- Email: [email protected]
A distributed multicast tree using share link migration scheme for wireless asynchronous transfer mode network
Computers and Electrical Engineering 29 (2003) 395–406 www.elsevier.com/locate/compeleceng
A distributed multicast tree using share link migration scheme for wireless asynchronous transfer mode network H.-C. Chao
a,*
, Ching-Rong K. Lay b, J.L. Chen
c,1
a
c
Department of Electrical Engineering, National Dong Hwa University, Hualien 97401, Taiwan, ROC b Institute of Nuclear Energy Research, Taoyuan 32546, Taiwan, ROC Department of Computer Science and Information Engineering, National Dong Hwa University, Hualien 97401, Taiwan, ROC Received 9 June 2000; received in revised form 26 October 2000; accepted 3 May 2001
Abstract In recent years, wireless asynchronous transfer mode networks have become popular and are in widespread use for supporting various types of data transmission. To minimize the cost in setting up routes and to meet the QoS constraint in time-sensitive traffic, in addition to bandwidth capacity, this paper focuses on the optimization of the number of shared links to reduce the call-blocking rate and the handoff failure rate for intercell roaming. The average longest path length and the average of the average path length of the constructed multicast trees are shown to be stable when the network load becomes heavy. Ó 2002 Elsevier Science Ltd. All rights reserved. Keywords: Asynchronous transfer mode; Multicast; Migration; Wireless; Handoff
1. Introduction Integrating multicasting and mobility into internet architectures has become more and more popular [1,2]. Paired with the rapid developments in asynchronous transfer mode (ATM) networking technology, the rise of wireless communications signals the start of a new era in telecommunications. In this era, not only will users need higher bandwidth; they will also demand mobility. A typical future wireless user may be carrying a hand-held computer (PDA) with audio
*
Corresponding author. Tel.: +886-3-866-8500/2500x17001. E-mail addresses: [email protected] (H.-C. Chao), [email protected] (C.-R.K. Lay), [email protected] (J.L. Chen). 1 Tel.: +886-3-866-2500x22116. 0045-7906/02/$ - see front matter Ó 2002 Elsevier Science Ltd. All rights reserved. PII: S 0 0 4 5 - 7 9 0 6 ( 0 1 ) 0 0 0 4 0 - 4
396
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
and video conferencing capabilities and may demand high-speed data communication on the order of 10 megabits per second or higher. A wireless ATM network, which is designed to provide isochronous and asynchronous high speed communications for wireless users, is a good match for these demands. The wireless ATM network consists of base stations (BS), mobiles, and network interface equipments. A BS may contain several radio ports to support mobile users. A group of mobile users is connected to the same wireless ATM network interface equipment. This collection of mobiles is called a cell. The cell architecture is shown in Fig. 1. A mobile user might have a few simultaneous connections in the wireless ATM network. When a handoff occurs, these connections may need to be re-routed. Handoff is an important function of the network where users are mobile and is implemented by the network to give the users freedom of motion beyond a limited wireless coverage area while they are communicating. The handoff is the procedure by which a user’s radio link is transferred from one radio port to another through the network without an interruption to the user connection. We can divide the handoff event into two levels. One is the network level and the other is the radio level. The radio level handoff is the actual transfer of the radio link between two ports. The network level handoff supports the radio level handoff by performing re-routing and buffering. Some of the procedures used in the network level handoff are determined by the radio level handoff. There is a cell manager for control and management within each cell. We assume that the cell managers have some knowledge about the neighboring cells and the network addresses of neighboring cells are stored in a local lookup table and it is updated periodically by means of an update protocol. We also assume that these cells are interconnected by wireless ATM network switching nodes. According to these concepts, we investigated the handoff using two different situations: intracell handoff (Fig. 2) and intercell handoff (Fig. 3). In the intracell handoff, the user
Fig. 1. A cell architecture in wireless ATM network.
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
397
Fig. 2. Intracell handoff.
Fig. 3. Intercell handoff.
is roaming within the cell. The only routing/re-routing performed is in the wireless ATM network interface equipment within the cell. The cell manager in the BS is responsible for the correct update of ATM virtual circuit translation tables within the cell. This type of routing/re-routing does not require wide area ATM network switching. The intercell handoff takes place when the mobiles roam into a different cell. In this case, the routing/re-routing involves the wireless ATM network. An intercell handoff might require one or more wireless ATM switches, depending on the handoff location and topology of the network.
398
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
When a mobile terminal becomes active in the wireless ATM network, we must setup a connection between the source node and the destination node before starting to transmit the data. The mobile terminal must first look up the databases maintained in the node. At the same time, the source node broadcasts search messages to find the neighboring nodes that have existing routes toward the destination node. We used the two-level shortest feasible path routing algorithm to find a common node that belongs to the shortest sub-path of the connection between the source node and the destination node. The most efficient route will be selected and other routes will be released. Since the traffic in wireless ATM networks can be divided into two types: time-sensitive and throughput-dependent traffic. In a lightly loaded network, we can use the same routing scheme to construct the time-sensitive multicast trees and throughput-dependent multicast trees. While in a heavily loaded network, to guarantee QoS, we can use different routing schemes to construct the different multicast trees separately. Furthermore, if mobiles roam within the same cell (intracell), we do not need to rebuild a new connection to the ATM switch node. If mobiles roam into another cell (intercell), we must perform common node re-routing to setup a new connection between the common node and the destination node.
