University of Southern California Dissertations and Theses

MUNet: multicasting protocol in unidirectional ad-hoc networks

(USC Thesis Other)

MUNet: multicasting protocol in unidirectional ad-hoc networks

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content MUNET: MULTICASTING PROTOCOL IN UNIDIRECTIONAL AD-HOC NETWORKS by Noparut Vanitchanant A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (COMPUTER ENGINEERING) May 2006 Copyright 2006 Noparut Vanitchanant UMI Number: 3237705 3237705 2007 Copyright 2006 by Vanitchanant, Noparut UMI Microform Copyright All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, MI 48106-1346 All rights reserved. by ProQuest Information and Learning Company. ii ACKNOWLEDGEMENTS First of all, I would like to express my deepest gratitude and appreciation to my advisor, Professor John Silvester for providing me the unique opportunity to work in the area of mobile, wireless networks, for his expert vision and guidance, and for his understanding encouragement and unfailing support at all levels. I would also like to thank the other members of my dissertation committee, Professor Ahmed Helmy and Professor Christos Papadopoulos for their valuable time, guidance, feedback, and interest in this work. I have the privilege of receiving educational attention as well as professional guidance and direction from Professor Joseph Bannister, Professor Joseph Touch, Professor Timothy Pinkston, and Professor Monte Ung. Interactions with them throughout my study always give me insightful inputs. I am grateful to them. It has been a great pleasure to intellectually associate with a group of smart and nice fellow doctoral students in Computer Engineering department at USC. For the encouragement, assistance, and friendship, I thank Antonia Boadi, Yu Shun Wang, and especially Ka-Cheong Leung for his incredibly prompt and intelligent feedbacks on my publications. I was very fortunate to have the opportunity to learn about the structural framework aspect from Son Dao, Mohin Ahmed, and other research staffs at HRL Research Laboratories LLC. during 2000 - 2002. Our collaboration, supported by iii Defense Advanced Research Projects Agency (DARPA) Contract No. N00014-99-C- 0322, not only has inspired this work, but also has achieved a significant outcome. I am grateful for a pending patent. Saying "Thanks" is not enough to Annette Tessier, a retired professor and my godsent mentor. Her kind encouragement, unlimited patience, and wise guidance have kept me stay on course professionally and personally. I am truly in debt of gratitude to her. Also, I wish to thank Tess Reyes, Diane Demetras, and Tim Boston for their kind assistance and friendship. Last, but not least, I would like to thank my family for their love, encouragement, understanding, assistance, and support. Without them, this work could not have been completed. iv TABLE OF CONTENTS Acknowledgements................................................................................................. ii List of Tables........................................................................................................... vi List of Figures ......................................................................................................... vii Abstract ................................................................................................................... x 1 INTRODUCTION............................................................................................ 1 1.1 Motivation: Asymmetry-Aware Network Structure.................................. 1 1.2 The Problem: Asymmetric Network Communication ................................ 3 1.3 Contributions............................................................................................... 4 1.4 Dissertation Organization............................................................................ 5 2 BACKGROUND INFORMATION................................................................. 7 2.1 Ad-hoc Networks ........................................................................................ 7 2.1.1 What is an Ad-hoc Network?........................................................... 8 2.1.2 Why Utilizes Ad-hoc Networks? ..................................................... 11 2.1.3 Why Utilize Ad-hoc Networks?....................................................... 13 2.2 Multicasting Technique............................................................................... 14 2.2.1 Introduction ..................................................................................... 14 2.2.2 Problems and issues......................................................................... 15 2.2.3 Selected Traditional Multicast Routing Protocols........................... 17 2.3 Unidirectionality/Asymmetry ..................................................................... 26 2.3.1 Unidirectionality in General ............................................................ 26 2.3.2 Impact of Unidirectionality in Ad-hoc Networks ............................ 27 2.4 Summary ..................................................................................................... 29 3 ASYMMETRIC NETWORKS........................................................................ 30 3.1 Introduction................................................................................................. 30 3.2 Real Network Characteristics...................................................................... 31 3.2.1 Random Graph Model...................................................................... 35 3.2.2 Euclidean Graph Model ................................................................... 37 3.3 Simulation and Results................................................................................ 39 3.3.1 Unidirectional Link Density vs. Bi-directional Link Density.......... 40 3.3.2 Node Reachability............................................................................ 41 3.3.3 Depth of Shortest-path Root-based Tree.......................................... 45 3.3.4 Percentage of Pairs Connected......................................................... 48 v 3.3.5 Average Path Length........................................................................ 49 3.4 Summary ..................................................................................................... 51 4 RELATED WORK .......................................................................................... 52 4.1 Spatial Locality Exploitation....................................................................... 53 4.2 Topology-oriented Approach ...................................................................... 54 4.3 Unidirectional Routing Approach ............................................................... 55 4.4 Swarm Intelligence Concept ....................................................................... 57 4.5 GPS Enabled Approach............................................................................... 59 4.6 Summary ..................................................................................................... 60 5 MUNET: MULTICASTING PROTOCOL IN UNIDIRECTIONAL AD- HOC NETWORKS .......................................................................................... 61 5.1 Objectives of the Design ............................................................................. 62 5.2 Assumptions................................................................................................ 63 5.3 MUNet Protocol - Methodology ................................................................. 63 5.3.1 Path Establishment........................................................................... 64 5.3.2 Path Maintenance............................................................................. 69 5.3.3 Link Recovery.................................................................................. 81 5.4 MUNet Protocol - Enhancement................................................................. 83 5.5 Data Forwarding Mechanism...................................................................... 86 5.6 Summary ..................................................................................................... 86 6 COMPARATIVE EVALUATION.................................................................. 87 6.1 Protocol Evaluation Methodology .............................................................. 87 6.2 Evaluation Metrics ...................................................................................... 88 6.3 Impact of Unidirectionality......................................................................... 89 6.4 Impact of Protocol Enhancement ................................................................ 90 6.5 Summary ..................................................................................................... 91 7 CONCLUSIONS.............................................................................................. 95 7.1 Summary of Contributions.......................................................................... 95 7.2 Future Direction .......................................................................................... 97 BIBLIOGRAPHY................................................................................................... 98 APPENDICES......................................................................................................... 104 A. Data Structures ............................................................................................ 104 A.1 Control Packet Format ....................................................................... 104 A.2 Data Packet Format............................................................................ 106 A.3 Forwarding Table Structure ............................................................... 106 vi LIST OF TABLES 2.1 Comparisons of key features and characteristics of selected multicasting protocols…………………………………………………………...…………. 25 A.1 Control packet format………………...………………………….…………. 105 A.2 Data packet format……………………………………………….……..…... 106 A.3 Forwarding table data structure………………………………...….……..… 107 vii LIST OF FIGURES 1.1 An illustration of satellite broadcasting over ad-hoc nodes with terrain blockages…………………………………………………………..…….…..… 3 2.1 Example of an ad-hoc node…………………………………….………….… 10 2.2 Another illustration of satellite broadcasting over ad-hoc nodes with terrain blockages………………………………………………………..………….... 10 2.3 Multicast VS unicast technique…...….……………………………………… 15 3.1 Unidirectionality effect on optimal path……………………………..……… 32 3.2 Mathematical definition for Random Graph model………………………….. 36 3.3 Graphical explanation for Random Graph model……………………………. 36 3.4 Graphical explanation for Euclidean Graph model………………….………. 37 3.5 Unidirectional and bi-directional link densities from Random Graph….….... 42 3.6 Unidirectional and bi-directional link densities from Euclidean Graph…...… 42 3.7 Node reachability from Random Graph………………………………...……. 44 3.8 Node reachability from Euclidean Graph…………………………………….. 44 3.9 Depth of shortest path tree from Random Graph……………………..….…... 46 3.10 Depth of shortest path tree from Euclidean Graph………………………….. 47 3.11 Percentage of pairs connected from Random Graph……………………...… 48 3.12 Percentage of pairs connected from Euclidean Graph………………..…...... 49 3.13 Average path length in Random Graph…………………………………….... 50 3.14 Average path length in Euclidean Graph……………………………….....…. 50 5.1 Flowchart of path establishment protocol……...……………………….....….. 66 viii 5.2 Simple topology for path establishment……….......…………………………. 67 5.3 Sequence of actions in path establishment phase based on simple topology in Figure 5.2…………………………………………………………………..… 68 5.4 Flowchart of path maintenance protocol - part 1…..………………………… 71 5.5 Flowchart of path maintenance protocol - part 2…..…………………….…... 72 5.6 Topology before Circumstance Changes…………..………………………… 73 5.7 Sequence of actions for path maintenance under normal circumstance in Figure 5.6…………………………………………………………………….. 73 5.8.a Circumstance changes as active node moves away……..…………………. 76 5.8.a.1 Sequence of actions for circumstance changes as active node moves away in Figure 5.8.a…………………………………………………………….. 76 5.8.b Circumstance changes with line-of-sight blockage……..…………………. 78 5.8.b.1 Sequence of actions for circumstance with line-of-sight blockage in Figure 5.8.b……………………………………………………………….. 78 5.8.c Circumstance changes as inactive node is not okay due to……..…………. 79 1) line-of-sight blockage from upstream active node, and 2) moving out of range 5.8.c.1 Sequence of action for circumstance changes as inactive node is not okay in Figure 5.8.c…………………………………………………….…………. 79 5.8.d Circumstance with multiple inactive neighbor nodes exist…….………….. 80 5.8.d.1 Sequence of action for circumstance with multiple inactive nodes exist in Figure 5.8.d………………………………………….……………………. 80 5.9 Broken link recovery mechanism…………………………………...……… 82 5.10 Heuristic feature in path maintenance………………………………..……... 84 a) All neighbor nodes respond b) Only neighbor node 4 responds 6.1 Impact of unidirectionality - data delivery ratio………………………..…….. 92 ix 6.2 Impact of unidirectionality - normalized routing load…………………..……. 92 6.3 Impact of unidirectionality - overhead ratio………………………………….. 93 6.4 Impact of enhancement - data delivery ratio………………………………..... 93 6.5 Impact of enhancement - normalized routing load………………………….... 94 6.6 Impact of enhancement - overhead ratio…………………………………….... 94 x ABSTRACT The dissertation proposes a multicast routing protocol for unidirectional ad-hoc networks, called MUNet, based on the concept of spatial locality. Utilization of unidirectional links in ad-hoc networks is not generally considered by the mainstream research community because of the complexity and failure proneness. These weaknesses, which may be presented by a significant portion of unidirectional links can be overcome by the collaboration of neighbor nodes. The use of neighbor nodes significantly increases network resilience through path self-healing and path shortening. Potential link breakage can be detected and prevented by using neighbor nodes, prior to any costly, time-consuming link recovery. The advantageous impact of utilizing unidirectional links is studied using Random Graph and Euclidean Graph models. The Random Graph model, where node connection and link direction are the only concerns, serves as a theoretical benchmark for the study. The Euclidean Graph model, where more physical characteristics of the wireless network are captured, examines the ability of the network to provide network connectivity through unidirectional links. Similar to most routing protocols, MUNet consists of three main functions: path establishment, path maintenance, and link recovery. With the concept of spatial locality, it constructs a multicast meshed tree, enhanced by the inclusion of neighbor nodes. xi Because of the absence of essential feedback at the MAC layer, broadcast is the only applicable link layer protocol. The use of neighbor nodes without proper selection strategies can adversely affect the overall routing performance. In order to improve the routing efficiency and resource utilization, we proposed an enhanced version of MUNet which heuristically selects which neighbor nodes should participate. Instantaneously determined by local control traffic density, a node probabilistically sets a level of its responsiveness to participate in the routing protocol maintenance. Simulations using GlomoSim show that the packet delivery performance of MUNET is comparable to that of On-Demand Multicast Routing Protocol (ODMRP) but with significantly less routing overhead, in both unidirectional and bi-directional ad-hoc networks. The enhanced protocol shows further improvement in routing effectiveness along with a reduction in protocol overhead. CHAPTER 1 INTRODUCTION 1.1 Motivation: Asymmetry-Aware Network Structure For hybrid, multi-tiered networks, a combination of direct broadcast satellites with local redistribution, and access portals to the wired network, is envisioned as the next generation Internet (NGI) framework structure [AVD02]. The NGI is visualized to be a three-tier-network structure – the super-high-speed fiber-optic core network, the range-extension sub-networks, and the local access networks. The core network consists of fiber-optic backbones with gigabit to terabit bandwidth. The access networks, such as ad-hoc packet-radio networks, serve end-users and concentrate users’ traffic of moderate-bandwidth (hundreds of kilobits). The range-extension networks provide the bridge between NGI core and the access network with hundreds of megabit bandwidth. Since global coverage is crucial in NGI, GEO/LEO satellite networks, with their coverage nature, serve as the ideal bridge between the NGI core and the access network. Given this hybrid architecture, it is challenging to implement multicasting capability over the entire network, since packet radio network and satellite research and 2 development communities have so far been focusing only in routing in their own sub- network domain. Hybrid multicasting in such an environment is crucial to achieve continuous connectivity with great flexibility and survivability. A hybrid, multi-tiered framework structure enables a rich suite of interactive applications for mobile wireless users. It is particularly suited for tactical applications, such as information dissemination in battlefield networks, as well as potentially commercial ventures such as Internet smart cars, remote sensor networks, ubiquitous bio-terrorism alert network, emergency response units, etc. In order to support this framework, a multicast routing technique is needed to dynamically and effectively distribute relevant information to multiple mobile users. As depicted in Figure 1.1, the broadcast nature of the satellite medium, the global coverage of a GEO or LEO constellation, and the large cell size (spot-beam) make it advantageous to use a satellite network to deliver multicast packets for the NGI core or the access networks. Instead of sending duplicate multicast packets on multicast point-to-point branches as in traditional multicast-tree delivery on the ground-based wireline networks, multicast packets are forwarded onto a satellite network and broadcast in selected cells that have multicast membership. Since the satellite channel in each cell (spot-beam) is inherently broadcast, the bandwidth consumption is constant without regards to the membership size that receive the multicast traffic. Local gateway nodes eventually perform a limited range local multicast (redistribution) to deliver the packet to every group member in the cell. 3 Figure 1.1: An illustration of satellite broadcasting over ad-hoc nodes with terrain blockages. 1.2 The Problem: Asymmetric Network Communication It is a challenge to connect isolated groups of randomly scattered mobile wireless nodes by means of range-extending nodes: typically airborne communications nodes, or possibly low earth orbiting (LEO) or geostationary (GEO) satellites. Such an aerial range extension network can complement the terrestrial Mobile Ad-hoc NETwork (or MANET) to provide communications between land mobile nodes in the face of blockages, obstructions, channel degradations, etc. 4 The scope of this thesis covers the ground-level communication network. Only a subset of ground nodes can be deployed as the primary connectors between the aerial nodes and the ground mobile wireless nodes. Thus, these comparatively powerful mobile connector nodes or gateways, can act as bridges to interface several ad-hoc networks or isolated clusters of nodes. 1.3 Contributions A number of research activities has been on one-to-many communication in mobile wireless networks [CG98, GM99, LK00, LSG99, LSG02, RP99, RP00, SGC99]. However, these research efforts make unrealistic assumptions: strict reverse path forwarding, bidirectional links, or the existence of a central control component. Our contribution in this thesis is the study on impact of unidirectionality in ad- hoc networks. To better understand ad-hoc network characteristics and determine the advantageous impact of utilizing unidirectional links. We use Random Graph and Euclidean Graph models. The Random Graph model, where node connection and link direction are the only concerns, serves as a theoretical benchmark for the study. The Euclidean Graph model, where more physical characteristics of the wireless network are captured, examines the ability of the network to provide network connectivity through unidirectional links. The other contribution is in the design and evaluation of MUNet, a multicast routing protocol exploiting link directionality for mobile wireless networks. Due to the 5 existence of limited nodal capability, signal interference, and restrictive communication policy, link unidirectionality is an assumption, not an exception, in mobile wireless networks. With the concept of spatial locality, MUNet constructs a meshed multicast tree enveloped by neighbor nodes. While providing rich tree connectivity, neighbor nodes serve as boundary for scoped flooding at link recovery. As an extension to MUNet, a protocol enhancement is presented in order to improve routing performance. The enhancement heuristically sets a level of node' s participation based upon the intensity of its local activity. Redundancy in routing presumably increases network resilience. However, it proportionately diminishes routing efficiency due to signal interference and the lack of feedback at the MAC layer in unidirectional ad-hoc networks. 