An Efficient Multicast Routing Protocol in Wireless Mobile Networks

Wireless Networks 7, 443–453, 2001  2001 Kluwer Academic Publishers. Manufactured in The Netherlands.

An Efficient Multicast Routing Protocol in Wireless Mobile Networks YOUNG-JOO SUH ∗ Department of Computer Science and Engineering, Pohang University of Science and Technology (POSTECH), San 31, Hyoja-Dong, Pohang, Korea

HEE-SOOK SHIN EC Service Research Team, DACOM R&D Center, 34 Kajong-Dong, Yusong-Gu, Taejon 305-350, Korea

DONG-HEE KWON Department of Computer Science and Engineering, Pohang University of Science and Technology (POSTECH), San 31, Hyoja-Dong, Pohang, Korea Received January 2000

Abstract. Providing multicast service to mobile hosts in wireless mobile networking environments is difficult due to frequent changes of mobile host location and group membership. If a conventional multicast routing protocol is used in wireless mobile networks, several problems may be experienced since existing multicast routing protocols assume static hosts when they construct the multicast delivery tree. To overcome the problems, several multicast routing protocols for mobile hosts have been proposed. Although the protocols solve several problems inherent in multicast routing proposals for static hosts, they still have problems such as non-optimal delivery path, datagram duplication, overheads resulting from frequent reconstruction of a multicast tree, etc. In this paper, we summarize these problems of multicast routing protocols and propose an efficient multicast routing protocol based on IEFT mobile IP in wireless mobile networks. The proposed protocol introduces a multicast agent, where a mobile host receives a tunneled multicast datagram from a multicast agent located in a network close to it or directly from the multicast router in the current network. While receiving a tunneled multicast datagram from a remote multicast agent, the local multicast agent may start multicast join process, which makes the multicast delivery route optimal. The proposed protocol reduces data delivery path length and decreases the amount of duplicate copies of multicast datagrams. We examined and compared the performance of the proposed protocol and existing protocols by simulation under various environments and we got an improved performance over the existing proposals. Keywords: mobile computing, wireless networks, multicast routing, mobile IP

1. Introduction Rapid progress in hardware technology has made computers compact, powerful, and low-cost. Furthermore, the recent advance in data communication technology has spawned an increasing demand for various services over networks irrespective of users’ location. As a result, we have witnessed an explosive growth of research and development efforts in the field of wireless mobile networks [5,13]. While existing computing systems assume static devices and wired networks, a wireless mobile system allows users with portable devices to access a shared communication network independent of their physical location [10]. Providing multicast services to hosts is becoming popular since many applications, such as video/audio conferencing, distance learning, or multiparty games, require such services. Multicast is defined as one-to-many communication pattern where multiple destinations receive identical datagrams transmitted from the source. Without multicast routing protocols, the source node should transmit ∗ This work was supported in part by a grant from the Ministry of Educa-

tion of Korea through the BK21 program toward Electrical and Computer Engineering Division at POSTECH.

multiple one-to-one (unicast) communications for multiple destinations, which consumes valuable network resources, e.g., bandwidth. Thus, multicasting offers efficient multidestination delivery and robust unknown destination delivery [8]. In wireless mobile networking environments, users still require particular network applications, such as the dissemination of textual information, multipoint communications, and distributed systems functions, for which a multicast mechanism is more efficient. Many multicast protocols, such as DVMRP [19], MOSPF [15], CBT [3], and PIM [9], have been proposed to support the multicast service. However, the proposals were designed assuming static hosts, and thus they do not work well in mobile networking environments. In wireless mobile networks, the bandwidth is limited, wireless links are error-prone, mobile hosts frequently handoff, and battery life of a mobile device is limited. Thus, when we design a multicast routing protocol for wireless mobile networks, the characteristics mentioned above should be carefully considered. Several multicast routing protocols for wireless mobile networks have been proposed [6,11,17,18]. Although the protocols solve several problems inherent in multicast routing proposals for

