Ph.D. Dissertation Proposal: A Mobility-Based Framework for Adaptive Dynamic Cluster-Based Hybrid Routing in Wireless Ad-Hoc Networks

Transcription

1 Ph.D. Dissertation Proposal: A Mobility-Based Framework for Adaptive Dynamic Cluster-Based Hybrid Routing in Wireless Ad-Hoc Networks by A. Bruce McDonald B.S., Northwestern University, 1986 M.S., University of Pittsburgh, 1994 Submitted to the Graduate Faculty of Information Sciences in partial fulfillment of the requirements for the degree of Ph.D. University of Pittsburgh 1999

2 Ph.D. Dissertation Proposal: A Mobility-Based Framework for Adaptive Dynamic Cluster-Based Hybrid Routing in Wireless Ad-Hoc Networks A. Bruce McDonald, Candidate for Ph.D. University of Pittsburgh, 1999 PROPOSAL ABSTRACT This paper proposes a novel mobility-based framework for routing in wireless ad-hoc networks an emerging class of network architecture that is characterized by its highly dynamic topology and its limited resources. The framework supports adaptive clustering to dynamically organize the nodes of an ad-hoc network into clusters in which the probability of path failure due to node movement can be bounded over time. The cluster strategy provides the basis for the development of an adaptive hybrid routing strategy which dynamically balances the tradeoff between and routing responsiveness and efficiency according to localized mobility characteristics. The main objective of the proposed approach is to improve routing scalability, and support a wide range of wireless node mobility. According to the proposed scheme, routing within the clusters is managed on a proactive basis by a table-driven routing protocol that maintains up-to-date routing information regarding all cluster destinations; whereas, routing between clusters is managed on a demandbasis by constructing routes in a dynamic hierarchical fashion as a sequence of relatively stable clusters between the source and the destination. Efficient route search is enabled by the proactive maintenance of routes within each cluster, and reactive route repair is only needed if the source or destination departs from its original cluster, or if the next cluster along a route becomes unreachable. Most of the reaction to node mobility is handled locally within the clusters, hence, the far-reaching effects of topological changes are minimized. By adapting the characteristics of the cluster organization to localized node mobility patterns, the strategy is expected to perform well over a wide range of conditions. The proposed framework addresses several key, yet unanswered questions, which have been raised with respect to the routing problem in ad-hoc networks. Specifically, it has been argued that to achieve acceptable routing performance, multiple routing strategies may need to act cooperatively in the same network. However, this raises the question as to what those strategies should be, and how to effectively toggle between them. Furthermore, it has been proposed that mobility information can be used to select longer-lived routes, and to improve the efficiency of route creation and maintenance. However, no well-defined mobility based ii

3 metric has been proposed that reflects a quantitative measure of path stability. The framework proposed in this paper presents three novel and significant ideas which address each of the shortcomings described above, and builds them into a unified routing framework. Specifically, a mobility model is proposed that provides the basis for a well-defined mobility based routing metric. The metric, referred to as path availability, provides an upper bound on the probability of path failure due to node mobility. The analytical model for path availability represents the first well-defined routing metric based on node mobility. No other similar model has been proposed. The metric shows how mobility provides the basis for the cluster characterization, which is the core of the adaptive clustering framework. No other strategy has factored mobility information into the clustering decision process in this way. Finally, the cluster characterization provides the basis for an efficient distributed clustering algorithm which adaptively maintains a cluster organization, where the size and membership in each cluster is determined dynamically by the mobility characteristics of the local nodes. No other clustering strategy has been proposed that adapts dynamically to node mobility, nor has any proposed strategy adapted specifically to localized conditions. Hence, this work represents a significant, and sustainable contribution to the field. Finally, although it is assumed that node movement is the predominant factor affecting link failures, the framework is extensible in that it is designed to support any valid model for link availability. In the future it is expected that models will be developed to reflect factors including channel quality, energy, and node reliability. iii

4 Table of Contents 1 Introduction What are Wireless Ad-Hoc Networks? What Major Challenges Do Wireless Ad-Hoc Networks Pose? Ad-Hoc Network Design Challenges: The Physical-Layer Ad-Hoc Network Design Challenges: The MAC-Layer Ad-Hoc Network Design Challenges: The Routing Algorithm Research Objective Literature Review Proactive Ad-Hoc Routing Algorithms Demand-Based Ad-Hoc Routing Algorithms General Demand-Based Routing Location Stability Based Routing Signal Stability Based Routing Location Aided Route Search Cluster-Based Routing in Ad-Hoc Networks k-cluster-based Routing Adaptive Clustering for Mobile Wireless Networks Routing Using Minimum Connected Dominating Sets (Spine) Multimedia Support for Wireless Network System (MMWN) Virtual Subnet Routing (VSN) Zone Routing Protocol (ZRP) MANET Cluster Based Routing (CBR) Summary (; t)?cluster Framework Statement of Research Problem Framework Foundations Framework Overview (; t)?cluster Characterization (; t)?cluster Routing Methodology (; t)?cluster Algorithm Proposed Research Research Plan Path Availability Model Clustering Algorithm Routing Protocol Summary iv

5 4 Preliminary Work The Ad-Hoc Mobility Model Challenge Random Ad-Hoc Mobility Single Node Mobility Joint Node Mobility Discussion Random Ad-Hoc Link Availability Summary Addendum Research Objective Mobility Model Evaluation of Proposed Mobility Model: Research Plan System Parameters Determination of System Parameters Simulation Framework Performance Metrics System Parameters Experimental Design Conclusion Bibliography v

6 Chapter 1 Introduction The principal reasons for implementing wireless communications systems include support for terminal mobility [9], and more rapid, widespread access to communications services, without the need to construct and manage expensive cabling systems on the scale required for wired systems. Stated succinctly, the advantages of a wireless system are mobility, flexibility and cost savings. Most wireless networks currently in operation support untethered access for mobile communications devices by providing a wireless interface between the mobile devices and a fixed network of limited range base-stations (BS). Based on this infrastructured model for wireless communications, the air-interface consists of a single data-link terminating on a BS. Communications from that point is routed across a fixed network to its destination. Mobility is managed by allocating a limited set of communications frequency channels to each BS, and dynamically assigning a mobile device to a local channel as it moves from the coverage area of one BS to another. Channels are allocated to BSs such that they can be reused in a subset of the BSs in a network with minimal mutual interference. This infrastructured model is also referred to as cellular wireless communications, because the physical geography is subdivided into smaller partitions called cells, each supporting its own BS and allocated set of channels. The advantage of the infrastructured model for wireless communications is that existing wired networks can be leveraged to support access for mobile users without modifications to the network s control structure. The disadvantage is that it requires a fixed infrastructure constraining node mobility, limiting network deployability, and increasing installation and management costs. Infrastructured wireless networks are not well suited for rapid network deployment, temporary networking for mobile devices, or for environments where it is difficult to achieve adequate BS station coverage, or the installation of fixed infrastucture is not feasible. To address these shortcomings, a new class of wireless network architecture, referred to as Wireless Ad-hoc Networks 1, is emerging as a flexible and powerful wireless architecture that does not rely on a fixed networking infrastructure. 1 In this paper, Wireless Ad-Hoc Networks and ad-hoc networks will be used interchangeably. 1

7 2 This paper proposes a novel framework for routing in wireless ad-hoc networks an emerging class of network architecture that is characterized by its highly dynamic topology and its highly variable, and limited resources. The proposed framework is based on a dynamic cluster organization assuming that node movement is the predominant cause of the failure of wireless links in ad-hoc networks, the idea is to support an adaptive hybrid routing strategy that utilizes a node mobility model to place a lower bound on the probability of path failure over time. Thus, adaptively balancing the competing demands for network resources and routing responsiveness. By adapting the cluster organization to localized node mobility patterns, the strategy is expected to perform well over a wide range of operating conditions. No other clustering strategy proposed for ad-hoc networks has directly factored node mobility into the clustering process, and no other proposed routing strategy has adapted to a well-defined mobility-based metric. The remainder of this chapter is organized as follows. An overview of ad-hoc networks is presented in Section 1.1 The following questions will be answered: What is a wireless ad-hoc network? What are the unique features of this type of network? How did this architecture come about and how is it expected to evolve? What are the relevant applications for ad-hoc networks today and in the future? Next, the unique challenges which ad-hoc networks represent are presented in Section 1.2. This section describes the set of challenges that are uniquely characteristic to ad-hoc networks, thus differentiating ad-hoc network from other network architectures, both wired and wireless, and states the fundamental problem which must be addressed in order to ensure the future viability of the technology. The section continues with more detailed discussion of the design challenges which affect each component of the network architecture, with particular emphasis on issues central to routing, which is the topic of this research. Finally, the objectives of the proposed research are articulated in Section What are Wireless Ad-Hoc Networks? Ad-hoc networks are self-organizing, rapidly deployable, and require no fixed infrastructure [31, 12, 53]. They are comprised of wireless nodes, which can be deployed anywhere, and must cooperate in order to dynamically establish communications using limited network management [15]. Nodes in an ad-hoc network may be highly mobile, or stationary, and may vary widely in terms of their capabilities and uses [40, 16]. The primary objectives of this new network architecture are to achieve increased flexibility, mobility and ease of management relative to infrastructured wireless networks. This is achieved by eliminating the need for fixed BSs and routers; thereby, enabling instant infrastructure wherever ad-hoc nodes are activated, and eliminating many of the constraints to node mobility. An ad-hoc network is itself mobile because the network moves anywhere the nodes locate themselves. In effect, the end nodes themselves must act as mobile routers and BSs. Hence, an ad-hoc network is a dynamic entity which requires adaptive control algorithms in order to be responsive to node mobility, and to operate with minimal administrative intervention.

8 3 The fundamental property, which distinguishes ad-hoc networks from other wireless architectures, is that node mobility causes the network topology to be continuously reconfigurable. In a wireless ad-hoc network environment, transmission range is limited and variable due to numerous system and environmental factors, including transmission power, receiver sensitivity, noise and other channel effects, namely, path-loss, shadow fading, Raleigh fading, Doppler shift, and interference. Node mobility may exacerbate several of these capacity limiting effects. Furthermore, signal range may be limited by design in order to increase system throughput by minimizing channel access contention [67], and to increase battery lifetime by minimizing transmission power. In general, a node s transmission range is neither fixed, nor symmetric it demonstrates temporal and spatial variability. Consequently, the links of an ad-hoc network are not fixed entities their status changes over time and is dependent on the relative spatial location of the nodes, transmitter and receiver characteristics, and the signal propagation properties of the environment. Wireless channel effects and their impact on link status is not unique to ad-hoc networks, although the effects may be more pronounced when both ends of a wireless link are mobile. However, the crucial difference is that all the links in an ad-hoc network are wireless, potentially with rapidly moving end points. Hence, there is no fixed infrastructure in an ad-hoc network. Consequently, the links not only represent wireless end points, as in infrastructured wireless networks, they represent the network topology itself. Thus, as nodes move freely and independently, the topology of an ad-hoc network changes dynamically and arbitrarily. The lack of a fixed network, and the mobility of the nodes lead to two important features of adhoc networks, namely, multi-hop packet routing and mobile (end-system) routers. Unlike infrastructured networks, ad-hoc networks cannot rely on dedicated BSs and routers to forward traffic across fixed network segments between mobile users. Furthermore, direct communications between all nodes is infeasible due to limited transmission range and node mobility. Consequently, store-and-forward packet routing is required over multiple-hop wireless paths. Therefore, the mobile nodes themselves must cooperate in order to dynamically maintain routes and forward traffic on behalf of other nodes the mobile nodes themselves must be routers. In order to maintain communications subject to router mobility and the subsequent dynamic status of the wireless network links, the routers must implement adaptive algorithms that are responsive to dynamics in the network topology, without over-utilization of network resources. Ad-hoc networks evolved largely from the DARPA packet-radio network (PRNET) and related systems [39, 60, 35, 23, 63, 1]. These early multi-hop packet radio (PR) networks were developed to support the tactical requirements of advanced weapons and command and control systems the major attractions of the PR network architecture were rapid deployability and improved survivability, since there was no infrastructure that could be destroyed. Until recently, however, the development of PR technology remained limited to military applications, and few advances had been made in control algorithms since the mid 1980s. This was due largely to the cost and limitations of available hardware, and a lack of sufficient unlicensed radio spectrum.

9 4 Advances in transmission systems, microprocessor technology and power efficient portable communications devices have lead to a re-emergence of interest in PR networks. In particular, the Department of Defense (DoD) continues to pursue a research agenda which is aggressively investigating the possibility of evolving ad-hoc network technology to support the military s mobile communications needs [9]. It remains uncertain as to which network architecture can best suit DoD requirements infrastructured or infrastructureless. Current development and procurement policies in the DoD advocate substantial use of commercial off-the-shelf technology (COTS). Consequently, future wide-spread military deployment of ad-hoc network technology depends largely on civilian interest. Hence, the development of ad-hoc networks is currently driven by a vision focused as much on potential commercial, as well as military applications. Widespread interest in mobile mesh networking, later referred to as ad-hoc networks, started with the formation of a Birds-of-a-feather (BOF) session in a 1995 Internet Engineering Task Force (IETF) conference. Early discussions centered around military tactical networks, satellite networks and wearable computer networks, with specific concerns being raised relative to adaptation of existing routing protocols to support IP networking in such highly dynamic environments. By 1996 this work had evolved into the Mobile Ad-Hoc Network (MANET) BOF, and finally to the charter of the MANET working group (WG) of the IETF in The task of the MANET WG is to specify standard interfaces and protocols for support of IP-based internetworking over ad-hoc networks 2 [11]. More recently, the Ad-Hoc Wireless Networking/Computing Consortium was established, with the goal of coalescing the interests and efforts of industry and academics, in order to apply ad-hoc networking technology to applications ranging from home wireless, to wide area peer-to-remote networking and communications. Furthermore, sessions in major wireless networking conferences 3 and special issues devoted to ad-hoc networking issues are appearing in top quality journals 4, indicating growing advocacy and belief in the practical potential of this technology. Environments in which ad-hoc networks are expected to play an important role initially, include instant infrastructure and wireless LAN (WLAN) applications, particularly where mobile access to a wired network is either ineffective or impossible. However, due to their inherent flexibility, ad-hoc networks have the potential to serve as a ubiquitous wireless infrastructure capable of interconnecting thousands of devices [58], and supporting a wide range of applications. It is hoped that in the future, ad-hoc networks will emerge as an effective alternative to infrastructured WLANs, wired LANs, and even wide-area mobile networking services, such as Personal Communication Systems (PCS). In order to achieve this status, however, applications and services equivalent to those available in these environments must be made available to ad-hoc network users. Hence, control algorithms for ad-hoc networks must be scalable and capable of efficiently utilizing scarce network resources subject to expected high rates of topological change. 2 An ad-hoc network supporting IP-based internetworking services is referred to as a MANET. 3 See the IEEE Wireless Communications and Networking Conference (WCNC 99) 4 See IEEE Journal on Selected Areas in Communications (J-SAC), special issue on wireless ad-hoc networks, August 1999,

10 5 1.2 What Major Challenges Do Wireless Ad-Hoc Networks Pose? Recent interest in ad-hoc networks has come from two diverse communities, with different interests and objectives, the military and the Internet community. Despite a disparity in the potential application of this type of network, and the incumbent differences in a number of design objectives, there is a common set of issues which must be addressed to make ad-hoc networks an effective and usable tool for either community. Furthermore, changes in military procurement policies, and the need for increased interoperability between military and civilian systems have helped to bridge the gap in a number of technical areas. Consequently, advancement in ad-hoc technology with respect to a basic set of challenges will ultimately help achieve success for both potential user communities. These challenges are described in this section, along with a discussion of the resulting design issues. The successful implementation of wireless ad-hoc networking technology presents a unique set of challenges which differ from traditional wireless systems and wired networks. A set of five (5) properties are identified below, which are the basis for these challenges [31]: 1. There is no centralized authority for network control, routing or administration. 2. Network devices, including user terminals, routers, and other potential service platforms are free to move rapidly and arbitrarily in time and space. 3. All communication, user data and control information, is carried over the wireless medium. There are no wired communications links. 4. Resources, including energy, bandwidth, processing capacity and memory, that are relatively abundant in wired environments, are strictly limited and must be preserved. 5. Mobile node that are end points for user communications and process user applications must act cooperatively to handle network functions, mostly notably routing, without specialized routers. The challenges stemming from the properties described above affect every aspect of system design and performance from issues related to physical and MAC-layer design, to network-layer issues including routing, addressing and mobility management, and where real-time communications are required, connection admissions control and real-time resource management, and finally application-layer issues. Given these difficult challenges, the question remains as to what role ad-hoc networking technology will play in the future. Will this network architecture evolve to become a useful and important component of an increasingly diverse global networking infrastructure capable of delivering anywhere, anytime communications a Wireless Internet? Or, will ad-hoc networks remain a very specialized, albeit important, niche technology used for military and other limited instant-infrastructure application? To address this question, wireless

11 6 ad-hoc network designers are faced with the following broad, yet fundamental problem: Develop hardware components, scalable control strategies, and protocols that are capable of adaptively supporting equivalent applications and services to those available today, or envisioned for wired and mobile networks [15] in the presence of a wide range of mobility patterns. The remainder of this section presents a discussion of the various issues involved with meeting these challenges, with a particular focus on the problems and design choices required to implement routing algorithms capable of supporting the stated objectives Ad-Hoc Network Design Challenges: The Physical-Layer The focus of this paper is to address issues central to routing in ad-hoc networks. However, the unique challenges of ad-hoc networks suggest the need to revisit the strict separation of functionality and control that have been imposed by tightly layered network architectures. Specifically, issues including the highly reconfigurable nature of an ad-hoc network, the temporal and spatial variability of link quality, the mobility of the network routers, and the limited supply of power, suggest the that the notion of including physical information into the routing process may not only be effective, but necessary. To this end, it is important to understand the basic challenges relevant to physical-layer components of an ad-hoc systems. In particular, it is crucial to understand that there is a fundamental tradeoff that couples the physical-layer, MAC-layer and routing algorithm design and performance. Specifically, the range of the radio transceivers is chosen as a tradeoff between network connectivity, the reuse of available spectrum, and the power consumption. This tradeoff, which has a direct impact on channel contention, routing and battery lifetime, can be stated as follows: Power versus Bandwidth Tradeoff: In a wireless ad-hoc network, signals can be transmitted at lower power in order to reduce channel contention and conserve energy. Reducing channel contention can increase system utilization and mobile terminal battery lifetime. However, signal range will be reduced and channel effects such as path-loss may be increased. Therefore, nodes which could have communicated directly over a single-hop are forced to communicate over multi-hop wireless paths. Hence, average path-length will increase, and, for a given mobility pattern and environment, the link failure rate will increase. Consequently, more power and bandwidth will be consumed forwarding data packets and control information. This in turn may reduce system throughput and possibly hasten battery failure. Furthermore, a critical transmission range can be identified [62], which is the minimum range require maintain connectivity, below which the network will be partitioned. Achieving an effective balance between channel contention, power, and routing is a crucial challenge that spans the architectural layers in a manner that differs from infrastructured systems. The problem is made more difficult by the fact that mobility and environmental factors, such as physical obstructions, change signal transmission range dynamically. It is beyond the scope of this research to present a comprehensive discussion of physical-layer issues; however, due to this fundamental tradeoff, and the resulting

12 7 synergy between routing and the dynamics of the physical-layer, a brief discussion is presented with respect to the following three physical-layer challenges: 1. Coping with wireless channel effects which impose a fundamental limit on the capacity, quality and stability of wireless links and is affected by node mobility. 2. Increasing battery lifetime in order to improve network stability and connectivity, by either using energy more efficiently, or building better batteries. 3. Acquiring location and mobility information in order to adapt transmission parameters, or provide information to upper-layer entities. Wireless channel effects represent the first and most fundamental challenge to ad-hoc network system design. The effects of signal propagation impose a floor on the quality, the information carrying capacity, the stability, and the signal range of the wireless links [9]. The precise impact depends upon specific system factors, including, antenna design, signal transmission power and receiver sensitivity, modulation and detection schemes, use of signal processing, transmission bandwidth and carrier frequencies, the presence and dielectric characteristics of physical obstructions, and node mobility. Each of these factors will affect, in some way, the ability of an ad-hoc node to accurately transmit and receive information. The challenge is to design system components that can operate effectively in a range of environments and subject to expected mobility patterns. In the future it is likely that more systems will be designed with adaptive components, for example, that allow transmission power to be adjusted either to respond to physical-layer feedback, or via interaction through a higher layer entity such as the routing protocol. Furthermore, emerging technology for software radios is allowing transmission systems to be dynamically updated or changed depending on the network. Similar functionality could be envisioned in which the radio subsystem of an ad-hoc node could adaptively select any number of system parameters to best adapt to the current environment. The second major physical-layer challenge in ad-hoc network system design relates to energy consumption and battery lifetime. These issues represent a critical design factor in any wireless system, but particularly so in an ad-hoc network where there are no fixed infrastructure components. Two approaches have been proposed to address these issues, namely, either to build a more powerful battery, or to use existing batteries more efficiently. Advances in battery technology come about very slowly relative to advances in other areas of mobile terminal and system design [40, 16]. Consequently, more efficient power utilization becomes crucial. To achieve this goal, communications protocols should attempt to minimize energy consumption by optimizing channel access and eliminating unnecessary transmissions. Furthermore, it becomes effective to devise energy-based routing metrics that can minimize the effects of routing on power consumption [66]. The final challenge is to acquire and utilize mobility or location information in order to adapt physical-layer parameters to the dynamically changing environment, or to provide information to upper

13 8 layer (UL) entities for the purpose of enhancing system performance. For example, a strategy to adapt the physical-layer attributes to mobility information would be increase the transmission power as the speed of a mobile node increases. Hence, a multi-level signal strength architecture could be envisioned that is similar to the multi-level schemes proposed for cellular networks. The effect is to reduce the rate of link failure, which might otherwise become excessive for rapidly moving nodes if signal transmission range is very short. Furthermore, mobility and location information could potentially be used by a routing algorithm to select more stable routes or to improve the efficiency of the routing protocol. Chips for processing Global Positioning Systems (GPS) information are becoming very inexpensive, and hence practical to include in any portable computing or communications device. The accuracy of the GPS information has become very high, typically within a few meters [38] of the actual object location. In addition to GPS, motion detection devices are also being developed that can directly provide information regarding terminal speed and direction future ad-hoc terminals may include either GPS or motion sensing capabilities, if not both Ad-Hoc Network Design Challenges: The MAC-Layer The basic challenge to overcome when dealing with a shared transmission medium is how to control access to the communications channel in a fair and efficient manner. Medium-Access-Control (MAC) provides this functionality. Access to the channel can be accomplished in a static fashion, by assigning a fixed portion of the bandwidth to every user, or in a dynamic fashion using either a random access scheme on a per message basis, or by dynamically assigning a portion of the total bandwidth to each user for the duration of a communications session. Fixed allocation schemes are not appropriate for ad-hoc networks. However, dynamic channel allocation schemes can provide a user with probabilistic bounds on access delay and minimum bandwidth. These techniques can be used to support communications requiring probabilistic QoS guarantees in terms of throughput and delay. No such bounds can be guaranteed by a random access method, although random access techniques can increase system utilization. Early PRNETs relied upon a single shared channel and utilized random-access protocols such as ALOHA or CSMA/CD to arbitrate (or not) access to that channel. ALOHA in all its variants is highly inefficient due to the high probability of transmission collisions. Consequently, despite its simplicity, ALOHA is not an effective choice for ad-hoc networks. Carrier sensing techniques improve on ALOHA because devices listen to the channel prior to transmission, deferring their data if the channel is already in use. The effectiveness of CSMA based random-access depends on the ability of every terminal to hear the transmissions of every other terminal. In a wireless ad-hoc network this is typically not the case. Propagation effects vary randomly and transmission range is limited, variable and asymmetric. Hence, the MAC-layer must address two peculiar challenges, namely, the hidden terminal problem and the exposed terminal problem.

14 9 The hidden terminal problem can be explained as follows: Consider three nodes, A, B, and C. Assume that B is within range to hear transmission from both A and C, but neither A, nor C can hear each other s transmissions. In this case A and C are hidden terminals with respect to each other, if they both attempt to transmit to B at the same time, they will not hear each other and the transmissions will collide. The second challenge, the exposed terminal problem is the result of sender initiated channel sensing. It can be understood by considering a fourth node, D, which is within range to received transmissions from C, but is not close enough to hear B. Assume that node B begins transmitting data to A, and during that transmission, node C has data to send to D. Node C will hear the transmission from node A and will defer its transmission even though no collision would have resulted at the receiver, node D. In this case node C is an exposed terminal with respect to node B Ad-Hoc Network Design Challenges: The Routing Algorithm The routing problem characteristic of wireless ad-hoc networks presents a distinctive and unique set of challenges to system design and management. These challenges can be understood in terms of four basic attributes, namely, the routers are mobile, the routers are also end-nodes, the links are wireless, and the power supply is limited. Based on these factors, an ad-hoc network routing algorithm is expected to face a highly dynamic environment along several orthogonal dimensions. Furthermore, both end-nodes and intermediate nodes are expected to continuously and arbitrarily reorganize the network topology and connectivity. In the remainder of this subsection, the routing algorithm design challenge is developed and posed as a set of five questions, which are geared toward solving the fundamental ad-hoc network routing problem. This problem can be stated as follows: Wireless Ad-Hoc Network Routing: An ad-hoc network is a cooperative engagement of mobile hosts [52]. These hosts use wireless communications with constraints on signal transmission range, bandwidth and power. Peer-to-peer communications must be supported between arbitrary hosts without the need to involve specialized routers or requiring direct single-hop communications. Consequently, the mobile hosts must cooperate in order to establish and maintain routes between arbitrary end-points. All the links are expected to be wireless, and any intermediate node or end-point is free to move arbitrarily in time and space. Consequently, routes must be adapted rapidly to node movement and variability in link quality without overutilizing network resources. The wireless ad-hoc network routing problem presents a very difficult challenge that can be posed as a classic tradeoff between responsiveness and efficiency. This tradeoff must balance the need to rapidly adapt the network to node mobility and changes in link quality, against the overhead associated with responding to frequent topology changes. In a wireless ad-hoc network overhead is primarily measured in terms of bandwidth utilization, power consumption and the processing requirements on the mobile nodes. Finding a strategy for efficiently balancing these competing needs forms the basis of the routing challenge.

15 10 This raises the question as to whether or not existing routing protocols are sufficient to meet this challenge, and if not, how to architect an effective routing framework that can. Question 1 Is it feasible to adapt traditional routing protocols that were designed for infrastructured network to operate effectively in wireless ad-hoc networks? Traditional routing algorithms tend to exhibit their least desirable behavior under highly dynamic conditions [54]. Routing protocol overhead typically increases dramatically with increased network dynamics. If the protocol overhead is unchecked, it can easily overwhelm network resources. Furthermore traditional routing protocols require substantial inter-nodal coordination or global flooding in order to maintain consistent routing information and avoid routing table loops. These techniques increase routing protocol overhead and convergence times. Consequently, although they are well adapted to operate in environments where bandwidth is plentiful and the network links are relatively stable, the efficiency of these techniques conflict with routing requirements in ad-hoc networks. It therefore appears that new routing strategies are required for ad-hoc networks that are capable of effectively managing the tradeoff between responsiveness and efficiency. The following definitions present the most commonly used means of classifying routing protocols that have been designed for ad-hoc networks: Definition 1.1 Proactive Routing is defined as a strategy in which routes are continuously maintained for all reachable network destinations. This approach requires periodic dissemination of routing updates to reflect the up-to-date state of the network. Definition 1.2 Reactive Routing is defined as a strategy in which routes are established and maintained on a demand basis only if they are needed for communications. This approach requires procedures to acquire new routes and to maintain routes following topology changes. Definition 1.3 Hybrid Routing is defined as a strategy which selectively applies either proactive or reactive routing techniques based upon either predefined or adaptive criteria. Reactive routing, which is also referred to as demand-based routing, represents a new class of routing algorithm has been proposed as a means of achieving a better balance between responsiveness and efficiency. The objective of reactive routing is to minimize the reaction of the routing algorithm to topology changes by maintaining a limited set of routes those required for on-going communications. A given node, i, requires a route to destination node j if it is either the source of data for node j, or if it lies along a path to node j, from another node with data for node j. The idea is that by selectively limiting the set of destinations to which routes are maintained, less routing information needs to be routinely exchanged and processed. Consequently, less bandwidth is consumed by routing information, less computation is required to process routing information, and less memory is consumed by routing tables. Based on this technique,

16 11 routing is expected to converge more rapidly following topology changes, and additional network resources are expected to be available for the transmission and processing of application data. In a reactive routing strategy, paths are maintained on a demand-basis using a query-response process. This involves a variation of controlled flooding referred to as a directed broadcast, in which a query, or route request packet is selectively forwarded along multiple paths toward a target destination. The search process dynamically constructs one, or multiple paths from the source node to the destination. This strategy limits the total number of destinations to which routing information must be maintained, and, consequently, the volume of control traffic required to achieve routing. The shortcomings of this approach include the possibility of significant delay at route setup time, the large volume of far reaching control traffic required to support the route query mechanism, and reduce path quality. Furthermore, despite the objective of maintaining only desired routes, the route query could propagate to every node in a network during the initial path setup causing each node to establish paths even when they are only required by certain sources. Finally, most reactive strategies do not discover optimal paths, and the paths typically become increasingly less optimal following each topology change. Pro-active routing protocols take a different approach from reactive schemes in that they periodically distribute routing information throughout the network in order to pre-compute paths to all possible destinations. Hence, each node maintains a-priori routing information to all destinations, regardless as to whether or not a given node actually needs to reach each such destination, or lies along a path of another node that does. Although this approach can ensure higher quality routes 5 in a static topology, it does not scale well to large, highly dynamic networks. This routing strategy is also referred to as table-driven routing, because protocols that adopt this strategy attempt to maintain consistent information in the routing tables distributed throughout the network. Several proactive routing protocols have been proposed for ad-hoc networks, these are reviewed in Section 2.1. Question 2 Routing algorithms that have been proposed for wireless ad-hoc networks are typically classified according to their route acquisition policy as either being proactive or reactive. A third possibility is to combine the features of both policies into a hybrid routing strategy. If new routing algorithms are required in order to establish an effective balance between routing algorithm response and efficiency, the question arises as to whether the routing algorithm should be designed to acquire routes proactively, reactively, or using a hybrid approach? Maintaining routes proactively has important advantages over demand-based approaches most notably, it does not incur the overhead and latency of a demand-based route discovery process. Furthermore, proactive routing protocols generally compute shortest-path routes, whereas demand-based protocols 5 Higher-quality refers to the ability to establish paths based upon some optimization objective, e.g. shortest-path, least-delay, most-reliable, etc..

17 12 generally suffice to discovering a path, rather than the best path, to a given destination. Finally, proactive schemes can more easily be adapted, using the same update information to modify routing to all affected destinations. Many of the proactive schemes execute logically separate instances of the protocol for each destination; consequently, a given link failure must be reflected in multiple path maintenance operations a case of the left hand not knowing what the right hand is doing. For these reasons, researchers continue to explore methods for adapting proactive routing strategies for operation in ad-hoc networks. Reactive routing protocols adopt the philosophy that it is too costly to maintain a-priori information to all destinations it is believed that it is more efficient to incur the one-time initial cost of discovering a path to a newly required destination. It is clear, however, that when resources are plentiful the performance of this approach can be enhanced significantly by maintaining a-priori information. In the tightly resource constrained environment of ad-hoc networks this is not the case; therefore, on-demand protocols seems to offer an attractive solution. However, the problem identified above must be addressed, as efficient route creation and minimization of reaction to topological change are critical issues in order to achieve efficient routing in ad-hoc networks. It is very difficult to achieve this objective with a pure on-demand protocol. New techniques are needed to improve the efficiency of the route search process. Furthermore, reactive protocols typically must react multiple times to the same topology change once for each affected destination. Consequently, better techniques are needed to manage the process of route maintenance in highly dynamic environments. Despite the apparent advantages of reactive routing protocols, they have some substantial shortcomings and do not perform well under all possible circumstances. The question of which technique is better suited for ad-hoc networks depends on a number of factors. Each technique has its advantages under certain circumstances, but may not perform well in other situations. The criteria which determine the approach to favor in a given environment depend upon the following factors: the size of the network, the patterns and rates of node mobility, the distribution of traffic among the nodes, the efficiency of the route search process, the speed of route repair following node movement, and the far reaching effects of algorithm actions following node movement. Based on this set of dynamic factors, it is evident that neither strategy will perform optimally, or even adequately under all circumstances. Consequently, the idea of hybrid routing strategies have been advocated by several researchers [29, 1, 61]. Specifically, it has been suggested that multiple routing algorithms may need to be implemented to operate in different physical or logical domains, or on different time-scales within the same network. These parallel strategies may differ in many ways, including the route acquisition policy, routing metrics, and path maintenance algorithms. The idea is to divide the routing task in order to strike a better balance than can be achieved by a single algorithm. Hence, the following question is raised with respect to implementing a hybrid routing strategy: Question 3 Hybrid routing techniques have recently been advocated to improve the performance of routing in ad-hoc networks by selectively using different routing strategies under different circumstances. The ques-

18 13 tion, therefore, arises as to how the network can effectively toggle between each technique to best utilize the strengths of each approach in a manner that is consistent and if possible complimentary? A flat-routed ad-hoc network requires the construction and maintenance of routes that may span a large number of unreliable wireless links. Consequently, in periods of high node mobility, a path may frequently suffer multiple simultaneous failures. Although reactive protocols have been designed to improve routing efficiency, a constantly changing topology is likely to put excessive stress on any flat-routed network that involves more than a few tens of nodes. Consequently, network architects have attempted to devise strategies that dynamically organize ad-hoc networks into clusters the cluster topology is then leveraged in order to limit far reaching reactions to topology dynamics, typically through hierarchical or hybrid routing techniques. The idea of dynamic cluster-based routing, and its variations, was originally conceived as a means of achieving network scalability [63, 1] in PR networks. Early adaptations were intended to achieve to same objectives as hierarchical routing in fixed networks [36]. However, unlike the cluster organization of a fixed network, the assignment of nodes to clusters in PRNETs and ad-hoc networks must be a dynamic process. In a clustered ad-hoc network, the network is dynamically organized into partitions called clusters with the objective of maintaining a relatively stable effective topology [41]. The membership and characteristics of each cluster may change dynamically over time in response to node mobility and is determined by the criteria specified by the clustering algorithm. Clustering in ad-hoc networks can be used to achieve several different objectives, namely, to support hierarchical routing, to make the route search process more efficient for reactive protocols, to support hybrid-routing in which different routing strategies operated in different domains, or levels of a hierarchy, or to provide more control over access to transmission bandwidth. Section 2.3 reviews several dynamic clustering strategies that have been proposed for ad-hoc networks any or all of the objectives just described may underly the design of each of these strategies. In a fixed network, the assignment of nodes to clusters remains fixed. However, in an ad-hoc network clusters must be dynamically formed as node mobility continuously alters the connectivity and spatial relationships among the nodes. Thus, a clustering framework designed for an ad-hoc network must specify an algorithm for dynamically assigning nodes to clusters, and for responding to node mobility. It must also answer the following question: Given a dynamic cluster organization, how do nodes discover and maintain intra and inter-cluster routes? The argument made against dynamic clustering is that the rearrangement of the clusters and the assignment of nodes to clusters may require excessive processing and communications overhead, which in turn may outweigh its potential benefits. Furthermore, if the clustering algorithm is complex or can not quantify a measure of cluster stability, its support for routing may also be questioned.