1.4 Dissertation Organization Chapter 1 of the thesis has presented the introduction, the motivation, the problem, and the contributions. Chapter 2 gives background information as a framework for the study. It covers an overview of ad-hoc networks, multicast routing, and unidirectionality. Chapter 3 presents a study of asymmetric networks characteristics, particularly on link unidirectionality. The advantageous impact of utilizing unidirectional links is determined by exploring different models of connectivity (i.e. Random Graph and Euclidean Graph models.) 6 Chapter 4 is a review of relevant literature in the field of routing protocols in ad-hoc networks. It addresses the following topics: spatial locality exploitation, topology-oriented approach, unidirectional routing approach, swarm intelligence concept, and GPS enabled approach. Chapter 5 presents the methodology used in the study, including the objectives of the study; the assumptions; the protocol design; the protocol enhancement; and the data forwarding mechanism. Chapter 6 presents the evaluation methodology and discusses the results of the study, particularly with the impact of unidirectionality, and the impact of protocol enhancement. The dissertation ends with concluding remarks and future direction in Chapter 7. 7 CHAPTER 2 BACKGROUND INFORMATION To provide a framework of the study, this chapter covers topics on ad-hoc networks, unidirectionality, and multicasting, in general. Characteristics and applicability of ad-hoc networks are surveyed. Then, the mechanism of multicast routing, along with traditional multicast routing protocols, are included. In addition to the cause of unidirectionality, the impact of utilizing unidirectional links in ad-hoc networks is presented. 2.1 Ad-hoc Networks The purpose of this section is to present various aspects of ad-hoc network. As wireless communication technology has been progressively investigated, researched, and developed, multi-hop mobile wireless networks, or called ad-hoc network, have gained more interest from the military and commercial communities, in addition to the research community. Due to its wireless mobility attributes, and capability to relay signal for peer-to-peer communication, such network is envisioned to have dynamic, sometimes rapidly changing, unpredictable, multi-hop topologies. Especially in a harsh 8 environment, an ad-hoc network is considered robust and autonomous because of its potential mesh-like interconnection. This section consists of three sub-sections. Section 2.1.1 answers the “What is an Ad-hoc Network?” question. Following Section 2.1.2 “Why utilize Ad-hoc Networks?”, Section 2.1.3 explores “What Ad-hoc Networks are for?” 2.1.1 What is an Ad-hoc Network? A mobile ad-hoc network is defined by MANET [CM99] as an autonomous collection of mobile network devices communicating with each other over a wireless medium, likely in a multihop manner, without fixed infrastructure support. A mobile network device --herein simply referred to as “ad-hoc node”--is a portable network device with the capability to transmit and receive wireless signals using antennas, which may be omni-directional or highly directional. Ad-hoc nodes may be located in or on vehicles (in the air, on the ground, in the sea), in small appliances, perhaps on people. An ad-hoc node functions both as an end user, being a source or a destination, and as a router, forwarding signals destined for other nodes. Characteristics of ad-hoc networks in general: - Ad-hoc node logically consists of one or more end devices, a router, and one or more wireless interfaces, shown in Figure 2.1. Each node is free to move about arbitrarily. Ad-hoc networks interconnect through wireless media . 9 Multi-hop networks. A node’s transmission range reaches only some other nodes but a node is able to relay signals. No existing fixed network infrastructure such as base stations. A set of nodes, which may be combined routers and end devices, themselves form the network routing infrastructure in an ad-hoc fashion. No preplaced fixed infrastructure or connections to the existing fixed infrastructure is required. No centralized access point. All nodes have comparable features and status, yet are independent from other nodes. Energy-constraint. Due to battery limitations, a node is likely to have a short life (until it can be recharged). Likelihood of rapidly changing and unpredictable topology due to node mobility. An ad-hoc network may serve as an extension of existing wired Internet fixed infrastructure or may be self-sufficient. Self-organization. An ad-hoc network has ability to instantly connect a wide variety of mobile wireless devices without access to routers, base stations, or Internet service providers. Bandwidth-constrained, variable capacity links. High error rate due to fading, multi-path, signal interference, environmental adversity. 10 Limited security. The security threats include eavesdropping, denial-of-service attacks, and signal jamming. Physical attack is possible as a device can be pinpointed if an enemy detects its signal. Figure 2.1: Example of an ad-hoc node. Figure 2.2: Another illustration of satellite broadcasting over ad-hoc nodes with terrain blockages. Ad-hoc networks can be further developed into multi-tiered hybrid ad-hoc networks—multiple ad-hoc networks interconnected via a satellite network, as depicted in Figure 2.2. The broadcast nature of the satellite medium, the global coverage of a GEO or LEO constellation, and the large cell size (spot-beam) make it advantageous to Satellite-capable node Mobile wireless node Terrain blockage 11 use satellite network to deliver multicast packets. Instead of the sending duplicate multicast packets on multicast point-to-point branches as in traditional multicast-tree delivery on the ground-based wired networks, multicast packets are forwarded onto a satellite network and broadcasted in selected cells that have multicast membership. Since the satellite channel in each cell (spot-beam) is inherently broadcast, the bandwidth consumption and delay is constant without regards to the membership size. A cell that receives the multicast traffic will perform a limited range local multicast (redistribution) to deliver the packet to every group member in the cell. 2.1.2 Why Utilizes Ad-hoc Networks? Despite its limited temporal and spatial extent, ad-hoc networks are convenient, flexible and cost-effective to use due to their mobility, multi-hop, fixed infrastructure independence and self-organization features. In addition to giving low cost in network setup and maintenance, system implementation time is low. As mentioned in [Per00], several salient design issues for ad-hoc networks include: Network Size Network size usually refers to the number of network nodes, but it can also refer to the geographical area covered by the network; both are critical parameters for coordinating network actions with distributed control mechanisms. Connectivity 12 Connectivity may refer to the number of neighbors that a node can link to directly, over unidirectional or bidirectional links. Connectivity may also refer to the link capacity between two nodes. Network Topology User mobility can directly affect how fast the node connectivity and hence the network topology changes; thus, it influences how and when the network protocol must adapt to changes. User Traffic The characteristics and types of user-generated traffic heavily influence the design of ad-hoc networks. What is the combination of user traffic? Short? Bursty? Delay- bound? How tolerant of loss? Priority type? Operational Environment Operational environment refers to the terrain (urban, rural, maritime, forest, etc.) that may prevent the line of sight operation. It also refers to potential sources of interference in the radio channel. Energy Since there are no fixed base stations in an ad-hoc network, all nodes have roughly equal status and the energy burden cannot be transferred to energy-advantaged nodes such as fixed base stations. However, if fixed infrastructure nodes exist, they will have an energy advantage over those that are battery operated. 13 2.1.3 Why Utilize Ad-hoc Networks? Ad-hoc networks support an array of potential applications for military and commercial use. Some interesting applications are as follows. Conferencing When mobile computer users gather outside the normal office environment, the need for collaborative computing is important due to the missing business network fixed infrastructure support. Home Networking Ad-hoc networking offers the reachability to all the nodes at home regardless of their “normal” point of attachment. Moreover, the computers can be taken to and from the office work environment and on business trips without reachability difficulty. Emergency Services When the existing fixed infrastructure is damaged or out of service due to loss of electricity or natural disasters, ad-hoc networking can help to overcome network impairment during these emergencies. Sensor Networks Sensor networks facilitate data acquisition. They offer detailed information about terrain or environmental dangerous conditions. Situations where wiring is not an option 14 For historic buildings, cable wiring might cause damages to the structure or the intention with aesthetic. In a battlefield, there is no such fixed infrastructure to support communication but communication is needed and has to be sufficiently fast and reliable. This includes any situation where wiring is not cost-effective. 2.2 Multicasting Technique The purpose of this section is to discuss traditional multicast routing protocols and their application to ad-hoc networks. The section is organized as follows. In Section 2.2.1, multicasting is introduced. Section 2.2.2 investigates problems and issues of multicasting in general and particularly in ad-hoc networks. Section 2.2.3 surveys and provides comparisons of the selected traditional multicast routing protocols. 2.2.1 Introduction Multicasting is a technique developed to send datagram packets across the Internet to a group of destinations in one-to-many communication. It is desirable to avoid or minimize packet duplication. Only one packet is sent out from a source and is replicated as needed in the network to reach many end-users as needed, as depicted in Figure 2.1. A multicasting protocol is typically composed of route discovery phase and route maintenance phase. 15 Regardless of the network environment, multicasting provides an important service for improving the efficiency and robustness of distributed systems and group- oriented applications. The benefits of multicasting are to enhance Internet performance, to support distributed applications, to reduce the cost to deploy applications, to increase productivity, to ease scalability, and to increase application availability. Figure 2.3: Multicast VS unicast technique. 2.2.2 Problems and issues Even though multicasting significantly conserves scarce resources (e.g. bandwidth and server and network processing) compared to traditional unicast routing protocols, several design concerns arise. The limited temporal and spatial extent of ad- hoc networks imposes additional design concern for a multicasting protocol. Problems and issues for multicasting protocol in general and those of ad-hoc networks in particular are briefly mentioned hereafter. General problems and issues of multicasting can be characterized or relating to: the physical node itself; network as a whole; and, the routing scheme both operation and administration. Resource usage efficiency is concerned with respect to memory requirement for router state, processing power requirement at router, and bandwidth 16 consumption for multicast path establishment, maintenance, and failure recovery. In addition to the time delay for multicast path establishment, maintenance, and failure recovery, other routing metrics such as hop count, distance, bandwidth, and signal strength are taken into consideration. Dependency upon underlying unicast routing protocol, IGMP (Internet Group Management Protocol) is as relevant as interoperability with other multicast networks with different multicasting protocol and routing policy is. Routing algorithm complexity matters as well. Whereas robustness (number of points of failure) and loop-free path are of critical protocol design goals, traffic concentration indicates the traffic capacity of a protocol. Multicast membership administration is another operational issue. Last, but not least, privacy and security have theirs significance increased along with the exponential growth of computer networking deployment. Problem and issues of multicasting in ad-hoc networks, in particular, are mostly concerned with the mobility and wireless attributes of ad-hoc node. That starts with inherent broadcast capability and error-prone connection due to wireless medium. Mobility speed and pattern have significant impacts. Nodes with matching speed and direction preserve the network topology and connection. Otherwise, multicast path maintenance is tremendously challenging because of the continuously changing network connectivity and topology. The limited battery constraint imposes power and bandwidth limitation. Merging of several ad-hoc network partitions as they move forward toward each other is rare in traditional networks. Last, but not least, is 17 asymmetric link characteristic due to different node capability and the harsh nature of the fading channel. 2.2.3 Selected Traditional Multicast Routing Protocols In order to determine key design features for a multicast routing protocol for ad- hoc networks, several existing multicasting protocols are compared as shown in Table 2.1. These are Distance Vector Multicast Routing Protocol (DVMRP) [SC90], Protocol Independent Multicast (PIM) [DEF+96], On-Demand Multicast Routing Protocol (ODMRP) [CG98, LSG99, LSG02, SGC99], Forwarding Group Multicast Protocol (FGMP) [CGZ98, CM98], Core-Assisted Mesh Protocol (CAMP) [GM99] and Multicast Ad-hoc On-Demand Distance Vector (MAODV) routing protocol [RP99, RP00]. The key features, basic operation of procedures and their applicability to ad-hoc networks are described. 2.2.3.1 Distance Vector Multicast Routing Protocol (DVMRP) Pioneers in the IP multicasting area, S. Deering and D. Cheriton proposed DVMRP [DC90], a multicasting protocol, which is an extension to a common routing algorithm – distance-vector routing protocol – to provide multicasting service. Based on a routing table built by the Bellman-Ford algorithm [Bel57], a source-based shortest path tree is constructed and is used to delivery multicast packets from a sender to the members of the multicast group. Each sender periodically broadcasts a multicast packet. Based on the reverse shortest path forwarding mechanism (RPF) [DM78] and its variants, a router 18 selectively forwards a multicast packet, only if the packet arrives from its upstream node, to its downstream nodes that have members of the group. Non-member leaf nodes and nodes without any downstream members send prune messages upstream to prune off branches to non-member nodes. After timeout, pruned branches become active again and get flooded with multicast packets. In order to speed up the membership registration process, a new member receiver can also send a join message to its upstream node. Categorized as a table-driven protocol, DVMRP is designed for use in wired networks, where the topology is static and link interfaces are well defined. In addition to Bellman-Ford shortest path tree maintenance routine, DVMRP’s occasional broadcast behavior severely limits scalability, especially in a sparse network. Even though DVMRP is simple and widely implemented, including in the MBone, it is not desirable for ad-hoc networks for several reasons. Despite low delay due to pre-discovered routes and their associated metrics, scarce bandwidth is quickly exhausted in flooding for route and membership maintenance. Dynamic topology and wireless characteristic raise critical issues such as the reverse path forwarding (RPF) and the leaf node detection. Since packets are accepted only from the shortest path, if the shortest path changes and no packets are forwarded in the new path, the node will be disconnected from the tree. There is no concept of leaf node in ad-hoc networks because all nodes are routers and there is no explicit link interface information to determine the leaf node status. 19 2.2.3.2 Protocol Independent Multicast (PIM) To efficiently establish a distribution tree across wide area internets, where many groups are sparsely represented, PIM, proposed by S. Deering, et al. in [DEF+96], suggests the concept of a rendezvous point (RP). Based on a routing table built by any independent unicast routing protocol, PIM is able to deliberately construct either a per-group RP-rooted shared tree or a source-based shortest path tree, based on the traffic concentration level. The receiver periodically sends out a join message toward the RP. Associated with un-expired multicast forwarding entry, RP forwards multicast packets down to receivers along the RP-rooted shortest path tree. To announce its existence, the sender sends out a register message piggybacked on a data packet to the RP. The RP periodically sends out an RP-reachability message. When the received data rate is higher than a certain threshold, an RP-rooted shared tree is reconfigured to be a source-based shortest path tree. The receiver sends out a join message toward the source. An intermediate router, if a received join message destined to an outgoing interface link that differs from that of link toward RP, sends a join message toward the source and sends a prune message toward the RP afterward. Classified as a receiver-initiated, table-driven protocol, PIM is intended to constrain the data distribution in sparse wired static networks. With the ability to switch from RP-root shared tree to source-based shortest path tree, PIM avoids a single point of failure and traffic concentration at the RP. With RP-root shared tree, however, 20 shortest path is not ensured. Note that route discovery is pre-performed by an underlying unicast routing protocol. PIM seamlessly supports interoperability with non-PIM-capable domains. Soft state mechanism is used to adapt to underlying network conditions and group dynamics. Although PIM’s innovative design improves scalability, robustness, and flexibility, it is unsuitable for ad-hoc networks because of its design mainly for static and wired networks. However, the idea of a rendezvous point is practical for ad-hoc networks. An RP could act as a connecting access point between network tiers in multi- tiered hybrid networks. In addition to typical tree-based path maintenance, RP-root shared tree causes suboptimal path length. 2.2.3.3 On-Demand Multicast Routing Protocol (ODMRP) Primarily designed for ad-hoc networks, ODMRP by S.