Opportunistic multihop communication using mobile platforms for very sparse infrastructures
FRANC ¸ OIS WILLAME
Master’s Degree Project Stockholm, Sweden 2005
Opportunistic multihop communication using mobile platforms for very sparse infrastructures
FRANC ¸ OIS WILLAME
Master’s Degree Project March 2005 TRITA–S3–RST–NONO ISSN 1400–9137 ISRN KTH/RST/R--NO/NO--SE
Radio Communication Systems Laboratory Department of Signals, Sensors and Systems
Abstract In this thesis, a mobile communication system based on a sparse infrastructure is addressed. The cost of a wireless infrastructure has been shown to be proportional to the service area covered and to the data-rate provided to the users. Therefore, if an operator wants to increase the data-rate per user while maintaining a low cost, the number of Access Points deployed has to be reduced. The Infostations concept which was already used to provide high data-rate to small disjoint areas, is used and extended with Multihop Capable Nodes. These nodes have additional memory capacity dedicated to “store-and-forward” the messages until their delivery. Considering a non delay sensitive service but requiring high data-rates (a delay tolerant multimedia game), the investigation is conducted in both a linear and a sparse-hexagonal environments with simple models of traffic and mobility. The linear environment can be compared to a street or highway where the nodes have a constrained mobility whereas the sparse-hexagonal is a partially covered hexagonal environment which resembles a cellular network, but with lower AP (Access Point) density than what required for providing complete coverage. The goal of the thesis is to determine how low the coverage can be but still provide interesting performances. System performances are evaluated through the delivery rate (messages successfully delivered) of messages, relative delay to deliver a message, memory occupation and number of hops performed by a message to be delivered. Simulation results show that the system can provide good performances for delivery rate and relative delay in some scenarios, but at the expense of higher memory occupation. The system uses the multihopping and the mobility of the nodes differently to combat the lack of coverage depending on the environment and the user density.
iii
This page intentionally contains only this sentence.
Acknowledgements First of all, I would like to thank my advisor Pietro Lungaro for his time, his advice and valuable ideas during this thesis. I am highly thankful to my examiner, Prof. Dr. Jens Zander, for giving me the opportunity and the resources needed to pursue this thesis. Thanks as well to everyone else in the RST and Wireless@KTH departments that have helped me in anyway. I would like to extend a warm thank you to my family, my girlfriend Nina Norrg˚ ard, and my friends Jeremy Lain´e, M´elanie Debouche, Jonas Johansson, Rochdi Zouakri and Estifanos Haile for their support and encouragements throughout my studies. I could not have done it without you. I am really grateful to your help. Finally, I would like to thank my opponent Thomas Ess´en for his suggestions to improvements and help to find typos.
v
This page intentionally contains only this sentence.
Contents 1 Introduction 1.1 General background . . . . . . 1.2 Advantages . . . . . . . . . . . 1.2.1 Economical advantages 1.2.2 Applications . . . . . . 1.3 Related work . . . . . . . . . . 1.3.1 Multihop Capable Node 1.3.2 Multiuser Diversity . . . 1.3.3 Infostations Concept . . 1.3.4 Other related works . . 1.4 Problem Definition . . . . . . . 1.5 Thesis outline . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
1 1 2 2 3 5 5 5 5 6 6 7
2 Context and system models 2.1 Sparse infrastructure and AP (Access Point) . . . . 2.2 Environmental settings . . . . . . . . . . . . . . . . . 2.2.1 Linear environment model . . . . . . . . . . . 2.2.2 Sparse-cellular model . . . . . . . . . . . . . . 2.3 Users’ terminals . . . . . . . . . . . . . . . . . . . . . 2.3.1 MCN (Multihop Capable Node) . . . . . . . 2.3.2 Nodes mobility . . . . . . . . . . . . . . . . . 2.4 Messaging model . . . . . . . . . . . . . . . . . . . . 2.4.1 Service . . . . . . . . . . . . . . . . . . . . . . 2.4.2 Traffic in simulation . . . . . . . . . . . . . . 2.4.3 Transmission . . . . . . . . . . . . . . . . . . 2.4.4 Routing . . . . . . . . . . . . . . . . . . . . . 2.4.5 Structure of a message . . . . . . . . . . . . . 2.5 Reliability and limits of the model . . . . . . . . . . 2.5.1 Dependency on the time of simulation . . . . 2.5.2 Dependency on the initiation of the simulator 2.5.3 AP-MCN before MCN-MCN transmissions . 2.5.4 Risk of flooding or Memory overflow . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . .
9 9 9 9 10 11 11 11 12 12 12 12 13 14 14 14 15 15 15
3 Performance measures 3.1 Successful Transmission 3.2 Relative delay . . . . . . 3.3 Buffer occupation . . . . 3.4 Number of hops . . . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
17 17 18 18 18
. . . .
. . . .
. . . .
. . . .
vii
. . . . . . . . . . .
. . . .
. . . . . . . . . . .
. . . .
. . . . . . . . . . .
. . . .
. . . . . . . . . . .
. . . .
. . . . . . . . . . .
. . . .
. . . . . . . . . . .
. . . .
. . . . . . . . . . .
. . . .
. . . . . . . . . . .
. . . .
. . . . . . . . . . .
. . . .
. . . . . . . . . . .
. . . .
. . . . . . . . . . .
. . . .
. . . .
viii
Contents
4 Simulation method and results 4.1 Simulation method . . . . . . . . . . . . . . . . . 4.1.1 AP density . . . . . . . . . . . . . . . . . 4.1.2 MCN density . . . . . . . . . . . . . . . . 4.2 Simulation results . . . . . . . . . . . . . . . . . 4.2.1 Successful delivery rate . . . . . . . . . . 4.2.2 Relative end-to-end delay . . . . . . . . . 4.2.3 Multihopping versus “physically carrying” 4.2.4 Buffer occupation . . . . . . . . . . . . . . 4.2.5 Number of hops . . . . . . . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
. . . . . . . . .
19 19 19 19 20 20 22 24 24 28
5 Conclusion and future work 31 5.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 5.2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 References A Opposition Report
33 37
List of Tables 2.1
Values of the normal distributions of the mobility profiles . . . .
12
4.1 4.2
Required coverage values for 50s and 100s delay profiles . . . . . Buffer occupation for the 50s and 100s delay profiles . . . . . . .
22 26
ix
This page intentionally contains only this sentence.
List of Figures 1.1 1.2 1.3
Overview of Range and Data Rate of Wireless Technologies. . . . Mercury Wideband Network Radio (WNR) . . . . . . . . . . . . System description . . . . . . . . . . . . . . . . . . . . . . . . . .
2 4 6
2.1 2.2 2.3
Street environment model . . . . . . . . . . . . . . . . . . . . . . Sparse-cellular environment model . . . . . . . . . . . . . . . . . Categorization of Ad-Hoc Routing Protocol . . . . . . . . . . . .
10 10 13
4.1
Successful delivery rate for the linear and sparse-hexagonal models, MCN and SCN (Single hop Capable Node). . . . . . . . . . . Relative end-to-end delay for the MCN and SCN cases. . . . . . Multihopping versus “physical carrying” . . . . . . . . . . . . . . Average and maximum buffer occupation . . . . . . . . . . . . . Distribution of the number of hops performed to deliver a message
21 23 25 27 29
4.2 4.3 4.4 4.5
xi
This page intentionally contains only this sentence.
List of Acronyms AP . . . . . . . . . . . . Access Point GPS . . . . . . . . . . Global Positionning System ISM . . . . . . . . . . . Industrial, Scientific, and Medical MCN. . . . . . . . . . Multihop Capable Node SCN . . . . . . . . . . Single hop Capable Node TTL . . . . . . . . . . Time-To-Live
xiii
This page intentionally contains only this sentence.
Chapter 1
Introduction 1.1
General background
Ubiquity of Computing and increasing demand With the increasing ubiquity and sophistication of mobile devices (mobiles phones, PDA or laptops), a demand in network and multimedia related applications can be expected. End users will probably expect from this “Wireless Internet”, a service comparable to the Quality-of-Service provided by their broadband internet access on their home personal computers. The existing solutions and their limits As shown in Figure 1.1, the existing cellular systems will not be able to fulfil the requirements of very high data-rates services. These cellular communication systems are mainly designed to provide telephony or messaging services. The services they provide are based on very low latency or real-time, bi-directional and interactive communications. They do not require high bandwidth but instead continuity and high coverage, in order to minimize delay and avoid interruptions during a phone calls. Services requiring low to medium data-rates such as e-mailing with attached files, sending pictures, videos or music, files sharing, or news reading can already be provided by those 2G-3G cellular systems. A solution: a Multihop ad hoc network One solution to provide higher data-rates is to adopt a Multihop ad hoc architecture. A Multihop ad hoc network consists of nodes which can communicate with each other. It relies on other nodes to transmit/relay the informations allowing a message to be delivered to its end-user via multiple hops. An appropriate definition of a mobile ad hoc network can be found in the RFC 2501 [2]. A differentiate traffic for a differentiate service Instead of being “available everywhere” (ubiquity of the service), the service is then provided in some areas but with the advantage of providing a higher data-rate (“many-time, many-where” [3]). The purpose of this infrastructure is 1
2
Chapter 1. Introduction
Figure 1.1: Overview of Range and Data Rate of Wireless Technologies [1].
therefore not to provide real-time and ubiquity of service but instead located services focused on exchange of large amount of delay-insensitive data.
1.2 1.2.1
Advantages Economical advantages
Assuring ubiquity of coverage and providing high data-rate for a cellular system is expensive, in terms of deployment of the network and also because it may require expensive licensed bands. Operators may as well want to reduce their expenditures after the high prices paid to obtain the 3G licences in some countries and to get or remain competitive on the market with a good quality/price ratio. Less infrastructures deployed for a lower cost According to [4], the costs (Csystem ) of a wireless infrastructure can basically be broken down into the following factors : Csystem ≈ c · NAP ≈ c0 · Nusers · Buser · Aservice · f (Q)
(1.1)
where NAP is the number of Access Points and Nusers is the number of users. Operators do not want to loose customers but instead increase their numbers of clients (Nusers ) as it is also proportional to to their income. If the operators want to increase the data-rate per user (Busers ), while maintaining the same order of magnitude for the infrastructure costs (Csystem ), or the Quality-of-Service has
1.2. Advantages
3
to be lowered (decrease (f (Q))), or the service area covered has to be reduced (smaller Aservice ). Reducing the service area covered can be done by reducing the number of AP deployed, but this will affect the Quality-of-Service (f (Q)). The services will only be available on some sparse areas where the operator has decided to deploy AP. In such sparse infrastructures, a reduction in the cost may then be achieved by compromising on the offered Quality-of-Service (f (Q)) to the end-user. But this has to be done in such a way that the service is still desirable to the user. This can be done by taking advantages of “human misperceptions” so as to cover the service’s defaults (see 2.4.1). Less required resources A multihop network requires less resources as compared to a cellular system: • no or less planning in advance of capacity/coverage. • reduced sites investments (less AP reduces the costs of construction, rents or maintenance of the sites). • more flexibility concerning the deployment or the administration. • With its distributed architecture, It requires less or no central control and avoids planning of resources. Its self-deployment and adaptive capabilities make this network easier and quicker to setup. • use of the free ISM (Industrial, Scientific, and Medical) radio bands to reduce the cost. • use of less expensive radio components (IEEE 802.11 a/b/g radio components are now popular and quite cheap). New market entities With less resources and lower investment required, this solution has a potential faster and higher ROI (return on investment) which might interest companies. But with the emergence of these new market players (smaller operators, local authorities, see as well 1.2.2), multiple operators will compete on the same area (higher competition). On the other side, it could be a good solution for wider operators (national size) to improve weak coverage points or outsource a localized high-data rate service. Through agreements between the national and the local operators, these hotspots are then integrated in a larger overlaying network and keep the low data-rate traffic (voice, telephony) for these overlaying cellular network (traffic differentiation).
1.2.2
Applications
Military - tactical Telecommunications are critical targets in military situations. Moreover, for military applications, they need to be reliable, flexible and secure. The attackant will probably destroy its enemy’s telecommunications infrastructures so as to make him unable to communicate. In the same way the
4
Chapter 1. Introduction
GPS (Global Positionning System) was established to localize and coordinate troops, the Mercury Wideband Network Radio (WNR) [5] can be used to setup a rapid and flexible local network. WNR relies on an ad hoc networking technology and gives the advantage to communicate on the enemy’s own territory (see Figure 1.2). On the defensive’s side, the infrastructures have most probably been damaged or destroyed and need to be replaced rapidly. If they remain, they are most probably spied by the attackants’ Intelligence Services.
Figure 1.2: Mercury Wideband Network Radio (WNR) [5].
Emergency services or developing countries Multihop ad hoc networks can also be used to quickly build up a primary or emergency informations structures in some public applications. For example, Police, rescue teams or fire department units sometimes have to operate in areas where no information infrastructures are present and operations still need to be coordinated. It can also be deployed for developing countries [6] or after a disaster used by emergency services.
Multimedia Almost any mobile device (PDA, mobile phones . . . ) now has miniaturized and sophisticated multimedia capabilities (camera, microphone . . . ). The movies, pictures or MP3 files are often transfered to a more usable or comfortable device (screen of the TV or the PC). This could be done using an ad hoc technology [7].
1.3. Related work
5
Automation Automation can also benefit from ad hoc technology. Robots are more and more mobile and could profit from ad hoc network to be independent of a fixed communication environment, which will increase their mobility. Other applications Multihop ad hoc networks can also have commercial oriented applications. The interest is not to replace the existing infrastructures, which are more advantageous in terms of bandwidth and security, but instead adding flexibility and avoid wires. Instead of providing internet access points, airports, train stations, fairs or conferences centers can provide a punctual and localized service to their clients, or in the case of offices [7] to the employees. For information or advertisement applications, motorway companies might want to inform about traffic conditions (warning of accidents [8], traffic-jam, slow-down of the traffic . . . ) or shopping centers who want to send advertisement to the customers within the malls. Another applications is the field of biological surveys [9, 10] or provide a communication system to nomadic populations [11, 12].
1.3
Related work
1.3.1
Multihop Capable Node
As shown in [13], a model based on “store-and-forward” capable terminals can extend the AP coverage and, as it allows self-organization and self-configuration of multihop cells, this architecture could also provide a rapid and easy way to install and set up an infrastructure.
1.3.2
Multiuser Diversity
One important factor of the multiuser diversity effect [14] is the mobility of the terminals. For immobile terminals (nodes), Gupta and Kumar have demonstrated in [15, 16] that when the number of nodes n increases, the throughput decreases with a rate √1n (the network could provide a per-node throughput of √
c0 n log n
bits/sec and that even in the best possible placement conditions of
the nodes, this network could not provide a per-node throughput of more than c00 √ bits/sec), whereas for mobiles nodes, Grossglauer and Ts´e have shown in n [17, 18] that the average long-term throughput per source-destination pairs can be kept constant.
1.3.3
Infostations Concept
The Infostations concept was introduced by the Rutgers Wireless Information Network Laboratory (WINLAB). Infostations are low-cost and low-power AP providing strong signal quality (and so high-bit connections) to small disjoint areas. Mobility of terminals is exploited by allowing the nodes to transmit at times with larger signal quality to get the improved throughput at the expense of potentially larger delays [3, 19, 20]. In [21], the performances of a highway
6
Chapter 1. Introduction
mobile infostation networks are studied but without multihop capabilities and extension of memory. In this thesis, a similar concept to the Infostations is used but with multihop and “store-and-forward” capabilities.
1.3.4
Other related works
The feasibility of a “store-and-forward” messaging service has been studied in [7]. However this study has been conducted using specific devices (Cybiko Computers), specific indoor environment (school and office style buildings) and a pseudo human mobility model. In this thesis, outdoor environment is considered. The terminals are not specific and both vehicular and pedestrian mobility models are used. In [9], a Shared Wireless Infostation Model (SWIM) is proposed for the biological survey of whales population. SWIM allow additional improvement in the capacity-delay tradeoff through a moderate increase of the storage capacity. The contact rates between the whales is used to calculate the storage requirements. Here as well, the thesis differs mainly from the environment and the mobility models studied.
1.4
Problem Definition
This thesis will focus on delay insensitive services but requiring high data-rates in a multihop ad hoc network. The service considered is a non-realtime multiparty multi-player game, with rather low interactivity between the players. A good example is a strategic game with high multimedia features (sounds, dynamic maps . . . ) and so requiring high data-rates and with several players located in a given area (a group-team of players). In such a strategic game, players have to play by turn, one after the other, and then wait for other players’ responses. Once a player has decided of an action, the information has to be communicated to the other players, or at least, the next-turn player and a central database (to keep track of the players actions).
Central Database
Access Point Player Player
Player
Player
Player
Player Player
Figure 1.3: System description
Player
1.5. Thesis outline
7
One way to reach the central database, is to transmit the information by multiple hops throw the players’ devices. The disadvantages of this multihop Ad Hoc networks are the time taken to propagate the information over the network and the risk of “isolated or unreachable” groups of players. The time of reflection needed by a player to decide of an action can be used as an advantage to “hide” the delay introduced by the spreading over the network. As well, several other “human misperceptions” can be used as advantages to cover the absence of full coverage (see 2.4.1). We assume as well the possibility of extending the memory of the nodes in order to increase the number of replicas (copies) of the message in the network. Increasing the number of replicas of the message in the network may increases the probability of successfully deliver the message to an AP and therefore reduce the delay. As the message needs to be diffused and replicated in other nodes memory, this will require larger storage capacity in the terminals, but it can be considered technically feasible taking into account the reduction of the cost and the size of memory (“ Moore’s Law”). However, if the nodes are far from the AP, or if there is only one node in the network, the message will have to be “physically carried” within an AP range. The goal of this thesis is to evaluate how sparse the infrastructure could be so that the service described above can still be provided and desirable by the players. For this purpose, we will evaluate the system on three points: • the reliability. The system will be considered satisfactory if 95% of the packets generated are received. • the end-to-end delay - The time taken to transfer a packet from the terminal who generated it until its final delivery to an AP. • the memory cost. The increase of memory that the users’ terminals will require to handle the lack of AP.
1.5
Thesis outline
The system that is to be studied is described along with the assumptions in Chapter 2. We define how we will measure the performances of the system studied in Chapter 3. Chapter 4 shows and evaluates the results obtained from the simulations. In Chapter 5, we summarize the conclusions of this study and give some ideas for future work.
This page intentionally contains only this sentence.
Chapter 2
Context and system models In this chapter, the system models will be described with the common parameters followed by the specificities of each model and their limitations.
2.1
Sparse infrastructure and AP (Access Point)
The AP (Access Point) are low power and low coverage range base-stations, providing high data-rate. With a low density of these AP, and as their coverage is low, the network is not completely covered. As the density gets lower, we come from situation of “lack” or “holes” in the coverage to just “spots of coverage”, partitioning the network in “islands of coverage”. This scenario defines the sparsity factor which will be studied in the rest of the thesis, varying from 5% (very sparse) to 100% (full) coverage.
2.2
Environmental settings
Two different environmental settings are considered for the simulations : • a street or motorway like model, referred as the linear model. • a cellular based model, referred as the sparse cellular model. The modelling of these environmental settings is described below.
2.2.1
Linear environment model
Environment description The linear model is a one dimensional environment and has the most restricted mobility model out of all the environment studied. It is based on a similar model described in [21], but the street length is fixed and the AP are placed equidistantly from each others. With this AP distribution and wrapping, a MCN moving with a constant speed will see an AP at a fixed intervals of time. 9
10
Chapter 2. Context and system models
MCN mobility and distribution At the initiation of the simulation, MCN are randomly placed along the street and affected with the motion of one of the two mobility profiles (pedestrian or vehicular). Both forward and reverse traffic are considered and MCN are wrapped around at the extremities of the street (taking the street in the reverse direction).
v x −R
−2R
−r
0
r
R
+2R
Figure 2.1: Street environment model
2.2.2
Sparse-cellular model
The hexagonal model has the less constraint of the three models studied.
800 600 400 200 0 −200 −400 −600 −800
−1000
−500
0
500
1000
Figure 2.2: Sparse-cellular environment model
Environment description This environment √ is based on a classical hexagonal cellular network, with cells as hexagons of 1002 3 meters side. The network used for the simulation is composed of 91 cells and wrapped around. The AP are randomly distributed at the interstices of the hexagonal cells, with only one AP possible by interstice.
2.3. Users’ terminals
11
MCN distribution and mobility The MCN have less constraint on their mobility in this model. At the initiation of the simulation, they are randomly distributed over the whole area and affected with a given speed according to the pedestrian or vehicular profile they belong to and move according to the random walk mobility model. Validity and limitations There is an extensive literature on mobility models in ad hoc networks [22, 23, 24, 25, 26, 27, 28], specially on the walk model. The possibility of users stopped (sitting) is not considered, even thought they might be the most consumers at this time, and users in a vehicle are taken into account even thought it should be forbidden for security reasons.
2.3 2.3.1
Users’ terminals MCN
MCN (Multihop Capable Node) are terminals which can communicate with each other outside the coverage area of an AP by relaying. MCN can store messages and then forward them when they come within the communication range of an AP or another MCN. The system studied relies on the “users diversity” and following characteristics of the MCN: • their mobility. • their capability of “store-and-forward” a message. • their multihop capability.
2.3.2
Nodes mobility
In such a network, the mobility of the MCN has a high influence on the performances of the system in a sparse infrastructure. Additionally, the MCN are not aware of the location of the AP and then do not move according to where there is coverage. Two types of mobility profiles are considered: A MCN with a low mobility may stay in or outside a coverage area for a long period of time. Consequently the message it generates will be directly delivered to the AP or it will take longer time depending on if it can transmit it to another MCN via multihop in the vicinity or have to “physically carry” the message via moving to an AP if it is alone in the area. The MCN with low mobility profile are referred in the rest of the thesis as pedestrian MCN. On the other hand, MCN with a higher mobility will move with a higher speed and may come more often in covered areas or meet more AP on their way and consequently deliver faster the messages in their buffer. They may also meet other MCN and act as “collectors” of the messages of the low mobility MCN. High mobility MCN are faced to a memory trade-off between collecting other MCN’s messages and delivering messages faster. The MCN with high mobility are referred as vehicular MCN.
12
Chapter 2. Context and system models Mobility pedestrian vehicular
Mean 3 km/h 50 km/h
standard deviation 0.3 km/h 2.5 km/h
Table 2.1: Values of the normal distributions of the mobility profiles
2.4 2.4.1
Messaging model Service
The services which can be provided are limited by occurrences of delays and interruptions inherent to the system, but can provide heavy bandwidth consumption applications. Typically, services dealing with voice, or direct communication can not be offered with the same quality provided by cellular systems. During a phone call for example, interlocutors do not tolerate delay and interruptions while talking. The system is not suitable for services such as telephony, video telephony or video conferencing. If human perception is a limiting factor in this case, it could be taken as an advantage for other applications, focusing users’ attention while continuing downloading the rest of the application. Such applications can be listening to music (starting and listening a MP3 while downloading the rest of the file), reading news (reading the headlines while downloading the rest of the contents). It could be possible to wait until a message or an email is completely downloaded before notifying the user. Services which do not require ubiquity of coverage, which are not delay sensitive but require high data rates can be provided by the system. Typical commercial applications are messaging service, movies or news content provider or advertisement.
2.4.2
Traffic in simulation
The packets are assumed to be of constant size and arrive according to a Poisson process with total (external) rate λ. For N nodes in the network, each node i generates packets with the rate : λ where i ∈ {1, 2, . . . N } (2.1) N No considerations are made about the fragmentation of the data or on the packets size optimization. All further measurements referring to packets are in “units” or number of packets. λi =
2.4.3
Transmission
It is most likely that this kind of networks will use the unlicensed bands. Some of the existing standards could be used: IEEE 802.11 x These were developed by the IEEE. The IEEE 802.11 b/g use the 2.4GHz band and the IEEE 802.11 a uses the 5GHz band and provide throughput from 2 Mbit/s up to 54 Mbits/s. HiperLAN/2 HiperLAN/2 was established by the ETSI. A cell of a HiperLAN/2 AP typically extends to 30 (office indoor) to 150 metes.
13
2.4. Messaging model
The coverage ranges of AP and MCN are assumed to be 100 meters and 50 meters. Transmissions are assumed error-free and without collisions. Resubmissions of packets are not considered. The transmission time is neglected and all the packets are assumed to be transmitted at once. If we consider the disadvantageous case of two vehicular MCN moving toward each others at 50km/h, a sufficient amount of data can still be transmitted. These two vehicular MCN are in radio contact during 3.6s and can exchange 180Mbits (at 50Mbits/s).
2.4.4
Routing
In a multihop ad hoc network with mobiles nodes, the “available routes” to deliver a message are changing rapidly and dynamically according to the MCN’s mobility. Several routing protocols have already been proposed and studied: AODV (Ad hoc On Demand Distance Vector) [29], DSR (Dynamic Source Routing) [30], OLSR (Optimized Link State Routing Protocol) [31], TBRPF (Topology Dissemination Based on Reverse-Path Forwarding) [32], DSDV (Destination Sequenced Distance Vector Routing) [33]. A review of the current routing protocols can be found in [34, 35]. Ad-Hoc Routing Protocols
Table Driven
DSDV CGSR
WRP
Source-initiated On-Demand Driven
AODV
DSR
LMR
ABR
TORA
SSR
Figure 2.3: Categorization of Ad-Hoc Routing Protocols [34]
In this thesis, the routing algorithm adopted is a “blind” broadcasting or epidemic routing [36, 37]. Once a MCN generates a message, it broadcasts it to its neighboring MCN. Upon receiving a message, there are two possible actions to be taken: • If the receiving MCN has no previous knowledge of the new message, the message is kept in the MCN buffer and its number of hops flag is increased of 1. The message, as well as all the other messages in the MCNs buffer, is then re-broadcasted to the other MCN at the following iteration. • If the MCN has already the received message in its buffer, it simply discards it. The flags are kept to the values of the message already in the buffer (no increase of the number of hops). The neighbors in turns then transmit to their neighbors and so on until the message has been propagated over the entire network. Once the message is delivered to an AP, the message is erased from all the MCN’s memory (totally erased from the network).
14
Chapter 2. Context and system models
2.4.5
Structure of a message
The message has several flags in order to keep track and share information with other replicas. the identity number flag contains the identity number of the message. It is a unique number attributed when the message is created and will be shared with the other replicas. the initiation time indicates at which instant the message is generated. This flag is also common to the other replicas. the initiation mobile indicates the identity number of the MCN which have created the message. This flag is also common to the other replicas. the number of hops indicates the number of hops performed by the message and each of its replicas. It is different from one replica to another. Conventions • Only the original message in the buffer of the mobile which created it has its number of hop equal to 0. • If a message is directly transmitted to an AP, its number of hop equal to 1. Direct transmission is regarded as a one hop transmission. • Replicas of a message received by MCN which is already in the MCN’s buffer (same identity number) are discarded. The MCN has already received this specific message in the past so it is kept in the MCN’s buffer and its corresponding number of hops performed is not updated.
2.5
Reliability and limits of the model
2.5.1
Dependency on the time of simulation
By implementation, the simulation tool is dependant on the time of simulation. During the simulation time (ti to tf ), a certain number of packets are generated, according to the rate λ. The first packets are received at ti + δti . At tf , the simulation is stopped. The remaining packets received after (between t f and tf + δtf ) are not taken into consideration, in order to respect the constant rate λ. 6
simulation ti
δti
tf
δtf
2.5. Reliability and limits of the model
2.5.2
15
Dependency on the initiation of the simulator
As evoked in [28], the initial random distribution of the MCN is “not representative of the manner in which nodes distribute themselves when moving”. Note that [38] also proposes methods to solve this problem. Instead, the method adopted in this thesis is to run simulation long enough (1 hour) to minimize both the problem of the initiation of the mobility of the MCN and the sides effects of the packets discarded after the end of the simulation evoked in the previous paragraph.
2.5.3
AP-MCN before MCN-MCN transmissions
The transmission between AP and MCN occurs before transmissions between MCN, meaning in the “worst case”. If the MCN-MCN transmissions would have occur before the AP-MCN, the AP’s range would have potentially been extended of the range of the (AP+MCN)’s range.
2.5.4
Risk of flooding or Memory overflow
With the flooding technique adopted, the node transmits a message to all of its neighbors. Hence the performances will depend on the average number of neighbors (neighbor degree) and the traffic in simulation (risk of congestion of the wireless medium). An efficient flooding scheme or a different garbage collector policy based on the time-to-live or the number of hops performed, could have been adopted also. In this thesis, the densities of users and the traffic in simulation are sufficiently low to avoid these problems. The garbage collector policy is to erase only the delivered messages. For information, the performances of efficient flooding techniques are compared in [39].
This page intentionally contains only this sentence.
Chapter 3
Performance measures This chapter extends the description of the parameters studied. The first two have been selected as well in [12] and referred as message delivery ability, message delivery rate.
3.1
Successful Transmission
The criteria selected for a successful transmission of a packet is when one of the replicas of the original packet is delivered to an AP. Once a replica is delivered to an AP, the further delivered replicas are discarded. The replicas of a delivered packet which remain in the network need to be erased for several reasons: • to avoid delivering a packet several times, as they are discarded when arriving to an AP. This saves time and resources. • to reduce the traffic load implied by the spreading of replicas of an already delivered packet. A MCN which has no knowledge that the packet is delivered continues spreading it, provoking unnecessary traffic. • to optimize the MCN’s buffer occupation, by erasing the replicas of a delivered packet, a part of the MCN’s memory will be free. This can be done in two different ways: • By changing the flag in the packet’s header and spreading it over the network again. Only the information in the header remains to alert the other MCN that the message has be delivered. The body of the packet is discarded. This solution has the advantage of being independent (not relying on other network), but the drawbacks of the multihop network remains (delay implied by the spreading and unreachable areas risk). • By relying on a overlaying network, typically a cellular network. This solution has the advantage to avoid spreading again. With this method, the “controlling part” rely on the cellular network, and the transmission of volume of data is insured by the multihop network. The replicas are erased quickly and over the whole network. 17
18
Chapter 3. Performance measures • In a distributed way: the users’ devices take individual decision on how to manage the packets. The decision criteria can be based for example on the TTL (Time-To-Live) (see 4.3) or the number of hops (see 4.5).
In the thesis, the second method is adopted, assuming the existence of an overlaying network in charge of broadcasting an information once a message is delivered. Moreover, the cost of this overlaying network is not taken into account in these thesis.
3.2
Relative delay
The delay is measured as relative time for each successfully delivered packet, this means the time taken from the generation of the original packet at the MCN’s side until the first of its replicas is delivered at a AP’s side. This relative delay measurement can be used to set a TTL while generating the packet. Once the original message or its replicas have exceed this TTL, it is erased from the MCN’s memory. If the message or one of its replicas reach an AP within this TTL, the message is successfully delivered, otherwise the message is lost. A new message has to be generated again and transmitted again.
3.3
Buffer occupation
During the whole simulation, the buffer occupied by the messages is recorded for each MCN. As seen above, once a message is delivered, all its corresponding replicas are erased over the whole network in all MCN’s memory so the recorded buffer occupation includes only packets which are not delivered. Two values are extracted from these recorded data: • the average buffer occupation. The average buffer occupation of each MCN’s devices is computed during the 60 minutes of simulation. The results show on the graphs are the average of these values over all the MCN. • the maximum buffer occupation. The maximum buffer occupation of each MCN’s devices is recorded during the 60 minutes of simulation. The average of this maximum values is then computed over all the MCN.
3.4
Number of hops
For each successfully delivered packet, the number of hops performed by the replica which first reach an AP is stored as well. This number of hops can be used to determine a number of hops to live. The number of hops to live is a the number of hops a packet and its replicas are allowed to perform. The number of hops to live can be set in respect to a probability that the message can be delivered within a given number of hops. It can be used for example if one want to set a garbage procedure so that the message is erased from the MCN’s buffer once it has exceed a given number of hops (this garbage procedure was not implemented in the thesis).
Chapter 4
Simulation method and results 4.1 4.1.1
Simulation method AP density
The network sparsity varies from 5% to 100%. This network sparsity is defined as the ratio between the total area covered by the AP and the total area considered for each model. The simulations are performed for 1 hour to avoid the problems inherent to the time of simulation and the initiation of the simulator (see 2.5). For the street model, the coverage percentage is computed as following 2 · number of AP · range of an AP total length of the street
(4.1)
Here the length of the street is fixed to 2 km and the range of an AP is 100 m. For the sparse hexagonal model, the coverage percentage is computed as following: number of AP · area of an hexagonal cell total area of the sparse hexagonal model
4.1.2
(4.2)
MCN density
For each AP density and model, three users densities were studied (20, 50 and 100 users/area considered). These users densities are shared equally between the two mobility profiles groups (“pedestrian” and “vehicular”) for all the simulations. The user densities chosen are quite low values for a better evaluation of the user diversity/mobility. For too high values of user density, the messages will spread too fast over the network, as a result of the multihop effect, and will not let a proper evaluation of the “physically carried” messages (messages delivered thanks to the mobility of the users). 19
20
Chapter 4. Simulation method and results
4.2
Simulation results
The results are obtained by means of simulations. Results for the successful transmissions and relative end-to-end delay are presented for the three users densities in the MCN and SCN cases. Results concerning the buffer occupations and the number of hops are shown for 20 and 100 users in the network to get clearer results. The messages external arrivals rate is assumed to be λ = 6 msg/min. This parameter does not affect the delivery rate and the delay, but plays a role in the buffer occupation. On the x-axis of each graph, the coverage in percentage can be founded.
4.2.1
Successful delivery rate
We first investigate the successful delivery rate shown in Figure 4.1. The successful delivery rate is the percentage of packets successfully delivered after 1 hour of simulation. Our results obtained are coherent with the percentage of delivery rate in [36]: we also have successful transmission probabilities close to 100%, independently of the AP density. Nevertheless, it must be stressed that these results are dependent on the time of simulation, as explain in 2.5. For a shorter time of simulation, some packets would not have been spread over the network through the other MCN and thus delivered to an AP. Effect of the AP density on the successful delivery rate By comparing the delivery rate for the MCN and the SCN in terms of AP density, two observations can be made: • In very sparse infrastructures, the MCN’s delivery rate outperforms the SCN’s one. The delivery rate is improved of 5 points at 20% coverage and over 12 points at 10% coverage in both linear and sparse-hexagonal models. • The performances for denser coverage are quite similar and reach high delivery performances. Both MCN and SCN based systems deliver more than 95% of the generated packets for coverages higher than 40%. However, these high performances are not significant because the delay and the memory occupation are not taken into account. The delay and the memory occupation will be discussed in the next sections, and highlight the cost in delay and memory used to provide this high data-rate. Effect of the user density on the successful delivery rate From figure 4.1, it can be seen as well that higher density of users increases the probability to deliver a message. This is because messages have more facilities to spread in the network by performing hops on more users. However, as explained above, these high delivery rate will be nuanced by the delay and the memory used to deliver the messages.
21
4.2. Simulation results
percentage of messages successfully delivered (street model, 60 min) 1
nb msg delivered / nb msg generated
0.95
0.9
0.85
0.8
0.75 20 users MCN 50 users MCN 100 users MCN 20 users SCN 50 users SCN 100 users SCN
0.7
0.65
0
10
20
30
40
50
60
70
80
90
100
coverage in % percentage of messages successfully delivered (hexagonal model, 60 min) 1
nb msg delivered / nb msg generated
0.95
0.9
0.85
0.8
0.75 20 users MCN 50 users MCN 100 users MCN 20 users SCN 50 users SCN 100 users SCN
0.7
0.65
0
10
20
30
40
50
60
70
80
90
100
coverage in %
Figure 4.1: Successful delivery rate for the linear and sparse-hexagonal models, MCN and SCN.
22
Chapter 4. Simulation method and results
4.2.2
Relative end-to-end delay
Figure 4.2 represents the 95% percentile of the relative end-to-end delay, meaning the time needed to deliver 95% of the messages generated during the 1 hour simulation. Two delay profiles of 50 and 100 sec. are considered in this section. These two delay profiles are used to determine a TTL of the messages in the network. A first observation shows that the SCN need a higher delay to deliver the messages. It performs with a delay of 200 sec. at 50% coverage for the linear model and more than 300 sec. at 60% coverage for the sparse-hexagonal. This highlights the trade-off mentioned in the delivery performances: the SCN can achieve high delivery rate but at the expense of a much higher delay to deliver the messages. The reason why the MCN outperform the SCN is because of the multihop capabilities of the MCN, which SCN lack. The high delivery rates of the SCN are only due to their mobility and their capacity to store their own messages. The SCN infrastructure is therefore not a good candidate for a lowcost service. The SCN will remain as a benchmark in the rest of the comparison to evaluate the improvement offered by the MCN. The density of users has a significant impact on the relative delay. This end-to-end delay is longer for lower number of users. In the linear environment, the end-to-end delay for 20 users is twice the delay for 50 users, and 4 times longer than the delay for 100 users. The same applies to the sparse-hexagonal environment, with a delay of 100 sec. for 100 users and 200 sec. for 50 users at 10% coverage. With lower user density, each user has less chances to come in the communication range of another user. This means from the message point of view, less opportunities to spread or less opportunities of multihops. As the relative end-to-end delay increases with the lower users density, it comes to a situation that the users are dependent on each other to get a low delay. A operator will have to consider this density of user/delay trade-off in the areas where he wants to provide the service. The MCN infrastructure can be interesting for an operator willing to provide a low-cost service along a motorway. By covering 20% and fixing the TTL of the messages in the network to 100 sec. (Fig. 4.2), the messages are delivered with a probability higher than 99% (Fig. 4.1). The same TTL of 100 sec. and delivery-rate over 99% will have require a coverage of 65% for the SCN. This difference in coverage (20% compared to 65%) represents an important economy for the operator.
linear model Coverage for a delay of 50 sec. Coverage for a delay of 100 sec.
M CN20 40% 15%
M CN50 20% 9%
M CN100 14% 8%
SCN 75% 65%
sparse-hexagonal model Coverage for a delay of 50 sec. Coverage for a delay of 100 sec.
M CN20 82% 68%
M CN50 72% 48%
M CN100 48% 15%
SCN 88% 82%
Table 4.1: Required coverage values for 50s and 100s delay profiles
23
4.2. Simulation results
95% percentile of the relative delay (street model, 60 min) 300
MCN 20 users SCN 20 users MCN 50 users SCN 50 users MCN 100 users SCN 100 users
relative delay in sec
250
200
150
100
50
0
0
10
20
30
40
50
60
70
80
90
100
coverage in % 95% percentile of the relative delay (hexagonal model, 60 min) 300
MCN 20 users SCN 20 users MCN 50 users SCN 50 users MCN 100 users SCN 100 users
relative delay in sec
250
200
150
100
50
0
0
10
20
30
40
50
60
70
80
90
coverage in %
Figure 4.2: Relative end-to-end delay for the MCN and SCN cases.
100
24
4.2.3
Chapter 4. Simulation method and results
Multihopping versus “physically carrying”
When the network coverage is low and the number of users is not sufficient to use them as multihop-relays, some of the messages are “physically carried” by the users themselves. This situation is illustrated in figure 4.3. Some precisions have to be made about the messages “physically carried”. The messages considered as “physically carried” are only the messages delivered and generated by the same user. The other condition is that this user did not have radio contact (not in the coverage of an AP) when he generated the message. The message is stored in the buffer of the user’s terminal until the user moves and gets in radio contact with an AP. Any message which has been delivered by performing one or more hops is not considered “physically carried”, but delivered via multihop. In the linear environment, the effect of multihop is dominant in the 100 users case. In this case, messages are delivered using multihop to cope with the lack of coverage instead of using the mobility of the users. Less that 7% of the messages are physically carried. The case of the 20 users in the linear model and the 100 users in the sparsehexagonal model are similar. Messages are delivered with the combined effect of multihopping and the user mobility. The messages delivered thanks to the mobility of the users reach its maximum at 26% of the messages delivered at 15% of coverage in the linear for 20 users and the maximum for 100 users is 22% at 30% coverage in the sparse-cellular environment. In the sparse-hexagonal environment with 20 users, the multihops and the mobility of the users are also combined to deliver the messages but the mobility has more important impact on how the messages are delivered. There are more messages “physically carried” than delivered by multihops for a coverage higher than 15%. The percentage of messages delivered thanks to the mobility of the users reaches its maximum of 42.7% at 20% coverage.
4.2.4
Buffer occupation
The results of the simulations for the average and maximum buffer occupation are shown in figure 4.4.These results are for both MCN and SCN with 20 (left) and 100 users (right) in the linear and sparse-hexagonal models. Effect of the user density on the buffer occupation For coverage higher than 15%, the effect of the user density is different for the linear and the sparse-hexagonal environments. For the linear model, the buffer occupation is higher for higher number of users. At 15% coverage for example, the buffer of the terminals are occupied by 11.6 messages on average (maximum 69.1 messages) for the 20 users, and 21.7 messages on average (max 103.6 msg) for 100 users. Whereas, for the sparse-hexagonal model, the buffer occupied is lower for higher number of users. At the same coverage as above (15%), the average buffer occupation is 12.3 messages (max 90.8 msg) for 20 users and 7.8 messages (max 80.1 msg) for 100 users.
Percentage of messages "physically carried" (street model, 20 users, 60 min)
Percentage of messages "physically carried" (street model, 100 users, 60 min) 1
nb msg physically carried / nb msg generated
20 users, physically carried 20 users, multihops
0.8
0.6
0.4
0.2
0
0
10
20
30
40
50
60
70
80
90
0.8
0.6
0.4
0.2
0
100
100 users, physically carried 100 users, multihops
4.2. Simulation results
nb msg physically carried / nb msg generated
1
0
10
20
30
40
coverage in % Percentage of messages "physically carried" (hexagonal model, 20 users, 60 min) 1
60
70
80
90
100
Percentage of messages "physically carried" (hexagonal model, 100 users, 60 min) 1
20 users, physically carried 20 users, multihops nb msg physically carried / nb msg generated
nb msg physically carried / nb msg generated
50 coverage in %
0.8
0.6
0.4
0.2
100 users, physically carried 100 users, multihops
0.8
0.6
0.4
0.2
25
0
0
10
20
30
40
50
60
70
80
90
100
0
0
10
20
30
coverage in %
40
50 coverage in %
Figure 4.3: Multihopping versus “physical carrying”
60
70
80
90
100
26
Chapter 4. Simulation method and results
linear model M CN20 M CN100 SCN sparse-cellular model M CN20 M CN100 SCN
delay
coverage
50 s 100 s 50 s 100 s 50 s 100 s
40% 15% 14% 8% 75% 65%
delay
coverage
50 s 100 s 50 s 100 s 50 s 100 s
82% 68% 48% 15% 88% 82%
delivery rate 99.7% 98.9% 99.5% 98.8% 99.8% 99.6%
Buffer occupation average maximum 2.2 msg 23.3 msg 11.6 msg 69.1 msg 21.8 msg 105 msg 100 msg 300 msg 0.6 msg 8.2 msg 1.3 msg 115 msg
delivery rate 99.6% 99.5% 99.6% 99.1% 99% 99%
Buffer occupation average maximum 1.2 msg 21 msg 2 msg 26 msg 1.5 msg 27 msg 7.8 msg 80 msg 1.5 msg 15 msg 2.2 msg 21 msg
Table 4.2: Buffer occupation for the 50s and 100s delay profiles
This difference of buffer utilization between the two environments can be explained by the contact rate between the users while moving and the delay to deliver a message (see 4.2.3). In the linear environment, the users movement is bounded to a street. Hence, the probability to get in radio contact with other user, moving in the same or opposite direction, increases as the density of users on the road increase. With its bounded movement, a single user has also more chances to get in radio contact with an AP, thus delivering its own messages and the other messages collected from other users. In the sparse-cellular environment, the users have more freedom in their mobility and the contact rate with the other users and AP is lower. The messages generated have less chances to spread through the other users, specially with lower user density, thus less chances to be delivered. The messages not delivered remain in the buffer. Additionally, the delay is longer for lower user densities in the sparse-cellular environment (fig. 4.2). This means that for lower user densities, we have a larger accumulation of created messages. Effect of the AP density on buffer occupation Table 4.2 illustrates the tradeoff between the coverage and the required buffer on the terminals to provide a low-cost service. For a coverage of 8% in the linear environment, the terminals’ buffer will be filled of 100 messages on average with burst of 300 messages. Considering the arbitrary arrival density of 6 msg/min, the results are very specific. However, it quantifies how large the dedicated memory of the terminals should be. From the operator’s side, assuming that the cost of an infrastructure is proportional to the coverage, and on the user’s side, the cost is proportional to the amount of memory invested in its terminal, part of the cost of this service is shifted from the operator to the users’ side.
buffer occupation of the terminals (street model, 20 users, 60 min) 200
buffer occupation of the terminals (street model, 100 users, 60 min) 200
MCN maximum buffer SCN maximum buffer MCN average buffer SCN average buffer
number of packets
number of packets
150
100
50
0
4.2. Simulation results
150
MCN maximum buffer SCN maximum buffer MCN average buffer SCN average buffer
100
50
0
10
20
30
40
50
60
70
80
90
0
100
0
10
20
30
40
coverage in % buffer occupation of the terminals (hexagonal model, 20 users, 60 min) 200
60
70
80
90
100
buffer occupation of the terminals (hexagonal model, 100 users, 60 min) 200
MCN maximum buffer SCN maximum buffer MCN average buffer SCN average buffer
MCN maximum buffer SCN maximum buffer MCN average buffer SCN average buffer
150
number of packets
150
number of packets
50 coverage in %
100
50
100
50
27
0
0
10
20
30
40
50
60
70
80
90
100
0
0
10
20
30
coverage in %
40
50 coverage in %
Figure 4.4: Average and maximum buffer occupation
60
70
80
90
100
28
4.2.5
Chapter 4. Simulation method and results
Number of hops
Figure 4.5 illustrates the distribution of the number of hops for the linear and sparse-hexagonal models. Results are shown for 20 and 100 users in the network. The size of the histograms represent how the messages are delivered. The counting of number of hops performed is limited to 10 hops. Effect of the AP density on number of hops It was intuitive that the number of hops performed to deliver a message would increase with the sparsity of the network: as the users have less possibility to deliver their messages directly to an AP when the network get more sparse, they have to use multihopping to deliver their messages. Effect of the user density on the number of hops The number of users has an important impact on the number of hops performed to deliver a message. For a coverage less than 20%, a message performs more hops in environment with higher density of users. As shown in figures 4.5, most of the messages are delivered within 4 hops for 20 users in both linear and sparse-hexagonal models, whereas for 100 users, it exceeds 10 hops in the linear model and reaches 7 for the sparse-hexagonal model. The operator willing to provide a low-cost service along a motorway (example mentioned before) will have to consider has well the number-of-hops-to-live. By covering the 20% of the motorway, he will have to fix a number-of-hops-to-live superior to 10 hops.
Distribution of the number of hops perfomed (street model, 20 users, 60 min) 1
0.8
% of packets
0.6
0.4
0.2
0
10 hops 9 hops 8 hops 7 hops 6 hops 5 hops 4 hops 3 hops 2 hops 1 hops direcly
0.6
0.4
0.2
0
10
20
30
40
50
60
70
80
90
0
100
0
10
20
30
40
coverage in %
60
70
80
90
100
coverage in %
Distribution of the number of hops perfomed (hexagonal model, 20 users, 60 min) 1
Distribution of the number of hops perfomed (hexagonal model, 100 users, 60 min) 1
10 hops 9 hops 8 hops 7 hops 6 hops 5 hops 4 hops 3 hops 2 hops 1 hops direcly
0.8
0.6
0.4
0.2
10 hops 9 hops 8 hops 7 hops 6 hops 5 hops 4 hops 3 hops 2 hops 1 hops direcly
0.8
% of packets
% of packets
50
4.2. Simulation results
0.8
% of packets
Distribution of the number of hops perfomed (street model, 100 users, 60 min) 1
10 hops 9 hops 8 hops 7 hops 6 hops 5 hops 4 hops 3 hops 2 hops 1 hops direcly
0.6
0.4
0.2
29
0
0
10
20
30
40
50
60
coverage in %
70
80
90
100
0
0
10
20
30
40
50
60
coverage in %
Figure 4.5: Distribution of the number of hops performed to deliver a message
70
80
90
100
This page intentionally contains only this sentence.
Chapter 5
Conclusion and future work 5.1
Conclusion
In this thesis, a communication system based on sparse infrastructures has been studied. The Infostations concept already focusing on sparse infrastructures, is extended with mobile Multihop Capable Nodes with store-and-forward capabilities. These mobiles have dedicated memory which allows them to collect and store own created messages and messages relayed from other multihopping terminals, until delivery to an Access Point. Using Epidemic Routing, the performances of this infrastructure, were evaluated through the delivery rate, the end-to-end delay, the average and maximum memory occupation and the number of hops performed by the messages that are successfully delivered. The contribution of the messages delivered via multihops and via the mobility of the nodes was also evaluated. Assuming that the system cost is proportional to the number of Access Points that need to be deployed, the results show that the system can provide significant cost reduction for some specific types of services in some scenarios, specially in a linear environment (motorway). The cost of the system can be lowered by reducing the coverage and the system can still achieve delivery rates higher than 95%. To compensate this reduction of coverage, multihop and/or mobility of the nodes need to be exploited to deliver the messages. In a sparse linear model, the contribution of the multihop is dominant for high density of users. The majority of the messages are delivered via multihops and can even perform more than 10 hops before being delivered. This increases substantially the probability of being delivered to an AP, and reduces the expected delay. The counterpart in this situation is that the spreading of the messages increase the demand for higher memory capacity. In a sparse-hexagonal model, the messages are delivered with a combined effect of the mobility of the users and the multihops, but the contribution of the mobility can reach 40% of the messages delivered for a low user density. When the messages are conveyed by the mobility of the nodes, the drawback is a longer delay until the messages are finally delivered. 31
32
5.2
Chapter 5. Conclusion and future work
Future work
The current study was limited to the “uplink” meaning from one MCN to an AP. The downlink should be studied in a future work. Due to the complexity of human mobility, the mobility model could be approached in a more realistic way. The linear model gave interesting results because of the predictability of the mobility of the users. One can study in a future work if an improved sparsecellular system but with AP located in strategic places (such as subway exits . . . ) and using a Group mobility model can give performances close to the linear model. The Epidemic routing can be improved with the purpose of reducing the memory occupation. A future work should focus on metrics, such as the speed or the direction, for selecting users for carrying the messages.
References [1] J. Gacnik, “Technologies and successful applications for direct and multihop ad hoc networks,” Seminar on New Network and Information Technologies and Infrastructures in SS04 hosted by the Institute for Information Systems Research at the University of Hannover, Germany. [2] S. Corson and J. Macker, “Mobile ad hoc networking (manet): Routing protocol performance issues and evaluation considerations,” RFC 2501, Internet Engineering Task Force, January 1999. [3] D. Frenkiel and T. Imielinski, “Infostations: The joy of ”many-time, manywhere” communications,” Technical Report WINLAB-TR-119 119, WINLAB - Rutgers University, April 1998. [4] J. Zander, “Affordable QoS in future wireless networks myth or reality ?,” Personal, Indoor and Mobile Radio Communications, 2001 12th IEEE International Symposium on, vol. 1, pp. C–39–C–43 vol.1, 2001. [5] “Mercury wideband network radio (WNR).” http://www.acd.itt.com/pdf/WNR.pdf, (last seen : 2004-12-15). [6] A. Pentland, R. Fletcher, and A. Hasson, “DakNet: rethinking connectivity in developing nations,” Computer, vol. 37, no. 1, pp. 78–83, 2004. [7] R. Iglesias, “Feasibility of store and forward messaging service with cybiko computers,” Master’s thesis, KTH, 2003. [8] L. Briesemeister and G. Hommel, “Role-based multicast in highly mobile but sparsely connected ad hoc networks,” Mobile and Ad Hoc Networking and Computing, 2000. MobiHOC. 2000 First Annual Workshop on, pp. 45– 50, 2000. [9] T. Small and Z. J. Haas, “The shared wireless infostation model: a new ad hoc networking paradigm (or where there is a whale, there is a way),” in MobiHoc ’03: Proceedings of the 4th ACM international symposium on Mobile ad hoc networking & computing, pp. 233–244, 2003. [10] P. Juang, H. Oki, Y. Wang, M. Martonosi, L. S. Peh, and D. Rubenstein, “Energy-efficient computing for wildlife tracking: design tradeoffs and early experiences with zebranet,” in ASPLOS-X: Proceedings of the 10th international conference on Architectural support for programming languages and operating systems, pp. 96–107, 2002. 33
34
References
[11] A. Doria, M. Uden, and D. P. Pandey, “Providing connectivity to the saami nomadic community.” Development by Design Conference 2002. http://www.snc.sapmi.net/Project-docs/Saami-Network-Connectfinal.pdf (last seen: 2004-12-15). [12] A. Lindgren, A. Doria, and O. Schel´en, “Probabilistic routing in intermittently connected networks,” in Proceedings of The Fourth ACM International Symposium on Mobile Ad Hoc Networking and Computing (MobiHoc 2003), june 2003. [13] H. Li, M. Lott, M. Weckerle, W. Zirwas, and E. Schulz, “Multihop communications in future mobile radio networks,” Personal, Indoor and Mobile Radio Communications, 2002. The 13th IEEE International Symposium on, vol. 1, pp. 54–58 vol.1, 2002. [14] W. Yuen, R. Yates, and S.-C. Mau, “Exploiting data diversity and multiuser diversity in noncooperative mobile infostation networks,” INFOCOM 2003. Twenty-Second Annual Joint Conference of the IEEE Computer and Communications Societies. IEEE, vol. 3, pp. 2218–2228 vol.3, 2003. [15] P. Gupta and P. Kumar, “The capacity of wireless networks,” Information Theory, IEEE Transactions on, vol. 46, no. 2, pp. 388–404, 2000. [16] P. Kumar, “A correction to the proof of a lemma in ”the capacity of wireless networks”,” Information Theory, IEEE Transactions on, vol. 49, no. 11, p. 3117, 2003. [17] M. Grossglauser and D. Tse, “Mobility increases the capacity of ad-hoc wireless networks,” INFOCOM 2001. Twentieth Annual Joint Conference of the IEEE Computer and Communications Societies. Proceedings. IEEE, vol. 3, pp. 1360–1369 vol.3, 2001. [18] M. Grossglauser and D. Tse, “Mobility increases the capacity of ad hoc wireless networks,” Networking, IEEE/ACM Transactions on, vol. 10, no. 4, pp. 477–486, 2002. [19] D. Goodman, J. Borras, N. Mandayam, and R. Yates, “INFOSTATIONS: a new system model for data and messaging services,” Vehicular Technology Conference, 1997 IEEE 47th, vol. 2, pp. 969–973 vol.2, 1997. [20] R. Frenkiel, B. Badrinath, J. Borres, and R. Yates, “The infostations challenge: balancing cost and ubiquity in delivering wireless data,” Personal Communications, IEEE [see also IEEE Wireless Communications], vol. 7, no. 2, pp. 66–71, 2000. [21] W. H. Yuen, R. Yates, and C. W. Sung, “Performance evaluation of highway mobile infostation networks,” Global Telecommunications Conference, 2003. GLOBECOM ’03. IEEE, vol. 2, pp. 934–939, 2003. [22] A. McDonald and T. Znati, “A path availability model for wireless ad-hoc networks,” Wireless Communications and Networking Conference, 1999. WCNC. 1999 IEEE, pp. 35–40 vol.1, 1999.
References
35
[23] P. Santi and D. Blough, “An evaluation of connectivity in mobile wireless ad hoc networks,” Dependable Systems and Networks, 2002. Proceedings. International Conference on, pp. 89–98, 2002. [24] K. Nakano, M. Sengoku, and S. Shinoda, “Effect of mobility on connectivity of mobile multihop wireless networks,” Vehicular Technology Conference, 2002. VTC Spring 2002. IEEE 55th, vol. 3, pp. 1195–1199 vol.3, 2002. [25] M. Zonoozi and P. Dassanayake, “User mobility modeling and characterization of mobility patterns,” Selected Areas in Communications, IEEE Journal on, vol. 15, no. 7, pp. 1239–1252, 1997. [26] A. McDonald and T. Znati, “A mobility-based framework for adaptive clustering in wireless ad hoc networks,” Selected Areas in Communications, IEEE Journal on, vol. 17, no. 8, pp. 1466–1487, 1999. [27] D. Shukla, “Mobility models in adhoc networks,” Master’s thesis, Kanwal Rekhi School of Information Technology, IIT Bombay, november 2001. [28] T. Camp, J. Boleng, and V. Davies, “A survey of mobility models for ad hoc network research,” WIRELESS COMMUNICATIONS AND MOBILE COMPUTING, 21 April 2002. http://toilers.mines.edu. [29] C. Perkins, E. Belding-Royer, and S. Das, “Ad hoc on-demand distance vector (AODV) routing,” RFC 3561, Internet Engineering Task Force, July 2003. [30] D. B. Johnson, D. A. Maltz, and Y.-C. Hu, “The dynamic source routing protocol for mobile ad hoc networks (DSR),” Section 10 of RFC 2026 draft for RFC 2026, Internet Engineering Task Force, 19 July 2004. [31] T. Clausen and P. Jacquet, “Optimized link state routing protocol (OLSR),” RFC 3626, Internet Engineering Task Force, October 2003. [32] R. Ogier, F. Templin, and M. Lewis, “Topology dissemination based on reverse-path forwarding (TBRPF),” RFC 3684, Internet Engineering Task Force, February 2004. [33] C. E. Perkins and P. Bhagwat, “Highly dynamic destination-sequenced distance-vector routing (DSDV) for mobile computers,” in ACM SIGCOMM’94 Conference on Communications Architectures, Protocols and Applications, pp. 234–244, 1994. [34] E. Royer and C.-K. Toh, “A review of current routing protocols for ad hoc mobile wireless,” Personal Communications, IEEE [see also IEEE Wireless Communications], vol. 6, no. 2, pp. 46–55, 1999. [35] Y. Chun, L. Qin, L. Yong, and S. MeiLin, “Routing protocols overview and design issues for self-organized,” Communication Technology Proceedings, 2000. WCC - ICCT 2000. International Conference on, vol. 2, pp. 1298– 1303 vol.2, 2000. [36] A. Vahdat and D. Becker, “Epidemic routing for partially connected ad hoc networks.” Duke Technical Report CS-2000-06, July 2000. http://www.cs.ucsd.edu/ vahdat/pubs.html.
36
References
[37] A. Khelil, C. Becker, J. Tian, and K. Rothermel, “An epidemic model for information diffusion in MANETs,” in MSWiM ’02: Proceedings of the 5th ACM international workshop on Modeling analysis and simulation of wireless and mobile systems, pp. 54–60, 2002. [38] J. Yoon, M. Liu, and B. Noble, “Random waypoint considered harmful,” INFOCOM 2003. Twenty-Second Annual Joint Conference of the IEEE Computer and Communications Societies. IEEE, vol. 2, pp. 1312–1321 vol.2, 2003. [39] Y. Yi, M. Gerla, and T. J. Kwon, “Efficient flooding in ad hoc networks: a comparative performance study,” Communications, 2003. ICC ’03. IEEE International Conference on, vol. 2, pp. 1059–1063 vol.2, 2003.
Appendix A
Opposition Report Opponent: Thomas Ess´en
Organization and structure The report is divided into five parts: Introduction, Context and system models, Simulation method and results, Performance measurement, Conclusions and future work. It is easy to find in the report because the structure of the report has the form that can be expected from a technical report.
Background and previous work A good introduction is given where multihopping is described and compared to other transmission techniques, advantages are discussed and applications are proposed. This gives a good background, but I would like to have a better explanation of Previous work. One important purpose of a master thesis is to make a contribution to the research in the subject. Without a good overview of Previous work in the report it is difficult to understand the contributions of this work. You have a section were you write about related work, it is good but I think you should be even more precise and give efforts to explain what have been done and what your thesis will contribute with. On the other hand a big number of references are given and all context in the report is well rooted in the scientific literature. The “Related Work” section has been moved before the “Problem Definition” section as suggested by the examiner, Jens Zander. The section has been reworked to give a better overview of the previous works and emphasize the similitudes and differences with the thesis.
Method Simulations have been done to study the performance of the networks. The methods have been explained in a systematic way and the assumptions have been motivated. 37
38
Appendix A. Opposition Report
Results and Conclusions The results from the simulations are plotted in clear and very good figures, but it’s a shame the text is hard to read. I would recommend you to increase the font of the text in the figures in the results part. In Appendix A you introduce an “Urban environment model”, and in Appendix B you have plotted the performance in an environment like that. Why not have the results of the “Urban environment model” in the result part in the report, and not in an Appendix? Some of the plots are superfluous and could be removed, for example it is not necessary to have one plot in the report, and the same plot in one Appendix with only a few revisions. The figures of the “Results” section were put in the “Appendix” section in larger size in order to improve the readability. The figures of the “Results” section are now shown on pages of floats for the 2-by-1 figures and landscape view for the 2-by-2 figures (to keep the comparison view). As those figures are now easier to read, all the superfluous and uncommented figures in the “Appendix” sections have been removed. The “Urban environment model” was presented in “Appendix” and not in the “Results” section because the results obtained with this model did not match my expectations. I was expecting results between the “linear model” and the “sparse-hexagonal model”, considering the “Urban environment model” as a model with less mobility constrains than the “linear model” but more than the “sparse-hexagonal model”. The description of this model were therefore given in “Appendix”. This part has been removed as well, as the description, the uncommented results and figures did not give additional informations to the thesis.
Language and layout I have made suggestions of improvements in the language in a printed version of the report that you’ll get with a few comments of the report are given as well. The report has a really good layout. But I don’t see a point in having that many blank pages that you have. It’s ok to have it after the front page but I don’t think it’s necessary to have it between all sections. The blank pages are due to the LATEX RST style template. This template was written by Magnus Lindstr¨ om. These blank pages can not be removed unless modifying the code of the template.
More comments • The delay tolerant multimedia game is considered. As you mention in the part Future work, it maybe isn’t realistic to use the mobility model with pedestrians and vehicles. When I think about playing a multimedia game I would imagine that I wouldn’t move around that much. Can you motivate your mobility model? Is a multimedia game really the right application for a system that you have simulated? As you suggested in your presentation, the military could benefit from your results. Write about that!
39 As required by the title of the thesis, mobility of the terminals had to considered. But as you mentionned, the multimedia game service considered was probably not the best example, especially in combination with a pedestrian model. The multimedia game was used to illustrate an example of application and give a better understanding to the reader. The important point was to get an application which was delay-tolerant and requiring transfer of important volumes of data. Without getting too specific in the type of multimedia game it could be, the passengers of a buss or children at the back seats of a car could be an illustration of the vehicular mobility model. Another application could have been the upload of camera recordings in busses to a server. These videos can be used later for police investigations in case of aggressions. In the case of the pedestrian model, the attention is differently focused and may generate less traffic. With a MP3 player for example, one can listen to a selected playlist and download the missing MP3 from the network while walking. The military can benefit from multihop systems as shown in the introduction with [5]. Form the picture, one could think that the mobility models adopted in the thesis could fit the military usage. However, I do not think that the pedestrian mobility model corresponds to the mobility of soldiers, neither does the mobility of vehicles with tanks. Moreover, military have other requirements like security and reliability of the data transmitted. Considerations about these requirements are out of the scope of this thesis work. • A list of abbreviations would be appreciated. What is SCN for example? This has been corrected. A list of abbreviations has been added. • One thing I didn’t find out in the report was the transmission capacity of the MCN. How long time does it take for a MCN to resubmit a data packet? Do you neglect the transmission time? The transmission time is neglected. All the packets are transmitted at once. The transmission are considered error-free and resubmission in case of collision or error are not considered. A part on the transmission has been added in the thesis to explain better the transmission model. • Here epidemic routing is considered. That’s probably not the most efficient routing scheme. It’s outside of the scope of your master thesis but still it would be interesting to hear about your opinion how much the system can be improved with a better routing algorithm. The epidemic routing should be considered as an “upper bound” because of its bad management of resources. Other routing algorithms have better management of resources. However, as mentionned in [36], “while the existing ad hoc routing protocols are robust to rapidly changing network topology, they are unable to deliver packets in the presence of a network partition between source and destination.” The epidemic routing was adopted in the thesis because of network partition. Improvements can be gained by selected MCN moving faster, thus having a higher probability of reaching
40
Appendix A. Opposition Report an AP. They will act as “collectors” of messages. Moreover, we can suppose than these MCN moving faster are vehicles, and thus may have larger memory capacity than the pedestrian MCN’s terminal.
