A Performance Comparison of On-Demand MulticastRouting Protocols for Ad Hoc NetworksJorjeta G. Jetcheva and David B. JohnsonDecember 15, 2004CMU-CS-04-176School of Computer ScienceComputer Science DepartmentCarnegie Mellon UniversityPittsburgh, PA 15213This work was supported in part by NASA under grant NAG3-2534 at Rice University; by NSF undergrants CNS-0209204, CNS-0325971, CNS-0338856, and CNS-0435425 at Rice University; by a gift fromSchlumberger to Rice University; and by the Air Force Materiel Command (AFMC) under DARPA con-tract number F19628-96-C-0061 at Carnegie Mellon University. The views and conclusions contained hereare those of the authors and should not be interpreted as necessarily representing the official policies orendorsements, either express or implied, of NASA, NSF, Schlumberger, AFMC, DARPA, Rice University,Carnegie Mellon University, or the U.S. Government or any of its agencies.Keywords: wireless, ad hoc networks, routing, performance comparison, multicast, mesh networks.AbstractIn this paper, we present a comparative performance evaluation of three general-purpose on-demand multicast protocols, namely ADMR, MAODV, and ODMRP, focusing on the effects ofchanges such as increasing number of multicast receivers or sources, application sending pattern,and increasing number of nodes in the network. We use mobile networks composed of 100 or200 nodes, with both a single active multicast group and multiple active multicast groups in thenetwork, in a wide range of multicast scenarios. Although some simulation results for these pro-tocols have been published before, the three protocols have not been compared, and prior studieshave focused on smaller networks using a small set of simulation scenarios, many with only asingle active multicast group. We focus here on the effects of the protocols’ relative degree ofon-demand behavior and their performance in different multicast scenarios.1. IntroductionRouting in ad hoc networks requires the discovery of multi-hop paths between the wireless mobilenodes in the network that wish to communicate. In multicast routing, packets from one sendermust in general be delivered to multiple receivers making up a multicast group; any node cansend packets to a group at any time, and any node can join or leave a multicast group at any time.Multicast routing is a hard problem in wired networks, and is even more challenging in ad hocnetworks, due to the dynamic topology changes in the network due to node motion and wire-less propagation variability, and due to the limited wireless network bandwidth and node energyresources available.Because of these factors, multicast routing protocols designed for the Internet are not well suitedfor the ad hoc networking environment. In recent years, a number of multicast protocols for ad hocnetworks have been proposed (e.g., [1, 4, 6, 8–11, 16–20]), some of which rely on reactive (on-demand) mechanisms and discover routes only when they are needed for current communication,and others which rely on proactive mechanisms such as periodic neighbor sensing or flooding, orperiodic routing table exchanges.In this paper, we present a performance comparison of three on-demand multicast routing proto-cols for ad hoc networks: the Adaptive Demand-Driven Multicast Routing protocol (ADMR) [9],the Multicast Ad Hoc On-Demand Distance-Vector protocol (MAODV) [16], and the On-DemandMulticast Routing Protocol (ODMRP) [17]. All of these protocols contain a significant on-demand(reactive) component, but they differ in how reactive and proactive mechanisms are combined tomake the complete protocol: ADMR uses source-based trees and does not utilize any periodiccontrol packet transmissions, MAODV uses a shared group tree and uses periodic Hello messagesfor link break detection and periodic group leader floods for group information dissemination, andODMRP uses a group forwarding mesh for packet forwarding and utilizes periodic flood-responsecycles for multicast state creation and maintenance.In this performance evaluation, we focus on the effects of changes such as increasing numberof multicast receivers or sources, application sending pattern, and increasing number of nodesin the network. We use mobile networks composed of 100 and 200 nodes, with both a singleactive multicast group and multiple active multicast groups in the network. We study a wide rangeof multicast scenarios representative of a variety of multicast applications, such as conferencingand single-source vs. multi-source groups. Although some simulation results for these protocolshave been published before, the three protocols have not been compared, and prior studies havefocused on smaller networks using a small set of simulation scenarios, many with only a singleactive multicast group. We focus here on the effects of the protocols’ relative degree of on-demandbehavior and their performance in different multicast scenarios.Section 2 of this paper summarizes related work. The operation of the three protocols westudied, ADMR, MAODV, and ODMRP, is summarized in Section 3. Section 4 discusses ourevaluation methodology. We present results in Section 5 and conclude with Section 6.2. Related WorkA number of evaluations of ADMR, MAODV and ODMRP have been presented in the literature,all using ad hoc networks composed of 50 or fewer nodes [3, 6, 9, 13–16]. The only exceptionis the work of Kang et al. [12] who present the Scalable Multi-source Multicast Routing Protocol(SMMRP) and compare its performance to that of ODMRP in a network of 150 nodes. Theirperformance evaluation does not include ADMR and MAODV, and is only based on simulations1of one multicast scenario: 1 group with 10 sources and 10 receivers. In contrast, in this work, wepresent an extensive evaluation of ADMR, MAODV and ODMRP, in networks with 100 and 200nodes, in a wide range of multicast scenarios.Lee et al. [15] compare the performance of ODMRP, CAMP [6], AMRoute [1] and Amris [20].Their simulations are based on 50-node networks with a variable number of multicast senders andreceivers, all part of a single multicast group. Lee et al. [14] propose a new multicast protocolcalled Neighbor-Supporting Multicast Protocol (NSMP), and compare its performance to that ofODMRP in a 50-node network with a single active multicast group as well. Similarly, Royer etal. [16] present an evaluation of MAODV in a 50-node network with only one multicast group,and Kunz et al. [13] compare the performance of