-J. Lee, et al. [CG98, LSG99, LSG02, SGC99] suggests a mesh-based, rather than a conventional tree-based multicast scheme and uses a forwarding group concept, to achieve robustness and scalability. A forwarding group, or a mesh, is basically a subset of nodes that forwards the multicast packets via scoped flooding. The mesh provides richer connectivity among multicast members compared to trees. Flooding, scoped within the mesh, helps overcome node displacements, the exchange of routing table, and channel fading. Hence, unlike trees, frequent reconfigurations are not required. In ODMRP, group membership and multicast routes are dynamically established and updated by the source on demand. Periodically, the source floods a join 21 packet to refresh multicast routing entries. Intermediate nodes along the shortest path constitute the forwarding group. Basically, the forwarding group is an aggregate of intermediate nodes of all source-based shortest path trees. After the route discovery phase, packets can be forwarded via any forwarding group members, preferably via ones along the shortest path. Categorized as a sender-initiated, on-demand protocol, ODMRP functions satisfactorily in dense ad-hoc networks, with respect to bandwidth and storage overhead. While providing remarkable robustness and scalability, the forwarding group mechanism could adversely consume scarce resources, such as bandwidth and memory storage, in a sparse scenario. ODMRP can operate, not only with any unicast routing protocol, it can also function as a multicast routing protocol. Since some ad-hoc networks are intended for vast area network coverage, ODMRP’s flooding mechanism in the route discovery phase and the forwarding group membership maintenance are not very efficient. The mesh concept, however, increases robustness in connectivity, route and membership maintenance. A major problem of ODMRP is that it assumes bidirectional links, which is not true for all links in an ad- hoc network. 2.2.3.4 Forwarding Group Multicast Protocol (FGMP) Stemming from the same research group, FGMP [CM98, GM99] is the predecessor of ODMRP [SGC99]. ODMRP is described in the Related Work chapter. 22 Since both share the forwarding group concept and fundamental mechanisms, only the main differences are noted here. The FGMP routing scheme is table-driven [CGZ98] and on-demand [CM98]. Similar to other table-driven protocols, the table-driven FGMP poses scalability problems as the multicast group grows. It is both sender-initiated and receiver-initiated. Periodically, either the sender globally broadcasts its information or the receiver globally floods a join request. The idea of the forwarding table establishment and maintenance is the same for both the table-driven and on-demand FGMPs. In order to reduce the channel overhead, improve the delivery efficiency and generally enhance scaling and mobility support, a modified version called the on-demand FGMP has been proposed. On-demand FGMP is sender-initiated. 2.2.3.5 Core-Assisted Mesh Protocol (CAMP) CAMP [GM99], proposed by J.J. Garcia-Luna-Aceves and E. Madruga, generalizes the notion of core-based trees into multicast meshes that have much richer connectivity. CAMP classifies nodes in the network as cores, duplex members, simplex members, or non-members. Cores, as central points, are used to limit the control traffic needed for receivers to join multicast groups. They need not be part of the mesh of their group. Acting as mesh members, duplex members forward any multicast packet for the group. Simplex members forward packets only from senders to the rest of the group. Simplex members can later become duplex members. The use of cores in CAMP is claimed to eliminate the need for flooding, assuming availability of routing information from a unicasting protocol. 23 CAMP uses a receiver-initiated method for routers to join a multicast group. A join request is sent to a nearest duplex member toward a nearest core. When the nearest duplex member, or in worst-case scenario the nearest core, replies with an ACK, the multicast routing table is updated. Periodically, a receiver sends out a heartbeat message toward each source to ensure that the mesh contains all reverse shortest paths from all receivers to all senders. A router, including a core, leaving a multicast group explicitly sends out a quit message to its neighbors. Classified as a receiver-initiated, on-demand protocol, CAMP supports multicasting in very dynamic ad-hoc networks. Similar to ODMRP, it provides robustness with the idea of mesh. The use of mesh, instead of trees, in a dynamic environment avoids frequent repairs of branch reconfiguration and the progressively faster exchange of routing tables as mobility increases. Cores act as representatives of senders and receivers. Even though CAMP seems to overcome scaling limitations, it relies on an underlying unicasting protocol, which in turn directly affects CAMP’s scalability. Even though CAMP relies on several assumptions that are not applicable to ad- hoc networks, its key idea of using core nodes in directing and limiting traffic flows, as well as the mesh concept are quite practical for ad-hoc networks. A selected core node could serve as a connecting access point between network tiers in multi-tiered hybrid networks. The number of mesh membership increases the connectivity’s robustness. Unfortunately, the size of core nodes imposes synchronization problems. Similar to ODMRP, CAMP assumes bidirectional links, which cannot be assured in ad-hoc networks. 2.2.3.6 Multicast Ad-hoc On-Demand Distance Vector (MAODV) Routing Proposed by Elizabeth M. Royer and Charles E. Perkins, Multicast Ad-hoc On- Demand Distance Vector Protocol (MAODV) [RP99, RP00] is a novel multicasting algorithm stemming from the Ad-hoc On-Demand Distance Vector (AODV) [PR99] 24 for the operation of ad-hoc networks. It creates bidirectional shared multicast trees connecting multicast sources and receivers. These multicast trees are maintained as long as group members exist within the connected portion of the network. Each multicast group has a group leader whose responsibility is maintaining the group sequence number, which is used to ensure freshness of routing information. In MAODV, either a sender or a receiver can join the tree at its own discretion. A tree member can graft a branch toward a requesting node. A tree member keeps track of its upstream tree member, its downstream neighbor(s) and its downstream multicast neighbor(s). Started at the group leader, a tree member forwards a Multicast Activation packet (MACT) to its multicast neighbor(s) upon receiving a MACT. To leave the tree, a multicast member can simply be silent if it has no downstream multicast neighbor. Otherwise, it becomes just a tree member. Classified as an on-demand multicasting protocol, MAODV would work well in very dynamic ad-hoc networks. However, due to the bidirectional link assumption, the group leader selection mechanism and the expanding ring search mechanism, MAODV is not quite practical for ad-hoc networks that utilize unidirectional links. 25 MAODV On-demand Wireless Sender/Receiver Flat Yes Tree Yes No No No Shortest-path length Directly Yes Shortest path between senders and receivers No No CAMP On-demand Wireless Receiver Flat Yes Mesh Yes Yes Yes No Shortest-path length Via core No Shortest path between senders and receivers No No ODMRP On-demand Wireless Sender Flat Yes Mesh Yes Yes No No Shortest-path length Directly Yes Shortest path between senders and receivers No No PIM Table-driven Wireline Receiver Flat Yes Tree Yes No Yes Yes. At RP Suboptimal path length Strictly via RP No Shortest path between RP and senders/receiver s No No DVMRP Table-driven Wireline Sender Flat No Tree Transient Loops No No No Shortest-path length Directly No Shortest path between senders and receivers No No Protocols Category Network Environment Multicasting Initiated Routing Philosophy Soft State Mechanism Configuration Loop Free Path Multiple Route Appointed Core Traffic Concentration Path Length from Sender to Receiver Traffic flow between Independency on Routing Metrics QoS Awareness Unidirectional Link Table 2.1: Comparisons of key features and characteristics of selected multicasting protocols. 26 2.3 Unidirectionality/Asymmetry 2.3.1 Unidirectionality in General Unidirectionality, an asymmetric characteristic of ad-hoc networks, is simply caused by different radio capabilities, signal interference, selection of connection configurations, and strategic communication policy. Unidirectionality can be temporary or permanent. The existence of a unidirectional link makes the whole network unidirectional. • Different radio capabilities. Radio devices within a network can have different transmission power or receiver sensitivities. Moreover, power consumption rate and power conservation policy temporally vary the capability of radio devices. Node with high transmission power can transmit messages to nodes with low transmission power, but not vice versa. Nodes with high receiver sensitivity can receive messages, while nodes with low receiver sensitivity may not be able to do so. This is commonly found in battlefields or emergency response services where man-packed and vehicular radios exist. • Signal interference. Even a network of radio devices with identical communication capabilities can have unidirectional links. Signal interference induced either by hostile jammers or by friendly “co-sites” will reduce a nearby receiver’s 27 sensitivity. If only one receiver is sensitivity-impaired, unidirectional links may result. One classical example is the “hidden terminal” problem [TK75]. • Selection of connection configuration. For wide-area information broadcast, satellite-based transmitters essentially provide high bandwidth links over large geographical areas. They have been used for the forward links while the return links use alternative paths, due to the high cost of satellite up-link devices. • Strategic communication policy such as emission-controlled (EMCON) operations. An extreme instance—applicable only in military networks—is when some cannot transmit due to impending threat. In such cases, it may be necessary to have some other node provide a comparable route in response to a route discovery request. (Obviously, an EMCON node cannot participate in two-way communications, but it still needs to receive data.) 2.3.2 Impact of Unidirectionality in Ad-hoc Networks Unidirectionality in ad-hoc networks has been studied in [VS03], using Random Graph and Euclidean Graph models [Wes00]. The Random Graph model serves as a theoretical benchmark for the study. Only node connection and link directionality are considered. The Euclidean Graph model simulates wireless network scenarios and captures the network dynamics including node mobility and transmission power. The basic idea is to apply to a 100-node network, with a variety of network density and connectivity, and generate connections between nodes and a shortest-path root-based tree. The shortest-path root-based tree is defined as a tree network topology, rooted at 28 one node that connects to most, if not all, of the nodes, over the smallest number of hops. The rationale is that in ad-hoc networks where connectivity is highly transient and evidently unidirectional, the ability to transmit messages to most, if not all, of the nodes within a small number of hops on the provision of the minimal network connectivity is crucial. The advantages of unidirectional links over bi-directional links in ad-hoc networks are examined in [VS03]. Specifically, the work explores several effects of the exclusive use of bi-directional links, and the use of all links (i.e. both bi-directional and unidirectional links) on reachability and connectivity at the node level and the network level. More detail is in Chapter 3. The study shows the presence of a significant proportion of inherent unidirectional links, compared to that of bi-directional links, in ad-hoc networks. The use of available unidirectional links increases node reachability, shortens the depth of shortest-path root-based tree, raises the percentage of pairs connected and cuts down the average path length. In other words, the use of unidirectional links improves network performance by increasing network connectivity and reducing path length. 29 2.4 Summary While being limited spatially and temporally, ad-hoc nodes can collectively yield a robust and autonomous multi-hop mobile wireless network by utilizing the potential mesh-like interconnections. They can be organized into a multi-tiered network. Multicasting is an economical and efficient way of one-to-many communication, conserving scarce network resources. Traditional multicasting protocols are categorized as table-driven or on-demand, sender- or receiver-initiated, sender- or core-based, and with tree or mesh topology. Nonetheless, all assumes bi- directional networks when designed whereas ad-hoc networks are unidirectional. Unidirectionality -an integral characteristic of ad-hoc networks- is resulted from signal disparity between a sending node and a receiving node. The signal discrepancy could result from unavoidable physical limitations, signal interference and attenuation, or strategic design and policy. 30 CHAPTER 3 ASYMMETRIC NETWORKS 3.1 Introduction The objective of this chapter is to study ad-hoc network' s asymmetric characteristics, particularly on link unidirectionality. Although unidirectionality in ad- hoc networks has been a concern, no performance analysis has been carried out or available in the public domain. The study offers a better understanding on the future routing protocol design for ad-hoc networks. By using unidirectional links, in addition to bi-directional links, the routing performance is expected to improve greatly. The advantageous impact of utilizing unidirectional links is studied using Random Graph and Euclidean Graph models. The Random Graph model, where node connection and link direction are the only concerns, serves as a theoretical benchmark for the study. The Euclidean Graph model, where more physical characteristics of the wireless network are captured, examines the ability of the network to provide network connectivity through unidirectional links. 31 We will investigate the improvement of the routing performance with the use of unidirectional links, in addition to bi-directional links, in ad-hoc networks by examining the following questions: What is the portion of unidirectional links in the network? How can the use of unidirectional links, in addition to bi-directional links, improve the routing performance? 3.2 Real Network Characteristics To study asymmetric network characteristics, particularly on link unidirectionality, a set of experiments has been carried out to quantitatively measure the impact of utilizing all available links (both unidirectional and bi-directional links) in comparison to that of utilizing bi-directional links only. An example is illustrated in Figure 3.1. In a connected ad-hoc network, depicted in Figure 3.1(a), where both unidirectional and bi-directional links exist, routing over only bi-directional links is likely to establish non-optimal paths, as in Figure 3.1(b). On the other hand, routing over all possible links (both bi-directional and unidirectional links) can give out an optimal path, which is the shortest path, as in Figure 3.1(c). To measure the impact, two models, namely Random Graph and Euclidean Graph [Wes00] are employed. The Random Graph model serves as a theoretical benchmark for the study. Only node connection and link directionality are considered. The Euclidean Graph model simulates wireless network scenarios and captures more 32 physical characteristics of the network. The basic idea is to apply to a 100-node network, with a variety of network density and connectivity, and generate connections between nodes and a shortest-path root-based tree. A connection between two nodes (i.e. i and j) of n hops is denoted as d(i, j: n) where d(i, j: n+1) = min{d(i, k,: n) + d(k, j:1)} and i j k. The shortest-path root-based tree is defined as a tree network topology, rooted at one node, connects to most, if not all, of the nodes, within a smallest number of hops. The rationale is that in ad-hoc networks where connectivity is highly transient and evidently unidirectional, the ability to transmit messages to most, if not all, of the nodes within a small number of hops on the provision of the minimal network connectivity is crucial. a) A simple network b) Tree established over bidirectional links c) Tree established over unidirectional links Figure 3.1: Unidirectionality effect on optimal path. Node Source Node Destination Node Bi-directional Link Unidirectional Link S D S D D a) Use only bidirectional links b) Use all available links S D D S D D 33 For evaluation purposes, the definitions of graph properties are as follows: Link Density (LD) is defined as the percentage of directional links from the underlying fully connected network. A fully connected network has a link from every node to every other node. A bi-directional link consists of two directional links in forward and backward direction. Link Density can be written as 1 N N L LD x 100 % where L is a number of directional links, N is a number of nodes in the network. Unidirectional Link Density (ULD) is defined as the percentage of unidirectional links from the underlying fully connected network. 1 N N L ULD u x 100 % where L u is a number of unidirectional links, N is a number of nodes in the network, L = L u + L b , and LD = NLD + BLD. Bi-directional Link Density (BLD) is defined as the percentage of bi- directional links from the underlying fully connected network. 1 N N L BLD b x 100 % where L b is a number of bi-directional links, N is a number of nodes in the network, 34 L = L u + L b , and LD = NLD + BLD. Node reachability (NR) level is defined as the percentage of all reachable nodes included in a shortest-path tree. N N NR r x 100 % where N r is a number of all reachable nodes included in a shortest-path tree, N is a number of nodes in the network. Depth of the shortest-path root-based tree (D) is defined as the maximum number of hops it takes from a root node to most, if not all, of the other nodes on the provision of the minimal network connectivity. It can also be referred as the height of the shortest-path root-based tree. Pairs Connected (PC) is defined as the percentage of all connections between nodes, if reachable. ) 1 ( N N C PC ij x 100 % where Cij is a number of connections between node i and node j such that d(i, j: n+1) = min{d(i, k,: n) + d(k, j:1)} and i j k. Average Path Length (APL) is defined as the average length of connections between nodes in hops. The evaluation is performed at the network level and the node level. At the network level, the evaluation metrics are the link density including the unidirectional link density and the bi-directional link density, the node reachability, and the depth of 35 the shortest-path root-based tree. At the node level, the evaluation metrics are the percentage of pairs connected, and average path length. 3.2.1 Random Graph Model A Random Graph is a graph in which properties such as the number of graph vertices, graph edges, and connections between them are determined in some random way. In order to serve the study purposes, the randomness of its properties is limited with respect to directionality and connection mode. Directed Graph (G) which is theoretically defined in [Wes00] as a triple consisting of a vertex set V(G), an edge set E(G), and a function assigned each edge an ordered pair of vertices. The first vertex of the ordered pair is the tail of the edge, and the second is the head; together they are endpoints. For instance, (A, B) denotes a unidirectional link from node A to node B. In order to have a bi-directional link between node A and node B, both (A, B) and (B, A) are required. Random Graphs are generated in two modes, namely the entirely randomly connected and the weakly connected. For an entirely randomly connected Random Graph, a directed link between any two nodes is randomly generated such that the number of links generated fulfills the specified link density. It is possible that a node may be detached from the rest. A weakly connected graph is similar to an entirely randomly connected graph, with an additional constraint that its underlying graph is connected, regardless of the link direction. Mathematic definition and graphical explanation are shown in Figure 3.2 and 3.3 respectively. 36 Directed graph G := (V, A) ; where V is a set of nodes, A is a set of ordered pairs of nodes, e.g. (x, y) is directed from x to y. Entirely randomly connected mode Weakly connected mode ) 1 ( * * , , 1 , 1 , 2 , , , , , ,..., 1 1 1 1 N N LD l j i N j N i N b V y x for y x y x A N V N l j i N b ) 1 ( * * , , 1 , 1 , , , ,..., 1 N N LD l j i N j N i V y x for y x A N V l j i Figure 3.2: Mathematical definition for Random Graph model. 1. Entirely randomly connected mode 2. Weakly connected mode Figure 3.3: Graphical explanation for Random Graph model. 37 Then, two shortest-path root-based trees are constructed; one with only bi- directional links, and the other with both unidirectional and bi-directional links. To ensure the best possible shortest-path tree, the selected root node is the node that yields the minimum tree depth, where all nodes are examined. 