Improving performance in delay/disruption tolerant networks through - - PDF document

SLIDE 1

Improving performance in delay/disruption tolerant networks through passive relay points

Saeed Shahbazi • Shanika Karunasekera • Aaron Harwood

Published online: 30 September 2011 Springer Science+Business Media, LLC 2011

Abstract In this paper, we study the case of a limited number of mobile nodes trying to communicate in a large geographic area, forming a delay/disruption tolerant net- work (DTN). In such networks the mobile nodes are disconnected for significantly long periods of time. Tradi- tional routing protocols proposed for mobile ad hoc net- works or mesh networks, which assume at least one path between each source and destination, are ineffective in

• DTNs. One approach to improve communication is through

gossip based protocols because these protocols do not rely

• n a fixed path. Another approach is to control the move-

ment of the mobile nodes and/or use special mobile nodes called ferry nodes. Others try to employ a fixed infra- structure including stationary relay points. One scheme in stationary relay point approach is to use base stations as relay points which need their own power supply. In this paper, we study a passive approach where mobile nodes deposit/retrieve messages to/ from known stationary loca- tions in the geographic region. Messages are delivered from a source by being deposited at one or more locations that are later visited by the destination. A proposed implementation of our approach using read/writable pas- sive Radio Frequency Identification (RFID) tags, one per point location, is considered in this work. Passive RFID technology is desirable because it operates wirelessly and without the need for attached power. Our simulation results indicate that our approach can achieve competitive mes- sage delay and delivery rates. We also demonstrate several techniques for optimizing the stationary relay node place- ment, namely relay pruning, probability based relay dis- tribution and a genetic algorithm; the genetic algorithm is shown to provide the best solutions to this problem. Keywords Ubiquitous network connectivity Delay/disruption tolerant networks Performance evaluation RFID tags 1 Introduction The number of connections present in a mobile network at any one time is an important topological property because such connections allow communication to take place. We categorize mobile networks as delay/disruption tolerant network (DTN) depending on the degree to which con- nections are available. In DTNs, the number of nodes per unit area, or the node density, is small and the nodes do not frequently connect. The network may remain partitioned into individual nodes for relatively long periods of time. DTNs arise naturally from applications such as wildlife tracking [25], vehicle-based disruption-tolerant networks (VDTN) [12, 31, 38], rural kiosks in developing countries [47], delay-tolerant mobile sensor network [57], and environmental monitoring including metropolitan areas [27] and underwater [1, 40], or from fragility and failures in the network itself due to disasters, jamming and noise, and power failure. DTNs are also referred to as sparse mobile networks, extreme wireless networks, or intermit- tently connected networks in the literature. In DTNs, if two nodes are within the broadcast range of each other and the link between them is up then we say they are connected.

• S. Shahbazi (&) S. Karunasekera A. Harwood

The University of Melbourne/NICTA, 3.08, 111 Barry St., ICT Building, Carlton, VIC 3053, Australia e-mail: shahbazi@csse.unimelb.edu.au

• S. Karunasekera

e-mail: shanika@csse.unimelb.edu.au

• A. Harwood

e-mail: aaron@csse.unimelb.edu.au

123

Wireless Netw (2012) 18:9–31 DOI 10.1007/s11276-011-0384-1

SLIDE 2

In the literature, the connected link is referred to as a contact [15]. Perur et al. [44] define sparseness of a network by measuring connectivity of the network, i.e., the probability that the network graph forms a single connected compo- nent, if it is less than 0.95, then it is referred to be as a sparse network. From another perspective, we consider a mobile network as a DTN for a given time interval, if the average number of contacts is less than 5% of all potential contacts, i.e., all pairs of nodes, over the interval. The details of calculating this threshold is referred to -1. As there is no fixed infrastructure in DTNs to deliver the messages from source nodes to destination nodes, mobile nodes have to participate in routing and they have to act as a router similar to mobile ad hoc networks (MANETs). However, traditional routing protocols proposed for MANETs [24, 41–43, 48] are ineffective in DTNs as they typically make an assumption that the underlying network is connected. A connected network in this context means that there exists at least one (possibly multi-hop) path between each pair of nodes and that exists for a long- enough period of time to allow a packet to traverse it. Furthermore, these protocols assume that if the path is disrupted it can be repaired or replaced in a relatively short period of time. The assumption of connectivity is clearly ineffective in DTNs because of the lack of instantaneous end-to-end paths in such networks which prevents estab- lishing any routes to forward the data packets. In order to overcome the lack of instantaneous end- to-end paths in DTNs, a routing protocol can use a store-and-forward paradigm. Therefore, a new class of routing protocols, referred to as store-carry-and-forward [10, 23, 25] has emerged. This class of routing protocols exploit the mobility of the nodes in the network to forward the data packets by relaying packets to intermediate nodes. The intermediate nodes then keep the data and deliver it to the final destination or to another intermediate node. Therefore, the data is incrementally distributed throughout the network, i.e., in the intermediate nodes, leading to facilitate the data delivery process. Recently there has been focus on augmenting DTNs with some low cost, easily deployable fixed relay nodes. We refer to this emerging class of protocols as stationary relay point approaches. In these approaches some station- ary relay nodes are added to the network in order to improve connectivity [6, 22, 49, 66]. This class of proto- cols can increase contact opportunities among mobile nodes, consequently improving DTNs performance. For example, Banerjee et al. [5] show experimental results collected from the UMassDieselNet DTN [10] that adding a fixed relay node, called throwbox, to the network improves the packet delivery by 37% and reduces the message delivery latency by at least 10%. Stationary relay point approaches can be categorized as active and passive based on the type of relay nodes. If the relay node can initiate the communication we refer to the approach as an active relay point; otherwise, we refer it as a passive one. Further, in active approaches [6, 22, 66], stationary relay nodes have their own supply of power while in passive approaches [49], they are powered by readers, i.e., mobile nodes. Also, in active approaches, the number of relays is less than the passive approaches while their broadcasting range is usually bigger. In this paper, we propose a passive stationary relay point based protocol to improve the delivery performance of the

• DTNs. In contrast to the active approaches proposed in the

literature, in our protocol, the stationary points do not require their own power supply. Specifically, we propose an alternative approach where mobile devices deposit/ retrieve messages to/from known point locations in the geographic region. The point locations act like ‘‘mail- boxes‘‘. Mobile nodes are assumed to know the position of the mailboxes. As a mobile node moves around the region, it checks the mailboxes that it meets. Messages can be retrieved from a mailbox and copied into another, subse- quently met mailbox. This mechanism allows all nodes in the network to help push messages over the geographic region and thereby expedites message delivery. In fact, in

