to support inter-switch handoff in wireless ATM networks. It employs path extension for each inter-switch handoff,. followed by path optimization if necessary.
Path Optimization for Inter-Switch Handoff in Wireless ATM Networks W. S. Vincent Wong, Henry C. B. Chan and Victor C. M. Leung Department of Electrical and Computer Engineering University of British Columbia 2356 Main Mall, Vancouver B.C., Canada V6T 1Z4 {vincentw, chunc, vleung}@ece.ubc.ca
Abstract Recently, a two-phase handoff protocol has been proposed to support inter-switch handoff in wireless ATM networks. It employs path extension for each inter-switch handoff, followed by path optimization if necessary. In this paper, we propose a number of path optimization initiation schemes to facilitate the two-phase handoff protocol. The design objective is to determine the time to invoke path optimization such that the average cost per connection is minimized. Three path optimization schemes, namely: periodic, exponential, and Bernoulli, are proposed. A discrete-event simulation model is developed to compare their performance. Simulation results indicate that the Bernoulli path optimization scheme outperforms the other two schemes by providing a lower average cost per connection.
1. Introduction The phenomenal growth in cellular telephony over the past few years has stimulated active research in providing broadband multimedia to the mobile users. The advent of wireless ATM (Asynchronous Transfer Mode) [1] is intended to provide direct wireless extensions to the core ATM networks and services. In order to support mobility in ATM networks, a number of technical challenges need to be resolved [2]. Our work focuses on handoff control in wireless ATM networks. Handoff occurs when a mobile terminal moves from one base station to another. ATM handoff differs from traditional voice handoff in that a mobile user may have several active connections with different bandwidth requirements and quality-of-service (QoS) constraints. The handoff function should ensure that all these ongoing connections are rerouted to another access point in a seamless manner. In other words, the design goal is to prevent service disruptions and degradation during and after the handoff process. Recently, several connection rerouting protocols to facilitate inter-switch handoff have been proposed in the literature [3]-[8]. Two methods that are based upon partial re-establishment are the path extension and the path rerouting schemes. The rationale behind path extension is
This work was supported in part by the Natural Sciences and Engineering Research Council of Canada under a Postgraduate Scholarship and Grant OGP0044286; and by the Fessenden Postgraduate Scholarship from the Communications Research Centre, Industry Canada.
to extend the original connection to the switch where the new base station is located. The switch to which the original base station is connected is usually referred to as the anchor switch, and the switch to which the new base station is connected is called the target switch [4]. The path extension method extends the connection from the anchor switch to the target switch during handoff. The minimum hop path is usually chosen as the extended path. The path extension scheme is fast and simple to implement. QoS degradations such as cell loss, duplicate cells, and mis-sequence cells do not occur. However, since the extended path is longer than the original one, certain QoS requirements, such as cell transfer delay and cell delay variation, may not be guaranteed after handoff. Path rerouting can be considered as a generalization of the path extension scheme. In path extension, the anchor switch extends the original connection to the target switch, while in path rerouting, any switch along the original connection can be selected to set up a branch connection to the target switch. The switch chosen to perform this function is usually referred to as the crossover switch [5] or the handoff switch [6]. Despite the fact that path rerouting has the potential of creating an optimal path, the time to determine the location of the crossover switch and the time for branch connection set up may exceed the stringent handoff delay. Recently, a two-phase handoff protocol has been proposed [7] which combines the advantages of path extension and path rerouting schemes. It employs path extension to process handoff request immediately, followed by path optimization if necessary. The advantage of the two-phase handoff protocol is that path extension ensures a fast handoff while path optimization increases the network utilization by using a more efficient path. Since the mobile terminal is still communicating over the extended path via the current base station while path optimization takes place, this gives enough time for the network to perform the necessary functions while minimizing any service disruptions. To implement the two-phase handoff protocol, two important issues need to be addressed: 1. What are the procedures required for path optimization such that service disruptions are minimized? 2. When and how often should path optimization be performed? The first issue is concerned with network signaling to maintain ATM cells sequencing and minimize cell loss. Related work can be found in [9]. In this paper, we focus
Remote terminal
the execution of path optimization, path extension can still be used to extend the connection to the target switch. Although path optimization can increase the network utilization by rerouting the connection to a more efficient route, transient QoS degradations such as ATM cell loss and an increase in cell delay variation may occur. In addition, path optimization increases the signaling load of the network. Thus, path optimization after each path extension may not be necessary or desirable.