3.2.2 Euclidean Graph Model Euclidean Graph is a graph whose nodes have Cartesian coordinates in the plane as depicted in Figure 3.4. Each point corresponds to a node. Each edge between any two points represents associated parameters of a link between the two corresponding nodes. The Euclidean Graph model shows the ability of the network: data dissemination, connectivity resilience. The link parameters include link existence and link directionality. Link existence depends on transmission power, receiver sensitivity and distance between two corresponding nodes. To make the graph directed, every link is transformed to two opposite (directed) links. N 1 (X 1 , Y 1 ) N 2 (X 2 , Y 2 ) N 3 (X 3 , Y 3 ) Parameters of edges (L) between two nodes: • L(N 1 , N 2 ) = {a 12 , b 12 , c 12 , …} • L(N 1 , N 3 ) = {a 13 , b 13 , c 13 , …} • L(N 2 , N 3 ) = {a 23 , b 23 , c2 3 , …} Figure 3.4: Graphical explanation for Euclidean Graph model. 38 We consider a wireless network with 100 stationary wireless nodes. Each node with omni-directional antennas is randomly placed in a square area (A), and assigned transmission power (TP). Then, network size, area and transmission power parameters are denoted as network density (ND) and relative transmission power (RTP). For example, network A has one node with TP of 1, located in a one unit area and network B has four nodes, each with TP of 2, located in a four unit area. Both networks yield the same ND (= 1) and RTP (= 1). Formally, the network density (ND) and relative transmission power (RTP) can be defined as follows. Network Density (ND) is normalized as an average number of nodes ( n ) per area unit, which can be written as A n ND where A is the area of coverage e.g. A 100-node network with node density of 1 is placed in a 10 unit by 10 unit region. Relative Transmission Power (RTP) is defined as the normalized transmission power with respect to an area length, which can be written as AL TP RTP where TP is the transmission power in diameter, AL is the length of the coverage region. A AL . e.g. A node with TP of 2 in a 10 unit by 10 unit region has a RTP of 0.2. 39 3.3 Simulation and Results With 15 runs on different random seeds, simulation experiments were carried out to determine the impact when unidirectional links, in addition to bi-directional links, are used, where two models have been used. Interestingly, both models show the similar outcome. This section comparatively demonstrates the network performance when all links, not only bi-directional links, are used, under the Random Graph model and the Euclidean Graph model. Performance metrics are the unidirectional link density, bi-directional link density, node reachability, depth of the shortest-path root- based tree, percentage of pairs connected, and average path length. With 100 nodes, more than 1,500 directed Random Graphs have been generated in two modes, namely the randomly connected and the weakly connected. A Random Graph is constructed with link densities between 1 and 60. Then, node connections and shortest-path root-based trees of each mode are constructed using both unidirectional and bi-directional links and only bi-directional links. Of the same number of nodes, 500 Euclidean Graphs have been generated. Nodes are randomly placed with network densities of 0.005, 0.01, 0.05, 0.1, 0.5, 1, and 5, and with relative transmission powers of 0.01, 0.05, 0.1, 0.2, 0.4, and 0.6. Two different power transmission modes are used, namely the fixed transmission power at the maximum transmission power, and the uniformly distributed transmission power in 40 the range (0, maximum Tx power * 2], whose mean value is the maximum transmission power. 3.3.1 Unidirectional Link Density vs. Bi-directional Link Density The proportion of unidirectional links is significant enough such that the use of unidirectional links, in addition to bi-directional links, will greatly improve the routing performance. The portions of unidirectional links and bi-directional links used from all generated directional links are shown in Figures 3.5 and 3.6. Figure 3.5 shows no significant differences between the randomly connected graphs and the weakly connected graphs. When the link density is less than 20 percent, most directional links are unidirectional, whereas bi-directional links constitute less than 5 % out of all links considered. The bi-directional links show up at a higher probability when the graph gets more connected. With 60% link density, half of the directional links are bi-directional. Assuming the use of omni-directional antenna in the Euclidean Graph, Figure 3.6 shows that nodes with identical transmission power produce a completely bi- directional network. Networks of nodes with the fixed maximum transmission range contain only bi-directional links. Network of nodes with random transmission power has unidirectional links almost as much as bi-directional links. Due to the uniform distribution of the transmission power level, the disparity in the transmission power 41 level of a source node and of a sink node, which causes unidirectional links, occurs as equally as the similarity, which creates bi-directional links, does. The different results between two models can be explained. The unidirectional link density from the Random Graph model is twice as much as the unidirectional link density of network of nodes with the random transmission power using the Euclidean Graph model. The bi-directional link density of network of nodes with the random transmission power from the Euclidean Graph model is twice as much as that using the Random Graph model. This is due to the nature of omni-directional transmission range. With a directional antenna, where the transmission range can be beamed in one particular direction, not all directions, the bi-directional link density of network of nodes with random transmission power will be lower. 3.3.2 Node Reachability As the link density increases, more nodes are connected, that results in higher node reachability. Figure 3.7 illustrates that all nodes from the Random Graph model can be reached via unidirectional links, in addition to bi-directional links, at the mere 4% link density. One with both unidirectional and bi-directional links in the weakly connected mode performs slightly (1 percent) better than that in the randomly connected mode. Formed by only bi-directional links, a network is able to cover all the 42 nodes at 20 percent link density. For bi-directional links, there is no difference between the weakly connected and randomly connected modes. 0 5 10 15 20 25 30 35 0 10 20 30 40 50 60 Link Density (% ) Categorized Link Density (%) Bi, weakly connected Bi, randomly connected Uni, weakly connected Uni, randomly connected Figure 3.5: Unidirectional and bi-directional link densities from Random Graph. 0 5 10 15 20 25 30 0 5 10 15 20 25 30 35 40 45 50 Link Density (%) Categorized Link Density (%) Bi-directional with fixed Tx power Bi-directional with random Tx power Unidirectional with fixed Tx power Unidirectional with random Tx power F igure 3.6: Unidirectional and bi-directional link densities from Euclidean Graph. 43 With the fixed transmission power, network connected via both unidirectional links and bi-directional links, discovering all nodes at the link density of 9 percent, is slightly better than via only bi-directional links, at the link density of 10 percent. With the random transmission power level, unidirectional network reaches all nodes at the link density of 8 percent, while bi-directional network may be able to approach all nodes with the link density of higher than 40 percent. Non-uniform transmission power levels among nodes can greatly worsen bi-directional node reachability, but considerably improve unidirectional node reachability. Based on Figure 3.7 and Figure 3.8, the use of unidirectional links, in addition to bi-directional links, can ease routing, as most nodes can be reachable with minimal network connectivity. This is crucial in ad-hoc network where the link connectivity is unstable. In order to reach 90 percent of the network for both approaches, the connection using both unidirectional and bi-directional links needs only 3 – 8 percent link density, while the connection using only bi-directional links requires 8 – 25 percent link density. 44 0 20 40 60 80 100 0 5 10 15 20 25 30 35 40 Link Density (%) Node Reachability (%) Bi-directional, weakly connected Bi-directional, randomly connected Unidirectional, weakly connected Unidirectional, randomly connected Figure 3.7: Node reachability from Random Graph. 0 10 20 30 40 50 60 70 80 90 100 0 5 10 15 20 25 30 35 40 Link Density (%) Node Reachability (%) bi-directional with fixed Tx power bi-directional with random Tx power unidirectional with fixed Tx power unidirectional with random Tx power Figure 3.8: Node reachability from Euclidean Graph. 45 3.3.3 Depth of Shortest-path Root-based Tree The depth level of shortest-path root-based tree gives an idea of how fast a message can propagate across a network. The higher the depth of shortest-path tree is, the more the number of hops it takes a message to traverse across the network. However, a low depth level of shortest-path tree before its peak is meaningless, due to insufficient node reachability, as exhibited in Figure 3.7 and 3.8. With the node reachability level is insufficient, only a subset of network is reachable. Therefore a shortest-path root-based tree constructed from only a subset of nodes does not represent the whole network. A node is considered as reachable, if there exists a path from a root node onto the node itself. The peak depth level indicates a maximum number of hops to reach most, if not all, of the nodes with the least link density. Even though the weakly connected graph has a relatively lower depth level than the randomly connected graph does, the use of unidirectional links, in addition to bi- directional links, has much more impact on the depth of the shortest path tree than on the use of bi-directional links only. As demonstrated in Figure 3.9, with both unidirectional and bi-directional links, the peak depth level of around 10 is reached at 2 percent link density and at 4 percent link density for the weakly connected graph and the randomly connected graph, respectively. Whereas, with only bi-directional connection, the peak depth level of 10 is reached at 13 percent and 18 percent link densities for the weakly connected graph and the randomly connected graph, 46 respectively. In order to reach the depth level of 4 to reach most, if not all, of the nodes, unidirectional connection achieves at 8 percent link density, while bi-directional connection gets there at 30 percent link density. With more than 30 percent link density, there is no significant difference among all, as the depth level converges down to 3. 0 2 4 6 8 10 12 0 5 10 15 20 25 30 35 40 Link Density (%) Depth of Shortest Path Tree Bi-directional, weakly connected Bi-directional, randomly connected Unidirectional, weakly connected Unidirectional, randomly connected Figure 3.9: Depth of shortest path tree from Random Graph. Figure 3.10 illustrates how the transmission mode, in addition to link directionality, affects the depth level. At the link density of about 4 percent, each tree reaches its peak depth level, at different depth levels. The shortest path trees with both unidirectional and bi-directional links and with only bi-directional links, regardless of the transmission modes, have their peak depth levels at around 10 and around 8, 47 respectively. In order to reach the depth level of 4 to reach most, if not all, of the nodes, a tree with the random transmission power and unidirectional connection achieves at 15 percent link density. A tree with the random transmission power and bi-directional connection gets there at 35 percent link density. The rest arrives at around 25 percent link density. At more than 30 percent link density, there is no significant difference, as the depth level converges to 3. Both approaches show similar results as link density increases. The peak depth level of one with unidirectional links is almost 11, and one with bi-directional links is about 10. Figure 3.9 and Figure 3.10 illustrate that both approaches have a depth level of 3 at convergence. 0 2 4 6 8 10 12 0 5 10 15 20 25 30 35 40 Link Density (%) Depth of Shortest-Path Tree bi-directional with fixed Tx power bi-directional with random Tx power unidirectional with fixed Tx power unidirectional with random Tx power Figure 3.10: Depth of shortest path tree from Euclidean Graph. 48 3.3.4 Percentage of Pairs Connected The percentage of pairs connected indicates how well a node can reach all other nodes. Figure 3.11 shows that in the Random Graph model a node can reach 90% of all nodes at 5 and 15 percent link densities with both unidirectional and bi-directional links and with only bi-directional links respectively. The Euclidean Graph in Figure 3.12 illustrates networks with identical transmission power, regardless of directionality, pair up at less than half of the link density that those with random transmission power do. In order to reach 90% of all nodes, it takes 10%, 20% and 30% link densities, for networks with identical transmission power, unidirectional networks with random transmission power, and bi-directional networks with random transmission power respectively. Figures 3.11 and 3.12 confirm that using unidirectional links, in addition to bi-directional links, expedites network connectivity by 50 percent. 0 20 40 60 80 100 0 5 10 15 20 25 30 35 40 Link Density (%) Pairs Connected (%) Bi-directional Links Unidirectional Links Figure 3.11: Percentage of pairs connected from Random Graph. 49 3.3.5 Average Path Length In addition to how many connections nodes paired up together, the length of connections is important. It determines how fast for a message to arrive at a sink node. Illustrated in Figure 3.13, for Random Graphs, a node takes an average of 3 hops to reach all other nodes at only 5% link density using both unidirectional and bi- directional links, while it takes 3.7 hops at 15% link density using only bi-directional links. With Euclidean Graphs in Figure 3.14, a node with random transmission power takes an average of 3.2 hops to reach all other nodes at only 5% link density using both unidirectional and bi-directional links whereas it takes 4.3 hops at 15% link density using only bi-directional links. The average path length of a node with uniform transmission power, regardless of directionality, is in between. It takes an average of 4.7 hops at 5% link density. 0 20 40 60 80 100 0 5 10 15 20 25 30 35 40 Link Density (%) Pairs Connected (%) Bi-directional with fixed Tx power Bi-directional with random Tx power Unidirectional with fixed Tx power Unidirectional with random Tx power Figure 3.12: Percentage of pairs connected from Euclidean Graph. 50 0 1 2 3 4 5 0 5 10 15 20 25 30 35 40 Link Density (%) A verage Path Length Bi-directional Links Unidirectional Links Figure 3.13: Average path length in Random Graph. 0 1 2 3 4 5 0 5 10 15 20 25 30 35 40 Link Density (% ) Average Path Length Bi-directional with fixed Tx power Bi-directional with random Tx power Unidirectional with fixed Tx power Unidirectional with random Tx power Figure 3.14: Average path length in Euclidean Graph. 51 3.4 Summary The study on real ad-hoc network characteristics, particularly on link unidirectionality, offers a better understanding on the future routing protocol design for ad-hoc network. By using unidirectional links, in addition to bi-directional links, the routing performance will improve greatly. Both unidirectional and bi-directional links in the Random Graph and Euclidean Graph models are investigated. In order to accurately evaluate the use of unidirectional links, the Random Graph model serves as a theoretical benchmark, while the Euclidean Graph model shows the ability of the network to transmit messages to most, if not all, of the nodes within a small number of hops on the provision of the minimal network connectivity. The simulation results show the presence of a significant proportion of unidirectional links, and the improvement of network performance by increasing network connectivity and reducing path length. As the experiment in this chapter covers only stationary wireless nodes with different power transmission modes, it is possible to extend the study to take node mobility and its effect on link reliability into account. In addition to unidirectionality, other aspects of link characteristics such as bandwidth and error rate can be explored. The study of link unidirectionality in a large-scale network is also of interest. 52 CHAPTER 4 RELATED WORK Routing protocols can be categorized into static or adaptive, centralized or distributed, and minimal or non-minimal. Whereas static routing protocols are absolutely inapplicable in ad-hoc networks where network dynamics are transient, adaptive ones [JM96, JMQ00, Moy97, PB94, WPD88] incur inconsistencies arising from node mobility, node failures, and potential oscillations that lead to circular paths and instability. Centralized routing schemes have problems with scalability and a single point of failure while distributed ones [PR99, RP99, RP00, SGC99] do with inconsistencies among nodes. Minimal routing algorithms select only minimal cost paths (i.e. shortest path) while non-minimal ones takes other heuristics such as QoS awareness and load balancing into consideration. For performance improvement, several distinguishing features have been investigated, proposed and integrated into routing protocols for multi-hop, wireless networks. A selected set of features includes neighbor assistance [LK00, SG01], swarm intelligence [BDT99, DD97, LB03], mesh topology [CG98, LSG99, LSG02, SCG99], link unidirectionality [DDE97, DDI+01, NP00], and GPS capability [TR00, TR01]. A selected handful of the numerous routing protocols can sufficiently present their associated noteworthy features broadly categorized into four groups: spatial locality 53 exploitation, topology-oriented approach, unidirectional routing approach, swarm intelligence concept and GPS enabled approach. 4.1 Spatial Locality Exploitation The concept of spatial locality is demonstrated in Neighbor Supporting Ad-hoc Multicast Routing Protocol (NSMR) [LK00]. A multicast mesh of a group consists of sources, receivers, forwarding nodes, and links connecting them. The intermediate nodes partake in the request-reply packet exchange during the mesh creation are forwarding nodes. The nodes in a multicast mesh are called mesh nodes. Neighbor nodes - nodes within one-hop reach from a mesh node - actively participate in local route discovery and route maintenance. Expensive flooding route discovery is reserved only for an initial route establishment and a network partition repair. A soft state approach is used. Routes are built and maintained with basic hop-based route discover and reply messages. Categorized as a mesh-based, reactive multicast routing scheme, NSMR assumes bidirectional link communication for control packet exchange. While being reduced by the technique of node locality, the control routing overhead increases with the expanding ring search when a node joins. In the Neighborhood Aware Source Routing scheme (NSR) [SG01], a node maintains a partial topology of the network. Initially, nodes within a two-hop vicinity exchange routing information based on the link-state routing protocol [JMQ00]. To 54 build or repair a source route to a destination, a source node first performs a RREQ- RREP handshake with nodes in its neighborhood region. If the initial handshake attempt fails, a network-wide RREQ flooding becomes necessary. Then, a source route [JM96, JM99] is constructed. Unless link failure occurs within a two-hop neighborhood of a source route node, a RERR packet is sent along the reverse path, destined to the data source node. When the RERR packet reaches its destination, the source of the data packets updates its topology graph and re-computes its shortest-path tree. NSR requires the availability of reverse path and localized network information. Broadcast HELLO and RREQ packets are transmitted unreliably and it is assumed that link-level protocol can inform NSR when a packet cannot be sent over a particular link. 4.2 Topology-oriented Approach On-Demand Multicast Routing Protocol or ODMRP [CG98, LSG99, LSG02, SCG99] is a mesh-based multicast routing protocol that uses a forwarding group concept to maximize routing robustness and to minimize the transient route reconfiguration. A forwarding group – a subset of nodes that forwards multicast packets – is periodically configured through the network-wide flooding of the route discovery process. While Query control packets from a source are broadcast throughout the network, Reply packets initiated by a multicast group member are propagated back upstream to the source node using the reverse path forwarding mechanism [DM78] Nodes on the reverse path become the forwarding group. 