by two matically and show that the performance of the proposed ASN GWs. The resources are required in both ASN GWs. GRAC approache

s the lower bound. Therefore, those MSs will be counted twice in two ASN The rest of th is paper is organized as follows: Section 2 GWs by the AC algorithm. If there are many Anchored MSs, reviews the re lated work. The proposed GRAC is presented new incoming users will likely be rejected due to the lack of in Section 3. In Section 4, we propose an analytical model to resources. If the ASN GW relocation can be performed evaluate the p erformance bounds of the proposed GRAC. before the system becomes overloaded, the system may be The numerical results are discussed in Section 5. Section 6 able to accommodate more MSs. Therefore, a well-designed concludes this paper. AC algorithm should cooperate with the ASN GW relocation algorithm closely. 2 K In this paper, we propose Gateway Relocation AC (GRAC), which combines ASN GW relocation and AC algorithm to in mobile WiMAX have been studied [6], [7], maximize system capacity. In GRAC, the AC algorithm [21], [22], [23]. In [6], [7], [19], [20], the authors cooperates with the ASN GW relocation. When a new MS overview of the WiMAX technology and WiMAX

RELATED WOR Many issues [19], [20], provide an

LIU AND CHEN: DESIGN AND ANALYSIS OF THE GATEWAY RELOCATION AND ADMISSION CONTROL ALGORITHM I N MOBILE WIMAX... 7 network architecture. In [19], the authors discuss the ASN GW is not affected. Although the aforementioned mobility management in WiMAX networks and several techniques can reduce the l oad of the old serving ASN GW, optimization procedures for ASN Anchored Mobility manthe load of the new serving ASN GW is increased. Therefore, agement. For example, the data path is extended from the old only the Anchored MS needs to perform ASN GW relocation ASN GW to a new ASN GW to reduce the impact of the delay to reduce the load of the A nchored ASN GW. The load of the caused by IP-layer handover. In [21], the authors propose a

Serving ASN GW is irrelevan t. fast intra-network and cross-layer handover protocol to Admission Control (AC) is o ne of the resource managesupport fast and efficient handover in WiMAX. In [22], a ment techniques to limit ma ximum amount of traffic in the seamless IP mobility scheme is proposed and evaluated in network to guarantee servic e quality for subscribers. In the flat architecture of WiMAX networks. In [23], the authors wireless and mobile network s, the AC algorithms are much more complicated due to the movement of MSs. An MS propose an analytical model to study the cost of Anchor served in current network m ay move to another network. The Paging Controller (APC) reassignment in ASN GW for connection of the MS ma y be dropped if the required location update. The APC relocation problem has a great resources in the target net work cannot be supported. It is impact on signaling overhead for location tracking. generally agreed that keepi ng an ongoing connection unMoreover, the two-tiered mobility management defined broken is more important th an admitting a new MS. Therein WiMAX is similar to that in Hierarchical MIP (HMIP) fore, a handover MS is give n higher priority to access the [24]. In HMIP, the multiple levels of FA hierarchy can network resources. For this purpose, the overall resources are reduce handover latency and localize the MIP signaling

partitioned and some resour ces are preserved for the handtraffic. In addition, Xie et al. [25] propose a distributed over MSs only. This is call ed priority-based AC. Various dynamical regional location management to determine the priority-based AC algorithm s have been proposed [9], [10], size of a regional network based on the MS s traffic load and [11], [12], [13], [14], [15 ], [16], [17]. Here, we discuss two mobility patterns [26], the authors design a commonly used priority-base d AC algorithms: cutoff priority dynamical HMIP scheme for MIP networks. Each MS algorithm [10], [11], [12] and new call bounding algorithm [9]. dynamically determines the hierarchy of FAs according to Fig. 2 illustrates the resource allocation in the cutoff the call-to-mobility ratio. The MIP registration update is priority algorithm and new call bounding algorithm. In the only performed when a threshold is reached. Therefore, the cutoff priority algorithm, both new MS and handover MS signaling overhead incurred by MIP can be reduced can be admitted if the total number of new MSs and significantly. The similar idea of chain-like architecture is handover MSs in the network is equal to or less than a also applied to the location update in cellular networks [27]. [25]. In

predefined threshold, Tcp , which is less than the total In WiMAX, however, an MS is served at most by two ASN capacity C. Once the number of new MSs and handover GWs (FAs) simultaneously due to the specific two-tiered MSs in the network reaches Tcp, new MSs are blocked. Only mobility management procedures. handover MSs are admitted. Once the total number of MSs In the literature, some gateway relocation approaches or exceeds C, handover MSs are dropped. In the new call load control techniques for cellular networks or IP-based bounding algorithm, there i s a limit, Tncb, for the number of mobile networks have been proposed [28], [29], [30], [31], new MSs admitted into the n etwork, which is also less than [32]. In [31], the Serving Radio Network Controller (SRNC) the total capacity C. The h andover MSs use the resources in relocation is discussed for the Universal Mobile TelecomC Tncb first. If the numb er of new MSs is less than Tncb, munications System (UMTS). The SRNC in UMTS networks handover MSs can use mo re resources than C Tncb. is similar to the ASN GW in WiMAX. They all control and However, the number of new MSs is always less than Tncb manage the connections in the radio access network. When or the remainder resources the handover MSs have not an MS no longer connects to the BS under the RNC, which is used. This is shown as X < minTncb; C Y in Fig. 2b. To serving the MS currently, SRNC relocation is immediately show the difference betw een the two algorithms, we initiated by the new SRNC. Hence, the SRNC can decide

assume C equals 50, and bot h Tcp and Tncb are 30. We also assume in both algorithms, there are now 20 new MSs and when to perform SRNC relocation. However, in WiMAX, if 10 handover MSs. In the cut off priority algorithm, a newly the MS connects to the BS which is under another ASN, the arrived MS will be blocked and a handover MS will be MS only performs ASN Anchored Mobility. This is because admitted. In the new call b ounding algorithm, however, that if both ASN Anchored Mobility and CSN Anchored both a new MS and a handove r MS will be admitted. Mobility are performed simultaneously, the handover delay There are still many other AC algorithms. The ideas are will become too long. similar although they may h ave different names. NeverIn MIP, load balancing and load control mechanisms have theless, they cannot be applied to WiMAX networks been proposed [28], [29], [30], [32]. The idea is that according directly. As aforementioned discussion, due to the specific to different criteria, MSs are equally served by HAs or mobility management techniq ues in WiMAX, an MS may be Mobility Anchor Points (MAPs). However, if the approaches served by two ASN GWs simultaneously. Hence, the discussed in [28], [29], [30], [32] are used in WiMAX, the required resources of an An

