Load Balancing in the Call Admission Control of Heterogeneous ...

Load Balancing in the Call Admission Control of Heterogeneous Wireless Networks Kamil H. Suleiman

H. Anthony Chan

Mqhele E. Dlodlo

Department of Electrical Engineering Department of Electrical Engineering Department of Electrical Engineering University of Cape Town University of Cape Town University of Cape Town Rondebosch, South Africa Rondebosch, South Africa Rondebosch, South Africa +27 21 650 2813 +27 21 650 2788 +27 21 650 3441

[email protected]

[email protected]

ABSTRACT In the next generation wireless systems (NGWS) the end user will be able to connect to any of the different available access networks. The end user will also be able to roam seamlessly within the access networks through vertical handoffs. While these features of the NGWS are very advantageous for service providers as well as end users, technically they bring about many challenges to the joint call admission control (JCAC) algorithm. In addition to making decisions during new or handoff calls, the JCAC should perform load balancing among the different access networks. In this paper, we extend the hierarchical joint call admission control algorithm by proposing a scheme for load balancing. In the hierarchical approach, the work of the JCAC is divided into the algorithms horizontal call admission control (HCAC) and vertical call admission control (VCAC). A VCAC manages the inter-network admission of multiple access networks while in each access network there is an HCAC that controls its intra-network admission. In the proposed load balancing scheme, each HCAC in a heterogeneous environment periodically sends a load report to the VCAC. The VCAC makes a comparison of the load reports and makes a decision for load balancing.

Categories and Subject Descriptors C.2.1 [Computer: Communication Networks]: Architecture and Design – wireless communication.

Network

General Terms Algorithms, Design, Management, Performance.

Keywords Access networks, heterogeneous wireless networks, vertical handoff.

Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. IWCMC’06, July 3–6, 2006, Vancouver, British Columbia, Canada. Copyright 2006 ACM 1-58113-000-0/00/0004…$5.00.

[email protected]

1. INTRODUCTION Today, diverse wireless access networks including 2.5G, 3G, Bluetooth, WLAN and WiMAX exist each offering a different option in terms of its coverage area, data rate, utilization of the scarce radio resource, power consumption …etc. Besides, the trend shows that existing technologies are being improved and new ones are emerging to offer better performance and cost to end users and service providers. The idea of benefiting from integrating the different technologies has lead to the concept of beyond–3G wireless networks known as

the next generation wireless systems (NGWS). In this heterogeneous environment, the end user is expected to be able to connect to any of the different available access networks. The end user will also be able to roam seamlessly within these access networks through vertical handoffs. In the NGWS the global roaming of mobile terminals will be supported by the IP backbone [11]. Admission of users into the different access networks of the NGWS will be controlled by the joint call admission control (JCAC) protocol [2]. Although good models of call admission control algorithm exist for the current wireless networks, the heterogeneous nature of the NGWS requires a very different approach to the JCAC. A good JCAC protocol should maximize utilization of network resources by minimizing the blocking probability of new calls. It should also avoid dropping of handoff calls of users who are sensitive to service interruption. Meeting guaranteed quality of service (QoS) is one of the major targets of the JCAC. While the above aims of the admission control are common both to the current wireless networks and the NGWS, the heterogeneous nature of the NGWS will demand certain additional functions of the JCAC. Admitting users into the most preferable network is one of the additional tasks of the JCAC. In situations where some of the access networks serve so many users while others serve only a few, the JCAC should be able to decide to switch some of the users from heavily loaded networks to the lightly loaded ones through vertical handoffs. An interesting example is the heterogeneous environment consisting of a cellular network and a WLAN. At a certain time,

due to many departures the WLAN can be lightly loaded while the cellular network is heavily loaded. In this case, switching some of the users from the cellular network to the WLAN could be a good offer to the users both in terms of throughput and cost wise. The joint call admission control algorithm should play this interesting role by triggering vertical handoffs from the cellular network to the WLAN. The extra signaling for vertical handoffs function should be kept as low as possible so that it doesn’t consume a lot of network resources as well as mobile terminal’s power. Further, in [1] it was shown that frequent bandwidth switching among different bandwidth levels may be worse than a large QoS degradation ratio. Therefore, although we wish to make ideal load balance at any moment, we have to limit the frequency of load balancing. In [13] we proposed a hierarchical approach to the JCAC in heterogeneous wireless networks. It was intended to simplify the job of the JCAC by decentralizing some of its tasks. In this paper we propose the load balancing scheme as an extension to the above work. The rest of the paper is organized as follows. Section 2 describes the NGWS architecture and our assumptions. In section 3 we discuss some work related to our scheme. The proposed load balancing scheme is presented in Section 4. Finally, in section 5 we conclude the paper.

2. NGWS ARCHITECTURE AND ASSUMPTIONS There is a strong belief that access networks tend to adopt the IP backbone as a core network. Therefore, in this paper we use the loose coupling approach as shown in Figure 1 in which all the access networks connect to the IP backbone [5], [12], [14].

3. PREVIOUS WORK 3.1 Call Admission Control in Cellular Networks The simplest admission control discipline is first come first served. However, it offers no priority of service. In [6], the authors proposed an algorithm based on complete partition (CP). In this scheme, the bandwidth is divided into segments where each type of call can be admitted only in a segment assigned to it. The major problem associated with this algorithm is the waste of bandwidth. A call request to a full segment is rejected even if there is enough bandwidth in a segment assigned for lower priority service calls. In [4], a “movable” boundary system was used based on CP to support voice and data calls. In this algorithm, the boundary for the partition is adjustable to deal with traffic changes in the system. The above scheme was extended in [7] using the guarded channel policy to differentiate between new and handoff voice calls. In this scheme, called dynamic partition (DP) scheme, K1 out of the C channels of a cell are reserved for voice calls (new/handoff) and K2 channels are reserved for data calls . The remaining (C - K1 K2) channels are shared in a fair manner by both voice and data calls. New voice calls can only use K3 out of the K1 channels. According to the DP algorithm, when a new voice call arrives, a channel is sought in K3. If a channel is available there, it will be admitted. Otherwise, a channel is sought in the shared area. It will be blocked if there is no channel in both K3 and the shared area. When a handoff voice call arrives, a similar search is done in the voice only area and then the shared area. It will be dropped only if there is no channel available both in the voice only and shared area. A similar decision is made for a data call arrival by first searching in the data only area and then the shared area. Still in the DP algorithm, there is an efficiency problem in the cases where there is no bandwidth available for a voice call (new/handoff) while there is enough bandwidth in the data only area. This is a waste of radio resource since data calls should get lower priority than voice calls. In [17] and [8] the authors proposed the dual threshold bandwidth reservation (DTBR) scheme for voice/elastic data cellular networks. This scheme is based on complete sharing (CS) and the authors showed that it offers higher network utilization than the DP and meets guaranteed QoS.

Figure 1. IP backbone based NGWS architecture. RRM = radio resource manager, RNC = radio network controller, AR = access router, BSC = base station controller.

For simplicity, we assume that there are only two types of calls; voice and data calls. Voice calls are delay intolerant and a constant amount of bandwidth has to be reserved for a voice call before it is admitted to a certain access network. Data calls are delay tolerant and they can use flexible amount of bandwidth between a minimum and a maximum limit at times of bandwidth scarcity and abundance, respectively.

In DTBR, the C channels of the cell are divided into three regions by two thresholds K1 and K2 (K1 > K2) as shown in Figure 2. When the network occupancy level L is less than the threshold K2, both voice and data calls can be admitted into the system. When L is greater than K2, no data call can be admitted into the system. When L is greater than K1, no data call or new voice call can be admitted to the system. A handoff voice call will be dropped only if there is no channel available. One of the good features of the DTBR algorithm is that a higher priority call is never rejected while there is a bandwidth available for a lower priority call.

In this case, if all the data calls occupy their maximum bandwidth, and there is still some bandwidth left over more than that needed for handoff reservation, it will be wasted. Departure of calls in heterogeneous wireless networks should be handled in a different way from that of homogeneous networks. According to the per flow scheme in [15], when a call departs from the system, an attempt is made to assign the released bandwidth to the shortest call. And according to the work in [16], when a call departs, the per class scheme allows the highest priority class to use more bandwidth. Figure 2. DTBR algorithm.

3.2 Joint Call Admission Control in Heterogeneous Networks Some research has been done to solve the resource allocation challenges of the NGWS. The work in [10], [15], [16], [18] presents an adaptive policy based access management system to support heterogeneous networks. The authors in [2] presented a JCAC scheme for integrated UMTS-WLAN network. Because of the fact that WLAN technologies are short-range networks, they offer lower cost to users than UMTS. Thus, the vertical handoff from UMTS to WLAN was defined as a desirable handoff. On the other hand, the mobile terminal’s connection has to remain seamless as a user connected to the WLAN roams out of the WLAN domain. Thus, the handoff from the WLAN to the UMTS was defined in [2] to be a necessary handoff. The work in [2], gives more priority to necessary handoffs than desirable handoffs. In [15], a bandwidth adaptation scheme based on per flow degradation was proposed for heterogeneous wireless networks by defining a concept called degrade profile. In the cases where the bandwidth of an access network is full and the ongoing calls are using bandwidth more than the minimum needed bandwidth, degradation is done upon arrival of new or handoff calls. Here, in order to admit calls, this scheme degraded the longest calls in the system with a hope that those flows have bigger probabilities to quite the system and leave fewer degraded connections. The authors in [16] proposed a bandwidth adaptation scheme based on per class degradation. Here, in order to admit a call, the lower possible priority class calls are degraded. Ongoing higher priority class calls are not affected by arrival of lower priority calls. The performance analysis in [16] showed that the per class adaptation scheme is better in terms of fairness (by treating flows of the same class equally) and simplicity. But the per flow scheme is better in terms of resource utilization.

However, these methods only aim to allow the bandwidth released in an access network by call departures to be used by the ongoing calls. The above methods do not deal with avoiding high difference in network loads (i.e. creating load balance) between the different access networks that might result from departures of calls.

3.3 Hierarchical Joint Call Admission Control Algorithm 3.3.1 A High Level Description In the hierarchical JCAC proposed in [13], the task of the JCAC is divided into two; namely, horizontal call admission control (HCAC) and vertical call admission control (VCAC). The HCAC controls intra-domain admission policy in an access network. It is deployed in a decentralized mode. In each access network there will be an HCAC that works for it. The VCAC controls interdomain admission policy to distribute calls among the different access networks that are in the domain of the RRM. Hence, it is centralized in the RRM. The HCAC and VCAC communicate with each other at necessary times. According to the hierarchy described, these entities of the joint call admission control algorithm are located in the heterogeneous network environment as shown in Figure 3.

Departure of calls is also an important issue of radio resource management. Departure of calls from an access network can happen either because of end of call sessions or because of handoffs.

Figure 3. Distribution of the hierarchical joint call admission control components in a heterogeneous network environment.

Handling of departures of calls from homogeneous networks has been proposed by different researchers. According to the work in [7], when a call departs from a cellular network, the released bandwidth is shared by the ongoing calls until each data call takes a maximum bandwidth. However, since fixed bandwidth is allocated to voice calls, no change of bandwidth is done for them.

3.3.2 HCAC In [13], the dual threshold bandwidth reservation (DTBR) [8], [17] scheme was modified as the triple bandwidth reservation (TTBR) scheme as follows. The C channels of an access network are divided into four regions by three thresholds K1, K2 and K3 (K1 > K2 > K3) as shown in Figure 4. When the network

occupancy level L is less than the threshold K3, then both voice and data (whether new or handoff) calls can be admitted into the network. When L is greater than K3, no data call is admitted into the network. When L is greater than K2, no new voice call can be admitted. Vertical handoff voice calls are dropped if L is greater than K1. Horizontal handoff voice calls are dropped only if L is equal to C.

Let us define the virtual load level VLL as follows. The VLL of an access network is a measure of how much of the bandwidth is occupied by calls of a certain type out of the optimum amount it prefers as compared to other access networks. If two access networks have the same VLL at a certain time for a certain type of call, then both will have the same probability of accepting a call with this service type. The VLL is set with respect to the bandwidth in the access network. Assume that there are only four VLL thresholds with values 0, 1, 2 and 3. The VLLs 0, 1, 2 and 3 correspond to the 0, K3, K2 and K1 levels when using the TTBR algorithm. Although an access network has a single value for the true load level, it has two data bases for the VLLs. When a data call arrives the HCAC only looks at the data call VLLs (i.e. VLL(d)) and when a voice call arrives, it only looks at the voice call VLLs (VLL(v)).

Figure 4. TTBR scheme. Each access network is managed by an HCAC protocol that is assigned to it. The function of the HCAC is similar to the CAC that pre-4G wireless networks employ. However, the proposed HCAC needs to be able to do little more additional functions in order to comply with the heterogeneous environment. It replies for a query by the VCAC (to be explained later) whether or not it is possible to admit a new call and what the current status of the access network is. It also periodically reports on its own time to the VCAC to inform it about the network status. The information that the HCAC gives to the VCAC is determined according to the TTBR algorithm. The thresholds of the TTBR scheme are determined based on service providers’ preferences as to which type of calls they want to serve more.

3.3.3 VCAC Three conditions may trigger the VCAC algorithm. We will discuss two of them, call arrival and necessary handoff, in this section. We will discuss load balancing, which is the main topic of this paper in the next section. When a new call arrives, the VCAC requests traffic information from all HCACs in access networks that are in the vicinity of the user. From the replies the VCAC learns which access networks are able to accept the call and what the preference of each network is for the service request. It then compares the replies from the HCACs to find the most suitable network in the current situation. The flow chart in Figure 5 shows how the VCAC operates when a new call arrives. When a mobile user roams through the heterogeneous network, the HCAC might realize that the mobile terminal is about to be disconnected from the access network to which it is connected because of geographical limitation. In this case, the HCAC informs the VCAC so that it arranges for a vertical handoff using a similar method as in Fig. 3.

Figure 5. VCAC operation: i = an integer representing the type of service request.

Figure 6 illustrates the admission policy in a heterogeneous environment consisting WLAN, GPRS and UMTS. The WLAN is preferred to be always filled with data calls while the UMTS is

preferred to serve more calls (voice/data) than the GPRS in traffic conditions where not all the networks are full. In this figure, the WLAN is chosen to admit a new voice call only if there is no other network that is able to admit it.

The period over which the LRs are sent to the VCAC should not be very short so that the consequent signaling would not be high. It should not be very long either so that the VCAC will be able to quickly react to changes in the traffic conditions [3]. When the VCAC receives LRs, it compares the entries for the different access networks. If there is an unacceptable level of load imbalance, the VCAC attempts to adjust the loads and reduce the imbalance. A few of the most important entries of Table 1 are interpreted as follows. Even though there is no voice user in the WLAN, its VLL of voice service shows 1. Besides, the VLL for its data service is 0. This feature is important in situations where some network technologies are more preferable to serve certain applications with a specified QoS than others. For example, the current WLAN is more preferable to data users than voice.

Figure 6. Example of traffic conditions for three access networks.

When a call request arrives to the RRM domain, the VCAC issues a query to each HCAC asking whether or not the access network can admit the call and what the VLL of the access network is. The reply from the HCACs will have the form (b, VLL) where b takes the values 1 (for yes I can admit the call) and 0 (for no I can admit it). For example, in Figure 6 for a new data call arrival the reply from the HCACs in the WLAN, GPRS, and UMTS will be (1, 0), (0, 1), and (0, 1), respectively. Thus, the VCAC will choose the WLAN to admit the call because it is the only access network that can admit it.

4. THE PROPOSED LOAD BALANCING SCHEME In addition to the reply it makes for queries, the HCAC also reports the traffic condition to the VCAC on its own time. This is done to accomplish load balance among the different access networks in the domain of the RRM. Let us define the HCAC report as the load report (LR). It includes all the types of service that the access network supports, their respective value of VLL, and the number of users in the highest two VLL slots. Note that the highest VLL means the highest VLL value above which the network load is. The VCAC’s knowledge about the access networks looks like the one in Table 1. The LR is sent periodically in a synchronized time manner controlled by the VCAC. This ensures that the events that the VCAC is aware of are simultaneous values of the LRs sent by the different access networks. Table 1. An example of load report of an HCAC . The ordered numbers are VLL, number of users in the highest two VLLs and the number of users that an HCAC is able to admit in the Highest VLL. WLAN

GPRS

UMTS

Voice

1,0,0,0

0,0,0,20000

0,0,0,25000

Data

0,0,1000,200

1,0,8000,2000

1,0,10000,5000

Looking at the first row, the VCAC learns that VLL in the WLAN is the highest. It then looks at the number of users in the highest two VLLs. Note that this is the case where knowledge of the number of users in the first VLL is important. If the VCAC didn’t know this number, it wouldn’t know whether the reason that the number of users in the second level is 0 because the first level is full or only because the WLAN isn’t a preferred network to admit voice calls. The VCAC then concludes that there is nothing to be adjusted in the first row. Using the data in the second row, the VCAC wants to make an adjustment by taking 200 users from the GPRS and the UMTS and connecting them to the WLAN. However, the VCAC should know the location of the users that are connected to the GPRS and the UMTS. It gets this information from the visitor location registers of cells in the cellular networks [9]. In the simpler case, the total data users connected to the cellular networks in the domain of the WLAN are less than or equal to 200. In this case the VCAC simply decides to take these users and make vertical handoffs to the WLAN. If the total data users connected to the cellular networks in the domain of the WLAN are greater than 200, the VCAC randomly takes 200 out of these users and makes vertical handoffs.

5. CONCLUSION Load balancing is an interesting question that appears because of the heterogeneous nature of the next generation wireless systems (NGWS). It will be beneficial in terms of offering better throughput and cost to end users. In this paper we proposed a load balancing scheme for the joint call admission control algorithm of the NGWS. The main features that we attempted to gain in our proposed joint call admission control algorithm are simplicity and scalability. The proposed scheme works by periodically examining the traffic distribution among the different access networks and attempting to switch users from heavily loaded access networks to lightly loaded ones. Future work includes performance analysis of the proposed scheme analytically and in a simulation environment.

6. REFERENCES [1] C. T. Chou and K. G. Shin, “Analysis of combined adaptive bandwidth allocation and admission control in wireless networks,” Proc. IEEE INFOCOM, pp. 676-684, June 2002.

[2] O. E. Falowo and H. A. Chan, “Joint call admission control for integrated UMTS-WLAN network,” SATNAC, 2005. [3] M. Ghaderi, R. Boutaba, and G. W. Kenward, “Stochastic admission control for quality of service in wireless packet networks,” in Proc. 4th IFIP Networking Conference, pp. 1309-1320, May 2005. [4] Y. -R. Huang, Y. B. Lin, and J. M. Ho, “Performance analysis for voice/data integration on a finite mobile systems,” IEEE Trans. Veh. Technol., vol. 49, pp. 367-378, Mar. 2000. [5] W. Kellerer and H. Vogel, “A communication gateway for infrastructure-independent 4G wireless access,” IEEE Comm. Mag., pp. 126-131, Mar. 2002. [6] Y. C. Lai, and S. F. Tsai, “A fair admission control for large bandwidth multimedia applications,” Proc.GLOBECOM, Sam Antonio, TX, pp 317-322, Nov. 2002. [7] B. Li, L. Li, B. Li, K. M. Sivalingam, and X. –R. Cao, “Call admission control for voice/data integrated cellular networks: performance analysis and comparative study,” IEEE Journal on Selected Areas in Comm., vol. 22, pp. 706-718, May 2004. [8] L. –Z. Li, B. Li, and X. –R. Cao, “Performance analysis of bandwidth allocations for multi-service mobile wireless cellular networks,” Proc. IEEE WCNC, pp. 1072-1077, Mar. 2003. [9] Y. -B. Lin, “Reducing location update cost in PCS network,” IEEE/ACM Trans. Networking, vol. 5, pp. 25-33, Feb. 1997. [10] K. Muray, R. Mathur, and D. Pesch, “Adaptive policy based access management in heterogamous wireless networks,” Proc IEEE WPMC, Vol. 1, pp. 325-329, Oct. 2003. [11] C. Perkins, “IP mobility support for IPV4,” IETF RFC 3344, 2002. [12] A. K. Salkintzis, C. Fors, and R. Pazhyannur, “WLAN-GP)RS Integration for Next-Generation Mobile Data Networks,” IEEE Wireless Communications, October 2002. [13] K. H. Suleiman, H. A. Chan, and Mqhele E. Dlodlo, “A hierarchical approach to joint call admission control in heterogeneous wireless networks,” IST-Africa, South Africa, May 2006.

[14] P. Vidales, G. Mapp, F. Stajano, J. Crowcroft, and C. J. Bernardos, “A practical approach for 4G systems: deployment of overlay networks,” TRIDENTCOM, Trento, Italy, pp. 172-181, Feb. 2005. [15] X. G. Wang, G. Min, J. E. Mellor, K. Al-Begain, and L. Guan, “An adaptive QoS framework for integrated cellular and WLAN networks,” Elsevier Journal on Computer Networks, pp. 167-183, 2004. [16] X. G. Wang, G. Min, J. Mellor, and K. Al-Begain, “A QoS based bandwidth management scheme in heterogeneous wireless networks,” International Journal of Simulation Systems, Science and Technology, pp. 9-17, 2004. [17] L. Yin, B. Li, Z. Zhang, and Y. –B. Lin, “Performance analysis of a dual threshold reservation (DTR) scheme for voice/data integrated mobile wireless networks,” Proc. IEEE WCNC, pp. 258-262, Sept. 2000.

[18] F. Yu, V. Wong, and V. Leung, “A new QoS provisioning method for adaptive multimedia in cellular wireless networks,” Proc IEEE INFOCOM, pp. 2131-2142, Mar. 2004. Kamil H. Suleiman received his BSc in physics at Addis Ababa University in 2003. He then did his honours BSc in astrophysics and space science at the University of Cape Town in 2004. Currently he is a master’s student in Electrical Engineering at the University of Cape Town. His research interest is on call admission control in the next generation wireless systems. H. Anthony Chan received his PhD in physics at University of Maryland, College Park in 1982 and then continued post-doctorate research there in basic science. After joining the former AT&T Bell Labs in 1986, his work moved to industry-oriented research in areas of interconnection, electronic packaging, reliability, and assembly in manufacturing, and then moved again to network management, network architecture and standards for both wireless and wireline networks. He had designed the Wireless section of the year 2000 state-of-the-art Network Operation Center in AT&T. He was the AT&T delegate in several standards work groups under 3rd generation partnership program (3GPP). During 2001-2003, he was visiting Endowed Pinson Chair Professor in Networking at San Jose State University. In 2004, he joined University of Cape Town as professor in the Department of Electrical Engineering. Prof. Chan is Administrative Vice President of IEEE CPMT Society and had chaired or served numerous technical committees and conferences. He is distinguished speaker of IEEE CPMT Society and is in the speaker list of IEEE Reliability Society since 1997. Mqhele E. Dlodlo, an Associate Professor in the Electrical Engineering Department since January 2005, holds a doctor’s degree from Delft University of Technology in The Netherlands. He also holds a master’s degree from Kansas State University, USA, and two bachelor’s degrees in Electrical Engineering, Mathematics and independent studies in Engineering Management, from Geneva College, also in the USA. His research area is wireless and personal communication systems trend overview, performance analysis and applications to local needs. He published his doctoral dissertation as a book on the subject in 1996 and has continued collaborating with research colleagues in Southern Africa and abroad in conferences, symposia and seminars. He has been on conference technical committees and remains a referee in his field. Occasionally, he serves as an external examiner for post-graduate degree candidates. At UCT, he teaches courses in telecommunication networks and systems, participates in the Communications Research Group. A Member of IEEE since 1988 and a Fellow of the Zimbabwe Institution of Engineers since 2000, Mqhele has worked as lecturer, senior lecturer, principal lecturer and senior academic administrator in higher education institutions in Zimbabwe since 1983. In this period, he taught telecommunication, electronic and electrical technicians at the Bulawayo Polytechnic where he eventually chaired the Department of Electrical Craft. He also co-ordinated the design of the National Technician curricula for the certificate, diploma and higher diploma programs during that time.