55 Whereas the mesh concept provides richer connectivity and eases the route and membership maintenance, the ODMRP periodic flooding process can exponentially increase the routing overhead as the network size increases. The Multicast Ad-hoc On-Demand Distance Vector Protocol or MAODV [RP99, RP00] is analogous to AODV [PR99], which is a unicast operation. Multicast routes are discovered using a broadcast route discovery mechanism, initiated by a node wishing to join a multicast group. A RREQ message is broadcast across the network and is responded to by a node member with at least as great as recorded sequence number with a RREP packet. The RREP packet is then unicast back to the requested node, using the strict reverse path forwarding mechanism [DM78]. Since multiple routes could be discovered, the source node keeps and activates one with the optimal attributes by unicasting out an activation message, called MACT, along the selected route towards a multicast tree member that originated the RREP message. A multicast group leader is responsible for its members by beaconing Group Hello messages and maintaining the multicast group sequence numbers. Although MAODV works well in large network coverage and in very dynamic ad-hoc networks, its expanding ring search mechanism in the route discovery phase can cause extensive routing overhead. Moreover, its strict reverse path forwarding mechanism suppresses its operationality in unidirectional ad-hoc networks. 4.3 Unidirectional Routing Approach In order to use unidirectional links, in addition to bi-directional links, in routing decision in ad-hoc networks, several approaches have been proposed, which can be mainly categorized into two groups, namely the tunneling mechanism and the existing 56 protocol modification. However, they impose additional communication and storage overhead. The idea of the tunneling approach is to emulate bi-directional links through existing unidirectional links, as explained in [DDE97, DDI+01] for traditional wired networks, and in [NP00] for ad-hoc networks. In other words, it adds a virtual layer between the link layer and the network layer, functioning as a tunnel transparent to the routing module at the network layer. Upon detecting the existence of unidirectional links with the use of ACKs at the link layer, a sink, towards which a unidirectional link is directed, periodically sends out encapsulated link_inform messages destined to a source, i.e. the node on the other end of the unidirectional link. The link_inform message notifies the source about the existence of the unidirectional link towards the sink. When this packet is received by the source, the network layer and the data link layer make a note of the unidirectional property of the link. Hence, both the source and the sink of the unidirectional link become aware of its existence. The tunneling modifications are transparent to the routing module, thereby allowing any routing protocol to be used. By emulating bi-directional links, Nesargi’s approach for ad-hoc networks [NP00] is based on the tunneling mechanism at the link layer proposed for traditional wired networks in [DDI+01]. Benassy-Foch, et. al. [BCG02] suggested applying the tunneling mechanism at the link layer on DVMRP [WPD88]. To support unidirectional links in ad-hoc networks, one short-term solution is to modify existing routing protocols. With minimal routing algorithm modification, DSR 57 [JM96, JM99] can work on unidirectional links by establishing two directional paths (forward and backward) between the source and the destination. Prakask [Pra99] suggested the use of additional data structures and algorithms to DSDV [PB94] and AODV [PR94]. These modifications incur higher communication and storage overhead. 4.4 Swarm Intelligence Concept Swarm Intelligence - the emergent collective intelligence of groups of simple agents as demonstrated in certain insect species - is based on a principle called stigmergy [BDT99]. Stigmergy is a very clever strategy used by nature to communicate through the environment. It allows colonies of social insects to self-organize, to communicate with each other places to find food, to create sophisticated messaging systems and to build complex architectural structures. Self-organization is the main mechanism demonstrating how complex collective behavior may emerge from interactions among individuals that exhibit simple behavior. While interaction is based on primitive instincts with no supervision, the end result is accomplishment of very complex and apparently intelligent forms of social behavior including fulfillment of a number of optimization and other tasks [BDT99]. One of the most important features of social insects is that they can solve complex problems in a very flexible and robust way: flexibility allows adaptation to changing environments while robustness endows the colony with the ability to function 58 even though some individuals may fail to perform their tasks. Finally, social insects have limited cognitive abilities: it is, therefore, simple to design agents, including robotic agents that mimic their behavior at some level of description. The fundamental principle of swarm-based techniques used in computer networks lies with the use of mobile exploration agents. Mobile agents act like ants with pheromone deposited along traversed paths. The use of repeated and concurrent mobile agent process builds the solutions in an incremental way. The past collected information is used to direct future processes for better solutions. So far, existing routing schemes coupled with the swarm-based techniques assume bidirectional environments. AntNet [DD97] has two classes of ants: forward and backward ants. Forward ants regularly depart from a source node to a destination. They randomly select a next hop from the routing table probability entries for the destination, and leave behind a trail of pheromone-equivalent network delay information. Periodically launched from a destination, backward ants analyze unprocessed network delay information for trip times between any two nodes and update the routing table probabilities accordingly as they traverse throughout the network. Data flows are directed through a next hop node with the highest routing table probability. The ant-based network control system [LB03], called ABC, a variation of AntNet, has only one class of mobile agents. Similar to AntNet, ants are regularly sent from a source to a destination. They arbitrarily choose a next hop from the routing table probability entries for the destination. Other than that, ABC ants also update the routing 59 table probabilities as they move across the network, based on the life of the ants at the time of the visit. 4.5 GPS Enabled Approach Another approach is to use GPS. The Highly Mobile Network Routing [TR00] uses nodal GPS capability and a location tracking system. GPS-enabled node must register its location, through subscribe, update, and unsubscribe commands, with one location manager at any time. Having packets to transmit, a source node queries a location manager about a destination endpoint' s location information and sends packets accordingly with either non-predictive or predictive techniques. This approach assumes that a mobile node is highly likely to continue its direction as in the Random Waypoint model [MLTS97]. Spray routing [TR00, TR01] multicasts session traffic within the vicinity of the last known location of a destination endpoint. The spray is adjusted to the mobility level by using two parameters: width indicating the levels of neighbors to where the multicast traffic should be received, and depth which is the number of hops away from the destination the multicast begins. The trajectory mechanism [TR00] uses past trajectory information to predict future location. It estimates an endpoint' s location at a particular time and forwards data traffic directly to that location. Based on the limited history of an endpoint' s movement, 60 a location manager compiles trajectory information (i.e. distance, delay, future location manager affiliation, and departure time), and provides a sequence of locations and times to the source of traffic. Trajectory with Spray routing [TR00] combines the best feature of Spray routing and trajectory routing. Multicast session traffic is sprayed to a predicted location. With no flooding route discovery and a network partition repair, the routing overhead is significantly reduced, but it is increased by the need for location update and synchronization among location managers and nodes. 4.6 Summary A routing protocol attempts to enhance routing performance (i.e. throughput, delay, routing overhead), to efficiently utilize network resources (i.e. memory, processing power, bandwidth, and battery power), to increase network robustness, and to simplify routing algorithm in error-prone asymmetric ad-hoc network environments. However, tradeoffs among these conflicting protocol design objectives are inevitable. Thus, under various network characteristics, a routing protocol can fulfill only a different subset of protocol design objectives. 61 CHAPTER 5 MUNET: MULTICASTING PROTOCOL IN UNIDIRECTIONAL AD-HOC NETWORKS In this chapter, MUNet 1 is an on-demand end-to-end meshed-tree multicast routing protocol designed for unidirectional ad-hoc networks, where two-way nodal communication does not necessarily exist. Unidirectionality in ad-hoc networks has been disregarded by the mainstream research community due to its costly complexity [Pra99] and its failure proneness. As we saw in the last chapter, the study on unidirectionality 2 shows the advantageous impact of the use of unidirectional links in addition to bi-directional links. We take advantage of spatial locality borrowed from the Swarm Intelligence [BDT99] and Collaboration Discovery [PDM99], nodes in the close proximity of the critical path become neighbor nodes, fortifying the path. The protocol compensates for the unidirectional links’ weakness with neighbor 1 N. Vanitchanant, J. Silvester, and V. Vipanunt, “MUNET: Multicasting Protocol in Unidirectional Ad-hoc Networks,” IEEE CCECE 2004, Niagara Falls, Canada, 2(95):975-978, May 2004. 2 N. Vanitchanant and J. Silvester, “Unidirectionality in Ad-hoc Networks: A Simulation Study,” Proc. IEEE MILCOM 2003, Boston, MA, 22(1):1298-1304, October 2003. 62 collaboration. An enhanced version of MUNet 3 further improves the link utilization by applying a heuristic feature into the neighbor selection mechanism. 5.1 Objectives of the Design The objective, first and foremost, is to take advantage of unidirectional links, a key characteristic of ad-hoc networks. Unidirectional links may constitute the majority compared to bi-directional links in ad-hoc networks as we saw in Chapter 4 [VS03]. The second objective is to apply a neighbor collaboration concept into the proposed routing protocol. It is to overcome shortcomings of other multicasting protocols that employ the strict reverse path forwarding mechanism, and to increase routing efficiency in terms bandwidth usage and routing overhead. Thirdly, the design objective is to mitigate the severity of broadcast storms that result from the inapplicability of any hand-shake MAC protocol in unidirectional environments, by the exploitation of spatial locality. The forth objective is to limit unnecessary flooding during the path discovery and link repair phases. With the use of neighbor nodes, scoped flooding is possible. Lastly, the objective is to provide rich network connectivity with the use of mesh topology in combination with tree topology. 3 N. Vanitchanant, J. Silvester, and V. Vipanunt, “Optimization to Multicasting Protocol in Unidirectional Ad-hoc Networks,” IEEE WiMob 2005, Montreal, Canada, 3(44):329-325, August 2005. 63 5.2 Assumptions All nodes are assumed to be in a ready mode when located in the vicinity of a forwarding node. Instead of switching into the sleep mode when idle, they are fully aware of their surroundings (i.e. traffic flow, and node existence). They are willing to participate fully in the MUNet network protocols. They communicate with other nodes in the network without reservation. Each node is assumed not to have GPS capability and equipped with omni- directional antenna. The fact that we want to use unidirectional links justifies the implementation of the unreliable CSMA MAC protocol [BDSZ94, mac99, std99]. Without the ability to detect collisions in wireless medium, Carrier Sense Multiple Access (CSMA) is a probabilistic Media Access Control (MAC) protocol in which a node verifies the absence of other traffic before transmitting on a shared physical medium. 5.3 MUNet Protocol - Methodology The MUNet protocol is composed of three main functions: path establishment, path maintenance, and link recovery. The brief description of the protocol is as follows. Whenever a coordinator, assigned arbitrarily or by its appointed significance [ADK01], needs to send data to a multicast group without 64 prior routing information, it initiates a path establishment process in the source routing style [JM96, JM99, JMB01]. After a least-cost path (i.e. shortest path) to the group is acquired, the coordinator periodically sends out control packets containing complete hop-by-hop routing information. Each node en route – called an active node - is activated and ready for data forwarding. Recurring control packets reactivate active node from time to time. If no such packets are received after some time, the node becomes dormant and no longer performs forwarding. A control packet generated by the coordinator is identified by a wrap around sequence number. Potential link breakage can be detected and prevented by the assistance of collaborative neighbors, prior to any costly, time consuming link recovery. 5.3.1 Path Establishment To establish a multicast tree to an unknown multicast group, an appointed coordinator solicits its group members by broadcasting advertisement control packets across the network and waits for replies. The advertisement control packet format is shown in Table A.1. Each node, receiving non-duplicate advertisement packets, selects one with the best path (e.g. shortest-path), appends its own address to the forward Node List in the packet header and rebroadcasts the packet with random delay settings, unless it is a group member. Such a node also replies back with a reply packet which contains 65 the complete forward Node List (i.e. the forward hop-by-hop routing information.) Upon receiving reply packets, each node handles them the same way it does advertisement packets; it selects one with the optimal path, appends its own address to the backward Node List in the packet header and rebroadcasts the packet with random delay settings, unless it is the coordinator. After the coordinator receives reply packets, it selects one with the optimal path and converts it to a refresh control packet containing the forward and backward hop-by-hop routing information. Since only non- duplicate node IDs appear on the Node List for each direction, in addition to the use of a sequence number, loop formation is not possible. 66 Figure 5.1: Flowchart of path establishment protocol. Designated by refresh packet, each node on the critical path becomes active and ready for the forwarding duty. Unless an active node stays activated by periodic refresh packets, it gradually becomes dormant. Append its address in the received control packet and forward it. Upon receiving a advertisement or reply control message. Construct Forwarding Table based on the best possible path. Is a group member? Perform the reply process; Copy the packet and change the control mode to reply Place its address in the backward Node List Broadcast the reply packet out Append its address in the backward NodeList. Change the control mode to refresh. Rotate its address to the back of forward Node List and periodically broadcast it out. Yes No No Yes Is a coordinator node? In advertisement mode? In reply mode? Do nothing. Yes Yes No 67 If the coordinator does not receive a reply within a time limit, it rebroadcasts an advertisement. If the maximum number of attempts has been reached with no reply back, the session is cancelled due to the inexistence of members. The path establishment protocol is flowcharted in Figure 5.1. The sequence of actions for the simple topology, illustrated in Figure 5.2, is shown in Figure 5.3. Figure 5.2: Simple topology for path establishment. 2 5 C 4 M T1: advertisement T2: advertisement T1: advertisement T2: advertisement T3: advertisement T4: reply T6: reply T5: reply Selected path Node Coordinator Group Member C M Unselected path 68 Figure 5.3: Sequence of actions in path establishment phase based on simple topology in Figure 5.2. After the initial path is established, nodes along the multicast tree in both forward and backward directions complete their Forwarding Tables, so that data packets can be later transmitted. The data packet is formatted as in Table A.2. The Forwarding Table structure is shown in Table A.3. Sequence of Actions in Path Establishment based on Topology in Figure 5.2 1. Group coordinator C broadcasts a control packet with Mode: advertisement Multicast Group ID: <group_id> Group Coordinator: C Originator: C Forwarder: C Forward Node List: [C] Backward Node List: [NULL] Reconstructed Path Flag = OFF 2. Upon receiving a advertisement packet, node 2 modifies and relays the packet with Forwarder: 2 Forward Node List: [C, 2] Node 4 and node 5 perform similarly. 3. Member M selects the best forward path and sends out a control packet with Mode: reply Originator: M Forwarder: M Forward Node List: [C, 2, M] Backward Node List: [M] 4. Upon receiving a reply packet, node 5 modifies and relays the packet with Forwarder: 5 Backward Node List: [M, 5] Node 4 performs similarly. 5. Coordinator C selects the best backward path and prepares refresh control packets. The complete refresh control packet is ready to send out as Mode: refresh Multicast Group ID: <group_id> Group Coordinator: C Originator: C Forwarder: C Nodes in forward Direction: [2, M, C] Nodes in backward Direction: [M, 5, 4, C] 69 5.3.2 Path Maintenance The coordinator maintains an established path by periodically broadcasting refresh packets to its group members. A Refresh control packet contains both forward and backward Node List - round-trip hop-by-hop source routing information. Upon receiving a refresh packet, only the node designated by the refresh packet is activated for forwarding. The designated node is the node whose address is the first on the Node List, depending on the direction of the flow. It first rotates the Node List entries such that its address is moved from being the first to the last on the Node List of respective direction. Then, it caches and forwards the modified refresh packet. Finally, it refreshes the Forwarding Table entry and set the Forwarding Flag on. Having the inherent broadcast nature of ad-hoc node, a non-active node operates in a promiscuous mode, instead of switching into a sleep mode when idle. A node that can hear from both upstream and downstream active nodes is called a neighbor node for that respective direction. By monitoring control flow, a neighbor node can detect and prevent potential link breakage, in the absence of control flow. When an active node, moving out of range or becoming non-operational, is unable to partake in the critical path maintenance, a local link breakage results. A neighbor of the leaving active node, upon sensing the absence of the refresh control flow, suspects a link breakage and activates itself and thus takings over the forwarding duty. To preserve the established path and to prevent a potential link breakage, the 70 neighbor modifies refresh packet by replacing the absent active node address with its own address in its cached refresh packet, then sends the modified refresh packet out. It also updates its Forwarding Table and sets the Forwarding Flag on. The tasks are flowcharted in Figure 5.4 and 5.5. Unless its address is specified in future refresh packets, indicating it is part of the critical path, the activated neighbor is eventually timed out and becomes a regular neighbor node. Path maintenance protocols for normal circumstance (e.g. without node mobility) and under circumstances with node mobility and line-of-sight blockage are explained in Section 5.3.2.1 and Section 5.3.2.2 respectively. 