• ur protocol nodes never communicate messages directly

to each other, rather they communicate messages only via these mailboxes which are in known places. This helps saving energy because the transmission can be off when mobile nodes are not in the communication range of the relay nodes, which is a challenging issue in DTNs as the mobile nodes are battery powered. Banerjee et al. [5] show that using 802.11 radio to search for contacts in a DTN devotes 99.5% of the total energy of mobile nodes just to find other nodes which is leading to have a short network

• lifetime. Additionally, the possible communication between

mobile nodes is, by definition of a DTN, infrequent and therefore its effects are negligible. Although we augment the network with a passive unconnected infrastructure, mobile nodes still can make a network on the fly without a priori connected infrastructure. A proposed implementation of our approach using pas- sive read/writable Radio Frequency Identification (RFID) tags, one per point location, is considered in this work to evaluate the network performance; hence we refer to our protocol as a Tag-based Routing (TBR) Protocol. RFID technology is desirable because it operates wirelessly and without the need for attached power. This makes its deployment relatively easy and sustainable. It would be equally valid to consider the results of our work on a net- work where the point locations are wireless base stations (with no connections between the base stations), if such an infrastructure was feasible for the application. The increased

10 Wireless Netw (2012) 18:9–31

123

SLIDE 3

range of a wireless base station compared to the range of an RFID tag would serve to increase the effectiveness of our protocol in terms of message delivery rate and delay. The proposed protocol was studied in our previous works [49, 50]. In this paper we enhance the protocol to handle more realistic scenarios and we evaluate its per- formance with various scenarios. Further, we evaluate the performance of the proposed protocol under the effect of different mobility models and show how entity mobility models and group mobility models have different impact

• n its performance. Additionally, we present different relay

placement techniques namely relay pruning, probability based relay distribution, and a genetic algorithm based

• ptimization and we show their effect on the performance
• f the proposed protocol as well as comparing their

effectiveness with the existing placement techniques in the

• literature. Our simulation results show that our protocol
• utperforms the existing approaches in the literature and
• ur genetic algorithm based optimization placement tech-

nique is the superior relay placement technique in the lit-

• erature. In this paper, we mainly focus on our protocol

performance and the issues related to security, privacy, fault tolerance, etc. are outside of the scope of the paper. The main contributions of the paper are as follows: (i) We propose a novel DTN routing protocol and evaluate it under a verity of scenarios with one of the most widely used mobility models. (ii) We propose different relay placement techniques to

• ptimize the performance of our protocol given a

mobility model. Accordingly, we evaluate them with different mobility models including entity and group mobility models to identify the best placement technique. (iii) We present a comprehensive comparison between

• ur proposed protocol and well-known routing

protocols in DTNs namely Epidemic Routing [56], Message Ferrying [65], and Throwbox [66]. The remainder of the paper is organized as follows: Section 2 reviews the background and related works, Sect. 3 presents a detailed description of the proposed protocol,

• Sect. 4 defines different node mobility models used in this

paper, Sect. 5 shows different techniques for distributing relays in the region, Sect. 6 provides simulation results for

• ur protocol, Sect. 7 compares our work to representatives
• f both existing reactive and proactive approaches in Store-

Carry-Forward paradigms, and Sect. 8 concludes the paper. 2 Background A DTN routing protocol should be able to accommodate disconnections in the network without significant impact

• n message delivery performance. A pragmatic approach to
• vercome partitions in DTNs is by using longer transmis-

sion ranges and thereby maintaining persistent network connectivity [46]. Increased sparseness of the network then leads to increased power requirements, which is a clear shortcoming of this approach. Furthermore, using long radio range in some applications may not be possible or desirable. Recently there has been focus on developing routing protocols for DTNs. Al Hanbali et al. [2] classify these protocols based on the degree of knowledge of the mobile nodes about their future contacts with other mobile nodes. We categorize DTN routing into two major categories: Store-Carry-Forward paradigms and Stationary Relay Points approaches based on possibility of using fixed relay points. 2.1 Store-carry-forward The class of Store-Carry-Forward paradigms exploits the mobility of nodes to buffer data packets during network partitions and forward them when connections become

• available. They are divided into two categories: reactive

and proactive schemes. Reactive routing protocols [52, 53, 56] use mobility of the participating nodes to buffer and deliver messages across network partitions. While, proac- tive routing protocols [9, 18, 30, 59, 60, 65]) control the mobility of some nodes to accommodate disconnections. More details are provided in the related work section. 2.1.1 Reactive schemes Reactive approaches use mobility of the participating nodes to buffer and deliver messages across network par- titions without forcing any mobile node to change its tra-

• jectory. Vahdat et al. [56] propose a routing protocol for

partially-connected ad hoc networks called Epidemic

• Routing. In this protocol, every node exchanges all the data

packets stored in its buffer while encountering with other mobile nodes until meeting the destination(s). Spyropoulos et al. [52] tried to improve the performance of flooding- based routing schemes such as Epidemic Routing by bounding the number of data packet replicas that happens by spraying a limited number of identical messages by source node to the network using distinct relays and wait until one of those relays meets the destination to perform a direct transmission. They enhanced their approach in [53]. Also, Tournoux et al. [55] propose a measurement-oriented variant of the spray-and-wait algorithm called DA-SW (density-aware spray-and-wait) that can tune the number of message replicas in a dynamic manner. Gao et al. [17] improve Epidemic Routing scheme by reducing the data forwarding cost. They employ a multi-cast scheme to select

Wireless Netw (2012) 18:9–31 11

123

SLIDE 4

the relay nodes considering the forwarding probabilities to multiple destinations simultaneously. Some reactive approaches use historic information about node contacts, spatial information, etc. to calculate the delivery expectation of next hops indicating their proba- bility of being able to successfully deliver a data packet to select the message carriers. These approaches are including PROPHET [32], NECTAR [14], pattern-based Mobyspace Routing [28], location-based Routing [29], and context- based Forwarding [37]. Burgess et al. [10] tried to improve the performance of the network by deleting useless data packets and scheduling packets for transmission to other

• peers. More recently, Balasubramanian et al. [4] study

routing in DTNs as a resource allocation problem and propose a protocol to enhance the performance of a specific routing metric by using some heuristics. Yuan et al. [61] propose the Predict and Relay scheme which predicts the future contacts of specified nodes at a specified time. Then, source/intermediate nodes select a proper neighbor as the next hop to forward their messages using their estimations about the future contacts of their neighbors and the destinations. Other approaches try to make a network structure sim- ilar to social networks in order to find the nodes as message carriers with the highest centrality, i.e., the structural importance of the node, which typically have a stronger capability of connecting other network members together. The representatives of these approaches are SimBet [13] and BubbleRap [21]. Hossmann et al. [17] further evaluate SimBet and BubbleRap performance under real mobility traces. 2.1.2 Proactive schemes Proactive approaches control the mobility of some nodes to accommodate disconnections. Goldenberg et al. [18] use mobility control to reach optimality by moving relay nodes to their optimal positions. Li and Rus [30] propose the possibility of changing hosts trajectories to send messages in disconnected ad hoc wireless networks. Using motion information of destination node, they try to utilize coop- eration of the intermediate nodes to deliver messages by asking them to modify their trajectory while getting the messages. Zhao et al. [65] use a set of nodes, called ferries, responsible for carrying data for all nodes in the network (it is called Message Ferrying (MF)). Ferries act as a moving communication infrastructure for the network. Wu et al. [59] propose a logarithmic store-carry-forward scheme through a hierarchical structure of trajectory for ferries that controls the number of relays which ends up with reducing average delay which was very high in MF and they also utilize some new technical issues like