2. Algorithms IP multicasting is based on the concept of the host group [3]. As stated in Ref. [4], the shortestfeasible-path routing (SFPR) scheme always tries to find the shortest path among all of the feasible paths between the source and the destination nodes. This scheme is good when the performance is measured by the average route distance. However, this scheme has a higher blocking probability for future calls when the call requests are focused on some hot vertices or edges. The least loaded routing (LLR) scheme is proposed to overcome the SFPR drawback. A lower call-blocking rate might be obtained with this scheme. However, the tradeoff is that the average route distance is prolonged because the least loaded route may not be the shortest. Even under light loads, the LLR still changes the route dynamically and this is not necessary. The traditional multicast route has focused on reducing the bandwidth and delay cost in building the path. We considered how to allocate bandwidth in a multicast tree so as to optimize some global measure of performance. For supporting multimedia applications, which require realtime data transmission in addition to end-to-end delay, the route bandwidth must be reserved so that the minimum transmission rate is guaranteed [5]. When the traffic is time-sensitive and the user is a newly joined one, the connection should be setup as soon as possible. We first examine and distinguish the neighboring nodes that belong to the light-load node group (Gl ). If we can find such a node, it is involved in the shortest feasible path routing. If not, we shall search the nodes that belong to the heavy-load nodes group (Gh ). If none is available, we can suspend the throughput-dependent traffic for a while and try to meet the bandwidth requirements needed by the time-sensitive applications later. If the suspension of throughput-dependent traffic cannot meet the required bandwidth, we try other nodes that belong to Gh and these nodes are preferred to the shortest feasible path routing. In the mean time, if the time limit suspension expires, we can continue the throughput-dependent traffic in a round robin queue and search for the dynamic
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
399
alternative routing to avoid the indefinite postponement problems. If all of these methods are not feasible, the join request is rejected. If the neighboring nodes are available, we first check if there exists a path to the same server node. If the answer is positive, we can share the same path links if all links are available. In the mean time, we try the two-level SFPR algorithm to find the shortest path that has a common node to the multicast tree. If the bandwidth capacity cannot meet the required bandwidth, the user’s request is rejected. The two level shortest feasible path algorithm is described as follows: Step 1: Search the preferred neighboring nodes in Gl , if nodes in Gl are not available, then search nodes in Gh . If both are unavailable, we can suspend the throughput-dependent traffic optionally. If this still cannot meet the requirements, the request is rejected. Step 2: If neighboring nodes are available, search the common node, which is destined to the same server, from the source to the destination node with the two level shortest feasible path algorithm to construct the shortest path. At the same time, we use the two level shortest feasible path routing algorithm to find the shortest feasible sub-path to the multicast tree if the bandwidth capacity of every link is feasible. Step 3: Select the shortest path that has common node to the same multicast tree; if it is unavailable, select the shortest feasible sub-path.
3. Migration comparison for exclusive and shared links As stated in Refs. [6,7], if a mobile user roams into another cell (intercell), it uses the shortest feasible routing algorithm to setup the connection. Then the candidate node sends control messages to find the nearest common node to do the exclusive link migration. To improve the overall performance of the network, we used the two-level SFPR algorithm to setup the connection. In addition, to reduce the overhead incurred by the control signal broadcast messages in performing the exclusive link migration, we execute the migration and the connection setup at the same time. After roaming through all of the mobile users, it first uses the common node re-routing which has the shortest path to share the link with as many as possible. If sharing the shortest connection is not available, then a new shortest feasible connection is setup. The network efficiency is evaluated to compare with a threshold value. As more users are dispersed throughout a wide area ATM network, the bandwidth resources consumed by migration schemes for exclusive and shared links becomes an important factor, which can affect the efficiency of the network. The algorithm for the exclusive link migration scheme is stated briefly as follows. 3.1. Exclusive link migration scheme For v 2 T , T is the set of leaf nodes {Send search messages to all adjacent nodes u 2 U , U is the set of adjacent nodes; Update the search messages for each adjacent node; If (an eligible node is not found) {discard messages; release the channels occupied;} Otherwise {reserve channels for the connection;} Select the first confirmation message returned from u 2 U; Release the channels not used in this connection;
400
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
Inform the nearest common node 2 U to perform the connection chores if available Otherwise perform shortest feasible path routing; }
3.2. Proposed scheme We propose a distributed algorithm with or without a shared link migration scheme to compare with the exclusive link migration scheme. Instead of finding the nearest common node of the multicast tree, we broadcast ‘search’ messages to find feasible paths that attach to the same multicast tree. We then selected the shortest path among them. If this path is unavailable, we rebuild a new connection between the source and destination node. Once the path has been selected, the other paths will be released. The distributed algorithm of our scheme is stated below: Step 1: Find common nodes that have connections to the destination. Step 2: Select the most efficient path; release the others. Step 3: If both are unavailable, setup a new path using TSFPR. Similarly, the shared link migration scheme can be activated from the common node after the roaming user becomes stable. If the shared link migration results in network congestion, this migration will be aborted. We can maintain the status of each link in the nodes involved in the connection between the source and the destination node. After reducing the number of shared links, we expect that the blocking rate and handoff rate will be reduced. This means that the probability of joining and roaming in the network increases. We also anticipate that the average of the average path length (AAL) of these multicast trees becomes stable and the bandwidth network resource will not be consumed accordingly. The proposed share link migration algorithm is stated as follows: Step 1: The leaf node sends a control signal to the common nodes of this path. Step 2: The first common node then sends messages to find a shared path that is shorter than the old path. Step 3: If found then perform shared link migration otherwise search for the next common node. Step 4: If no common migration remained in the old path then go to Step 3.
4. Simulation model In this paper, a torus style network model with bi-directional links connecting all its neighboring nodes is considered (Fig. 4). The model for the network is an ATM switch-based backbone with a 10 10 size. Since the focus of our work is on the construction and improvement of multicast trees, which are built and based on a wired ATM network, we will ignore the wireless portions of this network. Each vertex represents a cell that is connected to an ATM switch and with a unique identifier. There are databases associated with each cell in the network to keep track of all transactions. The information that is recorded in these databases includes traffic type, routing/re-routing strategies, residual capacity, throughput-dependent data buffering, and so on.
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
401
Fig. 4. Example of the torus network.
We assumed that the transmission capacity of each link is 150 megabit per second. Since the transmission rate of wireless portions in wireless ATM network is about 2 megabit per second, we assumed that seventy-five multimedia services can fill up a link. With 50 megabit per second of capacity reserved for voice, data and control signaling, we have 100 megabytes per second for multimedia applications. The assumed maximum number of joined mobile terminals for each multicast service, such as video on demand and video conferencing, is limited as M. The destined server is randomly selected from the existing feasible servers for a newly joining request. When the associated multicast tree has a size less than the limitation M, we say that an existing server is available. We used the two level SFPR algorithm to establish the new join connection. When a new mobile joins the network, the two level SFPR algorithm is used to find a shortest feasible path from the source node to the node that is on the same multicast tree and that has the shortest path to the destination node. At the same time, we locate the shortest sub-path to the multicast tree that is destined to the same server. Then we select the most efficient path to be our new connection and release the others. We assume that there are 2000 users waiting to join the network. The arrival rate of new join requests is assumed to be a Poisson distribution with mean A. When the residual capacity of the path is exhausted or the limitation of the multicast tree M is exceeded, the join request of a mobile unit is rejected. The departure rate of joined mobiles is also assumed to be a Poisson distribution with mean D. We select R users among the active users randomly to roam within the network. If the roaming mobiles move within the same cell, we do not need to rebuild the connection. Otherwise, we must perform the re-routing algorithm to find an efficient path. In the simulation of our work, the following parameters were considered: (1) There are 2000 users waiting to participate in the network. (2) On average, there are ten mobile users which are randomly selected to depart from the network per unit time, D ¼ 10. (3) On average, there are one hundred mobile users who arrive and join the network per unit of time, A ¼ 100. (4) We utilized R ¼ 10 and 25 to investigate the behavior of the network. (5) We assumed that the maximum number of mobile users allowed to join in a multicast tree is twenty, M ¼ 20 and that there are 10 servers randomly selected from these network nodes.
402
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
In this network we applied the two-level SFPR scheme that finds the shortest path to the same multicast tree first and attempts to find the SFPR with the exclusive link migration scheme. Then we select the shortest feasible path as our new path. The shared link migration scheme was implemented to investigate the performance of the same network. For a generated mobile user request ri ð1 6 i 6 cÞ, if the request is accepted, we let accept ðri Þ ¼ 1; otherwise, accept ðri Þ ¼ 0. While the mobile unit roams, we let roam_accept ðri Þ ¼ 1 if the request of connection is accepted; otherwise, roam_accept ðri Þ ¼ 0. We also let distance ðri Þ equal the hop count of the established connection between the source and the destination node if the request ri is accepted; otherwise, distance ðri Þ is set to be infinity. We define the following metrics to evaluate the performance P of the migration schemes: (1) Call blocking rate: 1 ci¼1 accept ðri Þ=c, ri ¼ 0; c is the total users that want to join in the network. P (2) Handoff failure rate: 1 di¼1 roam accept ðri Þ=d, ri ¼ 0; d is the total users that want to roam within the network. P Pt p (3) AAL: Let AVG P ðiÞ ¼ i¼1 Path LengthðiÞ=p, we have the AAL as i¼1 AVG P ðiÞ=t. (4) ALP: the average of the longest path length in the constructed multicast trees; ALP ¼ (total of the length of the longest path in every constructed multicast tree)/(number of multicast trees).
5. Simulation results The performance of the three schemes on the torus network is shown in Figs. 5–10. Figs. 5 and 6 show the behaviors of the proposed scheme under R ¼ 10. It is clear that the benefit is not so attracting due to low roaming rate. Fig. 7 shows the blocking rates while the number of roaming users is twenty-five. Obviously, the scheme with share link migration can reduce the call-blocking rate. This means that the new joining mobiles will have more chances to build a connection successfully. Using our scheme without performing share link migration, the call-blocking rate is
Fig. 5. Call blocking rate with R ¼ 10.
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
Fig. 6. Handoff failure rate with R ¼ 10.
Fig. 7. Call blocking rate with R ¼ 25.
Fig. 8. Handoff failure rate with R ¼ 25.
403
404
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
Fig. 9. AAL with R ¼ 25.
Fig. 10. ALP with R ¼ 25.
better than that with the exclusive link migration scheme since it consumes less bandwidth resource. Owing to the hot node effect, the network will be saturated with the control messages for building a new connection. This accounts for the abrupt rising of the call-blocking rate and handoff failure rate curves. After that saturation point, the call-blocking rate increases slowly. In a wireless ATM network, if the number of roaming users is reasonable, usually R=A 6 30% and in our case R=A ¼ 25%, the overhead incurred by sending control messages does not affect the performance of the share link migration scheme that much. In Fig. 8, there is also a saturation region. However, our scheme without share link migration is still better than that using exclusive link migration scheme. This is because the probability of rejection decreases using the proposed scheme. In Fig. 9, the AAL of our scheme with share link migration is better than that with exclusive link migration scheme. Our scheme without share link migration also has a shorter AAL than the exclusive link migration scheme. Once we select the path in our scheme, it is the
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
405
shortest among all other paths. This is why our scheme, without performing further migration, has a shorter AAL than the exclusive link migration scheme. Our scheme has a good metric because a longer path length means more bandwidth resources consumption. While the roaming users increase, the probability of selecting a node that is destined to the same multicast tree increases. The tradeoff between the bandwidth that is occupied by control messages and the increased probability that a user can build a new connection makes the blocking rate grow slowly. In Fig. 10, the average of the longest path length in the constructed multicast trees tends to be stable. Our schemes with or without share link migration have shorter longest path lengths than the exclusive link migration scheme. This metric also accounts for the reason why our schemes spend less bandwidth resources and have lower call-blocking rates and handoff failure rates.
6. Conclusion Three algorithms for the construction of multicast trees on wireless ATM communication networks were studied and compared in this paper. The proposed heuristic algorithm with/ without share link migration shows that it has a lower blocking rate and handoff failure rate when the network is heavily loaded and a lot of roaming users. Finally, we demonstrated that when mobile users increase and the network load becomes heavy, the AAL and ALP metrics become stable and the proposed scheme with/without share link migration is better than the exclusive link migration scheme.
Acknowledgements Authors wish to thank S.W. Chang and M.T. Lin for their help with this research.
References [1] Polyzos GC. IP multicast for mobile hosts. IEEE Commun Mag 1997. p. 54–8. [2] Akyildiz IF, McNair J, Ho J, Uzunalioglu H, Wang W. Mobility management in current and future communications networks. Georgia Institute of technology. IEEE Network 1998: 39–49. [3] Deering S. Host extensions for IP multicasting. Internet RFC 1112, August 1989. [4] Huang N-F, Wu C-S, Wu Y-J. Some routing problems on broadband ISDN. Comput Networks ISDN Syst 1994;27:101–16. [5] Lee O, Im Y, Lee K, CHOI Y. A bandwidth and delay constrained minimum cost multicast routing algorithm. ICOIN-11 Proceedings, vol. 2. January 1997, Distributed algorithm; DA(2). p. 9c-4.1–4.5. [6] Huang N-F, Chen K-S. A distributed paths migration scheme for IEEE 802.6 based personal communication networks. IEEE J Selected Areas Commun 1994;12(8):1415–25. [7] Huang N-F, Chen K-S, Chao HC, Pan JY. A distributed multicast tree migration scheme for ATM-based personal communication networks. IEEE Globecom’96, 1996 Nov 18–22; London. p. 310.
406
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406 Han-Chieh Chao received the B.S.E.E. degree from National Cheng Kung University, Taiwan, R.O.C., in 1985, and the M.S. and Ph.D. degrees in Electrical Engineering from Purdue University, Indiana in 1989 and 1993, respectively. From 1994–1997, he was an associate professor of Department of Computer Science and Information Engineering at National Dong Hwa University, Hualien, Taiwan, R.O.C. Since August 1997, he has transferred to Department of Electrical Engineering and also served as the director of the University Computer Center. His current research interests include mobile networks, next generation Internet, and highspeed networks.
Ching-Rong Lai was born in Taiwan on December 15, 1962. He received the Master degree in Department of Computer Science and Information Engineering from National Dong Hwa University, Hualien, Taiwan in 1997. He used to work for Acer, ITRI, and CHTTL. Since 1999, he has joined the Institute of Nuclear Energy Research, where he is serving as a research assistant. His current research interests are directed at image retrieval and parallel computing. Jiann-Liang Chen was born in Taiwan on December 15, 1963. He received the Ph.D. degree in Electrical Engineering from National Taiwan University, Taipei, Taiwan in 1989. Since August 1997, he has been with the Department of Computer Science and Information Engineering of National Dong Hwa University, where he is an associate professor now. His current research interests are directed at mobility management and distributed computing systems.
A distributed multicast tree using share link migration scheme for wireless asynchronous transfer mode network H.-C. Chao
a,*
, Ching-Rong K. Lay b, J.L. Chen
c,1
a
c
Department of Electrical Engineering, National Dong Hwa University, Hualien 97401, Taiwan, ROC b Institute of Nuclear Energy Research, Taoyuan 32546, Taiwan, ROC Department of Computer Science and Information Engineering, National Dong Hwa University, Hualien 97401, Taiwan, ROC Received 9 June 2000; received in revised form 26 October 2000; accepted 3 May 2001
Abstract In recent years, wireless asynchronous transfer mode networks have become popular and are in widespread use for supporting various types of data transmission. To minimize the cost in setting up routes and to meet the QoS constraint in time-sensitive traffic, in addition to bandwidth capacity, this paper focuses on the optimization of the number of shared links to reduce the call-blocking rate and the handoff failure rate for intercell roaming. The average longest path length and the average of the average path length of the constructed multicast trees are shown to be stable when the network load becomes heavy. Ó 2002 Elsevier Science Ltd. All rights reserved. Keywords: Asynchronous transfer mode; Multicast; Migration; Wireless; Handoff
1. Introduction Integrating multicasting and mobility into internet architectures has become more and more popular [1,2]. Paired with the rapid developments in asynchronous transfer mode (ATM) networking technology, the rise of wireless communications signals the start of a new era in telecommunications. In this era, not only will users need higher bandwidth; they will also demand mobility. A typical future wireless user may be carrying a hand-held computer (PDA) with audio
*
Corresponding author. Tel.: +886-3-866-8500/2500x17001. E-mail addresses: [email protected] (H.-C. Chao), [email protected] (C.-R.K. Lay), [email protected] (J.L. Chen). 1 Tel.: +886-3-866-2500x22116. 0045-7906/02/$ - see front matter Ó 2002 Elsevier Science Ltd. All rights reserved. PII: S 0 0 4 5 - 7 9 0 6 ( 0 1 ) 0 0 0 4 0 - 4
396
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
and video conferencing capabilities and may demand high-speed data communication on the order of 10 megabits per second or higher. A wireless ATM network, which is designed to provide isochronous and asynchronous high speed communications for wireless users, is a good match for these demands. The wireless ATM network consists of base stations (BS), mobiles, and network interface equipments. A BS may contain several radio ports to support mobile users. A group of mobile users is connected to the same wireless ATM network interface equipment. This collection of mobiles is called a cell. The cell architecture is shown in Fig. 1. A mobile user might have a few simultaneous connections in the wireless ATM network. When a handoff occurs, these connections may need to be re-routed. Handoff is an important function of the network where users are mobile and is implemented by the network to give the users freedom of motion beyond a limited wireless coverage area while they are communicating. The handoff is the procedure by which a user’s radio link is transferred from one radio port to another through the network without an interruption to the user connection. We can divide the handoff event into two levels. One is the network level and the other is the radio level. The radio level handoff is the actual transfer of the radio link between two ports. The network level handoff supports the radio level handoff by performing re-routing and buffering. Some of the procedures used in the network level handoff are determined by the radio level handoff. There is a cell manager for control and management within each cell. We assume that the cell managers have some knowledge about the neighboring cells and the network addresses of neighboring cells are stored in a local lookup table and it is updated periodically by means of an update protocol. We also assume that these cells are interconnected by wireless ATM network switching nodes. According to these concepts, we investigated the handoff using two different situations: intracell handoff (Fig. 2) and intercell handoff (Fig. 3). In the intracell handoff, the user
Fig. 1. A cell architecture in wireless ATM network.
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
397
Fig. 2. Intracell handoff.
Fig. 3. Intercell handoff.
is roaming within the cell. The only routing/re-routing performed is in the wireless ATM network interface equipment within the cell. The cell manager in the BS is responsible for the correct update of ATM virtual circuit translation tables within the cell. This type of routing/re-routing does not require wide area ATM network switching. The intercell handoff takes place when the mobiles roam into a different cell. In this case, the routing/re-routing involves the wireless ATM network. An intercell handoff might require one or more wireless ATM switches, depending on the handoff location and topology of the network.
398
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
When a mobile terminal becomes active in the wireless ATM network, we must setup a connection between the source node and the destination node before starting to transmit the data. The mobile terminal must first look up the databases maintained in the node. At the same time, the source node broadcasts search messages to find the neighboring nodes that have existing routes toward the destination node. We used the two-level shortest feasible path routing algorithm to find a common node that belongs to the shortest sub-path of the connection between the source node and the destination node. The most efficient route will be selected and other routes will be released. Since the traffic in wireless ATM networks can be divided into two types: time-sensitive and throughput-dependent traffic. In a lightly loaded network, we can use the same routing scheme to construct the time-sensitive multicast trees and throughput-dependent multicast trees. While in a heavily loaded network, to guarantee QoS, we can use different routing schemes to construct the different multicast trees separately. Furthermore, if mobiles roam within the same cell (intracell), we do not need to rebuild a new connection to the ATM switch node. If mobiles roam into another cell (intercell), we must perform common node re-routing to setup a new connection between the common node and the destination node.
2. Algorithms IP multicasting is based on the concept of the host group [3]. As stated in Ref. [4], the shortestfeasible-path routing (SFPR) scheme always tries to find the shortest path among all of the feasible paths between the source and the destination nodes. This scheme is good when the performance is measured by the average route distance. However, this scheme has a higher blocking probability for future calls when the call requests are focused on some hot vertices or edges. The least loaded routing (LLR) scheme is proposed to overcome the SFPR drawback. A lower call-blocking rate might be obtained with this scheme. However, the tradeoff is that the average route distance is prolonged because the least loaded route may not be the shortest. Even under light loads, the LLR still changes the route dynamically and this is not necessary. The traditional multicast route has focused on reducing the bandwidth and delay cost in building the path. We considered how to allocate bandwidth in a multicast tree so as to optimize some global measure of performance. For supporting multimedia applications, which require realtime data transmission in addition to end-to-end delay, the route bandwidth must be reserved so that the minimum transmission rate is guaranteed [5]. When the traffic is time-sensitive and the user is a newly joined one, the connection should be setup as soon as possible. We first examine and distinguish the neighboring nodes that belong to the light-load node group (Gl ). If we can find such a node, it is involved in the shortest feasible path routing. If not, we shall search the nodes that belong to the heavy-load nodes group (Gh ). If none is available, we can suspend the throughput-dependent traffic for a while and try to meet the bandwidth requirements needed by the time-sensitive applications later. If the suspension of throughput-dependent traffic cannot meet the required bandwidth, we try other nodes that belong to Gh and these nodes are preferred to the shortest feasible path routing. In the mean time, if the time limit suspension expires, we can continue the throughput-dependent traffic in a round robin queue and search for the dynamic
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
399
alternative routing to avoid the indefinite postponement problems. If all of these methods are not feasible, the join request is rejected. If the neighboring nodes are available, we first check if there exists a path to the same server node. If the answer is positive, we can share the same path links if all links are available. In the mean time, we try the two-level SFPR algorithm to find the shortest path that has a common node to the multicast tree. If the bandwidth capacity cannot meet the required bandwidth, the user’s request is rejected. The two level shortest feasible path algorithm is described as follows: Step 1: Search the preferred neighboring nodes in Gl , if nodes in Gl are not available, then search nodes in Gh . If both are unavailable, we can suspend the throughput-dependent traffic optionally. If this still cannot meet the requirements, the request is rejected. Step 2: If neighboring nodes are available, search the common node, which is destined to the same server, from the source to the destination node with the two level shortest feasible path algorithm to construct the shortest path. At the same time, we use the two level shortest feasible path routing algorithm to find the shortest feasible sub-path to the multicast tree if the bandwidth capacity of every link is feasible. Step 3: Select the shortest path that has common node to the same multicast tree; if it is unavailable, select the shortest feasible sub-path.
3. Migration comparison for exclusive and shared links As stated in Refs. [6,7], if a mobile user roams into another cell (intercell), it uses the shortest feasible routing algorithm to setup the connection. Then the candidate node sends control messages to find the nearest common node to do the exclusive link migration. To improve the overall performance of the network, we used the two-level SFPR algorithm to setup the connection. In addition, to reduce the overhead incurred by the control signal broadcast messages in performing the exclusive link migration, we execute the migration and the connection setup at the same time. After roaming through all of the mobile users, it first uses the common node re-routing which has the shortest path to share the link with as many as possible. If sharing the shortest connection is not available, then a new shortest feasible connection is setup. The network efficiency is evaluated to compare with a threshold value. As more users are dispersed throughout a wide area ATM network, the bandwidth resources consumed by migration schemes for exclusive and shared links becomes an important factor, which can affect the efficiency of the network. The algorithm for the exclusive link migration scheme is stated briefly as follows. 3.1. Exclusive link migration scheme For v 2 T , T is the set of leaf nodes {Send search messages to all adjacent nodes u 2 U , U is the set of adjacent nodes; Update the search messages for each adjacent node; If (an eligible node is not found) {discard messages; release the channels occupied;} Otherwise {reserve channels for the connection;} Select the first confirmation message returned from u 2 U; Release the channels not used in this connection;
400
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
Inform the nearest common node 2 U to perform the connection chores if available Otherwise perform shortest feasible path routing; }
3.2. Proposed scheme We propose a distributed algorithm with or without a shared link migration scheme to compare with the exclusive link migration scheme. Instead of finding the nearest common node of the multicast tree, we broadcast ‘search’ messages to find feasible paths that attach to the same multicast tree. We then selected the shortest path among them. If this path is unavailable, we rebuild a new connection between the source and destination node. Once the path has been selected, the other paths will be released. The distributed algorithm of our scheme is stated below: Step 1: Find common nodes that have connections to the destination. Step 2: Select the most efficient path; release the others. Step 3: If both are unavailable, setup a new path using TSFPR. Similarly, the shared link migration scheme can be activated from the common node after the roaming user becomes stable. If the shared link migration results in network congestion, this migration will be aborted. We can maintain the status of each link in the nodes involved in the connection between the source and the destination node. After reducing the number of shared links, we expect that the blocking rate and handoff rate will be reduced. This means that the probability of joining and roaming in the network increases. We also anticipate that the average of the average path length (AAL) of these multicast trees becomes stable and the bandwidth network resource will not be consumed accordingly. The proposed share link migration algorithm is stated as follows: Step 1: The leaf node sends a control signal to the common nodes of this path. Step 2: The first common node then sends messages to find a shared path that is shorter than the old path. Step 3: If found then perform shared link migration otherwise search for the next common node. Step 4: If no common migration remained in the old path then go to Step 3.