5.3.2.1 Under Normal Circumstance (Without Node Mobility) A group coordinator periodically broadcasts a complete refresh control packet. On receiving a refresh control packet, a node, whose address is the first on the Node List, is considered as an active node, which means it is on forwarding duty, regardless of specified direction. It rotates its address on the Node List, forwards the packet and refreshes the Forwarding Table entry by setting the Forwarding flag to ON. If its address is not first on the Node List, it just refreshes the Forwarding Table entry. A simple scenario with inactive neighbor node is shown in Figure 5.6. Its sequence of actions is described in Figure 5.7. 71 Figure 5.4: Flowchart of path maintenance protocol - part 1. Upon receiving a packet. Is its address next on the Node List? In Forwarding Table: Set Forwarding Flag = ON Set Receptive-Inbound = Forwarder Set Receptive-Outbound = NULL Refresh the entry timer For packets: Rotate its address to the back of the control packet and broadcast it, or Broadcast data packets. In Forwarding Table: Set Forwarding Flag = OFF Set Receptive-Inbound = Forwarder. Refresh the entry timer. Set NextNode = the first address on the Node List Yes No Overhear a packet from NextNode within time limit? In Forwarding Table: set Receptive-Outbound = NextNode. Refresh the entry timer For cached control packet: Replace the next address on the Node List with its address. Set Reconstructed Path Flag ON Broadcast the control packet out In Forwarding Table: Set Forwarding Flag = ON Refresh the entry timer For data packet: Broadcast the cached data packet Yes No No Yes A Is Reconstructed Path Flag ON? Yes No Is data packet? Is Forwarding Flag ON? Yes Do nothing Receptive-Outbound is NULL? Set NextNode = Receptive-Outbound Yes No 72 Figure 5.5: Flowchart of path maintenance protocol - part 2. Is Forwarding Flag in Forwarding Table ON? Forward the packet. Refresh the entry timer. No No Do nothing. Be a coordinator or a member? A Yes Update the refresh control packet Broadcast it. Yes 73 Figure 5.6: Topology before circumstance changes. Figure 5.7: Sequence of actions for path maintenance under normal circumstance in Figure 5.6. Sequence of Actions for Path Maintenance under Normal Circumstances in Figure 5.6 1. From node 1, node 2 receives a control packet with Mode: refresh Multicast Group ID: <group_id> Originator: C Forwarder: 1 Nodes in forward direction: [2, 3, …, 1] Nodes in backward direction: […] Reconstructed Path Flag = OFF 2. Node 2, an active node, rotate its address to the last in the refresh packet before broadcasts it out. The control packet is modified as Forwarder: 2 Nodes in forward direction: [3, …, 1, 2] 3. Node 2 sets its Forwarding Table’s Forwarding Flag ON and refreshes the entry timer. 4. Node 4, when receiving broadcasted control packets from node 1 and node 2, updates and refreshes its Forwarding Table entry as Multicast Group ID: <group_id> Receptive-Inbound: 1 Receptive-Outbound: 2 Forwarding Flag: OFF 3 4 2 1 Node 1: Upstream active node Node 2: Downstream active node Node 3: Downstream active node (out of interests with respect to node 4) Node 4: Inactive neighbor node Note: Assume the forward path. 74 5.3.2.2 Under Circumstances with Node Mobility and Line-of-sight Blockage When a neighbor node fails to hear refresh control packets or data packets from both upstream and downstream active nodes within the time limit, it assumes a change in local circumstance and a possible link breakage. The change in local circumstance might result from node mobility, node failure, or line-of-sight blockage. Each cause requires different actions to preserve an established path. A node that is able to maintain its function and role under a change is considered as being "okay." Six possible cases for circumstance change are case 1) Active node moves away, neighbor node is in the vicinity. case 2) Active node and neighbor node are in the same vicinity, line-of- sight blockage exists. case 3) Neighbor node is out of range. case 4) Multiple neighbor nodes present in the vicinity. case 5) No neighbor node presents in the vicinity. In most cases, a neighbor node takes basic preventive action by sending out a modified refresh control packet with Reconstructed Path Flag ON. Eventually, when a group coordinator or a group member receives the modified refresh control packet, it modifies the Node List in the refresh control packet header accordingly. Duplicate packets are always discarded. Circumstance changes and relevant necessary preventive actions are described next. The starting topology is shown in Figure 5.6. Each circumstance change is 75 explained and depicted in Figure 5.8.a, 5.8.b, 5.8.c, and 5.8.d accordingly. Each of the corresponding sequence of actions is in Figure 5.8.a.1, 5.8.b.1, 5.8.c.1, and 5.8.d.1 respectively. For a combination of circumstance changes, several appropriate actions can be combined. Such a scenario might be when multiple neighbor nodes co-exist with the line-of-sight blockage. case 1) Active node moves away, neighbor node is in the vicinity. In this situation, the active node drifts out of its upstream active node’s range as depicted in Figure 5.8.a. Thus, it cannot relay packets. A neighbor node, however, notices the absence of its active downstream node’s transmission and takes over the forwarding duty until its Forwarding Table entry expires. Before it broadcasts its cached control packet, it replaces the first address with its own address in the Node List and sets the Reconstructed Path Flag ON. It then broadcasts cached data packets, if any. In Forwarding Table, the Forwarding Flag is set to ON and the entry timer is refreshed. The sequence of actions is described in Figure 5.8.a.1. The situation where an active node fails is identical to moving out of range as described here. 76 Figure 5.8.a: Circumstance changes as active node moves away. Figure 5.8.a.1: Sequence of actions for circumstance changes as active node moves away in Figure 5.8.a. Sequence of Actions For Circumstance Changes as Active Node Moves Away in Figure 5.8.a 1. Even though node 4 is not on the forwarding duty, it overhears a transmission from node 1 and from node 2 respectively. 2. As soon as node 4 hears from node 1 but not from node 2, node 4 assumes node 2 moves out of node 1’s range. 3. Node 4 modifies control packet by replacing node 2’s address with its address and setting the Reconstructed Path Flag ON as Forwarder: 4 FORWARD Node List: [4, 3, …, 1] Reconstructed Path Flag = ON In node 4’s Forwarding Table, Multicast Group ID: <group_id> Receptive-Inbound: 1 Receptive-Outbound: NULL Forwarding Flag: ON 4. Node 4 forwards packet until the entry timer expires. 5. Upon receiving control packets with Reconstructed Flag ON, node 3 rebroadcasts unmodified control packets. 3 4 2 1 3 4 2 1 77 case 2) Active node and neighbor node are in the same vicinity, Line-of-sight blockage exists. Even though the active node maintains its function and role, no neighbor node is able to hear from the active node due to line-of-sight blockage, as depicted in Figure 5.8.b. The neighbor node assumes that the active node has moved out of range. Hence, it takes over the forwarding duty until its Forwarding Table entry expires. However, downstream active nodes discard the neighbor node’s modified control packet due to packet duplication. The sequence of actions is described in Figure 5.8.b.1. 78 Figure 5.8.b: Circumstance changes with line-of-sight blockage. Figure 5.8.b.1: Sequence of actions for circumstance with line-of-sight blockage in Figure 5.8.b. Sequence of Actions for Circumstance with Line-of-sight Blockage in Figure 5.8.b 1. Node 2 and node 4 receives the same control packet from node 1 as Mode: refresh forward Node List: [2, 3, …, 1] Reconstructed Path Flag: OFF 2. When node 4 hears from node 1, but not from node 2, it performs the same task as if node 2 moves away (in case a). In control packet: forward Node List: [4, 3, …, 1] Reconstructed Path Flag: ON In Forwarding Table: Forwarding Flag: ON 3. Node 3 discards node 4’s packet because it already receives packet from designated downstream active node which is node 3. 4. Node 4 forwards the packets until the Forwarding Table entry expires. 3 4 2 1 3 4 2 1 79 case 3) Neighbor node is out of range (due to line-of-sight blockage from upstream node or its mobility.) When a neighbor node hears nothing from an upstream active node due to either line-of-sight blockage from upstream active node, illustrated in Figure 5.8.c.a, or moving away, depicted in Figure 5.8.c.b, it takes no action. When its Forwarding Table entry expires, the neighbor node becomes a regular node, not participating in multicast routing. The sequence of actions is in Figure 5.8.c.1. Figure 5.8.c: Circumstance changes as neighbor node is out of range due to 1) line-of-sight blockage from upstream active node and 2) moving way. Figure 5.8.c.1: Sequence of action for circumstance changes as neighbor node is out of range in Figure 5.8.c. Sequence of Action for Circumstance Changes as Inactive Node is not Okay in Figure 5.8.c 1. Node 1, 2 and 3 operate as they would under normal circumstance. 2. Node 4 does nothing. When its Forwarding Table Entry expires, it no longer involves in multicast routing. 3 4 2 1 3 4 2 1 a) Line-of-sight blockage from b) Moving out of range 80 case 4) Multiple neighbor nodes present in the vicinity. It is possible that multiple neighbor nodes co-exist, as illustrated in Figure 5.8.d. When noticing the absence of the active node, a selected set of neighbor nodes takes over the forwarding duty as mentioned earlier in a). The criteria in selecting the neighbor nodes can be arbitrary (e.g. random, proximity-bounded, or advantage- bounded.) The number of selected neighbor nodes varies from one to all neighbor nodes, depending on the desired level of network dependability. That is because the downstream active node may be out of some neighbor nodes’ range. To avoid collision, each node has a different delay setting before it reacts. The sequence of actions is described in Figure 5.8.d.1. Figure 5.8.d: Circumstance with multiple neighbor nodes exist. Figure 5.8.d.1: Sequence of action for circumstance with multiple neighbor nodes exist in Figure 5.8.d. 3 4 2 1 4 2 1 5 7 5 7 3 6 6 Sequence of Action for Circumstance with Multiple Neighbor Nodes Exist in Figure 5.8.d 1. When noticing the absence of node 2, selected neighbor nodes, that is node 4 and node 5, not node 6, take over the forwarding duty. (the same task as described in Box 2.b.) 2. Node 3 forwards the best packet selected from node 4’s and node 5’s. 3. Note that node 6’s transmission range does not cover node 3. 81 case 5) No neighbor node presents in the vicinity. When no neighbor node exists in the vicinity, an established path can no longer be maintained. The link recovery mechanism, fully described in Section 5.3.3, is required to repair the broken link. 5.3.3 Link Recovery The link recovery process starts when the coordinator misses an expected refresh control flow, or receives a member’s reply packet (as an alert for a missing refresh control flow). The process is similar to the path establishment process, but it puts a restriction on participants. The coordinator begins the recovery process by broadcasting an advertisement with ReconstructedPathFlag set. The members respond with a reply with ReconstructedPathFlag set. Intermediate nodes handle packets as they normally do. To reduce the flooding coverage area, only active nodes and neighbor nodes participate when ReconstructedPathFlag is set. If the attempt fails, the process repeats without the ReconstructedPathFlag, which effectively covers the entire network as in the initial path establishment process. The link recovery process is flowcharted in Figure. 5.9. 82 Figure 5.9: Broken link recovery mechanism. Not receive neither refresh nor data packet within time limit Receive advertisement packet? Receive reply packet? Be member with expired entry? Have sent out several advertisement or reply packets? Perform the reply process. Do the best-path forwarding. Respond as it would do to a regular reply Assume partition occurs. Is it in forward direction ? Change mode from refresh to repair Send it to group coordinator Wait for ADVERTISEMENT packet Is a group coordinator? Start the path establishment process. Forward packet, if any. No No No No Yes No No Yes Yes Yes Yes Yes 83 5.4 MUNet Protocol - Enhancement In unidirectional ad-hoc networks where link-level feedback does not exist due to physical limitation and strategic communication policy, signal interference and contention [KS78] are inevitable. To be selective on signal responsiveness, a heuristic feature 4 - an extension to MUNet - determines a node’s participation probability, based upon the intensity of local activity. Whenever a node’s participation probability has reached a preset reference point, it responds accordingly. Nonetheless, overall routing performance could be adversely affected, if all the nodes that meet the conditions unnecessarily respond to one same incident, as illustrated in Figure. 5.10.a). The heuristic feature allows only a subset of the nodes to respond as depicted in Fig. 5.10.b). 4 N. Vanitchanant, J. Silvester, and V. Vipanunt, “Optimization to Multicasting Protocol in Unidirectional Ad-hoc Networks,” IEEE WiMob 2005, Montreal, Canada, 3(44):329-325, August 2005. 84 By definition, a node’s participation probability (PP) is a normalized level of node participation, which can be written as T c L Sr PP * * (1) where Sr is signal reachability, as defined in (2), L is node participation level, as defined in (3), c is an arbitrary constant (usually c = 1), T is time in seconds. The signal reachability (Sr) is a normalized signal coverage density, which can be written as A N Tr Sr * (2) where Tr is a node’s average transmission range in meters, 3 2 4 1 3 4 2 1 5 6 5 6 Figure 5.10: Heuristic feature in path maintenance. a) All neighbor nodes respond. b) Only neighbor node 4 responds. a) b) 85 N is a number of nodes in the network, A is an area of network in meters 2 . The node participation level (L) is in inverse proportion to the traffic density of nearby active nodes, which can be written as R T L (3) where R is a number of refresh packets directionally overheard in T seconds. Traffic characteristics and density detected at a node signify the node status and its level of participation. When a non-active node can promiscuously listen to both its upstream and downstream active nodes, it is defined as neighbor node for respective direction. High traffic density indicates many active nodes in the vicinity, implying each neighbor node has less probability to respond. Specifically, the heuristic feature, applicable to both active nodes and neighbor nodes, is embedded in the link breakage prevention phase and the data forwarding phase. When a neighbor node detects a control flow missing and assumes a link breakage, it determines its probability in order to become active and to take over the forwarding duty. Activated neighbor nodes, temporally located in the same vicinity, might cause signal interference towards one another when relaying signal. The transmission redundancy can be reduced when they are probabilistically activated. 86 5.5 Data Forwarding Mechanism After a source node has a path to its multicast group members, it continuously injects data packets into the networks. An active node –a node with forwarding flag on- broadcasts non-duplicate data packets with random time delay, upon receiving them. Random time delay setting is necessary in unidirectional environments where the lack of feedback at the MAC layer can cause systematic signal contention. 5.6 Summary While sharing several characteristics with other routing protocols, MUNet regards link directionality as a feature, not an exception, and incorporates it as a primary design objective. The unavailability of link-level feedback weakens the network connectivity, however, is compensated with the network-level collaboration of neighbor nodes. After forward and backward paths are discovered in a source routing style, they are maintained by periodic refresh control packets from a coordinator, and prevented from link breakage by the assistance of the promiscuous neighbor nodes. Only a subset of neighbor nodes heuristically participates in the path maintenance phase because of their possible adverse signal interference. 87 CHAPTER 6 COMPARATIVE EVALUATION To demonstrate the benefits of MUNet over other multicasting protocols, we conduct packet-level simulation to explore the implications of some of our protocol features. This chapter describes the evaluation methodology and metrics, explores the impact of unidirectionality, and then investigates the impact of protocol enhancement. 6.1 Protocol Evaluation Methodology With ten runs on different random seeds, a set of simulation experiments was carried out to measure the effectiveness and the efficiency of the proposed enhanced multicast routing protocol. The protocol is compared with the original MUNet protocol and ODMRP, excluding MAODV. MAODV barely functions with the CSMA MAC protocol. Using Glomosim-2.02 [ZBG98], random networks with 50 uniformly distributed nodes were generated over a terrain of 1300x300 m 2 . An area in narrow rectangular shape is selected to magnify signal interference and to enforce higher hop counts. Destined to five destinations, four CBR flows are generated at 512 bytes per 88 second per flow for 300 seconds. Each node has a transmission range of 250 meters over a 2M bps wireless channel, using the two-ray path loss model [pro]. With 0, 45, 90, 180 and 360 pause times, a node moves with maximal velocity of 10 m/s guided by the Random Waypoint model [ML97]. The use of the omni-directional antenna and the unreliable CSMA MAC protocol [BDSZ94, mac99, std99] are assumed. To create a unidirectional environment, 50% of nodes are selected randomly and their transmission powers are reduced by half. Each experiment simulates the network for 360 seconds. 6.2 Evaluation Metrics The performance of MUNet as a function of pause time for two types of networks is measured with the following evaluation metrics. First, Data Delivery Ratio is the ratio of the number of received data packets to the number of data packets expected to be received. This metric measures the effectiveness of a protocol. Second, Normalized Routing Load is the number of total packets transmitted per data packet received. This metric gauges the efficiency of a protocol concerning bandwidth usage. Last, Overhead Ratio is the ratio of the number of control packets to the number of total packets transmitted. It measures the efficiency of a protocol with respect to routing overhead. 89 6.3 Impact of Unidirectionality To study the impact of unidirectionality on protocols, we run the simulation in bi-directional and unidirectional networks. The simulation shows how conventional multicasting protocols perform in directional environments, and how MUNet performs compared to other protocols. Only ODMRP, unlike MAODV along with other conventional multicasting protocols, is functional in unidirectional networks. MAODV, among others, is not operational because of its strict reverse path forwarding mechanism [DM78]. Our experiment on unidirectionality enforces the unavailability of guaranteed reverse path. For MUNet and ODMRP, protocol performance in unidirectional networks exhibits similarity to (less than 10% different) that in bi-directional networks. While both protocols yield a comparable data delivery ratio, at high node mobility MUNet successfully delivers data at 5 – 15% less than ODMRP does (Figure 6.1). One possible explanation is the increasing signal interference from the on-going path maintenance, specifically from the neighbor-turned-active-node process. MUNet is more efficient than ODMRP as it consumes 25 – 50% less bandwidth (Figure 6.2) and generates 25% less overhead (Figure 6.3) than ODMRP does. This is intuitive because MUNet’s control traffic is centralized around a critical path while ODMRP’s is 90 propagated throughout the network. When the network size increases, the difference is widened. The results indicate MUNet has exceedingly met its design objectives. By this we mean MUNet achieves a comparable level of routing effectiveness with much less routing overhead. 