• n-demand ferry solicitation, dynamic trajectory planning
• f ferries, rendezvous point placement, and adaptive ferry

migration and load balancing to enhance the network per-

• formance. Tariq et al. [9] extend the MF approach by

forcing ferries to follow fixed routes; therefore, they sim- plified designing complex routes where the ferry can con- tact the nodes with certainty which needs an on-line collaboration between ferry and the nodes in MF. Jeonghwa et al. [60] extends MF approach by proposing a mechanism to replace ferry nodes. Since the network

• peration relies on the ferries to provide connectivity in the

whole network, it can be a single point of failure. Also, to keep a balance among all mobile nodes (with limited resources) in the network it may be desirable to rotate this role with others after a fixed duration. A summary of some existing routing approaches for DTNs can be found in [63]. 2.2 Stationary relay points In Stationary Relay Points approaches such as [6, 22, 66], by adding some stationary relay points, the connectivity of the network is increased and the performance of the net- work is improved as a result. This class of intermittently connected MANET routing protocols works best in certain circumstances such as when we cannot expect a ferry to move on in an inhospitable terrain, or through the obstacles due to a disaster, etc. where missed contact opportunities may decrease network performance. Based on relay point type, we can divide the class of stationary relay points into two categories: active and

• passive. In active approaches [6, 22, 66], stationary relay

points can initiate the communication and have their own supply of power while in passive approaches [49], sta- tionary relay points are unable to initiate any communi- cations and they are powered by readers, i.e., mobile nodes. Further, in active approaches, the number of relay nodes is less than the passive approaches while their broadcasting range is usually bigger. One of the active approaches proposed in the literature is the use of Base Stations. This approach tries to increase the network connectivity based on a fixed infrastructure using base stations. Although Ibrahim et al. [22] introduce some powerful platforms to meet the requirements of base stations in DTNs, the possibility of making such an infra- structure may not be feasible or desirable. Zhao et al. [66] proposes an active approach for the case

• f

Stationary Relay Point. Relay nodes are called ‘‘throwboxes’’ and they are considered to be powered, wireless base stations that cannot interact with each other. However, they assume that relay nodes can initiate com-

• munication. Furthermore, in their work, a mixed integer

programming solution is proposed for the placement of throwboxes, which discretizes space and is NP hard.

12 Wireless Netw (2012) 18:9–31

123

SLIDE 5

Therefore, a greedy heuristic is used to place throwboxes. The throwbox work is further analyzed in [22]. Banerjee et al. [6] use different types of infrastructure to enhance mobile networks, namely untethered relays, base stations, and a mesh. In the case of base stations and a mesh, they propagate packets through an in-network proxy with storage which relaxes contemporaneous connections between mobile nodes; however, there is a trade-off between enhancing the network and the cost of the infra-

• structure. Use of untethered relays is similar to [66].

Our proposed protocol is very similar to the work done by Zhao et al. [66]; however, on the contrary to the throwboxes passive relay nodes in our work do not initiate communication and they are merely storages. Additionally, in their work relay nodes need their own supply of power while passive relay nodes in our approach can be powered by readers and consequently they can have a much longer lifetime in the network. Also, our approach could be more suitable for radio silent applications where relay nodes should not originate radio signals and should have a short radio range which makes using passive relay nodes, e.g., passive RFID tags, more economical. Furthermore, we have proposed continuous space placement techniques and in particular we provide a genetic algorithm approach for placement optimization which outperforms their greedy placement technique. Also, the number of throwboxes considered is relatively small compared to the number of relays we consider. 3 Proposed protocol In this section, we first present the motivation of our work and then we introduce our protocol and also the evaluation

• metrics. Since the protocol was initially considered for

passive RFID tags in prior work, we refer to our protocol as a Tag-based Routing or TBR Protocol. 3.1 Motivation There are approaches that try to reinforce connectivity on demand in DTNs by utilizing additional communication resources in the network. Examples of these resources include satellites1, base-stations [6], unmanned aerial vehicles [65], etc. Satellite communication needs a direct view of receiver/ sender to the satellite and is expensive as it requires all user stations such as handsets, portables, mobile stations, etc. to be equipped with specific opera- tional units to allow the communication to take place [34]. Further, satellite availability might be poor in noisy environments, e.g., when the operator is very close to large machineries. Blind spots is a well-known problem when using base-stations for communication due to the possible obstacles in the network [39]. Additionally, deploying base-stations might be difficult in areas which are difficult to reach and needs time/cost to take place. In addition, base-stations usually need to have a wired con- nection to the backhaul [39]. We consider a hypothetical scenario shown in Fig. 1 in which miners try to commu- nicate to improve mining productivity making a DTN

• together. Satellite communication fails in underground as

there is no direct line of sight. Using base-stations in underground like mines/tunnels suffers from the short communication range, deploy-ability, and cost effective- ness since UHF/VHF technology has a very short com- munication range in underground [62] and there could be blind spots due to the lack of line of sight transmission path of a wireless signal because of the existence of

• bstacles underground (e.g., curvy tunnels or branched

mines). Moreover, since mines might be extending, e.g., in gold mines to explore new source of golds, setting up new base-stations is time-consuming. For the above mentioned reasons and in order to arrive at a competitive protocol, we have considered the introduction of low cost and easily deployable fixed infrastructure. Miners can communicate to each other through passive relay devices, e.g., RFID tags, as well as communicating to the base station through the vehicles/others visiting the relays hit by the miners in the past. 3.2 Protocol description We consider a static distribution of Nrelay stationary relay nodes (relays) over the region where Nnode mobile nodes (nodes) will move. The region is defined by its extents, (Xmin, Xmax) and (Ymin, Ymax). Each relay contains a mes- sage buffer of size Brelay messages (each message has unit size), and each node contains a message buffer of size Bnode

• messages. A node can interact with a relay if it is within

distance r from the relay. Later in this section we show how a node can interact with a relay. The notation is summarized in Table 1. Units are always meters for distance, seconds for time, speed in meters/ second, unless otherwise noted. Messages are only transmitted between nodes and relays and vice versa, never between nodes and other nodes. Communication is only initiated by nodes since the relays are passive. To consider node-relay interactions we define relay connection and relay disconnection events in time. When a node moves from a distance[ r to within a distance