444

static hosts, they still have problems such as non-optimal delivery path, datagram duplication, etc. In this paper, we propose an efficient multicast routing protocol using a multicast agent in wireless mobile networks, where a mobile host receives a tunneled multicast datagram from a multicast agent located in a network close to it or directly from the multicast router in the current network, which offers suboptimal multicast delivery route to mobile hosts. While receiving a tunneled multicast datagram from a remote multicast agent, the local multicast agent may start multicast join process, which makes the multicast delivery route optimal. The proposed protocol reduces network traffic load by decreasing the number of duplicate copies of datagrams and reduces the multicast data delivery path length since datagrams are forwarded to mobile hosts by multicast agents which are located close to the current location of mobile hosts or located in the current network. We studied the performance of the proposed protocol by simulation and we got an improved performance over existing protocols. The remainder of this paper is organized as follows. In the next section, we briefly describe the IETF mobile IP and related works on multicast routing in wireless mobile networks. In section 3, we present the proposed multicast routing protocol. Section 4 presents the result of our performance study. Section 5 summarizes the paper.

SUH, SHIN AND KWON

(a)

2. Background and related work IETF mobile IP provides a unicasting mechanism that allows a mobile host to retain one address by transparently dealing with host movement from network to network while preserving the currently existing routing mechanisms [16,17]. Mobile IP uses home agents and foreign agents to achieve seamless communication. Datagrams destined for a mobile host are first delivered to the home agent. When the home agent receives datagrams destined for one of its mobile hosts registered away from home, it encapsulates the datagrams within new IP datagrams and then tunnels them to the mobile host’s current foreign agent. After the foreign agent receives them, it decapsulates them and then forwards the datagrams to the mobile host. The current IETF mobile-IP specification also briefly proposes two approaches for supporting multicast service to mobile hosts [16–18]: foreign agent-based multicast (referred to as remote-subscription) and home agent-based multicast (referred to as bi-directional tunneling) [6,11]. In foreign agent-based multicast, a mobile host has to subscribe to multicast groups whenever it moves to a foreign network (see figure 1(a)). It is very simple scheme and does not require any encapsulations. This scheme has the advantages of offering an optimal routing path and nonexistence of duplicate copies of datagrams. However, when mobile host is highly mobile, its multicast service may be very expensive because of the difficulty in managing the multicast tree. Furthermore, the extra delay incurred from rebuilding a multi-

(b) Figure 1. (a) foreign agent-based and (b) home agent-based multicast.

Figure 2. Multicast data duplication problem in home agent-based multicast.

cast tree can create the possibility of a disruption in multicast data delivery. In home agent-based multicast, data delivery is achieved by unicast mobile IP tunneling via home agent. When a home agent receives a multicast datagram destined for a mobile host, it encapsulates the datagram twice (with the mobile host address and the care-of address of the mobile host) and then transmits the datagram to the mobile host as a unicast datagram, as shown in figure 1(b). This scheme takes advantage of its interoperability with existing networks and

AN EFFICIENT MULTICAST ROUTING PROTOCOL

Figure 3. Tunnel convergence problem.

its transparency to foreign networks that a mobile host visits. However, the multiple encapsulation increases the packet size, and a datagram delivery path is non-optimal since each delivery route must pass through a home agent. Furthermore, if multiple mobile hosts that belong to the same home network visit the same foreign network, duplicate copies of multicast datagrams will arrive at the foreign network, as shown in figure 2. In [11], Harrison et al. proposed a home agent-based multicast protocol called MoM (Mobile Multicast), where a home agent is responsible for tunneling multicast datagrams to the mobile host. In home agent-based multicast schemes shown in figure 2, a home agent forwards a separate copy of multicast datagram for each mobile host even if all mobile hosts that wish to receive the multicast datagram are in the same foreign network. However, by MoM protocol, the home agent forwards only one copy of the multicast datagram to each foreign network that contains its mobile hosts. Upon receiving the multicast datagram, a foreign agent delivers it to mobile hosts using link-level multicasting. This scheme reduces the number of duplicate multicast datagrams and the additional load on low-bandwidth wireless links. But there still exists a problem, referred to as the tunnel convergence problem [6,11], resulting from the fact that multiple tunnels from different home agents can terminate at one foreign agent, as shown in figure 3. Thus, when multiple home agents have mobile hosts on the same foreign network, one copy of every multicast datagram is forwarded to the same foreign agent by each home agent. Therefore, the foreign agent suffers from the convergence of tunnels set up by each home agent. To solve this problem, the foreign agent appoints one home agent as the DMSP (Designated Multicast Service Provider) for the given multicast group. The DMSP forwards only one datagram into the tunnel, while other home agents that are not the DMSP do not forward the datagram, as shown in figure 4. MoM protocol reduces multicast traffic by decreasing the number of duplicate copies of datagrams. However, multicast datagrams from both the DMSP and a multicast router can cause a duplication since it is possible that local static hosts in the foreign network are members of the same group

445

Figure 4. DMSP selection in MoM protocol.