6.4 Impact of Protocol Enhancement To evaluate the impact of protocol enhancement, we simulated the conventional MUNet, the enhanced MUNet, and ODMRP in unidirectional network environments. The simulation shows how protocol modification enhances the routing performance. Only ODMRP is chosen out of traditional multicasting protocols because of its operability in unidirectional environments. Our experiment on protocol enhancement focuses on neighbor nodes’ signal contention in the path maintenance phase. With a threshold of neighbor participation probability at 0.01, the enhanced MUNet protocol exhibits an improvement over the conventional MUNet and ODMRP. All protocols yield a comparable (less than 5% different) data delivery ratio (Figure 6.4). The enhanced MUNet is more efficient than the conventional one and ODMRP as it consumes 20 – 60% less bandwidth per data received (Figure 6.5). This is instinctive because the enhanced MUNet’s control traffic is reduced by the less number of participating neighbors than that of the conventional 91 MUNet’s, and is centralized around the critical path, while ODMRP’s is propagated throughout the network. The effect is more evident in very large networks. With the same rationale, however, the enhanced MUNet incurs 5% higher overhead ratio than the conventional MUNet does and 15% lower overhead ratio than ODMRP does (Figure 6.6). The results show that protocol enhancement on MUNet has accomplished its goals. The enhanced MUNet reduces the adverse effect of potentially excessive neighbor support (i.e. signal interference), and it does sustain robust network connectivity with much less routing overhead. 6.5 Summary The protocol performance has been evaluated on the impact of unidirectionality and the impact of protocol enhancement, using simulation. The assumed unidirectionality of guaranteed reverse path has excluded most traditional multicasting protocols, except ODMRP, for the simulation. In both unidirectional and bi-directional environments, MUNet and ODMRP yield a comparable data delivery ratio. However, MUNet consumes much less bandwidth and produce less routing overhead than ODMRP does. With the protocol extension, only a subset of participating neighbor nodes is heuristically selected to respond. The enhanced MUNet performs in line with the conventional MUNet and ODMRP in terms of data delivery ratio, however, with much less consumed bandwidth. 92 This is intuitive because MUNet’s control traffic is centralized around a critical path while ODMRP’s is propagated throughout the network. 0 0.2 0.4 0.6 0.8 1 0 45 90 180 360 Pause Time (sec) Data Delivery Ratio MUNET - Bi-directional MUNET - Unidirectional ODMRP - Bi-directional ODMRP - Unidirectional Figure 6.1: Impact of unidirectionality - data delivery ratio. Figure 6.2: Impact of unidirectionality - normalized routing load. 0 2 4 6 8 10 12 0 45 90 180 360 Pause Time (sec) Normalized Routing Load 93 Figure 6.3: Impact of unidirectionality - overhead ratio. Figure 6.4: Impact of enhancement - data delivery ratio. 0 0.2 0.4 0.6 0.8 1 0 45 90 180 360 Pause Time (sec) Conventional MUNET Enhanced MUNET ODMRP Data Delivery Ratio 0 0.1 0.2 0.3 0.4 0.5 0.6 0 45 90 180 360 Pause Time (sec) Overhead Ratio 94 Figure 6.5: Impact of enhancement - normalized routing load. Figure 6.6: Impact of enhancement - overhead ratio. 0 0.1 0.2 0.3 0.4 0.5 0 45 90 180 360 Pause Time (sec) Overhead Ratio 0 2 4 6 8 10 12 0 45 90 180 360 Pause Time (sec) Normalized Routing Load 95 CHAPTER 7 CONCLUSIONS The ground level of the next generation Internet (NGI) framework structure serves as the network model for this dissertation. Data dissemination across isolated groups of randomly scattered ad-hoc nodes requires a routing protocol focusing on link unidirectionality. Research efforts on routing protocol in ad-hoc networks have been broadly categorized into the spatial locality exploitation concept, the topology-oriented approach, the unidirectional routing approach, the swarm intelligence concept, and the GPS enabled approach. 7.1 Summary of Contributions In this dissertation, we study network asymmetry characteristics and then design and evaluate MUNet, a multicast routing protocol in ad-hoc networks. With the concept of spatial locality, MUNet exploits the collaboration of promiscuous neighbor nodes, to increase network resilience. MUNet compensates the weakness imposed by unidirectional environments with the critical path fortification by neighbor nodes. Potential link breakage can be prevented collaboratively by reinforcing neighbor nodes, prior to any costly time-consuming link recovery. 96 In unidirectional ad-hoc networks where essential feedback does not exist at the MAC layer, neighbor collaboration, while increasing network resilience, can adversely affect the overall routing performance. In order to improve routing efficiency and resource utilization, the enhancement applies a heuristic feature into the original MUNet protocol. Instantaneously determined by local control traffic density, a node probabilistically sets a level of its responsiveness to participate in the routing protocol maintenance. Since most multicast routing protocols, except ODMRP, are nonfunctional in unidirectional ad-hoc network environments, MUNet was then comparatively evaluated along with ODMRP. With the evident omnipresence of unidirectional links in ad-hoc networks, we implemented MUNet in the GlomoSim simulator. This packet-level simulation was also used for exploring the impact of link unidirectionality and the impact of protocol enhancement. Our evaluation indicated that MUNet achieved significant routing effectiveness and outperformed traditional multicast routing scheme, like ODMRP, even with the non-enhanced path reinforcement. Specifically, our evaluation metrics were data delivery ratio, normalized routing load, and overhead ratio. These metrics indicated the effectiveness and the efficiency of routing protocols. While MUNet and ODMRP yielded a comparable throughput which was the data delivery ratio, MUNet produced much less routing overhead than ODMRP did. The improvement was much more evident with the enhanced MUNet. 97 7.2 Future Direction The routing scheme presented in this dissertation is only an attempt for the ground level of hybrid, multi-tiered networks. Bare necessity is the assumption of the protocol design. Given satellites or aerial nodes in NGI, all nodes in the large cell size (spot-beam) can potentially be coordinators, and be equipped with GPS receivers. The design has yet to take other available structural components into consideration. We perceive the future work of the dissertation on two topics. One topic is steered towards the unidirectional network characterization. The study is to be extended over large-scale networks and over node mobility and other link characteristics such as bandwidth and error rate. The other topic moves towards the routing protocol enhancement. The protocol could cover data and control message dissemination from aerial node such as satellite. Ad-hoc nodes could be GPS-enabled. The study could be extended over a range of network sizes and membership sizes. 98 BIBLIOGRAPHY [ADK01] M. Ahmed, S. Dao, and R. Katz, “Positioning Range Extension Gateways in Mobile Ad Hoc Wireless Networks to Improve Connectivity and Throughput,” Proc. IEEE MILCOM 2001, Vienna, VA, 1:331-315, Oct. 2001. [AVD02] M. Ahmed, N. Vanitchanant, and S. Dao, “A Collaborative Multicast Routing (CMR) Protocol for Hybrid Wireless Networks,” Technical Report supported by Defense Advanced Research Projects Agency (DARPA) Contract No. N00014-99-C-0322 at HRL Laboratories, LLC., 2002. [BEF+00] L.Breslau, D. Estrin, K. Fall, S. Floyd, J. Heidemann, A. Helmy, P. Huang, S. McCanne, K. Varadhan, Y. Xu, and H. Yu. Advances in network simulation. IEEE Computer, 33(5):59–67, May 2000. [BCG02] C. Benassy-Foch, P. Charron, and Y. Guinamand, “Configuration of DVMRP over a UniDirectional Link,” http://www.ietf.org/internet- drafts/draft-ietf-udlr-dvmrp-conf-02.txt, February 2002 (work in progress). [BDSZ94] V. Bharghavan, A. Demers, S. Shenker, and L. Zhang., “MACAW: A Media Access Protocol for Wireless LANs,” ACM SigComm ‘94, pp. 212-225, September 1994. [BDT99] E. Bonabeau, M. Dorigo, and G. Theraulaz, Swarm Intelligence: from Natural to Artificial Systems. Oxford University Press, ISBN 0-195- 13158-4, 1999. [Bel57] R. E. Bellman, “Dynamic Programming,” Princeton University Press, N.J., ISBN 0-486-42809-5, 1957. [CFD01] L. Costa, S. Fdida and O. Duarte, “Hop By Hop Multicast Routing Protocol,” SIGCOMM 2001, 31(4):249–259, San Diego, CA, 2001. 99 [CG98] C.-C. Chiang and M. Gerla, “On-Demand Multicast in Mobile Wireless Networks,” In Proceedings of IEEE ICNP 98, Austin, Texas, pp. 262- 270, October 1998. [CGZ98] C.-C. Chiang, M. Gerla, and L. Zhang, “Forwarding Group Multicast Protocol (FGMP) for Multihop, Mobile Wireless Networks,” ACM/Baltzer Cluster Computing, 1(2):187-196, 1998. [CM99] S. Corson and J. Macker, “RFC2501 Mobile Ad hoc Networking (MANET): Routing Protocol Performance Issues and Evaluation Considerations,” Internet Engineering Task Force, January 1999. [DC90] S. Deering and D. Cheriton, “Multicast Routing in Datagram Internetworks and Extended LANs,” ACM Transactions on Computer Systems, 8(2):85-110, May 1990. [DD97] G. Di Caro and M. Dorigo, "AntNet: a mobile agents approach to adaptive routing", Tech. Rep. IRIDIA/97-12, Universit Libre de Bruxelles, Belgium, 1997. [DDE97] W. Dabbous, E. Duros, and T. Ernst, “Dynamic Routing in Networks with Unidirectional Links,” Workshop on Satellite-Based Information Systems, Budapest, pp. 35-47, September 1997. [DDI+01] E. Duros, W. Dabbous, H. Izumiyama, N. Fujii, and Y. Zhang, “RFC 3077: A Link-Layer Tunneling Mechanism for Unidirectional Links,” Internet Engineering Task Force, March 2001. [DEF+96] S. Deering, D. Estrin, D. Farinacci, V. Jacobson, C. Liu, and L. Wei, “The PIM Architecture for Wide-Area Mulitcast Routing,” IEEE/ACM Transactions on Networking, 4(2) April 1996. [DM78] Y. K. Dalal and R. M. Metcalfe, “Reverse Path Forwarding of Broadcast Packets,” ACM Communications, 21(12):1040-1048, December 1978. [GM99] J.J. Garcia-Luna-Aceves and E.L. Madruga, “The Core-Assisted Mesh Protocol,” IEEE Journal on Selected Areas in Communications, 17(8):1380-1394, August 1999. [JM96] D. B. Johnson and D. A. Maltz, “Dynamic Source Routing in Ad Hoc Wireless Networks,” Mobile Computing, edited by Tomasz Imielinski and Hank Korth, chapter 5, pp. 153-181, Kluwer Academic Publishers, 1996. 100 [JM99] D. B. Johnson and D. A. Maltz, “Dynamic Source Routing in Ad Hoc Networks,” October 1999 IETF draft, http://www.ietf.org/internet- drafts/draft-ietf-manet-dsr-03.txt. [JMB01] D. B. Johnson, D. A. Maltz, and J. Broch, ”DSR: The Dynamic Source Routing Protocol for Multi-Hop Wireless Ad Hoc Networks. in Ad Hoc Networking,” edited by Charles E. Perkins, Chapter 5, pp. 139-172, Addison-Wesley, 2001. [JMQ00] P. Jacquet, P. Muhlethaler, and A. Qayyum, “An Optimized Link State Routing Protocol (OLSRP) for Ad Hoc Networks,” August 2000 IETF draft, http://www.ietf.org/internet-drafts/draft-ietf-manet-olsr-01.txt, August 2000 (work in progress). [KS78] L. Kleinrock and J.A. Silvester, “Optimum Transmission Radii in Packet Radio Networks or Why Six is a Magic Number” National Telecommunications Conference, Birmingham, AL, 4(3):1-5, December 1978. [KV99] Y.-B. Ko and N.H. Vaidya, “Geocasting in Mobile Ad Hoc Networks: Location-Based Multicast Algorithms,” In Proceedings of IEEE WMCSA' 99, New Orleans, LA, Feb. 1999. [LB03] D. Legge and P. Baxendale, “An Agent-Managed Ant-Based Network Control System”, the third symposium on Adaptive Agents and Multi- agent Systems (AAMAS-3), Aberystwyth, UK, 2003. [LK00] S. Lee and C. Kim, “Neighbor supporting ad hoc multicast routing protocol,” IEEE MobiHoc 2000, Boston, MA, pp. 37 – 44, August 2000. [LSG99] S.-J. Lee, W. Su, and M. Gerla, “On-Demand Multicast Routing Protocol,” In Proceedings of IEEE WCNC ‘99, New Orleans, LA, pp. 1298-1304, September 1999. [LSG02] S.-J. Lee, W. Su, and M. Gerla, “On-demand multicast routing protocol in multihop wireless mobile networks Mobile Networks and Applications, 7(6):441-453 December 2002. [mac99] IEEE Standards Association, “Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications: Higher-speed Physical Layer Extension in the 2.4 GHz Band,” http://standards.ieee.org/reading/ieee/std/lanman/802.11b-1999.pdf. 101 [Mal94] G. Malkin, “RFC 1723 RIP Version 2-Carrying Additional Information,” Internet Engineering Task Force, November 1994. [MLTS97] J.G. Markoulidakis, G.L. Lyberopoulos, D.F. Tsirkas, and E.D. Sykas, “Mobility modeling in third-generation mobile telecommunication systems,” IEEE Personal Comm., pp. 41-56, August 1997. [Moy97] J. Moy, “RFC 2178 OSPF Version 2,” Internet Engineering Task Force, July 1997. [NP00] S. Nesargi and R. Prakash, “A Tunneling Approach to Routing with Unidirectional Links in Mobile Ad-Hoc Networks,” Proceedings of the IEEE International Conference on Computer Communications and Networks (ICCCN), Las Vegas, pp. 12-23, October 16-18, 2000 [pac] HRL Research Staff, PackHunter, http://www.packhunter.com or http://www.hrl.com/TECHLABS/isl/FeaturedResearch/PackHunter/pack hunter.htm. [PB94] C. E. Perkins and P. Bhagwat, “Highly Dynamic Destination-Sequenced Distance-Vector Routing (DSDV) for Mobile Computers,” SIGCOMM Symposium on communications Architectures and Protocols, London UK, pp. 212-225, September 1994. [PDM99] D. Payton, M. Daily, and K. Martin, “Dynamic Collaborator Discovery in Information Intensive Environments,” ACM Computing Surveys (CSUR), 31(2er):8, June 1999. [Per00] C. E. Perkins, “Ad Hoc Networking,” Addison-Wesley, ISBN 0-201- 30976-9, December 2000. [PR99] C. E. Perkins and E. M. Royer, “Ad Hoc On-Demand Distance Vector Routing,” In Proceedings of WMCSA’99, New Orleans, LA, pp. 90- 100, February 1999. [Pra99] R. Prakash, “Unidirectional Links Prove Costly in Wireless Ad Hoc Networks,” In Proceedings of DIMACS Workshop on Mobile Networks and Computers, pp. 15-22, 1999. [pro] The CMU Monarch Project, The CMUMonarch Project’s Wireless and Mobility Extensions to NS. 102 [RP99] E. M. Royer and C. E. Perkins, “Multicast Operation of the Ad hoc On- Demand Distance Vector Routing Protocol,” ACM on Mobile Computing and Networking, pp. 207-218, 1999. [RP00] E. M. Royer and C. E. Perkins, “Multicast Ad hoc On-Demand Distance Vector (MAODV) Routing,” July 2000 IETF draft, http://www.ietf.org/internet-drafts/draft-ietf-manet-maodv-00.txt. [SEZ00] I. Stoica, T. S. Eugene Ng, and H. Zhang, “REUNITE: A Recursive Unicast Approach to Multicast,” IEEE INFOCOM pp. 1644-1653, March 2000. [SG01] M. Spohn and J. J. Garcia-Luna-Aceves, “Neighborhood aware source routing,” In Proceedings of ACM MobiHoc 2001, Long Beach, CA, pp. 11-21, October 2001. [SGC99] S.-J. Lee, M. Gerla and C.-C. Chiang, “On-Demand Multicast Routing Protocol,” In Proceedings of IEEE WCNC 99, New Orleans, LA, September 1999. [std99] IEEE Standards Association, “IEEE Std 802.11, 1999 Edition (ISO/IEC 8802-11: 1999) IEEE Standard for Information Technology - Telecommunications and Information Exchange between Systems - Local and Metropolitan Area Network - Specific Requirements - Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications,” http://standards.ieee.org/reading/ieee/std/lanman/802.11-1999.pdf. [TK75] F. A. Tobagi and L. Klienrock,” Packet Switching in Radio Channels: Part-II – The Hidden Terminal Problem in Carrier Sense Multiple- Access Models and the Busy-Tone Solution,” IEEE Trans. Comm., COM-23(12):1417-33, December 1975. [TR00] F. Tchakountio and R. Ramanathan , "Routing to Highly Mobile Endpoints: A theoretical and Experimental Study of Limits and Solutions" , BBN Technical Report no. 1280 , December 2000. [TR01] F. Tchakountio and R. Ramanathan , "Tracking Highly Mobile Endpoints", In Proceedings of ACM Workshop on Wireless Mobile Multimedia (WoWMoM01), Rome, Italy, July 2001. [VS03] N. Vanitchanant and J. Silvester, “Unidirectionality in Ad-hoc Networks: A Simulation Study,” Proc. IEEE MILCOM 2003, Boston, MA, 22(1):1298-1304, October 2003. 103 [VSV04] N. Vanitchanant, J. Silvester, and V. Vipanunt, “MUNET: Multicasting Protocol in Unidirectional Ad-hoc Networks,” IEEE CCECE 2004, Niagara Falls, Canada, 2(95):975-978, May 2004. [VSV05] N. Vanitchanant, J. Silvester, and V. Vipanunt, “Optimization to Multicasting Protocol in Unidirectional Ad-hoc Networks,” IEEE WiMob 2005, Montreal, Canada, 3(44):329-325, August 2005. [Wes00] D. B. West, “Introduction to Graph Theory,” Prentice Hall, ISBN 0- 130-14400-2, August, 2000. [WPD88] D. Waitzman, C. Partridge, S. Deering “RFC 1075: Distance Vector Multicast Routing Protocol,” Internet Engineering Task Force, November 1988. [ZBG98] X. Zeng, R. Bagrodia, and M. Gerla, "GloMoSim: a Library for Parallel Simulation of Large-scale Wireless Networks", In Proceedings of the 12th Workshop on Parallel and Distributed Simulations (PADS ' 98), pp. 154-161, Banff, Alberta, Canada, May 1998. 104 APPENDIXES Appendix A Data Structures The data structure section includes control packet format, data packet format and Forwarding Table structure. The control packet format is used in advertisement, reply, and refresh process. The data packet format is for framing data into an appropriate format suitable for transmission. The Forwarding Table keeps node’s status in order to proper perform path setup, path maintenance, and data forwarding. A.1 Control Packet Format As shown in Table A.1, control packet has three modes: - advertisement, reply, and refresh. The advertisement and reply modes are for path establishment phase. The refresh mode is for path maintenance phase. Globally unique class-D multicast address mechanism is used in Multicast Group ID. Sequence Number distinguishes control packets of a multicast group. Group Coordinator denotes the root node of a current network domain. Originator identifies end nodes (either group coordinator or group member.) Forwarder indicates the node transmitting the packet. The forward Node List progressively records all nodes in forward direction--along the path from a group 105 coordinator to a group member. Performance Metrics in forward Direction records performance metrics of its associated node. Those of backward direction do the same thing as these of the forward direction do, except nodes are of reverse path. Reconstructed Path Flag is to signify that the established path is altered. Field Value Notes Mode advertisement, reply or refresh Control packet mode Multicast Group ID group_id Sequence Number seq_no Group Coordinator cooridinator_id Root node Originator member_id End node: coordinator or member Forwarder node_id Transmitting node Forward Node List [cooridinator_id node_id_1, … , member_id] List of Nodes in forward Direction Backward Node List [member_id, node_id_1, … , coordinator _id] List of Nodes in backward Direction Reconstructed Path Flag ON or OFF (default) Indicate that the path is altered. Table A.1: Control Packet Format. 106 A.2 Data Packet Format Data packet starts being sent out after the first refresh control packet has been sent out or the Forwarding Flag in the Forwarding Table is set to ON. In order to conform to the network, data packet is formatted as shown in Table A.2. Multicast Group ID identifies a multicast group to which the data belongs. Sequence Number differentiates data packets. Group Coordinator distinguishes partitions. Data Field contains multicast data. Field Value Multicast Group ID group_id Sequence Number seq_no Group Coordinator coordinator_id Data Field data octet Table A.2: Data Packet Format. A.3 Forwarding Table Structure In Table A.3, Forwarding Table presents the status of a node, which further signifies its functions and role. A node can be active, inactive, or not involved in multicast traffic at all. Multicast Group ID identifies a multicast group to which the 107 Forwarding Table belongs. The Group Coordinator denotes the root node of a multicast tree. The Direction field indicates the traffic direction: forward or backward. The forward direction is from group coordinator to member, and vice versa for the backward direction. The Forwarding Flag denotes the duty in forwarding control packets and rebroadcasting data packets. Receptive-Inbound and Receptive-Outbound fields refer to nodes whose transmission range covers the node for incoming traffic flow, and outgoing traffic flow respectively. Their timestamp fields record the most recent in-range time for inbound and outbound nodes respectively. Field Value Notes Multicast Group ID group_id Sequence Number seq_no Sequence number Group Coordinator coordinator_id Root node of multicast tree Forwarding Flag ON or OFF ON if the node ID is in a control packet Receptive-Inbound Upstream node ID Sender of inbound traffic Inbound Timestamp timestamp Receptive-Outbound Downstream node ID Sender of outbound traffic Outbound Timestamp timestamp Table A.3: Forwarding Table Data Structure.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