4. Simulation model In this paper, a torus style network model with bi-directional links connecting all its neighboring nodes is considered (Fig. 4). The model for the network is an ATM switch-based backbone with a 10 10 size. Since the focus of our work is on the construction and improvement of multicast trees, which are built and based on a wired ATM network, we will ignore the wireless portions of this network. Each vertex represents a cell that is connected to an ATM switch and with a unique identifier. There are databases associated with each cell in the network to keep track of all transactions. The information that is recorded in these databases includes traffic type, routing/re-routing strategies, residual capacity, throughput-dependent data buffering, and so on.
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
401
Fig. 4. Example of the torus network.
We assumed that the transmission capacity of each link is 150 megabit per second. Since the transmission rate of wireless portions in wireless ATM network is about 2 megabit per second, we assumed that seventy-five multimedia services can fill up a link. With 50 megabit per second of capacity reserved for voice, data and control signaling, we have 100 megabytes per second for multimedia applications. The assumed maximum number of joined mobile terminals for each multicast service, such as video on demand and video conferencing, is limited as M. The destined server is randomly selected from the existing feasible servers for a newly joining request. When the associated multicast tree has a size less than the limitation M, we say that an existing server is available. We used the two level SFPR algorithm to establish the new join connection. When a new mobile joins the network, the two level SFPR algorithm is used to find a shortest feasible path from the source node to the node that is on the same multicast tree and that has the shortest path to the destination node. At the same time, we locate the shortest sub-path to the multicast tree that is destined to the same server. Then we select the most efficient path to be our new connection and release the others. We assume that there are 2000 users waiting to join the network. The arrival rate of new join requests is assumed to be a Poisson distribution with mean A. When the residual capacity of the path is exhausted or the limitation of the multicast tree M is exceeded, the join request of a mobile unit is rejected. The departure rate of joined mobiles is also assumed to be a Poisson distribution with mean D. We select R users among the active users randomly to roam within the network. If the roaming mobiles move within the same cell, we do not need to rebuild the connection. Otherwise, we must perform the re-routing algorithm to find an efficient path. In the simulation of our work, the following parameters were considered: (1) There are 2000 users waiting to participate in the network. (2) On average, there are ten mobile users which are randomly selected to depart from the network per unit time, D ¼ 10. (3) On average, there are one hundred mobile users who arrive and join the network per unit of time, A ¼ 100. (4) We utilized R ¼ 10 and 25 to investigate the behavior of the network. (5) We assumed that the maximum number of mobile users allowed to join in a multicast tree is twenty, M ¼ 20 and that there are 10 servers randomly selected from these network nodes.
402
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
In this network we applied the two-level SFPR scheme that finds the shortest path to the same multicast tree first and attempts to find the SFPR with the exclusive link migration scheme. Then we select the shortest feasible path as our new path. The shared link migration scheme was implemented to investigate the performance of the same network. For a generated mobile user request ri ð1 6 i 6 cÞ, if the request is accepted, we let accept ðri Þ ¼ 1; otherwise, accept ðri Þ ¼ 0. While the mobile unit roams, we let roam_accept ðri Þ ¼ 1 if the request of connection is accepted; otherwise, roam_accept ðri Þ ¼ 0. We also let distance ðri Þ equal the hop count of the established connection between the source and the destination node if the request ri is accepted; otherwise, distance ðri Þ is set to be infinity. We define the following metrics to evaluate the performance P of the migration schemes: (1) Call blocking rate: 1 ci¼1 accept ðri Þ=c, ri ¼ 0; c is the total users that want to join in the network. P (2) Handoff failure rate: 1 di¼1 roam accept ðri Þ=d, ri ¼ 0; d is the total users that want to roam within the network. P Pt p (3) AAL: Let AVG P ðiÞ ¼ i¼1 Path LengthðiÞ=p, we have the AAL as i¼1 AVG P ðiÞ=t. (4) ALP: the average of the longest path length in the constructed multicast trees; ALP ¼ (total of the length of the longest path in every constructed multicast tree)/(number of multicast trees).
5. Simulation results The performance of the three schemes on the torus network is shown in Figs. 5–10. Figs. 5 and 6 show the behaviors of the proposed scheme under R ¼ 10. It is clear that the benefit is not so attracting due to low roaming rate. Fig. 7 shows the blocking rates while the number of roaming users is twenty-five. Obviously, the scheme with share link migration can reduce the call-blocking rate. This means that the new joining mobiles will have more chances to build a connection successfully. Using our scheme without performing share link migration, the call-blocking rate is
Fig. 5. Call blocking rate with R ¼ 10.
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
Fig. 6. Handoff failure rate with R ¼ 10.
Fig. 7. Call blocking rate with R ¼ 25.
Fig. 8. Handoff failure rate with R ¼ 25.
403
404
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
Fig. 9. AAL with R ¼ 25.
Fig. 10. ALP with R ¼ 25.
better than that with the exclusive link migration scheme since it consumes less bandwidth resource. Owing to the hot node effect, the network will be saturated with the control messages for building a new connection. This accounts for the abrupt rising of the call-blocking rate and handoff failure rate curves. After that saturation point, the call-blocking rate increases slowly. In a wireless ATM network, if the number of roaming users is reasonable, usually R=A 6 30% and in our case R=A ¼ 25%, the overhead incurred by sending control messages does not affect the performance of the share link migration scheme that much. In Fig. 8, there is also a saturation region. However, our scheme without share link migration is still better than that using exclusive link migration scheme. This is because the probability of rejection decreases using the proposed scheme. In Fig. 9, the AAL of our scheme with share link migration is better than that with exclusive link migration scheme. Our scheme without share link migration also has a shorter AAL than the exclusive link migration scheme. Once we select the path in our scheme, it is the
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406
405
shortest among all other paths. This is why our scheme, without performing further migration, has a shorter AAL than the exclusive link migration scheme. Our scheme has a good metric because a longer path length means more bandwidth resources consumption. While the roaming users increase, the probability of selecting a node that is destined to the same multicast tree increases. The tradeoff between the bandwidth that is occupied by control messages and the increased probability that a user can build a new connection makes the blocking rate grow slowly. In Fig. 10, the average of the longest path length in the constructed multicast trees tends to be stable. Our schemes with or without share link migration have shorter longest path lengths than the exclusive link migration scheme. This metric also accounts for the reason why our schemes spend less bandwidth resources and have lower call-blocking rates and handoff failure rates.
6. Conclusion Three algorithms for the construction of multicast trees on wireless ATM communication networks were studied and compared in this paper. The proposed heuristic algorithm with/ without share link migration shows that it has a lower blocking rate and handoff failure rate when the network is heavily loaded and a lot of roaming users. Finally, we demonstrated that when mobile users increase and the network load becomes heavy, the AAL and ALP metrics become stable and the proposed scheme with/without share link migration is better than the exclusive link migration scheme.
Acknowledgements Authors wish to thank S.W. Chang and M.T. Lin for their help with this research.
References [1] Polyzos GC. IP multicast for mobile hosts. IEEE Commun Mag 1997. p. 54–8. [2] Akyildiz IF, McNair J, Ho J, Uzunalioglu H, Wang W. Mobility management in current and future communications networks. Georgia Institute of technology. IEEE Network 1998: 39–49. [3] Deering S. Host extensions for IP multicasting. Internet RFC 1112, August 1989. [4] Huang N-F, Wu C-S, Wu Y-J. Some routing problems on broadband ISDN. Comput Networks ISDN Syst 1994;27:101–16. [5] Lee O, Im Y, Lee K, CHOI Y. A bandwidth and delay constrained minimum cost multicast routing algorithm. ICOIN-11 Proceedings, vol. 2. January 1997, Distributed algorithm; DA(2). p. 9c-4.1–4.5. [6] Huang N-F, Chen K-S. A distributed paths migration scheme for IEEE 802.6 based personal communication networks. IEEE J Selected Areas Commun 1994;12(8):1415–25. [7] Huang N-F, Chen K-S, Chao HC, Pan JY. A distributed multicast tree migration scheme for ATM-based personal communication networks. IEEE Globecom’96, 1996 Nov 18–22; London. p. 310.
406
H.-C. Chao et al. / Computers and Electrical Engineering 29 (2003) 395–406 Han-Chieh Chao received the B.S.E.E. degree from National Cheng Kung University, Taiwan, R.O.C., in 1985, and the M.S. and Ph.D. degrees in Electrical Engineering from Purdue University, Indiana in 1989 and 1993, respectively. From 1994–1997, he was an associate professor of Department of Computer Science and Information Engineering at National Dong Hwa University, Hualien, Taiwan, R.O.C. Since August 1997, he has transferred to Department of Electrical Engineering and also served as the director of the University Computer Center. His current research interests include mobile networks, next generation Internet, and highspeed networks.
Ching-Rong Lai was born in Taiwan on December 15, 1962. He received the Master degree in Department of Computer Science and Information Engineering from National Dong Hwa University, Hualien, Taiwan in 1997. He used to work for Acer, ITRI, and CHTTL. Since 1999, he has joined the Institute of Nuclear Energy Research, where he is serving as a research assistant. His current research interests are directed at image retrieval and parallel computing. Jiann-Liang Chen was born in Taiwan on December 15, 1963. He received the Ph.D. degree in Electrical Engineering from National Taiwan University, Taipei, Taiwan in 1989. Since August 1997, he has been with the Department of Computer Science and Information Engineering of National Dong Hwa University, where he is an associate professor now. His current research interests are directed at mobility management and distributed computing systems.