Figure 5. Multicast data duplication problem and non-optimal delivery route in MoM protocol.

as the visiting mobile hosts (see figure 5). Moreover, this approach uses a non-optimal delivery route since a home agent (DMSP) forwards multicast datagrams to tunnels leading to each foreign agent. There are several other multicast proposals for mobile networks. In [1,2] a multicasting scheme for static membership of mobile hosts has been proposed. It attempts to prevent the occurrence of duplicate datagrams by using exactly one delivery scheme and it supports reliable data delivery. It suggests a two-tier scheme that shifts most of the communication, computation, and storage costs onto the wired field in consideration of the limited resources of the wireless field. [12] proposed a multicast scheme for mobile hosts using Columbia Mobile IP, and [4] proposed a multicasting method for dynamic group members in wireless networks with incomplete spatial coverage. Those proposals shows different goals for supporting multicast for mobile hosts with different viewpoints. 3. Proposed protocol 3.1. Protocol overview By a multicast routing protocol, a datagram is sent with a single group address by means of a logical address that is

446

independent of the real addresses of receivers. A multicast datagram is forwarded along a multicast delivery tree that connects all networks having group members, and thus a receiver can receive datagrams when it joins the multicast group. As described in the previous section, existing home agent-based multicast and MoM protocols have the problem of duplicate copies of multicast datagrams, which causes an increased amount of multicast data traffic. In addition, the mechanisms lie at the root of the forwarding scheme achieved by the home agent through tunneling. Consequently, they shows an increased multicast data delivery path length since multicast data are delivered to a mobile host through the tunnel from the mobile host’s home agent, even if there are distributed delivery tree nodes close to the mobile host. The proposed protocol tries to improve these inefficiencies of existing protocols. The proposed protocol, called MMA (Multicast by Multicast Agent), introduces the Multicast Agent (MA) and the Multicast Forwarder (MF). MAs provide multicast service to mobile hosts. Each MA has the information of a single MF selected among several MFs, per multicast group. MF of a MA (e.g., MA1) is the MA selected among MAs which are located near the MA1 and belongs to the multicast tree of a given multicast group. The MF is in charge of forwarding multicast datagrams to MA1. The MF of a MA may be the MA itself when its local network is included in the multicast tree, or the MF can be a MA in another network that belongs to the multicast group when its local network is not covered by the multicast tree. In the MMA protocol, two distinct methods are used depending on whether a mobile host’s visiting network belongs to a multicast tree or not. If the visiting network belongs to a multicast tree, the mobile host directly receives multicast data from the local multicast router in the network. If the visiting network does not belong to a multicast tree, multicast data are delivered to a mobile host through tunneling from a MA that is included in the multicast tree of a given multicast group and located in a network close to the mobile host’s visiting network. In former case, the MF of the MA is configured with the MA itself while, in latter case, the MF of the MA is changed to an optimal MF value by using the MF information of the mobile host. 3.2. Protocol description Initially, if a mobile host wants to join a group in a foreign network, subscription is done through a MA in the foreign network, which must be a tree node of the multicast group. If not, the MA starts tree joining process. This MA configures the MF value of the multicast group with the MA itself, and delivers multicast datagrams to the mobile host in the network. When a mobile host moves from a network (e.g., N1) to another network (e.g., N2), the mobile host sends its MF information to MA in N2 during registration, which is used by the MA for selecting new MF. If N2 belongs to the multicast

SUH, SHIN AND KWON

Figure 6. Operation of MMA protocol.

delivery tree, the MA itself becomes the MF. The MA and the mobile host update the MF information with the MA. If the MA in N2 does not belong to the multicast delivery tree and the MA has no MF information on the multicast group, the MF value that the mobile host had in N1 is used as the MF in N2. If the MA in N2 does not belong to the multicast delivery tree but the MA has MF information on the multicast group, there are two possibilities: The MF value that the mobile host had in N1 is used as the MF in N2 (oldest MF selection). Alternatively, the MA in N2 selects one that is closer to it, between the MF information that the MA currently has and the MF that the mobile host had in N1 (closest MF selection). In any case, the MA and the mobile host update the MF value with the selected MF. Figure 6 shows the basic operation of the MMA protocol. In the figure MA2 and MA3 are multicast tree nodes, while MA1 is not. Since MF of MA3 is MA3 itself, MH2 that is located in the same network as MA3 receives multicast datagram directly from multicast tree through MA3. On the other hand, since MA1 is not a multicast tree node, it receives a datagram through the tunnel from MA2 (MF of MA1) to itself, and then transmits the datagram to MH1. The multicast tree join process may be required by visiting mobile hosts according to the optional function of mobile hosts. Whenever a host moves to a new network and registers with a new MA, the new MA executes the multicast tree join processing, which is similar to that in the foreign agentbased multicast protocols. While setting up a connection to the multicast tree, a mobile host receives forwarded data


(a)

(b) Figure 7. Group Information table: (a) mobile host, (b) MA.

from its MF, and thus there is no service disruption period. When the join process finishes, multicast datagrams are delivered directly to the mobile host through an optimal path, just as in foreign agent based multicasting. This join process creates an overhead of reconstructing multicast tree and an extra time, but it presents an optimal delivery route. During the time delay, the disruption of multicast data delivery is reduced since datagrams are forwarded to the mobile host through tunneling from the MF. When a mobile host initially want to join a multicast group or when a mobile host that is using a join option moves to a new foreign network, a join process is performed by the current MA in the new foreign network. A quit process is executed when the local network wants to prune a multicast tree since no multicast group members exist in the local network. Therefore, for the quit processing, a MA checks the group membership each time a group membership change occurs, e.g., when a mobile host has left to another network, when a static host leaves the multicast group, or when a MA receives a forwarding stop message. The proposed MMA protocol offers better (sub-optimal) delivery route than home agent-based protocols since the forwarding pointer (i.e., MF) is generally located in a geographically adjacent network that is included in the multicast delivery tree. A mobile host moves among networks and these visited networks are geographically located close to each other. Therefore, in most cases, it is reasonable to assume that routing paths among these networks are relatively short. Consequently, a MA which receives the MF information from a visiting mobile host is able to have a sub-optimal delivery route since the MF is geographically located close to the current network. In the worst case, where there are

447

no multicast tree nodes among the networks visited by a mobile host, the MMA protocol shows a very similar path length to other mechanisms which use tunneling through home agents. With join option, the delivery route finally becomes optimal when join process completes. In addition, the MMA protocol reduces the number of duplicate datagrams and total amount of tunneling since multicast datagrams can be forwarded directly from the multicast router in the current network. There are some assumptions in the design of MMA protocol. First, the provided multicast service is not ordered and unreliable. The higher layer protocols are responsible for these problems. Second, both static and mobile hosts can be members of multicast groups, and hosts can send multicast datagrams to a multicast group whether they are members of the group or not. Third, mobile hosts can join and leave the multicast group at any time just as static hosts, even if they are away from their home networks. But there are no assumptions about the size of multicast groups, the distribution of multicast group members, the number of mobile hosts, the location of mobile hosts, and the rate of host mobility, since the proposed protocol aims to achieve scalability in the number of mobile hosts and the size of multicast groups, effectiveness of the delivery path length and the network traffic resulting from multicast data, and robustness to disruption of multicast data delivery resulting from host movement. 3.3. Data structure Figures 7(a) and (b) illustrate the group information table for a mobile host and a MA, respectively. Each mobile host has information on group ID, current MF, and JOIN_option. Each MA maintains the information on group ID, current MF, the number of static hosts of the multicast group (N_SH), a list of visiting mobile hosts that are members of the multicast group (MH_List), and a list of the foreign agents that receive multicast datagrams through tunneling from the MA (FA_List). The group membership is examined by checking the N_SH, MH_List, and FA_List. When a host moves into a new network and registers with a new MA, this group information for the mobile host may be sent to the MA using a previous foreign agent notification format that is included in the registration request message extension [16,17]. The information on the forwarding mechanism including forwarding REQUEST/STOP messages can be exchanged among MAs using the binding update message. This processing is based on the unicast routing of standard Mobile IP and can be implemented by the route optimization or other methods of Mobile IP implementations.

3.4. Algorithmic description In this subsection, we summarize the operations of the proposed MMA protocol. First consider when a mobile host is the sender. For a mobile host that wishes to send a multicast datagram, there are two possible schemes [6,11,17,18]. First, the mobile host can directly send datagrams through the current network. This scheme assumes that a multicast router exists in the current network. This is the simplest scheme for obtaining multicast

448

SUH, SHIN AND KWON

service, but the sender of high mobility may cause data disruption and overhead due to the reconstruction of the multicast tree. In the second scheme, the sender tunnels multicast datagrams to its home agent, and then the home agent forwards the multicast data to group members as if they are originated from the home network. This scheme is more efficient and reliable under high mobility conditions. The Mobile IP standard allows multicasts to be sent by either using a temporary address or via tunnels to the home agent [16,17] – that is, either a mobile host sends multicast datagrams in a foreign network directly or a mobile host sends encapsulated datagrams to its home agent and the home agent forwards them to the group members. Now, consider when a mobile host is a receiver. The operation of a mobile host is similar to that in Mobile IP. A difference is that message extension is needed for the processing of registration with a MA. A mobile host is responsible for managing and submitting the group information table. The operation of a mobile host (MH) can be summarized as follows: (a) When a mobile host that is a member of a multicast group enters into a new network. Receives Agent Advertisement message; Send Registration Request Message that contains (Group_ID, MF, JOIN_Option); Receives Registration Reply Message that contains MF info; IF(MF is different from MH’s current MF){ Modify MF with the received MF; }

(b) When a MF handoff occurs. Receive message notifying new MF info; IF(MF is different from MH’s current MF){ Modify MF with the received MF; }

The operation of a MA depends on various events, such as host arrival, host departure, control message arrival, multicast data arrivals, and join/quit process. The operation of a MA can be summarized using the event driven algorithms as follows: (a) When a mobile host that is a member of a multicast group arrives. Registration Request Message is received from MH Take the information (Group_ID, MF, JOIN_Option); IF(Group ID is already registered in Group Information Table){ Add MH to MH_List; Inform MH of its MF value; Select new optimal MF; IF(Selected MF equals to MF received from MH){ Send Fowarding REQUEST message to new MF; IF(datagram is forwarded from the MF before time-out occurs){ Send Forwarding STOP message to old MF; Notify new MF to all mobile hosts; Update its MF with new MF; } } }ELSE{ Make new group entry; Add MH to MH_List; Send Forwarding REQUEST message to MF of MH; Set up MF value with MF of MH; } IF(Multicast router exists && MH requests join process && MF is not MA itself){ Send join message for connection to multicast delivery tree; }

(b) When a mobile host that is a member of a multicast group leaves the current network. According to soft-state, the movement of a mobile host is detected by a time-out by the previous MA.


449

IF(Time-out occurs for MH){ Delete MH from MH_List in all group entries joined by MH; Check group membership; IF(Group member doesn’t exist in group entry anymore){ IF(MF equals to MA itself){ /* The network belongs to multicast delivery tree */ Send PRUNE message; }ELSE{ /* The multicast data is forwarded from MF */ Send Forwarding STOP message to MF; } Delete group entry from Group Information Table; } }

(c) When a control packet arrives from a foreign multicast agent. IF(Fowarding STOP message from a foreign MA is received){ Delete the sender MA from FA_List from the group entry; IF(Group member does not exist in group entry anymore){ Sends PRUNE message; } }ELSE IF(Fowarding REQUEST message from a foreign MA is received){ Add the sender MA to FA_List in group entry; }

(d) When a multicast datagram arrives. IF(Group entry exists in Group Information Table){ IF(MF is the MA itself){ Transmit multicast data to all MHs in MH_List; Transmit multicast data to all agents in FA_List through tunnels; }ELSE{ Transmit multicast data to all MHs in MH_List; } }

(e) When a connection to the multicast tree is completed. (When the JOIN processing is done.) IF(Group entry exists in Group Information Table){ Send Forwarding STOP message to MF in the group entry; }ELSE{ Make new group entry; } Set up MF with the MA itself; Notify new MF to all mobile hosts in MH_List;

4. Performance evaluation We have evaluated performance of the proposed protocol using a discrete-event simulation. We assumed that 400 LANs are located on the xy coordinate system, with the x and y coordinates are chosen uniformly at random for each LAN. This set of LAN locations is fixed for each simulation time.

In a randomly selected network model, the initial multicast tree is established for a randomly selected set of LANs and the initial tree has two modes: sparse and dense. In the sparse mode, group members are distributed sparsely across the overall network, while group members are centralized in the dense mode. Mobile hosts show diverse mobility rates ranging from 1 (low) to 5 (high). That is, the probabili-

450

SUH, SHIN AND KWON

Table 1 Simulation parameters. Parameter N T H MR α β Time

Description

Values

number of LANs number of initial tree nodes average number of mobile hosts per initial tree node host mobility rate service time join delay total simulation time

400 10–90 1–29 1–5 5 units 1 unit 1000 units

ties of mobile host handoff are 20, 40, 60, 80 and 100% in a unit time for mobility rates of 1, 2, 3, 4 and 5, respectively. We evaluated performance of the MMA protocol with various numbers of mobile hosts, various sizes of the initial multicast tree, and various rates of host mobility, in both the initial multicast delivery tree modes. We also studied the performance when tree join process is used and not used. The shortest path length between two LANs is measured by the shortest-path Euclidean distance, and we used generic time units. We assumed that there is a single source and it is selected randomly and is fixed during the simulation time. Table 1 summarizes the parameters used in our simulation study. Our simulation experiment compares the performance of the proposed MMA protocol, the MoM protocol, and the home agent-based multicast mechanism. The main features considered are multicast data traffic per unit time and average delivery path length of multicast data per mobile host. In addition, we observed a scaling characteristic of the protocols with multicast group size and compared DMSP handoff rate in MoM protocol with MF handoff rate in the proposed MMA protocol. The results are illustrated in figures 8–13. Total network traffic generated by a multicast delivery is the sum of the traffic occurred on the multicast tree and the traffic occurred by tunneling from the forwarding pointer to the mobile host. Thus we can compare the additional traffic by tunneling in the protocols. The number of tunneling is proportional to the number of mobile hosts in the home agent based multicast protocol, the number of foreign networks in which mobile hosts that subscribe a given multicast group are visiting in the MoM protocol, and the number of MAs which receive data forwarded by a MF in the proposed protocol. Figure 8 compares these in sparse mode as a function of H , when T = 10, 50, 90, and MR = 3 (figure 8(a)) and in dense mode as a function of H when MR = 1, 3, 5, and T = 50 (figure 8(b)). As shown in the figure, the proposed protocol shows an improved performance in the network traffic generated by a multicast delivery. It is due to the fact that the MMA protocol reduces the amount of tunneling since multicast datagrams are forwarded to a mobile host by a MF that is located close to the mobile host or directly by a multicast router in the current network. Figure 9 also compares total network traffic when join operation is

(a)

(b) Figure 8. Comparison of network traffic: (a) sparse mode, T = 10, 50, 90 MR = 3, H = (1–29), (b) dense mode, T = 50, MR = (1, 3, 5), H = (1–29).

used, where the same parameters that are used in figure 8 are also used. As shown in the figure, the proposed protocol shows a more improved performance. It is due to the fact that after a unit time, a MA that is currently receiving datagrams forwarded by a MF joins the multicast tree eventually, which decreases the number of nodes being forwarded. In the MMA plots, for relatively large initial tree sizes, there are peak points. This is due to the join operation of the MMA protocol. If the number of mobile hosts is relatively small, when a mobile host enters into a new network, the probability that the MA in the network is included in the multicast tree is small, and the MA will request forwarding service to the selected MF. Until the number of mobile hosts reaches to a certain point, the number of forwarding requests will increase as the number of mobile hosts increases. If the number of hosts further increases over the point, the probability that a MA is included in the multicast tree becomes large, since one or more other mobile hosts that enter into a network before a given mobile host may have been subscribing the same multicast group, which makes the MA a tree node. The following analysis, similar to that in [7], also illustrates the advantage of the MMA protocol over the home


451

(a) Figure 10. Comparison of average multicast delivery path length: T = (10–90), MR = (1, 3, 5), H = (1–29).

(b) Figure 9. Comparison of network traffic with join operation: (a) sparse mode, T = 10, 50, 90 MR = 3, H = (1–29), (b) dense mode, T = 50, MR = (1, 3, 5), H = (1–29).

agent-based tunneling mechanism and the MoM protocol: Let N be the number of networks in the system, N be the number of networks that belong to the multicast tree (N N), G be the number of multicast groups, c be the average number of MHs at each foreign network, k be the number of redundant DMSPs forwarding packets, and k be the number of MFs forwarding packets (k = 1). The number of multicast messages in the network is O(cN 2 G) by the home agent-based multicast mechanism, O(kNG) by MoM protocol, and O(k (N − N )G) = O((N − N )G) by the proposed MMA protocol. Figure 10 shows the average delivery path length of MoM and MMA protocols, relative to optimal path length, as a function of tree size, when the optimal path length (in foreign agent-based protocol) is 1 and MR = 1, 3, and 5. As shown in the figure, in both sparse and dense modes, the average delivery path length of MMA shows better performance than that of MoM. We divided the delivery path into a tree path and a tunnel path and compared them in detail. In figure 11, the top plot shows the difference of tree path length of MMA and that of MoM (that is, tree path length of MMA – tree path length of MoM), the bottom plot shows the difference of tunnel path length of MMA and that of MoM (that is, tunnel path length of MMA – tree path length of MoM), and the mid-

Figure 11. Relative multicast delivery path length: dense mode, T = 50, MR = (1, 3, 5), H = (1–29).

Figure 12. Comparisons of multicast delivery path length with join operation: dense mode, T = 50, MR = (1, 3, 5), H = (1–29).

dle plot shows difference of total path length of MMA and that of MoM (that is, total path length of MMA – total path length of MoM), as a function of H in dense mode when MR = 1, 3, 5, and T = 50. As shown in the figure, MMA shows slightly longer tree path length than MoM since multicast routers in foreign networks may join the multicast tree by MMA, which lengthens the tree path length. However, MMA shows much shorter tunnel path length than MoM. As a result, MMA shows shorter total path length than MoM,

452

SUH, SHIN AND KWON

called multicast forwarder, located a network close to the current network or directly from the multicast tree node in the current network. The protocol reduces network traffic by decreasing the number of duplicate copies of datagrams and reduces multicast data delivery path length since datagrams are forwarded to mobile hosts by multicast agents close to the current network or directly from the multicast router in the current network. We compared the performance of the proposed protocol with existing protocols by simulation under various wireless mobile networking environments and we got improved performance over the existing protocols. (a) References

(b) Figure 13. Comparison of handoff rate (sparse mode): (a) T = (10, 50, 90), MR = 1, H = (1–29), (b) with join operation, T = 50, MR = 3, H = (1–29).

and it becomes more dominant with the increase of host mobility rate as shown in the figure. Figure 12 shows relative multicast delivery path length when join option is used. As shown in the figure, the difference of delivery path lengths between the two protocols becomes larger than that of figure 11, and thus MMA shows much better performance in total path length than MoM. Figure 13 compares the number of MF handoffs in MMA protocol with the number of DMSP handoffs in MoM protocol as a function of H when T = 10, 50, 90, and MR = 1 (figure 13(a)) and when MR = 3, T = 50, and join option is used (figure 13(b)). As shown in the figures, MF handoff frequency in MMA is much less than DMSP handoff frequency in MoM. Frequent handoffs of DMSP may cause extra traffics in networks, which increases network overheads, and causes performance degradation due to the out-of-service period during a DMSP handoff.

5. Conclusion In this paper, we proposed an efficient multicast routing protocol supporting host mobility. The proposed protocol uses multicast agents, where a mobile host receives multicast datagrams through a tunnel from a selected multicast agent,

[1] A. Acharya and B. Badrinath, A framework for delivering multicast message in networks with mobile hosts, Mobile Networks and Applications (1996) 199–219. [2] A. Acharya, A. Bakre and B. Badrinath, IP multicast extensions for mobile internetworking, in: IEEE INFOCOM’96 (1996). [3] A. Ballardie, P. Francis and J. Crowcroft, Core Based Trees (CBT): An architecture for scalable inter-domain multicast routing, in: Proc. of ACM SIGCOMM 93 (September 1993) pp. 85–95. [4] A. Bartoli, Group-based multicast and dynamic membership in wireless networks with incomplete spatial coverage, Mobile Networks and Applications (1998) 175–188. [5] D.C. Cox, Wireless personal communications: what is it?, IEEE Personal Communications (April 1995) 20–35. [6] V. Chikarmane and C.L. Williamson, Multicast support for mobile host using mobile IP: design issues and proposed architecture, Mobile Networks and Applications (1998) 365–379. [7] V. Chikarmane and C.L. Williamson, Performance evaluation of the mom mobile multicast protocol, Mobile Networks and Applications (1998) 189–201. [8] S. Deering, Multicast routing in datagram internetworks and extended LANs, ACM Transactions on Computer Systems 8(2) (May 1990) 85–110. [9] S. Deering, D. Estrin, D. Farinacci and V. Jacobson, An architecture for wide-area multicast routing, in: Proc. of ACM SIGCOMM 94 (August 1994) pp. 126–135. [10] H. Forman and J. Zahorjan, The challenges of mobile computing, IEEE Computer 27(4) (April 1994) 38–47. [11] T. Harrison, C. Williamson, W. Mackrell and R. Bunt, Mobile multicast (MOM) protocol: multicast support for mobile hosts, in: Proc. of ACM MOBICOM’97 (1997) pp. 151–160. [12] J. Ioannidis, D. Dduchamp and G. Maguire, IP-based protocols for mobile internetworking, in: Proc. of ACM SIGCOMM’91 (1991) pp. 235–245. [13] B. Jabbari and G. Colombo, Network issues for wireless communications, IEEE Communications Magazine (January 1995) 88–98. [14] C.K. Miller, Multicast Networking and Applications (AddisonWesley). [15] J. Moy, Multicast routing extension for OSPF, Communications of ACM 37(8) (August 1994) 61–66. [16] C. Perkins, IP Mobility Support, RFC 2002, Mobile IP Networking Group. [17] C. Perkins, Mobile IP Design Principles and Practices (AddisonWesley). [18] G. Xylonmenos and G. Polyzos, IP multicast for mobile hosts, IEEE Communications Magazine (January 1997) 54–58. [19] D. Waitzman, C. Partridge and S. Deering, Distance vector multicast routing protocol, RFC 1075, BBN STC and Stanford University (November 1998).


Young-Joo Suh received the B.S. and M.S. degrees in electronics engineering from Hanyang University, Seoul, Korea, in 1985 and 1987, respectively, and the Ph.D. degree in electrical and computer engineering from Georgia Institute of Technology, Atlanta, Georgia, in 1996. He is currently an Assistant Professor in the Department of Computer Science and Engineering at the Pohang University of Science and Technology (POSTECH), Pohang, Korea. From 1988 to 1990, he was a Research Engineer at the Central Research Center of LG Electronics Inc., Seoul, Korea. From 1990 to 1993, he was an Assistant Professor in the Department of Computer Science and Engineering at the Chung-Cheong College, Korea. After receiving the Ph.D. degree, he worked as a Postdoctoral Researcher in the Computer Systems Research Laboratory in the School of Electrical and Computer Engineering at the Georgia Institute of Technology from 1996 to 1997. From 1997 to 1998, he was a Research Fellow of the RealTime Computing Laboratory in the Department of Electrical Engineering and Computer Science at the University of Michigan. His current research interests include mobile computing, networking protocols, and interconnection networks. Dr. Suh is a member of the IEEE, the IEEE Computer Society, and the IEEE Communications Society. E-mail: [email protected]

453

Hee-Sook Shin received the B.S. degree in computer engineering from Kyungbuk National University, Taegu, Korea, in 1998, and the M.S. degree in computer science and engineering from Pohang University of Science and Technology (POSTECH), Pohang, Korea, in 2000. She is currently a Research Engineer at the R&D Center of DACOM Inc., Taejon, Korea. Her research interests include multicasting in wireless networks and mobile computing. E-mail: [email protected]

Dong-Hee Kwon received the B.S. degree in electrical engineering from KAIST, Taejon, Korea, in 1997, and the M.S. degree in computer and communication engineering from Pohang University of Science and Technology (POSTECH), Pohang, Korea, in 2000. He is currently a Ph.D. student in the Department of Computer Science and Engineering at POSTECH. His research interests include wireless communication and mobile computing. E-mail: [email protected]