ANALYSING THE OVERHEAD IN MOBILE AD-HOC NETWORK WITH A HIERARCHICAL ROUTING STRUCTURE



Similar documents
Multi-Channel Opportunistic Routing in Multi-Hop Wireless Networks

Load Balancing Mechanism in Agent-based Grid

On Multicast Capacity and Delay in Cognitive Radio Mobile Ad-hoc Networks

Static and Dynamic Properties of Small-world Connection Topologies Based on Transit-stub Networks

Buffer Capacity Allocation: A method to QoS support on MPLS networks**

On the (in)effectiveness of Probabilistic Marking for IP Traceback under DDoS Attacks

A MOST PROBABLE POINT-BASED METHOD FOR RELIABILITY ANALYSIS, SENSITIVITY ANALYSIS AND DESIGN OPTIMIZATION

DAY-AHEAD ELECTRICITY PRICE FORECASTING BASED ON TIME SERIES MODELS: A COMPARISON

Finding a Needle in a Haystack: Pinpointing Significant BGP Routing Changes in an IP Network

TOWARDS REAL-TIME METADATA FOR SENSOR-BASED NETWORKS AND GEOGRAPHIC DATABASES

Large-Scale IP Traceback in High-Speed Internet: Practical Techniques and Theoretical Foundation

Local Connectivity Tests to Identify Wormholes in Wireless Networks

Automatic Search for Correlated Alarms

Comparing Dissimilarity Measures for Symbolic Data Analysis

Time-Cost Trade-Offs in Resource-Constraint Project Scheduling Problems with Overlapping Modes

Multiperiod Portfolio Optimization with General Transaction Costs

Characterizing and Modeling Network Traffic Variability

A Virtual Machine Dynamic Migration Scheduling Model Based on MBFD Algorithm

A Multivariate Statistical Analysis of Stock Trends. Abstract

An important observation in supply chain management, known as the bullwhip effect,

Design of A Knowledge Based Trouble Call System with Colored Petri Net Models

Web Application Scalability: A Model-Based Approach

The risk of using the Q heterogeneity estimator for software engineering experiments

Evaluating a Web-Based Information System for Managing Master of Science Summer Projects

Re-Dispatch Approach for Congestion Relief in Deregulated Power Systems

X How to Schedule a Cascade in an Arbitrary Graph

Service Network Design with Asset Management: Formulations and Comparative Analyzes

CRITICAL AVIATION INFRASTRUCTURES VULNERABILITY ASSESSMENT TO TERRORIST THREATS

An inventory control system for spare parts at a refinery: An empirical comparison of different reorder point methods

The Online Freeze-tag Problem

IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 29, NO. 4, APRIL Load-Balancing Spectrum Decision for Cognitive Radio Networks

Implementation of Statistic Process Control in a Painting Sector of a Automotive Manufacturer

An Analysis Model of Botnet Tracking based on Ant Colony Optimization Algorithm

Point Location. Preprocess a planar, polygonal subdivision for point location queries. p = (18, 11)

Service Network Design with Asset Management: Formulations and Comparative Analyzes

An Approach to Optimizations Links Utilization in MPLS Networks

An Investigation on LTE Mobility Management

STATISTICAL CHARACTERIZATION OF THE RAILROAD SATELLITE CHANNEL AT KU-BAND

Storage Basics Architecting the Storage Supplemental Handout

ENFORCING SAFETY PROPERTIES IN WEB APPLICATIONS USING PETRI NETS

Failure Behavior Analysis for Reliable Distributed Embedded Systems

Red vs. Blue - Aneue of TCP congestion Control Model

Proposition of a new approach to adapt SIP protocol to Ad hoc Networks

Project Management and. Scheduling CHAPTER CONTENTS

Saturation Throughput in a Heterogeneous Multi-channel Cognitive Radio Network

FIArch Workshop. Towards Future Internet Architecture. Brussels 22 nd February 2012

The impact of metadata implementation on webpage visibility in search engine results (Part II) q

A Novel Architecture Style: Diffused Cloud for Virtual Computing Lab

Machine Learning with Operational Costs

An effective multi-objective approach to prioritisation of sewer pipe inspection

TRANSMISSION Control Protocol (TCP) has been widely. On Parameter Tuning of Data Transfer Protocol GridFTP for Wide-Area Networks

Monitoring Frequency of Change By Li Qin

F inding the optimal, or value-maximizing, capital

The Magnus-Derek Game

Two-resource stochastic capacity planning employing a Bayesian methodology

The fast Fourier transform method for the valuation of European style options in-the-money (ITM), at-the-money (ATM) and out-of-the-money (OTM)

Concurrent Program Synthesis Based on Supervisory Control

A Novel Pathway for Portability of Networks and Handing-on between Networks

Branch-and-Price for Service Network Design with Asset Management Constraints

MetroNet6 - Homeland Security IPv6 R&D over Wireless

Mobile SCTP Transport Layer Mobility Management for the Internet

Receiver Buffer Requirement for Video Streaming over TCP

Stochastic Derivation of an Integral Equation for Probability Generating Functions

Managing specific risk in property portfolios

CABRS CELLULAR AUTOMATON BASED MRI BRAIN SEGMENTATION

A Method to Segment a 3D Surface Point Cloud for Selective Sensing in Robotic Exploration

An Efficient QoS Routing Protocol for Mobile Ad-Hoc Networks *

NUTSS: A SIP-based Approach to UDP and TCP Network Connectivity

Multistage Human Resource Allocation for Software Development by Multiobjective Genetic Algorithm

Performance Evaluation of AODV, OLSR Routing Protocol in VOIP Over Ad Hoc

Optimal Routing and Scheduling in Transportation: Using Genetic Algorithm to Solve Difficult Optimization Problems

On Traffic Fairness in Data Center Fabrics

An Efficient Method for Improving Backfill Job Scheduling Algorithm in Cluster Computing Systems

Secure synthesis and activation of protocol translation agents

Enhanced Security Key Management Scheme for MANETS

From Simulation to Experiment: A Case Study on Multiprocessor Task Scheduling

Effect Sizes Based on Means

Softmax Model as Generalization upon Logistic Discrimination Suffers from Overfitting

Pressure Drop in Air Piping Systems Series of Technical White Papers from Ohio Medical Corporation

This document is downloaded from DR-NTU, Nanyang Technological University Library, Singapore.

Alpha Channel Estimation in High Resolution Images and Image Sequences

Forensic Science International

Drinking water systems are vulnerable to

Risk and Return. Sample chapter. e r t u i o p a s d f CHAPTER CONTENTS LEARNING OBJECTIVES. Chapter 7

Risk in Revenue Management and Dynamic Pricing

Compensating Fund Managers for Risk-Adjusted Performance

Principles of Hydrology. Hydrograph components include rising limb, recession limb, peak, direct runoff, and baseflow.

Learning Human Behavior from Analyzing Activities in Virtual Environments

G.Vijaya kumar et al, Int. J. Comp. Tech. Appl., Vol 2 (5),

A Link-state QoS Routing Protocol for Ad Hoc Networks

Methods for Estimating Kidney Disease Stage Transition Probabilities Using Electronic Medical Records

Rummage Web Server Tuning Evaluation through Benchmark

The Economics of the Cloud: Price Competition and Congestion

EECS 122: Introduction to Communication Networks Homework 3 Solutions

Computational Finance The Martingale Measure and Pricing of Derivatives

Penalty Interest Rates, Universal Default, and the Common Pool Problem of Credit Card Debt

A Simple Model of Pricing, Markups and Market. Power Under Demand Fluctuations

A Modified Measure of Covert Network Performance

Synopsys RURAL ELECTRICATION PLANNING SOFTWARE (LAPER) Rainer Fronius Marc Gratton Electricité de France Research and Development FRANCE

Transcription:

AALYSIG THE OVERHEAD I MOBILE AD-HOC ETWORK WITH A HIERARCHICAL ROUTIG STRUCTURE Johann Lóez, José M. Barceló, Jorge García-Vidal Technical University of Catalonia (UPC), C/Jordi Girona 1-3, 08034 Barcelona, Sain, {johannl, joseb, jorge}@ac.uc.edu Keywords MAET, Routing, Clustering, Overhead. Abstract- Hierarchical routing is a common ractice in actual Internet networs. This ind of routing allows nodes and routers to get small routing tables and small routing overhead. On the oosite side, routing in MAETs is flat; therefore, nodes have to learn how to reach each node in the networ. It means large routing tables, imact in the loo u algorithm and large routing overhead. In this aer we analyze the advantages of using hierarchical routing in MAETs under secific scenarios, showing the reduction of routing overhead from a factor of to / (being the number of nodes and the number of subnets). I. ITRODUCTIO Currently mobile communications have become an imortant role for the eole. Since these inds of communications are considered less a luxury and more a necessity, the number of mobile users, services and devices grows quicly. Most of the researching efforts have been sent on the heterogeneous networs (integration of fixed networs with infrastructure Wireless etwors). However, during the ast few years, essentials efforts have been done in the researching field of Mobile Ad-hoc etworing (MAET) [1], which is basically a wireless networ without infrastructure. One of the main issues in MAET researching is the routing. The rincial results obtained into this area are the well-nown AODV [] (reactive) and the OLSR [3] (roactive) routing rotocols. In MAET, every node requires to articiate as a router, maintaining individually routes to other nodes. If the number of nodes grows largely, the number of routes will increase raidly. This effect together with the mobility of nodes may increase the overhead due to the maintenance of the routes, consuming the scarce bandwidth in the MAET and therefore reducing the throughut. If we do not tae into account the mobility, this scenario is something similar to what haened to the ARPAnet at the beginning. The logical solution to reduce this overhead would be to use a hierarchical structure lie the used in Internet, in which the nodes are aggregated into subnets to be handled as a single entity for routing uroses. However, this structure is difficult to aly in MAETs due to their dynamic and distributed nature. The main roblems to solve in such hierarchical structure are the address acquisition under mobility scenarios, the dynamic creation and removal of subnets and the maintenance of already established sessions when a node moves from one subnet to other and changes its IP address. We are roosing a scheme, which taes advantage of the fact that currently the nodes in MAETs can be groued following hysical or environmental constraints to aly this hierarchical structure. These formed clusters can be considered subnets of a MAET, giving the chance of reresenting multile routes of a large number of nodes by a single route. Since the routing information is cutting down, a reduction in the overhead may be obtained [5]. This scheme is useful in scenarios lie Wireless Mesh etwors (WM) with nodes connected to infrastructures but those are the edge-oints of a MAET, or in scenarios in which there are nodes with This wor was suorted by the Sanish Ministry of Education and Science; It was art of the And the Sixth Framewor Programme EUROGI P3/1

leader roles (e.g. an emergency networ). Figure 1 (extracted from [4]) shows an examle of an emergency situation in which a fire tro handle by the head quarters is sub-groued in rescue teams. Each rescue team is controlled by a leader (fire chief). Rescue teams also are divided in rescue units comosed of airs of firemen with a rescue leader. The firemen tro forms a hierarchical structure with communications between the head quarters with the chief and the chief with the rescue unit leaders. Each rescue unit forms a cluster with communications within its own cluster and his rescue leader (intra-cluster communications) and with other clusters (inter-cluster communications), see [4]. From the communications oint of view, each cluster is formed by a leader, usually it has more resources in terms of bandwidth, ower, gateway connectivity, etc, (e.g. a fire-truc) relatively static and nodes that move in the disaster area. In a natural manner, mobile nodes attached to the leader node may form a subnet. If the mobile node moves it can be attached to another leader node. ow the question is whether it is more efficient if that node ees the IP address or whether the node attaches and changes the IP address to one belonging to the new subnet. Figure1. Examle of an emergency situation, in which a MAET can assume a hierarchical structure. In this aer we evaluate analytically the exected overhead reduction generated by the routing rotocol when we adot a hierarchical routing scheme in MAET. We secifically evaluate the overhead incurred by the current routing rotocols (reactive and roactive) used in MAET, then, we modify the obtained models to include the arameters to reresent the hierarchical structure (subnet, address aggregation). Finally, we comare these models to get a reliminary idea of the reduction incurred by our roosal. The main objective of this aer is not to solve all the technological issues arisen in MAETs if subnetting solutions are aroached but to show that an effort in that direction is valuable. The rest of the aer is organized as follows: a generic descrition of the scenario characteristics and the challenges that we have to solve for integrating successfully the hierarchical routing to MAET are given in section II. Then, we resent the analytical model of the overhead generated by the routing mechanisms. Finally, we have the conclusions and future wor. II. SCEARIO DESCRIPTIO We are roosing a scheme which taes advantage of the fact that currently the nodes in a MAET can be groued following hysical or environmental constraints. Figure shows some of the following assumtions: odes form an area of size A around a leader node, and this area will corresond to an IP subnet of dynamic size, e.g. deends on the number of nodes attached to the area. odes form a multi-homed networ with nodes of the same area and with nodes of other areas. A node is attached to an area and therefore has an IP address corresonding to that IP subnet area. IP addresses are used lie host locators. When a node moves to other area has to acquire a new IP address corresonding to the new IP subnet. P3/

Every node nows each of its one-ho neighbours, e.g. using lin sensing or a Hello rotocol. A message sent by a node is received correctly within a finite time by all of its one-ho neighbours. L L L L Figure. MAET nodes groued to form subnets. Taing into account these assumtions, some challenges lie the subnet formation and address acquisition, the mobility of nodes between subnets, the intra-subnet routing, and the inter-subnet routing have to be solved. 1) Subnet Formation and Address Acquisition In our roosed scheme the node needs a mechanism to detect a leader node and to get attached to it. The leader node is resonsible to give an IP address to the node. In case the node roams to a new area, the node has to detect that is in a new area and must obtain an IP address from the leader node. Future wor will be required to define a rocedure that allows a node to get attached to an area and obtain an IP address within the subnet. There are several roosals in the literature about address acquisition schemes and leader selection algorithms (e.g. DHCP or dynamic stateless address acquisition [6], [7], [8], [9] and [14]). The main issues to solve by these ind of mechanisms are: Leader detection and selection ode Address Construction Dulicate Address Detection ) Mobility between subnets Mobility into the subnet is not a critical issue and it could be erformed by the current roosed MAET routing rotocols (AODV, or OLSR). Although, a mobility management between subnets loos necessary. The roosed mobility management between subnets relies on the use of an USER ID lie the node identification and the IP Address lie the location indicator of the node. This aroach breas the IP aradigm of uniqueness between location and identity. Additionally, this aroach allows suorting terminal and user mobility. The aroach could be based on the aforementioned Address Acquisition rocedure for terminal mobility, the Session Initiation Protocol (SIP), [10] and [11], for streaming multimedia alications, and the Stream Control Transmission Protocol (SCTP) [1] with an enhancement for dynamic address reconfiguration [13] for the alications based on reliable transort rotocol. It mainly rovides mobility in two layers: networ layer and the alication, and for these reason our roosed scheme is so called Multi-layer. 3) Intra and Inter subnet routing Some enhancements to the existing MAET routing rotocol are needed to allow interaction between them and the Address Acquisition mechanism. By default, reactive rotocols lie AODV does not require large changes because its standard [] includes address aggregation, allowing the adotion of this hierarchical scheme. For the case of hierarchical routing with roactive rotocols lie OLSR, the scheme requires a new routing mechanism to allow the address aggregation (inter-subnet routing). Since the subnets are assumed fixed (leader node is mainly static) in our scenario, the mechanism could tae benefit of this fact. Future wor will be required to define these mechanisms and modifications needed by the current routing rotocols to interact with our roosed hierarchical structure. P3/3

III. AALYTICAL MODEL To validate our roosal we use some well-nown analytical models of reactive and roactive routing rotocols [15], hierarchical routiing [17], and cellular systems [16]. First than all we will resent briefly each model. The model for MAET routing rotocols let us estimate the control overhead generated in a fixed networ, and the control overhead due to the mobility. After that, we aly some hierarchical routing (subnetting) arameters to the model, estimating the new control overhead generated by the routing rotocols for a fixed etwor. Then, with the exressions obtained and the cellular system models we can estimate the control overhead due to the subnetting and the mobility. Finally, we comare the estimations obtained to validate our roosal. A. MAET Routing Protocols Models These models were develoed and validated in [15], to comare the traffic control overhead and their bandwidth cost generated by reactive and roactive rotocols. It taes into account the networ density, the mobility, the traffic creation, and the traffic density. The arameters used to model the networ are: the number of nodes, the average live time of a lin T b (T b = 1/µ, where µ is the lin breaage rate) to model the mobility, the average length of a route L which deends mainly of the shae of the networ. Additionally, the model assumes that µ and L remain constants, and the networ always remain connected. 1) Reactive rotocol model Since the route creation in reactive rotocols deends mainly of the data traffic creation and diversity, the following arameters are used: the average number of route creation λ (Route Requests), and the average number of active routes er node a, such active route is a air where the source continuously sends acets to the destination. We also have to tae into account the arameters that deend on the rotocol, see [15]: the average number of emissions for a route request (could include the route rely messages) B r, and the route request otimization factor o r, which deends of the networ and the traffic arameters (o r = B r / ). For a ure flooding rotocol, if we include the Route Rely from the destination, we get that; o r = 1 + j /, where j is the number of rely messages. According to [5] the Route Requests are the resonsible for the majority of the routing overhead in networs running AODV, so, we could exect o r 1. In a Fixed etwor a reactive rotocol roduces λ route requests every second. These requests will generate λo r control acets in the networ. We can exress the control overhead due to the routing rotocol by: O ( ) ο = rλ (1) The mobility in the MAETs can be reresented by the lin breaage and creation. Being the lin breaage the most imortant factor to model the reaction of the routing rotocols to mobility, because the rotocols have to react quicly to the breaages when the lin is actively used to transmit data. Therefore, when a lin breaage is detected, the rotocol will basically issue a new route request to reair routes using that lin. If we have a routes in the networ, there will be al active lins, such that µ(al) acets will be generated by a lin breaage. Basically the reactive rotocol will issue a new route request to reair routes using that lin. This route request could be either initiated by the source of the route or by the node detecting the lin breaage. For the latter case this gain may be integrated in the arameter o r. For the more essimistic case the overhead due to the mobility is exressed by: ) Proactive rotocol model O( ) rµ al = o () A roactive rotocol mainly deends on the regular emission of control acets, having the advantage of not generating any overhead at route creation, because their fixed control traffic overhead includes the cost of the route creation. The arameters concerning to the roactive rotocol are, see [15]: the acets for roactively discovering the local toology (hello acets) emitted by a node er second h, and the toology broadcast acets emitted by a node during a second for allowing global nowledge of the toology t. These arameters are exressed in terms of rates. To react to toological changes the active next ho A is defined; it corresonds to the P3/4

average number of active lins er node, and it is used to evaluate what toology changes may trigger additional control traffic. Since roactive rotocols can tae advantage from their nowledge of the toology in order to otimize broadcasting, a broadcast otimization factor o is denoted (o = B /), where B denotes the average number of emissions to achieve a toology broadcast. In a roactive rotocol h hello messages are roduced er second in the entire networ, initiating t toology emissions, which generate t o acets. The overhead roduced can be reresented by O( ) = h + o t (3) Quantifying the overhead generated in reaction to mobility is similar as with reactive rotocols. A node detecting a lin breaage on a route will transmit an additional toology broadcast acet. In average a node is on al routes, and several routes may have the same outgoing lin. However, the robability that the next hos for these routes are the same is certainly greater than the robability that the destinations for these routes are the same. With the introduction of the average number of active lins of a node A arameter, the total overhead due to mobility in a roactive rotocol is exressed by O( ) µ = o A (4) With the exressions resented above, we can note that the number of nodes in the networ dominates the overhead in an exonential manner O( ) for both tyes of routing rotocols. The cost of these routing rotocols becomes rohibitive as the number of nodes in the networ gets large. B. Alying hierarchical routing (subnetting) to the routing rotocols In hierarchical routing any node ees comlete routing information about nodes that are close to it, and lesser information about nodes located further away from it. The reduction of the routing information is achieved artitioning the networ (creating subnets). We roose to form the subnets with a level hierarchical clustering based scheme [17], assuming all the nodes are uniformly distributed in the networ and in each subnet for the scoe of this analysis. To aly subnets a new arameter must be included in the model: the number of subnets in the networ. To guarantee full connectivity in the networ one node at least must maintain a route to each node into the same subnet, and a route to each subnet in the networ. From this, the networ must have (-1)+/ routes er node. Therefore, the first advantage in subnetting is the reduction of the routing table and the saving in the loo u algorithm. 1) Reactive rotocols with subnets When we introduce subnets to the reactive rotocol reactive rotocol model, we include the average number of route creation λ I by a node to other nodes into the same subnet, the average number of route creation λ E by a node to other subnets, and the average length of a route q to another subnet. We remind you that we are interested in nowing what is the overhead reduction (if there is) in a MAET using a reactive rotocol lie AODV, and not in solving the technological asects of AODV using subnets. We leave that study as a second ste. Alying the new arameters to the aforementioned model for a fixed networ (1), we have λ I route requests for routes to nodes into the same subnet roduced each second, and (-1)λ E route requests er second for routes to nodes located in other subnets. Therefore, the overhead generated by these requests can be exressed by ο r λi + οrλe ( 1) + q + S Where S reresents the cost of finding out the routes to other subnets (S = (-1) /). Assuming that the robability of having a route to a subnet is much higher (P(cache) 1) than having a route to a articular node, this cost is only aid few times, therefore during a eriod of time long this value will be small (because is not a rate). Since the q value is very small comared with the number of nodes in the networ, and S is constant during the time, we consider these two values does not dominate the overhead (are not relevant) for our analysis. From this we have P3/5

O ( ) ο rλi + οrλe ( 1) (5) Since the nodes are uniformly distributed, and assuming that λ I and λ E are indeendent from each other and have the same robability occurrence, we can reresent λ = λ I + (-1)λ E. Exressing (4) in terms of λ we obtain O ( ) οrλ (6) To model the overhead due to the mobility we divide it in two comonents O( ) = O ( ) O ( ) (7) s + The first term O S () reresents the overhead due to the mobility of the nodes into the subnet. And the second one O H () reresents the overhead due to the change of subnet. This overhead is roduced because each time that a node changes of subnet, it must change its address. The rocess to acquire a new address and to attach it to the node ID imlies a message exchange. The lin breaage is assumed in our model lie the main comonent of the overhead in reaction to mobility. After adding subnets to the reactive rotocol model we will have two cases. The first case aears when the broen lin corresonds to a lin between two nodes that are into the same subnet, and for this case the route reair will be local into the subnet. In that case a flooding in the subnet is enough to reair the route, therefore, it is not necessary to reair from the source. The second one, is resented when the broen lin corresonds to a lin between two nodes that are located in different subnets. For this case the route has to be reaired from the source of the route. If we have al active lins in the networ; the overhead O S () due to mobility could be reresented by o r µ al Os ( ) orµ al (8) Due to the difficult to quantify exactly the emissions of control acets in reaction to mobility, we decided to exress it between two quotes, corresonding each one to the aforementioned cases (the bottom and uer bounds). We have to note that often several routes have identical destinations and the majority can be reaired locally, meaning that the overhead will be close to the lower bound the mayor art of the time. ) Adding mobility between subnets to the model A subnet corresonds to a cluster, which is based on the roximity (# of hos) to a cluster head (leader node), forming an area around this cluster head. Desite of the variable size of this area (subnet), we assume a subnet of a fixed size, for this analysis. Therefore, we can assume that a node moving between subnets behaves lie a node moving in a cellular system. For our analysis the most imortant arameter related to the mobility is the robability of subnet (area) boundary crossing. According to [16], an aroximate evaluation for cellular system erformance is obtained through models based on fluid flow assumtions. The following conditions are needed to derive a simle exression: The nodes and their traffic are uniformly distributed over a given subnet (area or cell). The nodes have a mean velocity of v and their directions of movements are uniformly distributed over [0,π]. With the aforementioned conditions the subnet crossover rate ηis given by H P η = v π S Where P corresonds to the erimeter of the subnet, and S is the area of the subnet. If we denote the time sent by a node within a given subnet (dwell time) [18], [19]by the random variable T h. The mean dwell time E(T h ) can be calculated by nowing the subnet crossover rate [0], [1], that is P3/6

E ( ) = Which corresond to the average number of handoffs. T h Additionally, the dwell time is assumed exonentially distributed (the residing time of a node in the actual subnet is not related with the residing time of the same node in a revious subnet, memory less system). So, the robability of subnet boundary crossing is given by 1 η P( T h ) e t =η η If our Address Acquisition rocedure generates h acets in average, and if we have / nodes er subnet, we will have h(/) acets in average due to the change of subnet in the entire networ. ow, we can reresent the overhead due to change of subnet by O H ( ) = hη (9) Finally relacing (8) and (9) in (7), we can aroximate the overhead in the routing rotocol model with hierarchical subnetting due to mobility to 3) Proactive rotocol with subnets O( ) or µ al + hη When we aly hierarchical routing with a roactive rotocol, a mechanism to find out the routes to reach other subnets is needed. Within the scoe of this study we introduce the arameter to reresent the cost of finding out these routes, and leaving the descrition and evaluation (identify the value of the arameter ) of the aforementioned mechanism for a future study. With the roactive rotocol, each node emits h hello messages er second, and now each t is emitted to the nodes that are close to it (to all the members of the subnet). The overhead generated is given by O( ) = h + o t +ϕ (10) In addition, the arameter includes the cost of emitting the routes to other subnets to each node. Desite of all this routing information have a cost, we can exect this cost is much lower comared to the overhead without hierarchical routing, because this cost does not deend of a factor lie in the roactive and reactive rotocols, and because the rocedure can tae advantage of the artial nowledge of the toology and the characteristics of the scenario. The overhead generated in reaction to mobility is reresented by the exression (7) where the second comonent O H ()is the same exressed in (9), and the comonent O s ()will be exressed by Os ( ) = o µ A + ϕ (11) The first art corresonds to the overhead generated by a lin breaage, such that the toology is only udated locally. Here is evident that a local toology change does not have to generate control overhead in the entire networ, because certainly the nodes located further away are not affected by these changes. The main idea behind this assumtion is that if a node nows how to reach a subnet, it is not necessary to now how the toology in that subnet is. This assumtion is similar to the idea of Internet in which an Autonomous System (AS) does not now how is organized another AS and only nows how to reach that AS. Again, we have the arameter in the second art of the exression, to reresent the overhead generated in reaction to a lin breaage between subnets. Finally relacing (9) and (11) in (7) we obtain O( ) = o µ A + ϕ + hη P3/7

To acquire an order of magnitude of the reduction in the overhead generated by the MAET routing rotocols after alying hierarchical routing we draw two comarable exressions lie (1) and (6). Figure 3 shows a comarison between the overhead generated by a reactive rotocol and the same rotocol with 4 subnets, for the case of static nodes. Figure 3. Overhead comarison between a reactive rotocol and the same rotocol with subnets. The overhead reduction observed in the figure gives us a margin, which is enough to add some overhead generated by the modifications and new mechanisms needed to introduce hierarchical routing to MAET, and to ee an imortant reduction. In both cases, roactive and reactive rotocols with hierarchical routing, we observe an O( /)overhead, which deending of value could be significant, desite of the cost incurred by the changes and mechanisms generated by the addition of hierarchical routing to MAET. In addition we can exect a reduction of the toology acets (bandwidth) for the case of roactive rotocols, and a reduction in the delay incurred to acquire a route for the case of reactive rotocols. IV. COCLUSIOS We resented a scenario in which adding hierarchical routing to MAET roduces a reduction in the routing overhead. We also described the architecture (identify the modifications and the new mechanisms) needed to add the hierarchical routing. Finally we demonstrated analytically the exected reduction in the overhead generated by the MAET with hierarchical routing under a secific scenario. We notice that this study only obtains reliminary results, that allow us justify the current efforts to integrate the hierarchical routing to MAET. Some of the activities are: the subnet formation and address acquisition, the mobility of nodes between subnets, the intra-subnet routing, and the inter-subnet routing (modifications to the existent routing rotocols). We thin that an effort to integrate existing schemes with the corresonding modifications to suort subnetting in MAETs could be valuable. REFERECES [1] Mobile Ad-hoc etwors woring grou IETF URL: htt://www.ietf.org/html.charters/manetcharter.html. [] C. Perins, E. Belding-Royer, and S. Das. Ad hoc Vector (AODV) Routing. RFC 3561, July 003. [3] T. Clausen and P. Jacquet, "Otimized Lin State Routing Protocol", RFC 366, October 003. [4] H. Aiache et al, WIDES System Secification, Deliverable., IST WIDES, June 004 [5] C. Shiflet, E. M. Belding-Royer and C. E. Perins. "Address Aggregation in Mobile Ad hoc etwors." Proceedings of the IEEE International Conference on Communications (ICC), Paris, France, June 004. P3/8

[6] K. Weniger, M. Zitterbart. "IPv6 Auto configuration in Large Scale Mobile Ad-Hoc etwors", Proceedings of Euroean Wireless 00, Florence, Italy, Feb. 00. [7] M. Mohsin, R. Praash. IP Address Assignment in a Mobile Ad Hoc etwor, IEEE Military Communications Conference (MILCOM 00), volume,. 856-861, October 00. [8] S. esargi, R. Praash. MAETconf: Configuration of Hosts in a Mobile Ad Hoc etwor. Proceedings of IFOCOM, 00. [9] S. Thomson and T. arten, "IPv6 Stateless Address Auto configuration" in RFC46 (December 1998). [10] J. Rosenberg, H. Schulzrinne, E. Schooler, M. Handley, G. Camarillo, A. Johnston, J. eterson, R. Sars, Session Initiation Protocol, RFC 361 in IETF, June 00. [11] E. Wedlund, H. Schulzrinne, Mobility Suort using SIP. IEEE/ACM Multimedia conference WOWMOM 1999. [1] T. Dreibholz, A. Jungmaier, M. Tüxen, A new Scheme for IP-based Internet Mobility, The 8th Annual IEEE Conference on Local Comuter etwors (LC) Königswinter, Germany ovember 003. [13] R. Stewart, M. Ramalho, Q. Xie, M. Tüxen, I. Rytina, M. Belinchon, and P. Conrad, Stream Control Transmission Protocol (SCTP) Dynamic Address Reconfiguration. IETF, Transort Area Woring Grou, February 003. draft-ietftsvwg-addi-sct-07.txt, wor in rogress. [14] Y. Sun, E. M. Belding-Royer, A Study of Dynamic Addressing Techniques in Mobile Ad-Hoc etwors, Wireless Communications and Mobile Comuting, Aril 004. [15] L. Viennot, P. Jacquet, T. H. Clausen, "Analyzing control traffic overhead versus mobility and data traffic activity in mobile Ad-Hoc networ rotocols," resented at ACM Wireless etwors journal (Winet), 004. [16] B. Jabbari, "Teletraffic asects of evolving and next-generation wireless communication networs," Personal Communications, IEEE [see also IEEE Wireless Communications], vol. 3,. 4, 1996. [17] L. Kleinroc, F. Kamoun, "Hierarchical Routing for Large etwors: Performance Evaluation and Otimization," Comuter etwors, Vol. 1,. 155-174, 1977. [18] D. Hong and S S Raaort, "Traffic Model and Performance Analysis for Cellular Mobile Radio Telehone Systems with Prioritized and on-rioritized Hand-off Procedures," IEEE Trans Vehfc Tech, vol 35,no 3, Aug 1986,. 77-9. [19] S. S. Raaort, "The Multile-Call Hand-off Problem in High-Caacity Cellular Communications Systems," IEEE Trans. Vehic. Tech., Aug. 1991, vol. 40, no. 3,. 546-57. [0] G. Morales-Andres and M. Villen-Altamirano, "An Aroach to Modelling Subscriber Mobility in Cellular Radio etwors," Telecom Forum87, Geneva, Switzerland, ov. 1987. [1] R. Thomas, H. Gilbert, and G. Maziotto, "Influence of the Moving of the Mobile Stations on the Performance of a Radio Mobile Cellular etwor," Proc. 3rd ordic Seminar on Digital Land Mobile Radio Comm., Set. 1988. P3/9