0-7803-7016-3/01/$10.00 ©2001 IEEE Differential Destination Multicast–A MANET Multicast Routing Protocol for Small Groups Lusheng Ji and M. Scott Corson Institute for System Research University of Maryland College Park, MD 20742 e-mail: lji, corson@isr.umd.edu Abstract—In this paper we propose a multicast routing protocol for mo- bile ad hoc networks (MANETs). The protocol—termed Differential Desti- nation Multicast (DDM)—differs from common approaches proposed for MANET multicast routing in two ways. Firstly, instead of distributing membership control throughout the network, DDM concentrates this au- thority at the data sources (i.e. senders) thereby giving sources knowledge of group membership. Secondly, differentially-encoded, variable-length desti- nation headers are inserted in data packets which are used in combination with unicast routing tables to forward multicast packets towards multicast receivers. Instead of requiring that multicast forwarding state to be stored in all participating nodes, this approach also provides the option of stateless multicasting. Each node independently has the choice of caching forward- ing state or having its upstream neighbor to insert this state into self-routed data packets, or some combination thereof. The protocol is best suited for use with small multicast groups operating in dynamic networks of any size. Keywords—MANET, multicast routing, small group multicast I. I NTRODUCTION There are generally two mainstream approaches used for mul- ticast routing in fixed networks: Group Shared Tree (GST) and Source-Specific Tree (SST). Both construct data forward- ing paths interconnecting all group members, where a member is understood to be a multicast receiver. Data is firstly forwarded to the tree then along the tree paths to reach all group members. Several protocols include data sources as part of the forwarding tree as well. The GST approach builds one tree for the whole group regardless of where the data sources are located. The SST approach organizes multicast forwarding by data source. A “session” is associate with each source and is identified by the combination of group ID and source ID. For each session there is one distribution tree formed from the union of all short- est paths between the source and the group members. This form of multicasting usually results in lower end-to-end delay since the data is forwarded using the shortest source-receiver paths, but typically consumes a greater amount of network bandwidth than shared-tree approaches. A characteristic shared among these traditional multicast pro- tocols is that the multicast computation is distributed in the network. Not only are the forwarding states constructed and maintained by the network, group membership control (or lack thereof) is also distributed over the network. While this ap- proach improves scalability with respect to group size, it has certain drawbacks. Firstly, distributed membership management may make aspects of security more difficult due to the lack of admission control. Without natural support for end-to-end sig- nalling between sources and receivers, such approach needs to rely on external mechanisms for security management. At the same time, billing management becomes more complicated as the information property owner, typically the source, has no con- trol and knowledge over how and to whom its property (data) is distributed. Secondly, distributed per group forwarding state maintenance may result in large router resource usage. The number of possible multicast groups formed among members grows combinatorially. While there is no efficient way to aggre- gate multicast routing table entries, the linearly growing multi- cast routing table (with respect to the number of active multicast groups) may quickly become too much of a storage burden for routers. The issue is worsen by the number of routers involved in forwarding since all routers along the multicast forwarding paths need to participate in multicast routing state maintenance. To address this issue, recently there is an attempt of shifting towards stateless multicast routing for small groups. [1] and [2] lift multicasting out of routing layer so that multicasting no longer requires router support. Multicast data is encapsulated in unicast envelop and transmitted between end receivers. There- fore no multicast routing state is installed on routers. [3] and [4] propose connectionless, small group multicast as a “stateless” approach to multicast routing. In these approaches, variable- length destination lists are placed in packet headers that are self- routed towards the destinations using the underlying unicast for- warding tables. We consider the problem of multicast routing in Mobile Ad hoc Networks (MANETs). In MANETs all mobile nodes are equipped with wireless communication interfaces and can move at will. Here we assume their communications occur using omni-directional antennas over broadcast media. The combi- nation of node mobility and a wireless environment can result in MANET topologies subject to rapid and unpredictable changes. Because of these dynamics, and the fact that communication is carried over a bandwidth-constrained broadcast media, the mul- ticast routing problem in MANETs differs from that in fixed net- works. During recent years several multicast protocols have also been designed specifically for MANETs (e.g. CAMP [5], ODMRP [6], MAODV [7] and LAM [8]). These protocols all follow the traditional multicast approaches, i.e. distributed group membership management and distributed multicast rout- ing state maintenance. In addition to the security and resource use issues mentioned before, these approaches, especially when applied for use with small and sparsely distributed (potentially numerous) groups, may become even less efficient and more ex- pensive to function in MANETs due to bandwidth constraints, 1192 IEEE INFOCOM 2001