The importance of using domain knowledge in solving information distillation problems

PDF

Protocol evaluation in the context of dynamic topologies

PDF

Efficient media access and routing in wireless and delay tolerant networks

PDF

Internet security and quality-of-service provision via machine-learning theory

PDF

Orthogonal architectures for parallel image processing

PDF

Semantic heterogeneity resolution in federated databases by meta-data implantation and stepwise evolution

PDF

Probabilistic analysis of power dissipation in VLSI systems

PDF

An argumentation-based approach to negotiation in collaborative engineering design

PDF

The design and synthesis of concurrent asynchronous systems.

PDF

Iterative data detection: complexity reduction and applications

PDF

Three antibody-based immunotherapeutic modalities for malignancies

PDF

Colorectal cancer: genomic variations in insulin-like growth factor-1 and -2

PDF

Cathodoluminescence studies of the influence of strain relaxation on the optical properties of InGaAs/GaAs quantum heterostructures

PDF

Elevated inflammation in late life: predictors and outcomes

PDF

Spatial distribution of neuroendocrine motoneuron pools in the hypothalmic paraventricular nucleus

PDF

The effect of helmet liner density upon acceleration and local contact forces during bicycle helmet impacts

PDF

Adaptive routing services in ad-hoc and sensor networks

PDF

Initiation mechanism of estrogen neuroprotection pathway

PDF

Adaptive energy conservation protocols for wireless ad hoc routing

PDF

The problem of safety of steel structures subjected to seismic loading

Asset Metadata

Creator Vanitchanant, Noparut (author)

Core Title MUNet: multicasting protocol in unidirectional ad-hoc networks

School Graduate School

Degree Doctor of Philosophy

Degree Program Computer Engineering

Degree Conferral Date 2006-05

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag computer science,OAI-PMH Harvest

Language English

Contributor Digitized by ProQuest (provenance)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c17-113787

Unique identifier UC11350064

Identifier 3237705.pdf (filename),usctheses-c17-113787 (legacy record id)

Legacy Identifier 3237705.pdf

Dmrecord 113787

Document Type Dissertation

Rights Vanitchanant, Noparut

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA