Reliable Multicast Protocol with Packet Forwarding in Wireless Internet Taku NOGUCHI, Toru YOSHIKAWA and Miki YAMAMOTO College of Information Science and Engineering, Ritsumeikan University 1-1-1, Nojihigashi, Kusatsu-Shi, Shiga, 525-8577, Japan, noguchi@is.ritsumei.ac.jp Department of Communications Engineering, Graduate School of Engineering, Osaka University ABSTRACT One of the most important problems in reliable multicast communication is reducing generated redundant NAKs to avoid NAK implosion. A number of NAK suppression mechanisms which try to resolve this technical problem have been proposed thus far. In wireless multicast which includes mobile hosts with wireless access channel, packet losses occur because of not only low quality of wireless communication channel but also tentative shutdown of connection due to handover. When remote subscription which enables effective usage of network resources is applied, a host re-joins to a multicast group when handover occurs. This means handover will introduce long shutdown of connection, because not only handover process in layer 2 but also multicast join and multicast tree reconstruction is necessary. This causes loss of large amount of packets and finally causes NAK implosion. In this paper, we claim that packet forwarding which is originally applied to unicast communication can be applied to multicast communications and resolve this NAK implosion problem caused by handover. Our simulation results show that packet forwarding mechanism significantly improves scalability of NAK suppression-base reliable multicast protocol. Keywords: wireless multicast, reliable multicast, NAK suppression, handover, packet forwarding 1 Introduction Multicast communication can support dissemination of the same data to potentially large number of receivers. IP multicast[1]is the multicast protocol which supports network-layer multicasting in the Internet. In IP multicast, a source sends a single data packet whose destination IP address is the corresponding multicast group. A data packet is replicated at an adequate router and forwarded to an interface below which there are receivers. IP multicast uses UDP in the transport layer, so originally it does not support any reliability. To support reliability, reliable multicast protocol[2][3] should be implemented in the transport layer. In reliable multicast, retransmission of lost packet is necessary for maintaining reliability. In multicast communication, a lot of participants are included in general. This character of multicast communication easily leads to overload of retransmission process at the sender, receivers and network nodes. For example, concentration of control packets, acknowledgement packets(acks) or negative acknowledgement packets(naks) sent to the sender degrades performance of the sender. This problem is well known as the feedback implosion problem (ACK implosion or NAK implosion). A number of feedback suppression mechanisms have been proposed to deal with this problem. One of the most popular and effective solutions is NAK suppression mechanism used in SRM(Scalable Reliable Multicast)[4]. NAK suppression mechanism uses timer-based approach. Whenever a receiver detects a packet loss, it schedules a pending NAK transmission at a randomly chosen point of time in the future. If the receiver receives a NAK generated by another receiver, it cancels its NAK transmission. This mechanism is called NAK suppression. When a timer expires without receiving a NAK from other receivers, the receiver multicast a NAK. NAK suppression mechanism reduces the number of generated NAKs[6], which brings good scalability to reliable multicast protocol. So, we focus on reliable multicast protocol with NAK suppression mechanism. Today, it is natural to use wireless access for accessing the Internet and mobile communication environment is now becoming general. This technical trend makes wireless multicast attractive. When reliable multicast is served in wireless environment, several technical problems arise. One of the most serious problems is packet loss in wireless networks. Causes of packet loss in wireless network can be categorized into wireless-link-caused loss and handover-caused loss[7]. Wireless-link-caused loss is packet loss caused by low quality of wireless link. Handover-caused loss is packet loss caused by handover operation. The latter packet losses have more significant impact to reliable multicast in wireless communications, because handover causes long-period bursty loss. This bursty loss may cause tentative serious increase of generated NAKs and leads to performance degradation due to NAK implosion. This paper focuses on this handover-caused loss in wireless reliable multicast. In unicast communication, packet forwarding[8] can resolve handover-caused packet loss. When handover of mobile host is predicted, an access point caches some packets during handover period, and forwards this cached packet to a new access point of the corresponding mobile host. In wireless multicast, however, this packet forwarding is not easy to be applied. This is because mobile host will leave an old access point by leaving the multicast group and join to a new access point by re-joining to the corresponding multicast group. A new access point only manages multicast group, i.e. existence of at least one host which has an interest of receiving the corresponding multicast group datagram beneath it, and does not manage all hosts of a multicast group. This feature makes it a little difficult to apply packet forwarding mechanism to wire-
less reliable multicast. In this paper, we claim that packet forwarding mechanism brings significant performance improvement to wireless reliable multicast even though it has some disadvantage that an access point should manage all nodes beneath it. We think this kind of host management will resolve deployment problems of IP multicast[9] and be beneficial from the viewpoint of security and accounting reasons as well. Our simulation results show that packet forwarding in wireless reliable multicast improves scalability significantly, compared with reliable multicast protocol without packet forwarding. Remainder of the paper is structured as follows. Section 2 shows overview of reliable multicast and its technical problems. Section 3 explains mobile multicast and influence of mobility to reliable multicast. This section also includes brief introduction of packet forwarding. Section 4 presents packet forwarding mechanism in wireless reliable multicast. In section5, we evaluate packet forwarding in wireless reliable multicast and show its improvement in scalability. Finally, section 6 concludes the paper. 2 Reliable Multicast Protocol IP multicast provides an efficient way to disseminate the same data to potentially large number of receivers. However, IP multicast generally uses UDP in the transport layer, which means it does not provide any support for reliable data dissemination. To support reliability, retransmission control, congestion control and flow control should be implemented at the transport layer. A number of proposals for reliable multicast protocols have been proposed so far[2]. In this section, we briefly explain about reliable multicast protocols and show our simple protocol model which we focus on in the paper. 2.1 NAK-based or ACK-based There are two approaches for retransmission control, an ACK-based one and a NAK-based one. For an ACK-based approach, all receivers send an ACK to the source, so ACK implosion is a serious problem. For a NAK-based approach, only receivers suffering packet loss send a NAK to the source, so the number of feedback packets should be small than an ACK-based approach. Both approaches are mathematically analyzed and evaluated from the viewpoint throughput performance in [1] and from the viewpoint of delay performance in [6]. Performance evaluation results in these papers show that a NAK-based approach has better scalability than an ACK-based one. Simulation results in [5] show that an ACK-based approach has worse scalability even though a hierarchical structure, e.g. RMTP(Reliable Multicast Transport Protocol)[11], is applied. Even though a NAK-based protocol has better scalability than an ACK-based one, NAK implosion may cause performance degradation when multicast group size is large. To resolve this implosion problem, several approaches have been proposed. SRM(Scalable Reliable Multicast)[4] applies a timerbased approach. Whenever a receiver detects a packet loss, it schedules a pending NAK transmission at a randomly chosen point of time in the future. If the receiver receives a NAK generated by another receiver, it cancels its NAK transmission. This mechanism is called NAK suppression. When a timer expires without receiving a NAK from other receivers, the receiver multicast a NAK. With this NAK suppression mechanism, the number of generated NAKs is significantly reduced[6], which brings good scalability. Because of its good scalability, we focus on NAK-based approach, especially a NAK suppression protocol, in this paper. 2.2 NAK Suppression Protocol In this paper, we focus on wireless handover issues in wireless reliable multicast communications. It is not our purpose of this paper to develop a specific protocol, so we use a general protocol model for NAK-based protocol and applies it to mobile communication environment. General NAK-suppression protocol behaves as follows. The sender multicast all packets to all receivers. Whenever a receiver detects a lost packet, it schedules a pending NAK transmission at a randomly chosen point of time in the future. In the meantime, if the receiver receives a NAK(generated by another receiver) for this packet, it cancels the transmission of its own pending NAK(NAK suppression). If, at the scheduled time for a pending NAK transmission, a NAK for this packet has not been received since the pending NAK transmission was first scheduled, the receiver multicast a NAK to the sender an all other receivers, and starts a timer. When a receiver receives a NAK for a packet that it has not yet received and it has initiated the random delay prior to sending a NAK, the receiver sets a timer and behaves as if it had sent the NAK. The expiration of a timer without prior reception of the corresponding packet causes the receiver infers that a packet has been lost, and operates as it detects a packet loss (as in item2 above). Whenever the sender receives a NAK, it re-multicast a corresponding packet to all receivers. 3 Wireless Multicast In this section, we discuss about multicast communications in wireless networks and also describe about reliable multicast in wireless networks. 3.1 Multicast Communications in Wireless Networks In today s Internet, a multicast member can access to the Internet via not only wired access but also wireless access. Nomadic users require a new kind of network host that can
Figure 1: Wireless Mulitcast move from one access point to another easily. We call this kind of host as a mobile host(mh). For mobility management in the Internet, Mobile IP[12] has been proposed. When a MH moves to a new subnet(foreign network), care-of-address(coa) is assigned temporally to this MN. The network portion of CoA matches that of the foreign network, which means by CoA a datagram can be forwarded to a foreign network by ordinary IP routing. A MN also has its IP address in its permanent home network, home address. A CoA of MN is registered at the agent located at the home network, which is called home agent. When a correspondent node would like to communicate with a MN, it can forward a datagram to its home address. When this datagram reaches a home network, a home agent can identify the CoA of MN and forwards this datagram to a foreign network by encapsulation technique. For mobility management in wireless multicast, two techniques have been proposed, bidirectional tunneling and remote subscription[13](fig.1). Bidirectional tunneling looks like Mobile IP for unicast traffic. In bidirectional tunneling, a home agent receives multicast datagram and forwards it to CoA of MN by making use of unicast tunneling. By this new subnet, when a mobile host moves from a subnet where method, mobility of MN can be hidden to the sender(correspondent). wireless channel quality is bad and to a new subnet where However, it uses unicast tunneling from the home network to wireless channel quality is good, packet losses detected at a a MN, so redundant path caused indirect routing wastes network resources. When the source multicast these lost packets to all members, mobile host may be packet losses occurred at the old subnet. Remote subscription is a simple approach for mobility management in wireless multicast. When a MH moves to a new ets should receiver them, which leads to inefficient network receivers in this new subnet which correctly receive these pack- subnet, it obtains its CoA of this foreign network. The MH rejoins a multicast group by using this CoA. For re-join, modi- From these reasons, handover loss should not be recovered resource usage. fication of IGMP for Mobile IP, MLD(Multicast Listener Discovery)[14] is used. In remote subscription, delay for re-join point, we propose a new packet recovery mechanism for wire- globally and should be recovered locally. From this view- may be a little large because it needs not only general Mobile IP process but also multicast tree construction process. the next section. less multicast, packet forwarding, which will be explained in However, it can achieve efficient usage of network resource because it does not use unicast tunneling which is necessary 4 Wireless Reliable Multicast with Packet for bidirectional tunneling. So, remote subscription should be preferable approach for wireless multicast which potentially Forwarding includes many mobile multicast hosts. 3.2 Reliable Multicast in Wireless Networks When remote subscription is applied for wireless multicast communications, a mobile multicast receiver suffers from Figure 2: Scalable NAK Suppression Protocol with Packet Forwarding handover delay when it moves towards a new subnet. This is because not only handover operation, i.e. obtaining CoA, but also multicast joining operation such as join latency and tree reconstruction delay is necessary. In remote subscription, this delay highly depends on whether there exists another receiver(s) of the corresponding multicast group in a new subnet. When there exists another receiver in a new subnet, multicast tree has already a branch to this subnet and multicast datagram has already come there. In this case, handover delay is not so large. When there exists no receiver in a new subnet, multicast joining operation takes a little time period, which leads to bursty packet loss. In this paper, we call this kind of busrty packet loss a handover loss. When this handover loss occurs, NAK suppression mechanism may not work well. This is because during handover period a mobile node cannot receive not only a data packet but also NAK sent by other receivers. In this case, a NAK is multicast to all other members of the corresponding multicast group from the mobile node. Even in the case that another multicast member exists in a As an effective way to resolve handover loss discussed in the previous section, we claim that packet forwarding technique which is originally proposed for wireless TCP[8] can be a good solution(fig.2). General NAK suppression protocol applied packet forwarding behaves as follows.
The sender multicast all data packets to all receivers in a multicast group. A receiver detects a packet loss when there is a gap of sequence number of received packets and recovers this packet loss using NAK suppression protocol described in Section 2.2. When a mobile host detects packet loss during handover operation(exactly this can be detected just after the handover operation), it uses packet forwarding mechanism described below. When an access point predicts(or detects) handover of a mobile multicast host, it caches arrived packets of the corresponding multicast group. After handover operation completed, a mobile multicast host sends the transmission demand packet to the old access point. When the old access point receives this request, it forwards cached packets to the mobile multicast host. When the access point receives a demand packet and it does not have cached data of the corresponding multicast group, it sends a control packet notifying this. When a mobile multicast host receives this control packet, it starts ordinary NAK suppression protocol. In order to apply packet forwarding technique towards wireless multicast, an access point should identify which mobile host(mobile multicast host) exists beneath it. This is because it should identify mobility(handover) of each mobile multicast host in order to start caching mechanism. In general, an access point has only a necessity to manage existence of a least one multicast member. Thus, for packet forwarding technique, overhead for an access point increases, but we believe this overhead is acceptable because of the following two reasons. One reason is that as shown in the next section performance improvement obtained by packet forwarding technique is significant. The other reason is that for security and accounting reason management of mobile hosts will be managed by the network. For security and accounting, all members of each multicast group should be managed. When this kind of management is implemented, our proposed scheme can be easily deployed because management of all mobile hosts for each multicast group can be realized. Reliable multicast protocol is implemented in transport layer of end hosts and our proposed packet forwarding method is implemented in network layer of an access point. It is desirable that packet forwarding method is only applied to reliable multicast traffic. In order realize this operation, transport layer processing is necessary at an access point. Active network technology[17] which is a new network architecture and realizes higher layer operation than network layer, is a candidate technology which enables this transport layer operation at an access point. Figure 3: Parameters in Simulation Figure 4: Network Topology for Simulation (Tires Model) 5 Performance Evaluation We evaluate performance of packet forwarding technique applied for NAK suppression protocol by computer simulation. For evaluation, we especially focus on scalability. 5.1 Simulation Model For network topology model, we use a sophisticated model, Tiers model[15](fig.4). For LAN, we assume two kinds of networks, one is wired and the other is wireless. 5% of LANs in generated topology are assumed to be wireless LAN. Location of wireless LAN is located randomly in the network. Multicast hosts are connected to MAN router with probability of.2 and LAN(wireless and wired) with.8. We assume one static host as the source of a multicast group. Loss probabilities and other simulation parameters are shown in Fig.3. Packet loss probability is assigned to each Tier from measured results reported in [16]. For mobility of mobile multicast hosts, distribution of their mobility speed is truncated exponential distribution in 4 1[km/h] with average of 1[km/h]. For handover delay, 1[sec] handover delay is assumed for the case that another multicast member does not exist, and [sec] is assumed in the case that another member exists already in a new network. This means Layer 2 handover delay is assumed to be. And Layer 3 handover delay is assumed to be in the situation that another member exists already. It is easily predicted that our proposed packet forwarding method has better performance for higher handover delay. Thus, simulation results obtained under the condition that L2 handover is, will give the lower bound of the performance of our proposed scheme. When handover of a mobile multicast host is
predicted(detected), an access point caches arrived packet of the corresponding multicast group during 2[sec]. Reliable multicast protocol is implemented in transport layer of end hosts and our proposed packet forwarding method is implemented in network layer of an access point. It is desirable that packet forwarding method is only applied to reliable multicast traffic. In order realize this operation, transport layer processing is necessary at an access point. Active network technology[*****] which is a new network architecture and realizes higher layer operation than network layer, is a candidate technology which enables this transport layer operation at an access point. 5.2 Simulation Results Figure 5 shows average delay characteristics of NAK suppression protocol without Packet Forwarding mechanism and NAK suppression protocol with Packet Forwarding mechanism. Delay is normalized with one-way delay for ideal transmission, i.e. transmission delay of ideal case transmission with no queueing delay at a router. Horizontal axis shows the number of receivers in a multicast group. As shown in this figure, normalized delay is slightly improved with Packet Forwarding mechanism. This is because with Packet Forwarding lost packets are forwarded from not the source but the old access point which is generally located closer than the source. As shown in Fig.5, average delay characteristics of general NAK suppression protocol saturates when the number of receivers is around 4. With Packet Forwarding mechanism, delay performance saturates where the number of receivers is around 7. These saturations of delay characteristics are caused by NAK implosion at the sender. This means from the scalability viewpoint Packet Forwarding improves scalability of NAK suppression protocol significantly because NAK suppression protocol with Packet Forwarding mechanism can include more receivers. Figure 6 shows the total number of hops for data packets, NAKs and transmission demand packets(only for Packet Forwarding). This total number of hops can be used as performance metric for efficient usage of network resource. As shown in Fig.6, Packet Forwarding mechanism decreases this total number of hops. In NAK suppression protocol without Packet Forwarding mechanism, lots of NAKs are multicast to all members when handover loss occurs. However, with Packet Forwarding mechanism, a transmission demand packet is unicast to the old access point and lost packet which are cached at the old access point can be forwarded to the mobile multicast host by unicast. On the contrary, in NAK suppression mechanism, not only NAKs but also retransmitted packets are also multicast to all members, which leads inefficient usage of network resources. Figures 7 and 8 show similar characteristics of Figs. 5 and 6, but they show characteristics for only lost packets during handover period. For example, Fig.7 shows average delay characteristics measured only for lost packets during handover period. As shown in these figures, Packet Forwarding mechanism brings significant performance improvement to handover packet loss from the viewpoint of both delay performance and network resource usage. Figure 9 shows similar characteristics of Fig.8, but it shows characteristics for packets excluding lost packets(this means packets not averaged in Fig.8). As shown in Fig.9, even for not lost packets some performance improvement is obtained. This is because redundant packets generated for retransmission is reduced by our proposed packet forwarding, which leads to performance improvement for not lost packets. These results show that our proposed packet forwarding brings performance improvement not only for lost packets but also correctly received packets. In our performance evaluation thus far, we use homogeneous model for wireless networks, i.e. packet loss in wireless network occurs similarly in every network. Here, we evaluate our proposed packet forwarding mechanism in the situation that there are several WLAN in bad quality. We assign packet loss probability of % and 1% by turns. In this case, a mobile host may move towards good quality WLAN form bad quality WLAN. Figure 1 shows normalized delay characteristics of our proposed packet forwarding and general NAK protocol in this situation( heterogeneous in the figure). Simulation results for homogeneous model is also shown in the figure(this part is the same as Fig.5). As shown in this figure, general saturates with smaller number of receivers compared with performance results in homogeneous model. For our proposed packet forwarding mechanism, performance results almost the same as the homogeneous model. When a mobile host moves to WLAN with good quality, it receives further packets than its last received packet. However, our packet forwarding mechanism will forward these packets and prevent redundant generation of NAKs. 6 Conclusions In the paper, we discuss how bad handover packet loss is for wireless reliable multicast. We claim that packet forwarding which recovers packet loss caused by handover by forwarding cached packets from the old access point to the mobile multicast host, will bring significant performance improvement. This is because packet forwarding mechanism only needs unicast transmission of forward request packet and unicast transmission for lost packets(cached packets). On the contrary, general requires multicast transmission of NAKs and multicast retransmission from the source. Our simulation results show that Packet Forwarding mechanism brings significant performance improvement from the viewpoint of scalability and efficient network resource usage. REFERENCES [1] S.Deering, Host Extensions for IP Multicasting, IETF RFC 1112, Aug. 1989. [2] J.W Atwood, Classification of Reliable Multicast Protocols, IEEE Net., pp.24-34, Vo.13, No.3, May-June, 24.
Normalized Delay 1 8 6 4 1.5 ( 1 7 ) Total number of hop 1.5 Normalized Delay 8 6 4 2 2 2 4 6 8 2 4 6 8 2 4 6 8 Figure 5: Normalized Delay Characteristics 1.5 ( 1 6 ) Total number of hop 1.5 Figure 6: Total Number of Hops for Data, NAKs and Control Packets 1.5 ( 1 7 ) Total number of hop 1.5 Figure 7: Normalized Delay Characteristics (Only for Lost Packets) Normalized Delay 1 8 6 4 2 heterogeneous () heterogeneous () homogeneous () homogeneous () 2 4 6 8 2 4 6 8 5 1 15 2 Figure 8: Total Number of Hops for Data, NAKs and Control Packets(Only for Lost Packets) Figure 9: Total Number of Hops for Data, NAKs and Control Packets(Excluding Lost Packets) Figure 1: Normalized Delay Characteristics for Heterogeneous WLAN Environment [3] M. Yamamoto, Reliable Multicsat, The Journal of IE- ICE, pp.67-674, Vol.85, No.9, Sep. 22. [4] S. Floyd et al., A Reliable Multicast Framework for Light-Weight Sessions and Application Level Framing, IEEE/ACM Trans. on Net., Vol.5, No.6, pp.784-83, Dec. 1997. [5] K.Yamamoto et al., Performance Evaluation of ACKbased and NAK-based Flow Control Mechanisms for Reliable Multicast Communications, IEICE Trans. on Comm., vol.e84-b, no.8, pp.2313-2316, Aug. 21. [6] M.Yamamoto et al., A Delay Analysis of Sender- Initiated and Receiver-Initiated Reliable Multicast Protocols, IEEE INFOCOM 97, pp.48-488, Kobe, Apr. 1997. [7] M. Yamamoto and T. Matsuda, Recent Research Activities in Wireless TCP, Technical Report of IEICE, MoMuC23-58, November, 23. [8] G. Krishnamurthi et al., Buffer Management for Smooth Handovers in Mobile IPv6, IETF Internet Draft draft-krishnamurthi-mobileip-.txt, July 2. [9] M. Yamamoto, Multicast Communications -Present and Future(Invited Paper), IEICE Trans. on Commun., Vol.E86-B, No.6, pp.1754-1767, June 23. [1] S. Pingali et al., A Comparison of Sender-initiated and Receiver-initiated Reliable Multicast Protocols, ACM SIGMETRICS 94, pp.221-23, May 1994. [11] S. Paul et al., Reliable Multicast Transport Protocol (RMTP), IEEE J. Select. Areas Commun., Vol.15, No.3, pp.47-421, April 1997. [12] D. Johnson et al., Mobility Support in IPv6, IETF Internet Draft, draft-ietf-mobile-ipv6-24.txt, work in progress, June 23. [13] G. Xylomenos and G.Polyzos, IP Multicast for Mobile Hosts, IEEE Commun. Mag., pp.54-58, Vol.35, No.1, Jan. 1997. [14] S. Deering et al., Multicast Listener Discovery(MLD) for IPv6, IETF RFC 271, Oct. 1999. [15] M. Doar, A Better Model for Generating Test Network, 1996 IEEE GLOBECOM 96, pp.86-93, London, UK, Nov. 1996. [16] M. Yajnik et al., Packet Loss Correlation in the MBone Multicast Network, IEEE Global Internet, pp.94-99, London, UK, Nov. 1996. [17] D.J. Wetherall et al., Introducing New Internet Services: Why and How, IEEE Network Mag., vol.12, no.3, pp.12-19, July/August 1998.