chored MS are reserved in both loads of the anchored and serving ASN GWs are all affected. ASN GWs. Besides, the Ancho red MS will be counted twice The MSs may also need to perform both ASN Anchored in two ASN GWs in the AC al gorithm. Thus, when many Mobility and CSN Anchored Mobility during an inter-ASN MSs are served by two ASN G Ws in the system, a newly handover. The long handover latency and high packet loss arrived MS or handover MS may be easily blocked or will degrade the service quality. On the other hand, in dropped by the AC algorithm . Without considering ASN WiMAX, when performing ASN GW relocation, the load of GW relocation in the AC alg orithm, the network perforthe anchored ASN GW is reduced but the load of the serving mance will be degraded sign ificantly.

8 S ON MOBILE COMPUTING, VOL. 11, NO. 1, JANUARY 2012

IEEE TRANSACTION

Fig. 2. Resource alloca tion. (a) Cutoff priority algorithm. (b) New call bounding algorithm. 3 PROPOSED GATEWAY RELOCATION ADMISSION be deplo yed. Unlike voice traffic, data traffic tends to be bursty. Therefore, it is hard to estimate the resource required CONTROL (GRAC) in an AS N GW to fulfill the requirements of the MSs the The ASN GW relocation may be initiated at different times ASN GW i s currently serving. If the resource in one ASN with different reasons. For example, as aforementioned

GW is ov erprovisioned, the ASN GW may become a discussion, an MS may perform ASN Anchored Mobility performa nce bottleneck. Another approach is that the without performing CSN Anchored Mobility to reduce number o f BSs controlled by each ASN GW can be scaled handover latency. After the handover is completed (i.e., the down to prevent the resource overprovision. However, handover delay has been reduced), the MS may perform because the number of BSs controlled by each ASN GW is ASN GW relocation immediately so the number of reduced, this will cause many inter-ASN handovers. As a Anchored MSs can be kept small. However, it may not be result, this approach will incur high cost. In [22], the authors a good strategy always to relocate an Anchored MS so discuss the flat mobile WiMAX network architecture. The quick. For example, an MS may move fast and keep paper sh ows that the resource management problem in the changing its Serving ASN GW. In this example, it might be ASN GW h as a great impact on the performance of WiMAX better to keep the Anchored ASN GW unchanged. In some network architecture. Besides, in WiMAX, the AC algorithm other examples, if the system load is light, there is no can be d eployed in each ASN GW to limit the maximum emergent need to perform ASN GW relocation. However, number o

f MSs to ensure network service quality. Our goal when more and more MSs are served by two ASN GWs, the is to de sign a stand-alone algorithm such that each ASN GW system load will become heavy. New users may be blocked. can dete rmine when to request Anchored MSs to perform Handover users may be dropped as well. The network ASN GW r elocation. The proposed algorithm does not need performance may be reduced significantly. Therefore, to excha nge information between neighboring ASN GWs. It performing ASN GW relocation is essential. also do es not require centralized coordination and any In WiMAX standar ds [4], [5], it is specified that ASN GW assistan ce from extra servers. In addition, the proposed can decide when to perform ASN GW relocation. In this algorith m does not need to predict the movement of the paper, we consider that the system load is heavy so mobile s tations. It combines AC algorithm with a prediction Anchored MSs are forced to perform ASN GW relocation. techniqu e to determine when is necessary to perform ASN The proposed GRAC determines when to request Anchored GW reloc ation. Thus, it is called Gateway Relocation AC MSs to perform ASN GW relocation and how many (GRAC). The corresponding parameters used in this section Anchored MSs should be relocated. After ASN GW are list ed in Table 1. relocation, resources are released and system performance The prop osed GRAC consists of two components. The is improved. first on e is AC algorithm, which is discussed in Section 3.1. Because WiMAX is based on all-IP network architecture, The pred

iction algorithm based on Wiener Process (WP) is a variety of services, including voice and data services, can then pre sented in Section 3.2.

LIU AND CHEN: DESIGN AND ANALYSIS OF THE GATEWAY RELOCATION AND ADMISSION CONTRO L ALGORITHM IN MOBILE WIMAX... 9 TABLE 1 List of Parameters 3.1 New Call Bounding AC with ASN GW Relocation Algorithm 1. New call b ounding AC with ASN GW As discussed in Section 2, the ideas of AC algo rithms are relocation similar, although they have different names. Ba Require: A new or hando overall resources are partitioned and some reso urce are with the ASN GW at time t. preserved for the handover MSs only. The propos ed GRAC 1: if a new MS arrives then can work with any AC algorithm. In this section , we simply 2: C NH t then pick up the new call bounding algorithm. For si mplicity, 3: S is accepted. */ ch MS in 4: b; C NH t then one ASN GW is equal. The main point is not on a specific NS t NS t 1 /* The new M if NS t NAt < minTncb;

sically, the ver MS is requesting to connect

here we assume that the resource assigned to ea

else if NS t NAt minTn

5: dify an AC 6: 1 /* Requesting one of the

if NAt > 0 then AC algorithm. The focus is on how to mo

NAt

NAt

algorithm for the two-tier mobility management in WiMAX. The proposed GRAC with the new call boun ding Anchored MSs to perform ASN GW relocation. * / ithm 1, we 7: S is accepted. */ s in one mber of e number of 10: ork. Let 11: at 12: else if a handover MS arrives then time t. W t consists of NS t, NAt, and NH t, which 13: number of 14: ver MS is Anchored MSs, and the number of handover MSs, r NH t NH t 1 /* The hando if W t < C then represent the number of Serving MSs, the end if W t denote the total number of MSs in the ASN GW end if new MSs, which have been admitted into the netw 8: 9: algorithm is presented in Algorithm 1. In Algor NS t NS t 1 /* The new M limit the number of Serving MSs and Anchored MS else ASN GW. As shown in Fig. 2, C is the maximum nu The new MS is blocked. MSs in the network and Tncb is the limit for th

especaccepted. */ tively, at time t. As aforementioned discussion , a new MS 15: ng MS. eighborped. SN GW. y 1. 16: After the MS performs inter-ASN handover to a n The handover MS is drop ing ASN, the MS becomes an Anchored MS of the A 17: 18: end if end if Thus, NAt is increased by 1 but NS t is decreased b else admitted into the ASN GW is regarded as a Servi