• f any relay at time t1 then we say the node has connected

with the relay at time t1. When the node moves, say at time

1 Disruption Tolerant Networking. [Online]. Available: http://www.

darpa.mil/ato/solicit/DTN/. Wireless Netw (2012) 18:9–31 13

123

SLIDE 6

t2, outside distance r of a relay that it is connected to, then we say the node has disconnected from the relay at time t2. In this way, given the paths for all nodes, we can consider the set of distinct connection and disconnection events in

• time. To be complete, we also consider message arrival as

an event in time. Nodes move obliviously to each other and to any other aspects of the system that makes the imple- mentation of our protocol simpler and more scalable as no mobile node needs to know extra information about other mobile nodes. Our protocol is now defined by what happens at each of the three events, also depicted in Fig. 2. Relay Connection: the connecting node examines the relay’s buffer to see if any of the messages are destined to itself, if so then the message is said to have reached its destination at the time of the connection. The node then merges its buffer with the relay’s buffer according to Algorithm 1. Relay Disconnection: the disconnecting node examines its state to see if it is still in the merge operation with the relay, if so then it terminates the merge operation. Message Arrival: the node puts the newly arrived mes- sage into its buffer, the oldest message, i.e., the one with the earliest arrival time, is discarded if the buffer is already

• full. Merging does not happen at this event (even though

the node may be within range of a relay). Figure 2 provides a brief example of the movement of two nodes through a regular spaced field of relays. Node A deposits a message on relays 2,3 and 1. Assuming node B passes through relay 1 before node A, but passes through relay 3 after node A, then node B retrieves the message from A at relay 3 and deposits the message in relay 4. To describe node-relay interaction we define a merge

• peration between a node message buffer and a relay

message buffer. Algorithm 1 shows the merge operation. According to Algorithm 1, the mobile node acquires some meta-data regarding the current messages in the relay’s

• buffer. Based on the messages in its buffer it decides to

read/write messages one by one until it is disconnected from the RN or all messages are replicated. Therefore, the merge operation virtually combines both buffers into a single buffer with messages ordered according to their arrival times in the system. Then, the node and relay both keep as many of the latest messages, i.e., the ones with larger arrival times, as can fit in their respective buffer. Older messages are discarded. In general, it is not possible for a node to know at first face whether a message has been delivered to a destination or not (apart from the destination node itself), so delivered messages will continue to prop- agate until they are pushed out by newer messages. It is possible to increase the delivery performance by using more intelligent buffer policies, for example, Ma and Jamalipour [33] propose a fuzzy logic-based delivery framework called FLDF to facilitate the ranking of mes- sages based on their delivery preference in the future to increase the message delivery rate. Interactions are considered to be collision free; however, there could be two possible collisions. First, relays could be placed in such a way that a node falls into the communi- cation range of them simultaneously. In order to avoid this kind of collision, nodes can label the relays and send a request to communicate with the relay with the lowest ID first and then with the second lowest ID relay, and so on. Labeling the relays is possible because the mobile nodes know the positions of all the relays. However, in most cases relays are placed in such a way that this kind of

• Fig. 1 Mining scenario as an

application of our protocol Table 1 Notation in the paper Nrelay Number of relays Nnode Number of nodes X/Ymin/max Extents of the geographic region (m) d Regular spaced distance between relays (m) Brelay Relay message buffer size (messages) Bnode Node message buffer size (messages) r Relay range (m) t Time (s) q Message delivery rate (msgs/s) Lavg Average message delay (s) E Network efficiency 14 Wireless Netw (2012) 18:9–31

123

SLIDE 7

collision is highly unlikely to be occurred. Second, two or more mobile nodes may try to communicate with the same relay simultaneously. This kind of interference/collision should not be ignored as possibly two mobile nodes can communicate with a relay node respectively but with interference/collision. To handle this kind of collision we need collision detection and recovery/avoidance techniques in wireless networks [58] which is out of the scope of this

• paper. However, since two nodes by the definition of a

DTN is highly unlikely to access a relay simultaneously, the impact of this kind of collision on the performance of the protocol is negligible. 3.3 Evaluation metrics We evaluate our proposed protocol using four metrics defined below. 3.3.1 Message delivery rate The message delivery rate is defined as the ratio of the number of successfully delivered messages to total number

• f messages, i.e., identically generated messages by all

source nodes: q ¼ Mdelivered Mtotal : A low q indicates that the buffer sizes are not large enough to handle the rate of messages in comparison to average delay experienced by a message to get from the source node to the destination node. 3.3.2 Message delay Since a typical destination node may receive multiple copies of a message, we define message delay, Li, for message i, as the time between when a message is gener- ated to the first time the message is received by the desti-

• nation. The average message delay is then:

Lavg ¼ 1 Mdelivered X

Mdelivered i¼1

Li: 3.3.3 Network efficiency/inefficiency We define network efficiency given as E = q/Lavg. The goal is to maximize q and minimize Lavg simultaneously. Conceptually, network inefficiency could be defined as the fraction of 1 over E. In this paper, we focus on average message delay and delivery rate. We leave other metrics such as power consumption, transmission number, inter- action failures, etc. to future work.

connection event tag disconnection event mobile node movement mobile node r Ymax Ymin Xmax Xmin A B d d 1 2 3 4

(a) (b)

• Fig. 2 Aspects of the tag based routing model. a Movement of a node

through the circular range of a relay. b Example of two mobile nodes, A and B, moving through a field of relays spaced at regular intervals

• f distance d. In the example, node A reaches relay 2 and 3, but not

relay 1, before node B. Node A deposits a message on these relays, node B retrieves the message from relay 3 and deposits it on relay 4 Algorithm 1 Merge operation (MN,RN) MN and RN stand for mobile MN and relay MN respectively S1 list of message IDs in MN’s buffer S2 list of message IDs in RN’s buffer S ½S1S2 Sort(S, arrival time) counter 1 for all i 2 S do if i 62 S1 and len(S1) [ counter then Replicate msg i from RN to MN if MN’s buffer overflow then Evict oldest message from MN end if end if if i 62 S2 and len(S2) [ counter then Replicate msg i from MN to RN if RN’s buffer overflow then Evict oldest message from RN end if end if if IsDisconnected(MN,RN) then break; end if counter counter þ 1 end for Wireless Netw (2012) 18:9–31 15

123

SLIDE 8

4 Mobility models Mobility is a natural phenomenon in DTNs. Studying the mobility models in DTNs is important because it has a direct impact on the performance of network protocols including our proposed protocol. Therefore, in this section we review some widely used mobility models which are later used to evaluate the performance of the proposed protocol. There have been many mobility models proposed so far in the literature. Camp et al. [11] classified mobility models for ad hoc network research into two categories: entity mobility models and group mobility models. In entity mobility models, the movement of one node is independent to all other nodes, whereas in group mobility models, a set of mobile nodes move in a group. The widely used representatives of entity mobility models [7, 8] are Random Waypoint Model (RWP) and Random Walk Model (RWM), while Reference Point Group Mobility (RPGM) Model [20] is a well-studied representative of group mobility model. Figure 3(a, b) show the traveling pattern

• f a single mobile node using RWP and RWM models
• respectively. Figure 3(d) shows the traveling pattern of five

mobile node as a group using RPGM. Although, some of the above mobility models are artificial, they are widely used in DTNs [56, 65] to evaluate the performance of routing protocols. Therefore, in this paper we use them to evaluate the performance of our protocol and also to compare our protocol with some of the state-of-the-art protocols which are using similar mobility models. How- ever, there is other work that uses traced-based simulation. For example, Banerjee et al. [6] use traces from a bus route model to evaluate the effectiveness of their work. Additionally, we propose a restricted version of the RWP model where every mobile nodes can move only in a restricted area. Figure 3(c) shows the traveling pattern of 4 typical mobile nodes where the area is divided to 4 sub- areas with 40% overlap in each sub-area; however, we can divide the area based on the application to different sec-

• tions. According to Fig. 3(c), in two sub-areas there is only
• ne node while in other sub-areas there have two and no

nodes respectively. 5 Relay placement techniques Placement of relays plays an important role in our protocol as its performance is dependent to their positions. Relay node placement is already studied in VDTN [16]; however, in this section, we propose different relay placement strategies. 5.1 Uniform grid In this scheme, relays are placed on a regularly spaced grid with known distance d between two neighbor relays, such that the relay coordinates are (Xmin ? i d, Ymin ? j d) for i ¼ 0; 1; 2; . . . and j ¼ 0; 1; 2; . . .. This leads to Nrelay ¼ Xmax Xmin d

• þ 1

! Ymax Ymin d

• þ 1

! :

200 400 600 800 1000 200 400 600 800 1000

Y Position (m) X Position (m)

200 400 600 800 1000 200 400 600 800 1000

Y Position (m) X Position (m)

200 400 600 800 1000 200 400 600 800 1000

Y Position (m) X Position (m)

200 400 600 800 1000 200 400 600 800 1000

Y Position (m) X Position (m)

(a) (b) (d) (c)

• Fig. 3 Node mobility models.

a Traveling Pattern of a MN using theRandom Waypoint Model (50 steps). b Traveling Pattern of a MN using theRandom Walk Model (50 steps). c Traveling Pattern of 4 MNs using theRestricted Random Waypoint Model (50 steps). d Traveling Pattern of 3 MNs in a group using theRPGM (50 steps) 16 Wireless Netw (2012) 18:9–31

123

SLIDE 9

Although this placement technique is not optimal, for simplicity we mostly consider uniform grid placement as a baseline to evaluate the proposed protocol. 5.2 Relay pruning This is a simple pruning scheme where a certain percentage

• f relays are removed based on the number of messages

delivered by the relays when placed in a regular grid. Relays responsible for delivering least number of messages are removed until the specified percentage of relays are

• retained. The remaining relays will still remain at the ori-

ginal grid locations. 5.3 Probability based relay distribution By analysing the number of messages delivered by dif- ferent relays, when placed in a regular array, relays can be redistributed based on the probability distribution of the messages carried by the relays (higher relay densities in areas where the probability of a message being carried is high, and lower relay densities in areas where the proba- bility of a message being carried is low). We have imple- mented this scheme using an efficient re-sampling scheme given in Table 3.2 of [45], for repeating points, followed by jittering the points by adding Gaussian noise. 5.4 Genetic algorithm based optimization To compare the previous heuristics to a more general heuristic, we have implemented a basic genetic algorithm (GA). The purpose of using a GA is to explore a more general approach to finding the optimal placement, thereby to see if there exist better solutions than our deterministic

• approaches. We did not intend to undertake an extensive

study of GA solutions, but rather we intended to compare previous heuristics. GAs have been used before to find the

• ptimal location of base-stations (transmitters) in order to

satisfactorily cover subscribers [19, 35, 54]. There are many other non-linear optimization techniques, e.g., sim- ulated annealing, neural networks, etc., that could be explored but we leave them to future research. For these reasons, we have referred the details of our GA approach to Appendix -1. Our discussion assumes a general knowl- edge of GA terminology that can be found here [36]. We implemented a genetic algorithm, using MATLAB Genetic Algorithm and Direct Search Toolbox, to search for the locations of Nrelay relays in the region with the following objective function: minimize h

Lavg q

i . 6 Simulation results 6.1 TBR implementations In our simulation results the proposed protocol is consid- ered for passive RFID tags, however, it would be equally valid to consider the results of our work on a network where the point locations are other type of wireless devices with a capability of storing messages, if such an infra- structure was feasible for the application. A comparison between different implementation of Stationary Relay Point approach is shown in Table 2. The network designer can choose between them based on the application. 6.1.1 RFID MANET implementation Radio-frequency identification (RFID) is an automatic identification method, which is based on remote data storage/ retrieval using devices called RFID tags or tran-

• sponders. An RFID tag is a module which is used for

identification purposes using radio-waves. In this work we have considered its use for storing/retrieving messages in

• DTNs. Some commercially available tags in the industry

Table 2 Stationary relay points approaches

• Implement. Type

Cost Power resource Radio range Device size

• Commun. Initial.

WiMAX Base Stationa $24000 Stand-alone/high power 500 m–4 km [ .005 m3 Capable Active RFID tag b [ $20 Stand-alone/medium power Up to 100 m 35m3 Capable Throwbox $100 $300 Stand-alone Up to 250 m From 1.12-4 m3 Capable Passive RFID tag $:07 $:20 Powered by reader Up to 35 m 10-8 to 4.8-4 m3 Unable

a http://www.mobilesociety.typepad.com/mobile_life/2007/03/wimax_base_stat.html b http://www.gaorfid.com c Zhao et al. [66] introduce Intel Stargate as a device which can meet throwbox requirements. Stargate specification is available at http://www.

willow.co.uk/Stargate_Datasheet.pdf

d http://www.rfidjournal.com/faq and http://www.omni-id.com/products/RFID_tags-ultra.ph. Mojix, http://www.mojix.com/products, proposes a

solution to increase the range of reading from a tag up to 200m using their eNodes which is promising reaching longer radio ranges in future Wireless Netw (2012) 18:9–31 17

123

SLIDE 10

can be read from a distance of several meters away and beyond the line of sight of the reader; e.g. Mojix2 has recently developed a way to dramatically increase the range of passive RFID tags up to 200m, opening up many new applications for low-cost tags. Some tags are currently able to store kilobits of information; e.g. Fujitsu has recently reported the development of a 64 KByte High- Capacity FRAM RFID Tag3. RFID technology is rapidly improving and in the near future we expect to see more powerful RFID tags, which can be set-up more efficiently, with a wider range and higher memory capacity. Mobile nodes must be equipped with RFID readers. Recently, there has been significant attention in developing mobile RFID reader devices with Wi-Fi Network support and GPS4. This makes it possible for the applications where the mobile station cannot approach the RFID tags by itself, by using a smaller unit equipped with the mobile reader that can do the task of reading/writing from/to the mobile station from/to the tags. 6.2 Methodology We implemented an event driven simulation of TBR in

• MATLAB. We assumed that we have an ideal wireless link

layer that causes no collision with other transmissions during any transmission session. This simplification does not significantly effect our results because in a typical DTN, the density of mobile nodes in a specific area is usually very low (as discussed earlier in Sect. 3) Similarly, we assume communication from a tag to a mobile node and vice versa is instantaneous. In a real implementation, this assumption may not hold for some kinds of applications where MNs move relatively rapidly or cannot pause for the time taken to operate on an RN. However, there has been focus on designing low delay passive RFID reader systems which can accommodate this issue [3, 64]. Additionally, in Algorithm 1 we showed how we can prioritize the mes- sages based on their arrival time, to be exchanged between RNs and MNs for the time that they are being connected, using meta-data related to the messages. We leave the issue

• f read/write delay to future work.

Our simulation starts at time t = 0 with all tag buffers and node buffers empty. Nodes are placed randomly and messages arrive with exponentially distributed random inter-arrival times, at a global mean rate of k, such that each node generates messages at a mean rate k/Nnode. Each message has a single destination node picked uniformly at random from the nodes (not including the source node). The simulation is run for a fixed period of time, tmax. In the following section, we study the effect of different node mobility models and different relay placement strat- egies for our protocol. Furthermore, we evaluate our pro- tocol when using many small RNs and when using a super RN which is placed in the middle of the area. 6.3 TBR evaluation In this section we evaluate the effectiveness of TBR based

• n RWP mobility model and regular grid tag placement to

show the effect of node speed, node buffer size, tag buffer size, tag transmission range, and tag spacing on the per- formance of our protocol. Later, we show how we can significantly improve the performance of our protocol using a better tag placement technique. In addition, we show the required power consumption of nodes in terms of the required transmission number per second versus some

• f the latter parameters such as node speed and tag spacing.

Finally, we present the distribution of message delivery delay using the proposed protocol. 6.3.1 Control variables We use the following settings unless otherwise noted: Xmax - Xmin = 1, 000, Ymax - Ymin = 1, 000, d = 50 () Nrelay = 441), the average speed of mobile nodes is 5 (Smin = 2.5, Smax = 7.5) with no pause time in the Random Way Point model, Bnode = 1, 000, Brelay = 100, and r = 5. In most cases we have provided all simulations for Nnode = 10 with 0.5 message/second per node (k = 5). Tags are placed on a regularly spaced grid. All messages have equal size. 6.3.2 TBR effectiveness Figures 4 and 5 show the effect of Node Speed, Tag Buffer Size, Tag Transmission Range, and Tag Spacing on the delivery performance of the proposed protocol. According to the latter figures, in most cases the message delivery rate

• f our protocol is higher when there are less mobile nodes

in the network while the message latency is higher. The reason is that more mobile nodes in the network leads to more generated messages in the network and the proba- bility that mobile nodes overwrite the tags is higher as a result; therefore, messages have a lower probability to stay in the tag’s buffer for a longer time. Figure 5(a) shows that increased node speed signifi- cantly reduces the message delay. As the node speed increases the node meets more tags per message, however the node may also meet the same tags more than once and

2 http://www.mojix.com/products. 3 http://www.fujitsu.com/global/news/pr/archives/month/2008/20080

109-01.html.

4 http://www.rfid.net/product-listing/reviews/176-csl-cs101-handheld-

reader. 18 Wireless Netw (2012) 18:9–31

123

SLIDE 11

so higher speeds does not continue to decrease message delay with the same rate. Figure 4(a) shows the corre- sponding delivery rates. Similarly, meeting tags sooner greatly increases the delivery rate because messages have a smaller chance of being dropped if they exist in more

• buffers. Figures 4(b) and 5(b) show that increased tags

transmission range also significantly increases the message delivery rate and reduces the message delay; however, the speed up in delivery performance is not as high as the effect of increased average node speed. In general, with a

10 20 30 0.2 0.4 0.6 0.8 1

Average Node Speed (m/s) Message Delivery Rate

10 20 30 0.2 0.4 0.6 0.8 1

Tag Transmission Range Message Delivery Rate

N=5 N=10 N=20 N=50 500 1000 0.2 0.4 0.6 0.8 1

Tag Buffer Size Message Delivery Rate

N=5 N=10 N=20 N=50 100 200 300 400 500 0.2 0.4 0.6 0.8 1

Tag Spacing (m) Message Delivery Rate

N=5 N=10 N=20 N=50

(a) (d) (b) (c)

N=5 N=10 N=20 N=50

• Fig. 4 Message delivery rate

versus various parameters: a Average node speed, b Tag transmission range, c Tag buffer size, d Tag spacing

10 20 30 500 1000 1500

Average Node Speed (m/s) Average Message Delay

N=5 N=10 N=20 N=50 10 20 30 500 1000 1500

Tag Transmission Range Average Message Delay

N=5 N=10 N=20 N=50 500 1000 500 1000 1500

Tag Buffer Size Average Message Delay

N=5 N=10 N=20 N=50 100 200 300 400 500 500 1000 1500

Tag Spacing (m) Average Message Delay

N=5 N=10 N=20 N=50

(a) (d) (b) (c)

• Fig. 5 Average message delay

versus various parameters: a Average node speed, b Tag transmission range, c Tag buffer size, d Tag spacing Wireless Netw (2012) 18:9–31 19

123

SLIDE 12

larger tag transmission range, there is a larger chance that a mobile node meets more tags per unit time which results in meeting more tags per message and meeting tags sooner. Figures 4(c) shows that an increasing tag buffer size increases the message delivery rate much higher than the latter parameters. As the tag buffer size increases, the mobile nodes can replicate more messages, spread over more tags, and replicated messages can exist in more buffers for a longer period of time; therefore, messages could be delivered to the destination nodes with a higher chance instead of being dropped at an early stage. How- ever, according to Fig. 5(b) increased tag buffer size can lead to an increase in the average message delay. There are two factors which affect the average message delay: (1) an increased tag buffer size allows the messages to stay in the buffers for a longer time and those messages can be delivered in a longer period of time rather than being dropped when the buffer size is smaller, which thereby increases the average message latency; (2) an increased tag buffer size reminds that more buffers can hold the same messages which leads to a decrease in the average message

• delay. According to Fig. 5(b) (1) has a greater impact when

the delivery rate is low while (2) has a greater impact when the message delivery rate is approaching 1. Figure 5(d) show the delivery performance of the pro- posed protocol versus tag spacing, i.e., the total number of tags in the regular array, as given in Sect. 3. Significant variation in delay does not occur until the number of tags is quite small, i.e., tag spacing is quite big, and the tags are near the perimeter of the region. Message delivery rate is significantly affected by the tag spacing and the number of

• nodes. A smaller number of tags leads to a significantly

decreased rate. For large numbers of nodes, there is a greater total arrival rate of messages and so buffers are exhausted more quickly. 6.3.3 Transmissions per second Figure 6 shows the average transmissions per second that a mobile device uses as a result of TBR, for various tag spacing and speeds. The number of transmissions is directly proportional to the power requirements. Slower moving nodes and fewer tags both lead to fewer transmissions. 6.3.4 Power consumption versus number of relays Passive nature of relay nodes in our protocol requires communication cost from the mobile nodes. Assuming that RFID readers on average need equal amount of energy for each interaction with relay nodes, e.g., Klair et al. [26] present a table showing RFID readers power consumption, Equation shows the relation between the number of relays and the required power consumption for reading/writing the relays in the network, E.

• E ¼ Nnode

X

Nrelay i¼1

Pi½HEr=w; ð1Þ where Pi[H] is the the probability of hitting the relay i by a node per second and Er/w is the average required energy for each interaction between nodes and relays. Er/w depends on the storage space of the passive relay nodes and the traffic

• load. In priori work [50] we showed how to calculate the

average hit probability of a relay node by a mobile node using a conceptual mobility model. Further, in [51] we showed how we can extend this conceptual mobility model to RWP. In order to see the average number of relay hits by mobile nodes per second, we ran an experiment where Xmax - Xmin = 1, 000, Ymax - Ymin = 1, 000, the average speed of mobile nodes is 5 (Smin = 2.5, Smax = 7.5) with no pause time in the Random Way Point model, r = 5, and Nnode = 10 while we change the number of relay nodes placed on a regular grid. Figure 7 shows the corresponding

• results. According to Fig. 7 by placing 2,500 relays on a

grid, on average 1.26 of nodes would hit the relays per

• second. If we place 400 relays on average every 5 s one

mobile node would meet a relay and if we assume Er/w = 180 m watt for the case of Skye Module M1-Mini introduced in [26] then we need 36mwatt/sec for the read/ write operations in the system. 6.3.5 Distribution of message delay Figure 8 shows the cumulative distribution function of message delay, for delivered messages, versus various

• parameters. Additionally, we have fit these distributions to

the Generalized Extreme Value distribution using MAT- LAB dfittool:

50 100 150 200 250 300 350 400 450 500 10 -3 10 -2 10-1 100 101

Tag Spacing Message Transmissions per second (for each node)

S=50 S=40 S=30 S=20 S=10

• Fig. 6 Transmissions per second required by a mobile device

20 Wireless Netw (2012) 18:9–31

123

SLIDE 13

fðx; l; r; kÞ ¼ 1 r 1 þ kx l r

• 1

k1

e 1þkxl

r

ð Þ

1=k

; for 1 þ kxl

r [ 0.

The Generalized Extreme Value distribution was the

• nly distribution that consistently fit the data over various

ranges of parameters that we tested. Figure 9 shows an example message delay probability density function fit to the distribution. Further analytical work is required to substantiate this relationship. 6.4 Effect of tag placement and mobility models

• n TBR

In this section, we study the effect of different tag place- ment strategies on the performance of TBR. In this study, we also consider the effect of mobility models mentioned in Sect. 4 on TBR effectiveness. We compare our place- ment strategies with the ThrowBox algorithm proposed in [66]. In ThrowBox, base stations are greedily placed one by one so as on a grid of possible places in such a way to maximize the network throughput. In other words, given the traffic model and the mobility model of mobile nodes, all possible positions for a base station are considered. After placement of one base station, the algorithm con- tinues to the next base station excluding the previously taken positions. Since our optimization to find the optimal position, is based on E, in the case of ThrowBox we used the same objective function for optimization, which is different to the original objective function, i.e., increasing the delivery rate, that they used in their work. The simulation methodology is as described in Sect. 2. We also used the following configuration to evaluate the efficiency of each placement technique: the average speed

500 1000 1500 2000 2500 0.5 1 1.5

Average Hit Number/sec Relay Number

• Fig. 7 The average relay hit number per second versus the number of

relays placed on a regular grid

1000 2000 3000 0.5 1

Cumulative probability Message latency (sec)

d=25 d=50 d=100 d=200 d=250 d=500 200 400 600 0.5 1

Cumulative probability Message latency (sec)

100 200 300 400 0.5 1

Cumulative probability Message latency (sec)

100 200 300 400 0.5 1

Cumulative probability Message latency (sec)

BS=50 BS=100 BS=200 BS=500 BS=1000 BS=2000 r=5 r=10 r=20 r=30 r=40 r=50 S=10 S=20 S=30 S=40 S=50 S=60

(a) (c) (d) (b)

• Fig. 8 CDF of message delay

versus various parameters. a CDF of message latency fordifferent tag spacing, b CDF

• f message latency fordifferent

tag ranges, c CDF of message latency fordifferent mean node speeds, d CDF of message latency fordifferent tag/node buffer sizes

20 40 60 80 100 120 140 160 180 0.005 0.01 0.015 0.02

Message Latency (sec) Density

• Fig. 9 PDF of message delay fit to a generalized extreme value
• distribution. In this example, the distribution parameters are k =
• 0.104493, r = 24.1557 and l = 39.6301

Wireless Netw (2012) 18:9–31 21

123

SLIDE 14

• f MNs is 20 (Smin = 10, Smax = 30) and there are 16 RNs

in the network () Nrelay = 16). We show using a relay placement optimization technique we can still achieve competitive message delivery performance even using low number of RNs, i.e., 16. Other settings are as Sect. 3. Simulation time is 5,000 s and each result is based on 10 different runs to reduce the existing variation in the sim-