Crossover switch
3. Path Optimization Schemes
Mobile terminal Original VC
Base station
ATM switch
Partial VC
Optimal VC
Figure 1. Two-Phase Handoff Protocol on the second issue by proposing a number of path optimization initiation schemes. These schemes provide the means of optimizing the frequency of path optimization with respect to the average cost per connection. The rest of the paper is organized as follows. Section 2 explains the two-phase handoff protocol in more detail. The proposed path optimization initiation schemes are described in Section 3. The simulation model employed for the comparisons of the schemes are presented in Section 4. Section 5 presents the simulation results. Conclusions and further research directions are given in Section 6.
2. Two-Phase Handoff Protocol The two-phase handoff protocol consists of two stages, namely: path extension and possible path optimization. Referring to Figure 1, when a mobile terminal moves to a new base station connecting to a different switch, an interswitch handoff request is generated. To maintain the call connectivity, in phase 1, path extension is used to establish a branch connection between the anchor switch and the target switch. Although path extension can be completed within a short time period, it may lengthen the path or increase the number of hops. In certain situations, loops can also be created. Therefore in the second phase, path optimization may be invoked to set up an optimal path between the source and the destination. In general, the major steps during path optimization execution involves: 1. Determining the location of the crossover switch; 2. Setting up a new branch connection; 3. Transferring the user information from the old branch connection to the new one; 4. Terminating the old branch connection. Notice that path optimization described above is not restricted to the two-phase handoff protocol, but it can also be applied to other connection rerouting protocols where the location of the crossover switch is not the optimal one. When the mobile terminal moves to another switch during
In general, path optimization can be grouped into four types, depending on how it is initiated. These schemes are explained below:
3.1
QoS-based
As the name implies, QoS-based path optimization schemes trigger path optimization of each connection based on its current QoS measures. For example, path optimization can be initiated if the number of hops is greater than a certain number, or the end-to-end cell transfer delay bound is violated. To implement QoS-based path optimization schemes, information about the quality of the current path in terms of the defined QoS measure (e.g., hop count, current average delay) must be maintained by the network.
3.2
Network-based
Network-based path optimization schemes trigger path optimizations for a group of connections based on the existing traffic load or the utilization of the network. For example, the edge switch can initiate path optimization for a group of mobile connections whenever the new call dropping probability of a certain traffic class reaches a certain threshold.
3.3
Time-based
In this case, path optimization is triggered at time instants which are independent of the current QoS and the network load. The time instant can be deterministic or random. In this paper, we analyze two time-based schemes which are simple to implement. 1. Periodic path optimization scheme: The time to perform path optimization is periodic with period T . 2. Exponential path optimization scheme: The time to perform path optimization is modeled as an exponentially distributed random variable with mean 1 ⁄ ν . The periodic scheme has been proposed within the ATM Forum wireless ATM working group to facilitate the backward handoff protocol [10]. Both the periodic and the exponential schemes trigger path optimization independent of whether or not there is an inter-switch handoff. Thus, unnecessary path optimizations may be performed for stationary mobile connections.
3.4
Random-based
Random-based path optimization schemes trigger path optimization for each mobile connection based on some
Table 1: Summary of parameters used in simulation
100 90
item
80 Vertical Distance
number of nodes in the network 70
symbol N
value 20
100 × 100
size of the rectangular coordinate grid
δ max
60
maximum queueing delay (ms)
50
time between inter-switch handoff (min) 1 ⁄ λ
5
40
average call duration (minutes)
30
1⁄µ C PE C PO C link ν T p
20
signaling cost per path extension signaling cost per path optimization
20
link cost per unit time per link 10
path optimization rate
0 0
20
40 60 Horizontal Distance
80
100
path optimization period (minutes) path optimization probability
100
1 5 0.2 - 1 variable variable variable
Figure 2. A 20-node random graph with node degree of 4; minimum distance between any two nodes is 15.
denote as d ij , is equal to: d ij = d ( i, j ) + δ
random processes after each inter-switch handoff. In this paper, we also analyze the following scheme: • Bernoulli path optimization scheme: After each path extension, there is a probability p , 0 ≤ p ≤ 1 , such that path optimization is performed. For the Bernoulli scheme, a path optimization may only occur on the condition that there is an inter-switch handoff. Thus, no path optimization is triggered after each handoff when p = 0 . On the other hand, path optimization is performed after each path extension when p = 1 . In the remaining of this paper, we will compare the performance of the periodic, exponential, and the Bernoulli path optimization schemes, and determine how to assign the optimal values for ν, T and p given a set of call and mobility parameters.
where δ is a uniformly distributed random variable in the range 0 ≤ δ ≤ δ max . In equation. (1), the first term can be interpreted as the propagation delay of the link, and the second term as the queueing delay of the link. Note that if the second term is removed, then for any pair of nodes, the shortest delay path and the minimum hop path are the same. For the simulation, a 20-node random graph is generated from the above model. The size of the rectangular coordinate grid is 100 × 100 . The average node degree of the graph is 4. For equation (1), δ max is set to be equal to 100. The network model is shown in Figure 2. Each node represents an ATM switch and each edge represents a physical link connecting the two switches. For each simulation run, two nodes are chosen randomly as the source node and the destination node. Dijskstra’s algorithm [12] is used to compute the shortest delay path between these two nodes. The source node is assumed to be stationary. The destination node becomes the anchor switch of the mobile connection. The call duration and the time between inter-switch handoffs are modeled as exponentially distributed random variables with mean 1 ⁄ µ and 1 ⁄ λ , respectively. During each inter-switch handoff, the target switch is restricted to be one of the neighboring switches of the current anchor switch. Path extension is used to extend the connection from the anchor switch to the target switch. Subsequent path optimizations may be triggered based on different initiation schemes described in Section 3. The performance metric is the average cost of the call, which is defined as the sum of the link cost and the signaling cost due to inter-switch handoff. We let: • C link denote the link cost per unit time per link, • C PE denote the signaling cost for each path extension event, • C PO denote the signaling cost for each path optimization event.
4. Simulation Model In this section, we present the framework of the simulation model for the comparisons of different path optimization initiation schemes for the two-phase handoff protocol. An ATM network is modeled as a random graph based on the minimum spanning tree. The generation of the random graph consists of the following steps [11]: 1. N nodes are randomly distributed over a rectangular coordinate grid. Each node is placed at a location with integer coordinates. A node can be rejected if it is too close to another node. The Euclidean metric is then used to determine the distance d(i, j) between each pair of nodes (i, j). 2. A fully connected graph is constructed with the link weight equals to the Euclidean distance. 3. Based on the fully connected graph, a minimum spanning tree is constructed. To achieve a specified average node degree of the graph, edges are added one at a time with increasing distance. If node i and j are connected, then the link delay,
(1)
160
140
Clink = 1 Clink = .8 Clink = .6 Clink = .4 Clink = .2
120
Clink = 1 Clink = .8 Clink = .6 Clink = .4 Clink = .2
140 120 Average Cost
Average Cost
100
80
100 80
60
60
40
40 20
20 0
10 20 30 Path Optimization Period
40
The cost of the call is defined as: m
∑ τi li C link
0.1
0.2 0.3 Path Optimization Rate
0.4
Figure 4. Exponential path optimization scheme
Figure 3. Periodic path optimization scheme
Cost = n PE C PE + n PO C PO +
0
(2)
i=1
where n PE and n PO denote the number of path extension and path optimization, respectively; m is the number of events per call; τ i denotes the elapsed time between event i – 1 and i; and l i denotes the number of links between two end nodes during τ i . The total link cost of the call, which captures the amount network resources used per connection, is a function of time and the number of links. The total signaling cost due to path extension and path optimization captures the processing complexity at the base station and switches to perform those functions. Since we are only interested in the relative performance between different path optimization schemes, other signaling cost including the cost for call setup and termination, as well as the cost for location update and intra-switch handoff are not considered in this paper. In general, path optimization is more complex than path extension, so that the cost to perform path optimization is higher than the cost to perform path extension. Thus, it is reasonable to assume that C PO » C PE . In addition, for convenience, C link and C PO are normalized with respect to C PE . The simulation model described above can be extended to handle other network topologies (e.g., star, ring, and hierarchical), as well as other statistical distributions for the call duration and the time between interswitch handoff.
5. Simulation Results The baseline simulation parameters are summarized in Table 1. Figures 3 - 5 show the simulation results for the periodic, exponential, and the Bernoulli path optimization schemes, respectively. For each scheme, given a link cost
C link , there exists an optimal value or a range of values of T , ν , or p , such that the average cost per connection is minimized. For example, in Figure 3, when C link = 1 , the optimal path optimization period T is 8 minutes. These optimal values or the range of optimal values are the operating points or operating interval of that particular path optimization scheme. Results also show that the average cost per connection is sensitive to the changes in link cost C link . This indicates that the last term in equation (2) is the dominant term of the average cost per connection. Figure 6 shows the performance comparisons between different path optimization schemes. The Bernoulli scheme for p = 0 and p = 1 are also included for comparisons. For each path optimization scheme, its optimal value of T, ν , or p, is chosen to give minimum average cost per connection. As shown Figure 6, the average cost per connection for the Bernoulli p = 0 case is much higher than the other schemes once C link ≥ 0.3 . This justifies the importance of using path optimization. In general, the exponential scheme has a higher cost than both periodic and Bernoulli schemes. On the whole, the performance between the periodic and the Bernoulli schemes are very close. The Bernoulli scheme slightly outperforms the periodic scheme when C link ≥ 0.8 . The performances of the optimal- p and p = 1 Bernoulli schemes are almost the same once C link ≥ 0.6 , which indicates that the optimal p value for the Bernoulli scheme is 1. Other simulations results for different values of λ and µ also indicate that the Bernoulli scheme has a better performance over other schemes.
6. Conclusions In this paper, we have proposed a number of path optimization initiation schemes and analyzed three schemes in particular (namely: exponential, periodic, and Bernoulli) to facilitate the two-phase handoff protocol. The design objective is to determine the time to invoke
160
160 Clink = 1 Clink = .8 Clink = .6 Clink = .4 Clink = .2
Average Cost
120
Minimum Average Cost
140
Bernoulli (p = 0) Exponential Periodic Bernoulli (p = 1) Bernoulli
140
100 80 60
120 100 80 60 40
40
20
20
0 0
0.2
0.4 0.6 0.8 Path Optimization Probability
1
Figure 5. Bernoulli path optimization scheme path optimization such that the average cost of the connection is minimized. A discrete event simulation model has been developed to compare the performance of these schemes. Simulation results indicate that the Bernoulli path optimization scheme outperforms the other two schemes by providing a lower average cost per connection. Analytical work for the three path optimization schemes is in progress. Given the call and mobility parameters as well as the network costs, we believe there exists an optimal strategy to trigger path optimization after each path extension. Further work will include analyzing some other QoS-based and network-based path optimization schemes, as well as formulating the path optimization initiation problem as a stochastic optimization problem.
References [1] D. Raychaudhuri and N. D. Wilson, “ATM-Based Transport Architecture for Multiservices Wireless Personal Communication Networks,” IEEE Journal on Selected Areas in Communications, vol. 12, no. 8, pp. 1401-1414, October 1994. [2] IEEE Communications Magazine, Special Issue on Introduction to Mobile and Wireless ATM, vol. 35, no. 11, November 1997. [3] A. S. Acampora and M. Naghshineh, “An Architecture and Methodology for Mobile-Executed Hand-off in Cellular ATM Networks,” IEEE Journal on Selected Areas in Communications, vol. 12, no. 8, pp. 1365-1375, October 1994. [4] B. A. Akyol and D. C. Cox, “Signaling Alternatives in a Wireless ATM Network,” IEEE Journal on Selected Areas in Communications, vol. 15, no. 1, pp. 35-49, January 1997.
0
0.2
0.4
0.6
0.8
1
Clink
Figure 6. Comparisons between different path optimization schemes [5] C.-K. Toh, “Performance Evaluation of Crossover Switch Discovery Algorithms for Wireless ATM LANs,” Proceedings of IEEE INFOCOM’96, San Francisco, CA, pp. 1380-1387, 1996. [6] R. Yuan, S. K. Biswas, and D. Raychaudhuri, “A Signaling and Control Architecture for Mobility Support in Wireless ATM Networks,” Proceedings of IEEE ICC’96, Dallas, Texas, pp. 478-484, 1996. [7] M. Veeraraghavan, M. Karol, and K. Y. Eng, “A Combined Handoff Scheme for Mobile ATM Networks,” ATM Forum Contribution 96-1700, December 1996. [8] B. A. J. Banh, G. J. Anido, and E. Dutkiewicz, “Handover Re-routing Schemes for Connection Oriented Services in Mobile ATM Networks,” Proceedings of IEEE INFOCOM’98, San Francisco, CA, pp. 1139-1146, 1998. [9] W. S. V. Wong and V. C. M. Leung, “A Path Optimization Signaling Protocol for Inter-Switch Handoff in Wireless ATM Networks,” Proceedings of 1st International Workshop on Wireless Mobile ATM Implementation (wmATM’98), Hangzhou, China, pp. 92-98, April 1998. [10] The ATM Forum Wireless ATM Working Group, “Baseline Text for Wireless ATM Specifications,” BTD-WATM-01.07, April 1998. [11] M. B. Doar, “A Better Model for Generating Test Networks,” Proceedings of IEEE GLOBECOM’96, London, UK, pp. 86-93, 1996. [12] D. Bertsekas and R. Gallager, Data Networks, second edition, Englewood Cliffs, NJ: Prentice Hall, 1992.