10

IEEE TRANSACTI ONS ON MOBILE COMPUTING, VOL. 11, NO. 1, JANUARY 2012

To adapt the new call bounding algorithm into W iMAX 6:

/* Stage 1: Generating the expected drift rate and the networks, the algorithm is modified as:

standard deviation rate */

7:

^ is computed by using k samples, W t f NS t NAt < minT 0

i , and (3). I

NH t and a new ^

ncb; C0 8: is computed by using k samples, W t i , ^,

MS arrives; the new MS is accepted:

and (4).

9: /* Stage 2: Estimatin g the number of MSs */ where T 0

Tncb, C0 ose the value of T0 ncb

C. How to cho

ncb

10: Computing W by using ^, ^, and (2). and C0 will be discussed later. However,

11 : Computing

W t sing W , W t, and (1). When NS t NAt minT 0

t by u

ncb; C0

NH t

12:

/* Stage 3: Determining when and how many

and NAt > 0; one anchored MS is requested t o perform ASN GW relocation */

13 : if W t to perform ASN GW relocation: t > C then

14 : n C 1e dW^ t t

Because one Anchored MS is relocated, the new M S can 15 :

minn, NAt

be accepted. Otherwise, the new MS is blocked. Further16 :

Requesting n Anchored MSs to perform ASN GW more, if a handover MS arrives at time t, it is always

relocation. accepted unless W t C0.

17: d if

en

As aforementioned discussion, in this paper, we con18 : end if sider that the system load is heavy . Therefore, Anchored

Wiener Process has been proven effective in modeling MSs are forced to perform ASN GW relocation to

stochastic processes where the values of the ran dom accommodate new coming users. Based on this principle,

variables are affected by a large number of independent we can set T 0

ncb as Tncb and C0 as C. Thus, an Anchored MS

or weakly dependent factors, each with a relatively sma ll is requested to perform ASN GW relo cation only when no

impact [18]. The W t we want to model is impacted by a more resource is available for a new coming MS. The

large number of factors. These factors are either indep enproposed GRAC does not limit t he selection of other

dent or weakly dependent of each other. For example, W t parameters for other conditions.

is impacted by the arrival rate of new MSs, arrival rat e of 3.2 WP-Based Prediction Algorithm

handover MSs, average connection holding time, average In the above algorithm, we can set C0 as C because a new

network residence time, and so on. Based on the definit ions coming MS can be queued until the r esource is available

and properties of Wiener Process, W t is continuous and

after ASN GW relocation is com pleted. However, this

W ( W Wt Wt

t) follows normal distribuapproach cannot be applied to

handover MSs because

tion. However, according to the Central Limit Theorem handover MSs are sensitive to handover latency. The

(CLT) [33], we can use a normal distribution to approxi mate acceptable handover delay is much l ess than the queuing

W although W t we want to model is integer-valued. delay of a new MS. Assuming that a handover MS arrives

The reason is that the variation of he

W is large and t

and C is reached. If the handover M S needs to wait for the

probability distribution of al

W approaches to norm

ASN GW relocation of one Anchored M S, the handover

distribution. This suggests that Wiener Process can be used latency will be too high. Actually, if ASN GW relocation is

to estimate future value of W t based on current and past performed just when a handover MS a rrives, it is equivalent

samples of W t. By using Wiener Process, W t can be to performing both ASN Anchored Mobility and CSN mo deled as Anchored Mobility. The handover latency cannot be

pffiffiffiffiffffi

reduced. On the other hand, one may perform ASN GW

W Wt Wt

t t; 1 relocation much earlier than C is r

eached. However, this

where the

is a standard normal random variable and

t is

may force many Anchored MSs to perform ASN GW

prediction tim e interval. W is the variation from t

t relocation, which may not be p referable as already discussed earlier. Thus, for handov er MSs, it is critical to

to t. Thus, the quantity of n be

W can be computed.

W ca

perform ASN GW relocation at a n appropriate time.

further modeled as a normal random variable for any giv en Therefore, we propose a predictio n algorithm based on

t. The variation of Wiener Process in the follo wing Wiener Process (WP) which provides a systematic way to

equation allows the mean and standard deviation of

determine when to request Anchored MSs to perform ASN to be changed over time: GW relocation. In addition, the alg orithm can also estimate

pffiffiffiffiffffi how many Anchored MSs should be rel ocated. As we will

W t;

t 2 see later, the proposed algorithm i

s simple and accurate.

where s

and are constants. As shown in (2),

W become T

he algorithm is described in Algorithm 2.

a normal distributed pffiffiffiffiffffiom variable with mean

t and A lgorithm 2. WP-based prediction algorithm

st andard deviation

t. It suggests that the mean and Require: At each time interval.

the standard deviation of be

W , for any given

t, can

1: if the number of samples is equal to k then

ca lculated directly from and

. Therefore, 2:

and are the oldest sample is discarded.

referred to as the expected drift rate and the standard deviation 3: end if

rate of respectively. For any given time interval

W ,

, the 4: W t is recorded.

and

can be estimated based on the mean and variance of 5: if W t Twnr then

the sample values in previous k time intervals. The samples

LIU AND CHEN: DESIGN AND ANALYSIS OF THE GATEWAY RELOCATION AND ADMISSIO

N CONTROL ALGORITHM IN MOBILE WIMAX... 11 in the time interval t i i are W t bility of i 1. Hence, algorithm precan be estimated by ^ handover MSs, the WP-based prediction i W t Moreover, in order to reduce the dropping proba , for i 0; . . . ; k ;t

vents the handover MS from being droppe d. When W t Pk 1 W t t is expected to be larger than C, some Anchored M Ss i0 ^ k are requested to perform ASN GW relocat ion. In addition, 3 the number of Anchored MSs requested to perform ASN Wt Wt k : GW relocation is estimated as W t t C. On the other k hand, if W t is approaching C but the variation of the Also, ^, the estimation of i W t i

, is given by samples of W t is smooth, the WP-based prediction will

sffiffiffiffiffiffiffiffiffiffiffiffiff iffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffif fiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffii

not request Anchored MSs to perform ASN GW relocatio n.

Pk

W t i W t i This is because the system is not expected to be overloaded ^ ^ 2

i0

k although W t is approaching C. Thus, even if the syste m

load remains high, the Anchored MSs are not necessar y to By using (2)-(4), we can estimate W closely when k

relocate their traffic anchor points. Thus, signalin g over25 [34]. Therefore, W iener Process can estimate

W t

head can be reduced even though the system load is heavy. in the near future by using current W t and W .

In short, the proposed WP-based algorithm can tri gger

Algorithm 2 shows the prediction b ased on Wiener

ASN GW relocation at an appropriate time. It can als o Process. It is exe cuted at each time interval

. At the

estimate how many Anchored MSs should be relocated. sampling time t, the ASN GW records the sample, W t.

Thus, the dropping probability of handover MSs can be After that, only the latest k samples are recorded in the ASN

reduced significantly. GW. Also, we define a threshold, Twnr, to determine

In addition to the WP-based prediction algorithm, in our whether the WP-based prediction should be executed or

earlier paper [8], we also discussed some other pred ictive not. If W t Twnr, our algorithm uses the k samples and

and nonpredictive ASN GW relocation algorithms. Because (1)-(4) to estimate

W t

t to determine whether to

the WP-based prediction algorithm performs best, we adopt perform ASN GW relocation or not. We also use the

it in the proposed GRAC. difference between

W t

t and C to determine how

many Anchored MSs need to perform ASN GW relocation.

4 PERFORMANCE ANALYSIS

The proposed WP-based prediction algori thm directly

In this section, we propose an analytical mod el to estimates the average number of MSs,

W t

t, in the

investigate the performance of the proposed algorith m. In near future. According to (2), ed by W is determin

the analysis, the connection holding time is defined as the and t. In addition, by (3) and (4), and are affected by

time from an MS connects to the network unti l it is W t i , 8i 0; . . . ; k

1. Thus,

W is sensitive to the

disconnected. The network residence time is the time an MS is variation of the samples of W t and t. Therefore, when

served by an ASN GW. We assume each ASN GW has two W is increasing, the algorithm can easily predict that

arrival processes which are Poisson distributed with rate n W t t is going to be higher than C because

and h for new MSs and handover MSs, respectively. If a W t t W t W . Thus, it can request Anchored

new MS is admitted into the network, we assume the MSs to relocate their anchor points earlier. Also, by using

connection holding time and network residence time f ollow the difference between

W t lgorithm can

t and C, the a

exponential distribution with mean 1= c and 1= n, respe ceasily determine how many Anchored MSs should be

tively. For a handover MS, only network residence ti me is relocated. The calculations in Wiener Process are simple.

required. It is also assumed to be exponentially dis tributed The number of sampling data to be recorded is minimal. It

with mean 1= n. The corresponding parameters are al so is easy to implement. The proposed WP-based prediction

listed in Table 1.

To analyze the proposed GRAC, there are three major algorithm provides a systematic way to perform ASN GW

factors to be considered the number of Serving MSs, th e relocation when system load is heavy.

number of handover MSs, and the number of Anchored 3.3 Discussion

MSs. Intuitively, a 3-D Markov chain may be us ed to Comparing with traditional AC algorithms, the proposed

investigate the performance. Unfortunately, the comp uta-

tional complexity of a 3-D Markov chain will be incr eased GRAC decreases the blocking probability of new MSs in

dramatically when the number of MSs in the s ystem WiMAX networks. To see the reasons behind it, we use

becomes large. It is computationally infeasible to c alculate Flow (3) in Fig.

1 as an example. When deploying

the solution. Therefore, instead of solving the 3-D Markov traditional AC algorithms without considering ASN GW

chain, we derive the upper bound and lower bound by relocation, a new incoming MS is blocked when NS t

considering the following two extreme case s: NAt minTncb; C NH t in ASN GW A. However,

there may be some Anchored MSs served in ASN GW A,

1. Upper bou nd: If we assume each MS never performs that is, NAt > 0. In contrast, in the proposed GRAC, ASN

ASN GW reloca tion, it will always be served by two GW A will request one of the Anchored MSs to perform

ASN GWs. For each ASN GW, the average service ASN GW relocation to relocate the traffic anchor point from

time of n ew MSs is 1= c. That is, the MSs will stay in ASN GW A to ASN GW B. Therefore, NAt in ASN GW A

the ASN GW fo r the duration of whole connection

is decreased. Thus, NS t NAt < minTncb; C

NH t.

holding time. It will resul t in the highest blocking Therefore, a new incoming MS can be accepted by ASN

probabili ty for new MSs and dropping probability GW A. Besides, the W t in ASN GW B is not increased.

for handover MSs.

12 S ON MOBILE COMPUTING, VOL. 11, NO. 1, JANUARY 2012

IEEE TRANSACTION

Fig. 3. State tr ansition diagram for the new call bounding algorithm. 2. Lower bound: If each MS always performs ASN n hn1 c n2 npn 1;n2 npn 1 1;n2 hpn1 ;n2 1 GW relocation immediately after each inter-ASN n21 npn

1;n21 n11 cp 2 ;

5e handover, the average service time of new MSs

1 b; becomes 1= c n for each ASN GW. Thus, we

n1 < Tnc

1 n1: will have the lowest blocking probability for new MSs and dropping probability for handover MSs. In both of the above cases, the average service time of c n2 npn n2 1 1;n2 hpn1 ;n2

n2 < C

n1 npn 1; 1;

5f handover MSs is 1= n for each ASN GW. As we will see 1 n1 n2 C n1: later, the upper and lower bounds can be derived by a 2D Markov chain. Next, we calculate the upper bound and lower bound of cpn npn 1 1;n2 n2 1 npn1 ;n2 1; Tncb;

h n1 1;n2

5g the proposed GRAC. Because we aim at getting the upper

n1 Tncb; n2 0: and lower bounds, the WP-based prediction algorithm is irrelevant to the mathematical analysis. The simulation h n1 c n2 npn 1;n2 npn 1 1;n2 hpn1 ;n2 1 results of the WP-based prediction algorithm will be discussed in Section 5. T he following analysis is based on

n2 1 npn n1 Tncb; 5h

1;n21; that in [9].

n2 < C

n1:

We denote n as n= c f or the upper-bound analysis. In the lower-bound analysis, similarly, n equals n= c n.

Tncb C n1 In addition, h h= n. Let pn denote the steady state

1;n2 pn 5i probability that there ar e n1 Serving and Anchored MSs, and 1;n2 1

n10 n2 0

n2 handover MSs in the AS N GW for the upper-bound case. According to the new call bounding AC described above, By solving (5a)-(5i), th e steady-state probability distributhe state transition diag ram can be drawn as a 2D Markov tion can be obtained as follows: chain, as shown in Fig. 3 . Hence, for the upper-bound case, we can obtain the global balance equations as follows:

hn2

pn b;

nn1 0

n1

Tnc

1;n2

n1! 6

n2!

p0;0; n hpn

n2 1 npn 1;n2

1;n21

5a 0 C; n1 1 cpn n1 0; n2 0: n1 n2

11;n2 ;

where n hn2

npn hpn

X 1 1;n2 1;n2 1 n2 1 npn1 ;n2 1 nn1 hn2

5b p0;0 n1 1 cpn n1 0; 1 n2 < C: n1! n2! 11;n2 ; 0 1 Tncb ;0 n1 n2 C n

Tncb C n1 1

nn1 n2 n1 0; n2 C: :

hn2 npn hpn 5c

1;n2 1;n2 1;

n1!

n2!

n1 0 n20

Thus, we can obtain the blocking probability, pnb, of n hn1 1;n2 1 1;n2 n2 1 npn1 ;n2 1 cpn npn

5d new MSs and the dropping probability, phd, of handover n11 cpn 1 n1 < Tncb; n2 0: 11;n2 ; MSs as follows:

LIU AND CHEN: DESIGN AND ANALYSIS OF THE GATEWAY RELOCATION AND ADMISSION CONTROL ALGORITHM IN MOBILE WIMAX... 13 PC nTncb nn1 Tncb 1 hn2 hC n1 n2 0 Tncb ! n1 ! TABLE 2 pnb PT PC ; 8 ncb nn1 n1 hn2 Parameters for Simulation n1 0 n20 n1! n2! PT ncb nn1 hC n1 n1 0 n1! phd PC : ncb nn1 n1 hn2 n10 n1 ! n2 0 n2 ! 9 C n1 ! PT n2! Pn1 0 C n1 !

With the steady-state probabilities, the average serving rate, u, is given by Tncb n n1 n pn 10 u 1;n2 : n2

C n1 u

n10 n20 n For the lo wer-bound case, we can use the similar approach to obtain pn l by replacing c 1;n2 , pnb, phd, and analysis, we also provide simulat ion results for the with ( c n ) in (5a)-(5h) and n with n in (6)-(10).

proposed GRAC with WP-based predi ction. The paraWe also cons ider the signaling overhead generated by

meters and values used in simulations are listed in Table 2. executing A SN GW relocation. The average signaling

The following sections present th e results with various overhead is defined as the average number of ASN GW

performance metrics. The results are based on exponential relocation p erformed in the system. It is denoted as u and

distribution for connection holdin

g time and network l per-bound case and lower-bound case,

for the up

residence time. We have also conducte d simulations by respectively . As discussed above, the ASN GW relocation

using gamma distribution to model con nection holding time is never per formed in the upper-bound case. Thus, the

and network residence time with mean 1= c and 1= n. The average signaling overhead is given by results are similar to those shown in Figs. 4, 5, 6, 7, 8, 9, 10, u 0: 11 11. Due to space limitation, we only present the results by using expone ntial distribution. On the other hand, in the lower-bound case, the ASN GW relocation is immediately executed after an MS per5.1 Blocking Probability of New MS s forms inter-ASN handover. The average signaling overhead Fig. 4 depicts the blocking pr obability of new MSs when n is given by is varied from 0.01 (1/s) to 0.1 (1/s). We set h 0:04 (1/s) and 1= n 400 (s). As expected, for both upper-bound and Tncb C n1 lower-bound cases, the blocking probability increases

l 12

n n1 pn

1;n2 :

n10

n2 0 significantly when n increases. Nevertheless, Fig. 4 shows

that the blocking probability of the proposed GRAC is The computational complexity in the proposed analytical

close to t hat of the lower-bound case regardless of the models is low. In our analytic al model, n and c are two

value of t. This is because our algorithm can approimportant factors for the performance results. As men-

priately request Anchored MSs to perform ASN GW tioned above, the average serv ice times of a new MS for

relocation when a new MS arrives. upper-bound case and lowerbound case are 1= c and

We also investigate th e blocking probability with 1= c n, respectively. If c

n, becomes the

different mean network residence time, 1= n, as shown in dominating factor and the uppe r and lower bounds are

Fig. 5. In this case, we choose n 0:04 (1/s) and h 0:04 almost equal. Besides, if the difference between n and c in

(1/s). Whe n 1= n increases, the MSs will be served by the a new MS is large, the average service times of upper and lower bounds are relatively di fferent. Thus, the difference

ASN GW lon ger. Thus, they perform inter-ASN handover in the performance results bet ween upper bound and lower

less. Ther efore, the blocking probability in the lower-bound bound is large. Moreover, when n is increased but c

case and t he proposed GRAC is increased even if n and h keeps unchanged, the blocking and dropping probabilities

are fixed. On the other hand, because the new MSs never of the lower-bound case also b ecome relatively lower than

perform AS N GW relocation, the blocking probability of the that of the upper-bound ca se according to (8) and (9).

upper-bound case is irrelevant to 1= n. Therefore, it Therefore, our analytical m odel suggests that when to

remains co nstant. Comparing the upper-bound case with perform ASN GW relocation has a great impact on the

the lowerbound case, when 1= n is much lower than 1= c, performance of WiMAX networks.

many new M Ss become Anchored MSs. The incoming new

MSs can be accepted easily by requesting the Anchored MSs

to perform ASN GW relocation in the lower-bound case. 5 SULTS NUMERICAL RE

Thus, the difference of the blocking probability between This section provides the nume rical results for the analysis

upper-boun d case and lower-bound case is relatively large. presented in Section

4. The analysis is validated by extensive simulations by using Network Simulator-version 2

5.2 robability of Handover MSs

Dropping P (ns-2) [35]. The analytical re

sults of both upper-bound and

Fig. 6 ill ustrates the dropping probability of handover MSs

lower-bound cases are close to the simulation results. In

when n i s varied from 0.01 (1/s) to 0.1 (1/s). As that in addition to the upper-bound analysis and lower-bound

Section 5. 1, we set h 0:04 (1/s) and 1= n 400 (s). When

14 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 11, NO. 1, JANUARY 2012 Fig. 4. Blocking probability versus arrival rate ( n ) for new incoming MSs. Fig. 5. Blocking probability versus mean network residence time (1= n) for new incomin g MSs. n increases, i.e., there are more MSs in the system, the Anchored MSs are requested to perform ASN GW relocadropping probability increases too. The handover MS is tion earlier. However, if we set the value of dropped when C in the AC algorithm is reached. In the t to be a

relatively large value, e.g., t 1;000 (s), t becomes the proposed GRAC, however, the WP-based prediction is only dominating factor and the impact of the variation of sensitive to the variation of the samples. The Anchored MSs the samples is small. Thus, the WP-based prediction are requested to perform ASN GW relocation when the algorithm becomes useless. As we can see, when system is expected to be overloaded. Thus, the dropping equal to twice of the sampling interval, e result is very probability of handover MSs is reduced significantly. In close to that of the lower-bound case. This is sufficient for Fig. t is

, th

6, we show the simulation results of the proposed GRAC with the WP-based prediction algorithm. t 5 (s) and t 10 (s). We observe that

when

We also investigate the dropping probability with t 10 (s), the dropping probability of the proposed

different mean network residence time, 1= n, as shown in GRAC is very close to that of the lower-bound case. This is Fig. 7. In this case, we choose n 0:04 (1/s) and h 0:04 because the WP-based prediction is sensitive to the (1/s). When 1= n increases, the new MSs and handover variation of the samples of W t. According to (1)-(4), when MSs are served by one ASN GW longer. Thus, they W >0 and t increases, it is easier for the WP-based perform inter-ASN handover less. Therefore, the dropping prediction algorithm to determine whether the system will probability is increased even if n and h be overloaded or not in the near future. As a result, the are fixed.

However, unlike the blocking probability shown in Fig. 5,

LIU AND CHEN: DESIGN AND ANALYSIS OF THE GATEWAY RELOCATION AND ADMISSION CONTRO L ALGORITHM IN MOBILE WIMAX... 15 Fig. 6. Dropping probability versus new MS arrival rate ( n) for handover MSs. Fig. 7. D ropping probability versus mean network residence time (1= n) for handover MSs. the dropp ing probability of the upper-bound case is also of upper-bound case. This is because th e blocking and increased . This is because the handover MSs are also dropping probabilities in the upper-bound cas e are higher served b y one ASN GW longer. In addition, in the than those in the lower-bound case. Thus, les s MSs are

GRAC, the dropping probability of e that the

proposed t 10 (s) served in the upper-bound case. Please also not is lower than that of t 5 (s).

average serving rate of the proposed GRAC is ve ry close to

that of the lower-bound case. 5.3 Average Serving Rate

In Fig. 9, we investigate the average serving rate with The avera ge serving rate is defined as the average number of different mean network residence time, 1= n. We also set MSs serve d by an ASN GW per minute. It includes both new n 0:04 (1/s) and h 0:04 (1/s). Fig. 9 shows th at the MSs and handover MSs. Fig. presents the average reases. This is 8

average serving rate decreases when 1= n inc

serving r ate versus n, where n is varied from 0.01 (1/s) because both new and handover MSs perform int er-ASN to 0.1 (1 /s). We choose h 0:04 (1/s) and 1= n 400 (s). handover less. Besides, the average serving rate of the As shown in the figure, the upper-bound case and lowerproposed GRAC is close to that of the lower-b ound case. bound case are almost equal when n 0:02 (1/s). This is

5.4

Average Signaling Overhead

because t he blocking and dropping probabilities are small ases. However, when in both c n increases, the average Fig. 10 illustrates the average signaling overhead per serving r ate of lower-bound case increases faster than that minute versus n, where 0.01 (1/s) to n is varied from

16 NS ON MOBILE COMPUTING, VOL. 11, NO. 1, JANUARY 2012 Fig. 8. Average serving rate versus new MS arrival rate ( n). serving rate versus mean network residence time (1= n). 0.1 h 0:04 (1/s) and 1= n 400 (s). The gain,thesignalingoverheadoftheupperling traffic generated by executing CSN se is 0. For the lower-bound case, when 1= n is ty can be measured by the number of ASN he signaling overhead is relatively high because the erformed in the system. As shown in the more likely to perform inter-ASN handover. regardless of the variation of 1= n, the average

IEEE TRANSACTIO Fig. 9. Average (1/s). We set h 0:04(1/s).A amount of signa bound ca Anchored Mobili small, t GW relocation p MSs are However, figure, the sig

naling overhead of the upper-bound case is 0, signalin g overhead of the proposed GRAC almost remains because new MSs never perform ASN GW relocation in the constant. upper-bound cas e. In the lower-bound case, the signaling overhead is inc reased when n increases. However, the 6 signaling overh SUMMARY than that of th In WiMAX standards, an ASN GW can decide when to based predictio n, the proposed GRAC can request ASN GW perform ead of the proposed GRAC is always lower e lower-bound case. This is because with WP-

ASN GW relocation. In this paper, we consider that when the system is expected to be overem load is heavy, so Anchored MSs are forced to loaded as that discussed in Section 3.3. ASN GW relocation. We propose GRAC which also investigate the average signaling s admission control and ASN GW relocation jointly different mean network residence time, ove the performance of WiMAX networks. The in Fig. 11. We still set n 0:04 (1/s) and l AC algorithms cannot be used directly when the Furthermore, we consider overhead with to impr 1= n, as shown traditiona relocation only the syst perform

LIU AND CHEN: DESIGN AND ANALYSIS OF THE GATEWAY RELOCATION AND ADMISSION CONTRO L ALGORITHM IN MOBILE WIMAX... 17 Fig. 10. Average signaling overhead versus new MS arrival rate ( n). Fig. 11. Average signaling overhead versus mean network residence time (1= n). two-tiered mobility management is deployed in WiMAX numerical results show that the proposed algori thm can because some MSs may be served by two ASN GWs. If there effectively reduce the blocking probability, dr opping probare many Anchored MSs, new incoming users will likely be ability, and average signaling overhead. It als o increases the rejected due to the lack of resources. In the proposed GRAC, average serving rate. the AC algorithm cooperates with the ASN GW relocation. When a new MS arrives and there is no resource for the ACKNOWLEDGMENTS newly arrived MS, the proposed GRAC will request an Anchored MS to perform ASN GW relocation. In addition, This work was supported in part by the National Science for handover MSs, the WP-based prediction algorithm can Council (NSC) under Grant 98-2221-E-007-065-MY3 . trigger the ASN GW relocation at an appropriate time. It can also estimate how many Anchored MSs should be REFERENCES relocated. We develop an analytical model to investigate the [1] IEEE 802.16-2004 Std., Air Interface for Fixed Broadband Wireless performance of the proposed GRAC. The model analyzes Access Systems, IEEE, Oct. 2004. the performance bounds of the system. Extensive simula[2] IEEE 802.16e-2005 Std., Part 16: Air Inte rface for Fixed and Mobile Broadband Wireless Access Systems-Amendment

2: Physical and tions are also conducted to validate the analysis and Medium Access Control Layers for Combined Fix ed and Mobile evaluate the performance of the proposed GRAC. The Operation in Licensed Bands, IEEE, Feb. 2006.

18 ACTIONS ON MOBILE COMPUTING, VOL. 11, NO. 1, JANUARY 2012 [3] ://www.wimaxforum.org, 2011. [29] d H. Kameda, [4] um Std. rk Agents in Mobile IP Networks, re s, Reference Model and nos. 1-4, pp. 11-31, Apr. 2003. , WiMAX, Feb. 2009. , Z. Niu, and M. Ojima, [5] [30] A Hybrid

IEEE TRANS WiMAX Forum, http

A. Vasilache, J. Li, an Threshold-Based Load WiMAX For 1.0, Rev. 4, WiMAX Forum Netwo Balancing for Multiple Home Architectu (Stage 2: Architecture Tenet Telecomm. Systems, vol. 22, Reference Points) H. Deng, X. Huang, K. Zhang WiMAX forum Std.

1.0, Rev. 4, WiMAX Forum Network Architecture Load Balance Mechanism for Distributed Home Agents in Mobile (Stage 3: Detaile d Protocols and Procedures), WiMAX, Feb. 2009. IPv6, Proc. IEEE Int l Symp. P ersonal, Indoor, and Mobile Radio [6] L. Nuaymi, WiMAX: Technology for Broadband Wireless Access. John Comm. (PIMRC 03), pp. 2842-2 846, Sept. 2003. Wiley, 200 7. . Tsai, and P. Agrawal, [7] , [31] Serving Radio A.-C. Pang, Y.-B. Lin, H.-M K. Etemad Overview of Mobile Wi Network Controller Relocation

MAX Technology and for UMTS All-IP Network, IEEE

Evolution, mm. Mag., vol. 46, no. 10, pp. 31-40, Oct. 2008. ol. 22, no. 4, pp. 617-629, May 2004. [8]

IEEE Co

J. Selected Areas in Comm., v Z.-H. Liu, S.-Y. Pan, and J.-C. Chen, Access Service Network [32] , A Mobility-Based Load Control ocation Algorithms in WiMAX Networks, Scheme at Mobility Anchor Poi nt in Hierarchical Mobile IPv6 Proc. IEEE Int l Co nf. Comm. (ICC 08), pp. 2674-2679, May 2008. Networks, Proc. IEEE GlobeCom, p p. 3431-3435, Nov./Dec. 2004. [9] Y. Fang and Y. Zh ang, Call Admission Control Schemes and [33] Probability and Statistical Inference, IEEE Trans. seventh ed. Prentice Hall, 20 06. Vehicular Technol ogy, vol. 51, no. 2, pp. 371-382, Mar. 2002. [34] is, Time Series: Theory and Methods. P.J. Brockwell and R.A. Dav R.V. Hogg and E.A. Tanis, Performance Analy sis in Wireless Mobile Networks, S. Pack, T. Kwon, and Y. Choi (ASN) Gateway Rel

Springer Verlag, 1991. [10] D. Hong and S.S. Rappaport, Traffic Model and Performance

[35] , http://www.isi.edu/nsnam/ns, llular Mobile Radio Telephone Systems with

The Network Simulator - ns-2 Analysis for Ce

2011. Prioritized and N onprioritized Handoff Procedures, IEEE Trans. Vehicular Technol ogy, vol. 35, no. 3, pp. 77-92, Aug. 1986. [11]

Y.-B. Lin, S. Moh an, and A. Noerpel, gree from Assignment Strate gies for PCS Hand-Off and Initial Access, IEEE the Department of Computer Scien ce, National Trans. Vehicular Technology, vol. 43, no. 3, pp. 704-712, Mar. 1994. Tsing Hua University (NTHU), H sinchu, Taiwan, [12] B. Li, S.T. Chans on, and C. Lin, at Telcordia Priority Scheme f or Multiple Classes of Traffic in Multimedia Technologies, Piscataway, New Jersey, in 2009. Wireless Networks , Wireless Networks, vol. 4, no. 4, pp. 279-290, His research interests include m obility manage1998. Analysis of a Hybrid Cutoff in 2010. He was a summer intern Queueing Priority Channel Zong-Hua Liu received the PhD de

ment, admission control, resou rce management, [13] R. Ramjee, D. To wsley, and R. Nagarajan, On Optimal Call and performance evaluation of wireless netAdmission Contro l in Cellular Networks, er of the IEEE. vol. 3, no . 1, pp. 29-41, 1997. [14] M.D. Kulavarathar asah and A.H. Aghvami, Teletraffic Performance Evaluation Wireless Networks, works. Dr. Liu is a student memb

of Microcellular Personal Communication Networks (PCN s) wi th Prioritized Handoff Procedures, IEEE Jyh-Cheng Chen received the Ph D degree Trans. Vehicular Technology, vol. 48, no. 1, pp. 137-152, Jan. 1999. from the State University of N ew York at Buffalo [15] R. Garg and H. Sa ran, Fair Bandwidth Sharing Among Virtual in 1998. He is the director of the Institute of Networks: A Capac ity Resizing Approach, Proc. IEEE INFONetwork Engineering and a pr ofessor in the COM, pp. 2 55-264, Mar. 2000. Department of Computer Scie nce, National [16] J. Yao, J.W. Mark , T.C. Wong, Y.H. Chew, K.M. Lye, and K.-C. Chiao Tung University (NCTU), Hsinchu, TaiChua, itioning Resource Allocation for Multiclass wan. He was with Bellcore/Telcor dia TechnoloTraffic in Cellul ar Systems with QoS Constraints, during 1998Vehicular Technol ogy, vol. 53, no. 3, pp. 847-864, May 2004. 2001, and with Telcordia Techn ologies, Piscat[17] S.C. Borst and D. Mitra, Virtual Partitioning for Robust Resource away, New Jersey, during 2008-20 IEEE Trans. gies, Morristown, New Jersey, Virtual Part

10. He has Sharing: Computat ional Techniques for Heterogeneous Traffic, also been with the Department of Computer Science, National Tsing IEEE J. Selected Areas in Comm., vol. 16, no. 5, pp. 668-678, June Hua University (NTHU), Hsin chu, Taiwan, since 2001, as an assistant/ 1998. associate/full/adjunct pro fessor. He coauthored the book IP-Based [18] J.C. Hull, Option s, Futures, and Other Derivatives. Prentice-Hall, Next-Generation Wireless Ne tworks (Wiley, 2004). He has published 2000. more than 80 papers and is the holder of 19 US patents, six ROC [19] P. Iyer, N. Natar ajan, M. Venkatachalam, A. Bedekar, E. Gonen, K. patents, and four PRC paten ts. He received the 2000 Telcordia CEO Etemad, and P. Ta aghol, All-IP Network Architecture for Mobile Award, the 2001 SAIC ESTC (Executive Science and Technology WiMAX, Mobile WiMAX Symp., pp. 54-59, 2007. Council) Publication Award, t he 2004 NTHU New Faculty Research [20] F. Wang, A. Ghosh , C. Sankaran, P. Fleming, F. Hsieh, and S. Award, the 2006 NTHU Outsta nding Teaching Award, and the 2007 Benes, Mobile WiMA X Systems: Performance and Evolution, Best Paper Award for Young Scho lars, IEEE Communications Society IEEE Comm. Mag., vol. 46, no. 10, pp. 41-49, Oct. 2008. Taipei and Tainan Chapters & IEEE Information Theory Society Taipei [21] J.-H. Yeh, J.-C. Chen, and P. Agrawal, Fast Intra-Network and Chapter. He is a technical ed Proc. IEEE

itor of the IEEE Wireless Communications. Cross-Layer Hando ver (FINCH) for WiMAX and Mobile InterHe was a guest editor of the IEEE Journal on Selected Areas in net, obile Computing, vol. 8, no. 4, pp. 558-574, e on Communications special issu All-IP Wireless Networks, May Apr. 2009. 2004. He was the technical program cochair of the Ninth IFIP/IEEE [22] S. Sim, S.-j. Han , J.-s. Park, and S.-c. Lee, Seamless IP Mobility International Conference on Mobile and Wireless Communications Support for Flat Architecture Mobile WiMAX Networks, n Ireland, September 2007. He was Comm. Mag., vol. 47, no. 6, pp. 142-148, June 2009. the technical program co chair of the Third IEEE International IEEE Networks (MWCN 200707) held i IEEE Trans. M

Conference on Information Technology: Research and Education [23] Y.-B. Lin and Y.C. Lin, WiMAX Location Update for Vehicle

(ITRE 2005). He has been on the technical program committee of Applications, le Networks and Applications, vol. 15, no. 1, Mobi

numerous international conf erences, including IEEE INFOCOM 2005pp. 148-15 9, 2010.

2006, IEEE GlobeCom 2005-20 09, and IEEE ICC 2007-2010. He was a [24]

E. Fogelstroem, A . Jonsson, and C. Perkins, Mobile IPv4 Regional

tutorial speaker at IEEE GlobeCom 2002, 2003, and 2006 on the Registration, IET F RFC 4857, June 2007.

subject of next-generation wireless networks. He leads the development [25] nd I.E. Akyildiz, Regional J. Xie a A Distributed Dynamic

of WIRE1x, which is one of the most important implementations of the Location Managem ent Scheme for Mobile IP, Proc. IEEE

IEEE 802.1x supplicant. He is a senior member of the IEEE and a senior INFOCOM, vol. 2, pp. 1069-1078, 2002. member of the ACM. [26] W. Ma and Y. Fang, Dynamic Hierarchical Mobility Management Strategy for Mobile IP Networks, IEEE J. Selected Areas in Comm., vol. 22, no. 4, pp. 664-676, May 2004. [27] W. Ma an d Y. Fang, A Pointer Forwarding Based Local . For more informat ion on this or any other computing topic, Anchoring (POFLA) Scheme for Wireless Networks, IEEE Trans. please visit our Di gital Library at www.computer.org/publications/dlib. Vehicular Technologies, vol. 54, no. 3, pp. 1135-1146, May 2005. [28] J.P. Jue a nd D. Ghosal, Design and Analysis of a Replicated Server Architecture for Supporting IP-Host Mobility, Cluster Computing, vol. 1, no. 2, pp. 249-260, 1998.