• ulation. This configuration is chosen based on the results of

previous experiments reported in Sect. 3. In RPGM, there are 3 different groups moving together. One group includes 4 mobile nodes while the others include 3 mobile nodes. In the Random Walk model, a constant time interval t = 10 s is chosen to choose the new movement. In addition, in Restricted RWP model, the area is partitioned to 4 equal sized sub-areas which have an overlap equal to 40% of the area length and the mobile nodes are distributed uniformly in these sub-areas. Furthermore, to evaluate the effect of tag placement techniques on TBR, two scenarios are studied. In scenario 1, TBR performance is measured based on the same instances of traffic and mobility model which are used in each relay placement strategy. The results are shown in Table 3. According to Table 3, the GA approach exhibits the best performance among other approaches and the ThrowBox approach has the best second performance. The reason that the GA outperforms the ThrowBox approach is that the GA approach searches the area for a set of place- ment solutions than searching for a single tag position at a time as in ThrowBox. Placing the relays on a regular grid and the relay pruning approaches have the worst perfor- mance as they place the relays on a grid with no respect to the mobility models of the nodes. Accordingly, they do not place many relays in strategic locations. Although the relay pruning approach trims some of the relays responsible for delivering the least number of messages, its performance is very poor to make it a good alternative. The probability based distribution approach performs much better than the regular grid placement and the relay pruning; however, its performance is much worse than the GA approach. In scenario 2, new instances of traffic and mobility models are used in relay placement techniques. This sce- nario is more general to be used as it is independent to the traffic model. The results are shown in Table 4. According to Table 4, the GA approach is the best in ranking among

• ther placement techniques except for the Random Walk

mobility model where ThrowBox competes with the GA. For the same reason mentioned above, the GA approach can learn spatial characteristics of the relays in a more effective way than other approaches for all mobility models except the Random Walk which results in having a better performance than other approaches. More investigation is required to find out why the GA approach is unable to effectively learn the Random Walk characteristics. One hypothesis to describe this situation can be related to the limited iteration number of learning process used in the

• GA. As the Random Walk has the highest random behavior

Table 3 Comparison of tag placement techniques (same traffic/mobility instance) RWP Restricted RWP RWalk RPGM q Lavg Ea Nr

b

q Lavg Ea Nr

b

q Lavg Ea Nr

b

q Lavg Ea Nr

b

• Reg. Grid

.122 192.5 6.34 16 .102 214.2 4.76 16 .149 293.8 5.07 16 .123 172.4 7.13 16 Relay Prun. .113 168.3 6.71 6 .091 182.3 4.99 7 .151 293.2 5.15 16 .114 168.7 6.76 9

• Prob. based

.286 153.9 18.6 16 .305 118.7 25.7 16 .183 205.5 8.9 16 .256 146.7 17.5 16 GA .424 100.1 42.4 16 .491 71.7 68.5 16 .308 149.9 20.6 16 .402 99.0 40.6 16 Throwbox .396 114.7 34.5 16 .493 84.9 58.1 16 .313 157.3 19.9 16 .398 103.8 38.3 16

a

times 10-4

b

Nrelay Table 4 Comparison of tag placement techniques (different traffic/mobility instance) RWP Restricted RWP RWalk RPGM q Lavg Ea Nr

b

q Lavg Ea Nr

b

q Lavg Ea Nr

b

q Lavg Ea Nr

b

• Reg. Grid

.121 197.1 6.14 16 .099 218.4 4.53 16 .147 296.7 4.95 16 .125 183.3 6.82 16 Relay Prun. .111 172.3 6.44 6 .093 192.6 4.83 7 .142 297.6 4.77 14 .109 181.2 6.02 9

• Prob. based

.267 155.4 17.2 16 .29 119.2 24.3 16 .169 215.9 7.83 16 .241 152.31 15.8 16 GA .349 115.7 30.1 16 .445 82.1 54.2 16 .213 193.3 11 16 .346 116.5 29.7 16 Throwbox .348 120.6 28.9 16 .449 84.7 53 16 .222 196.4 11.3 16 .328 118.8 27.6 16 22 Wireless Netw (2012) 18:9–31

123

SLIDE 15

among the other mobility models, by adding a random traffic model the problem of learning spatial characteristics

• f the relays becomes more complicated and needs more
• bservations. The analytical modeling of the placement

techniques is out of the scope of this paper and can be found in our priori work [51]. 6.4.1 Discussion We initialize all the relay placement techniques by utiliz- ing uniform grid placement and later each relay placement technique optimizes the location of relays by replacing their positions based on the given mobility model. In other words, our relay placement optimization techniques learn the spatial characteristics of the mobile nodes through a learning phase and later they place the relays on strategic

• locations. We ran another experiment using probability

based

• ptimization

relay placement technique where the relay nodes initially are placed in random locations. We used the following setting: Ntag = 441, Nnode = 10, X/Ymin/max = 1, 000 m, Bnode = 4 9 Brelay, r = 5, Smin/max = [2.5, 7.5], and k = 0.5. Figure 10 shows the corresponding

• results. According to Fig. 10 the performance of the net-

work in both cases are very close in terms of message delivery rate and delay. The relay placement techniques proposed in this paper is based on simulation results. However, the connections between a mobility model and the relay placement strate- gies are presented in our priori work [51], where we pro- posed a generic analytical model in order to evaluate the performance of relay placement strategies in DTNs. 6.5 Buffer management policy In a typical DTN, the mobile nodes are disconnected for a relatively large period of time to any other mobile/relay

• nodes. During this time, due to their limited buffer size

they have to evict some messages to accommodate new

• messages. Additionally, when a mobile node forwards/

reads some messages to/from a relay node, both mobile and relay nodes may have to evict some other messages from their buffer to give space to the incoming messages. Therefore, a buffer management policy is quite demanding to define which messages should be dropped, if the buffer is full, when a new message has to be accommodated. One possible way to enhance our proposed protocol is by deleting the delivered messages if they exist in any other

• buffers. In this case, destination nodes delete the copies of

previously delivered messages from encountered relay

• nodes. Figure 11 shows that using this simple technique,

the performance of TBR in terms of message delivery rate is increased while the latency is almost at the same level. We also evaluated TBR in this paper based on the first in first out (FIFO) queuing Policy. In FIFO queuing Policy the message that first entered the queue is the first message to be evicted from the queue. This policy is easy to be implemented but it can be inefficient as it does not consider

• ther useful information about the probability of a message

reaching the destination. We leave improving buffer management policies to future work. For example, another queuing policy can be evicting the most forwarded mes- sages first (MOFO) in which nodes/relays keep track of the number of times each message has been forwarded and based on this information they evict those messages that have been forwarded the largest number of times. Conse- quently, nodes/relays can provide a higher chance of for- warding to the messages that have been forwarded fewer times. 6.6 Possibility of using a single base station to cover the area In this section, we study the possibility of using a single super tag placed in the middle of the area instead of using many small tags distributed over the area. Instead of using many small tags, one can use a few super tags to achieve similar delivery performance; however, there is a trade-off between the power usage and using longer radio range. In this section, we evaluate the the delivery performance of a super tag, i.e., a base-station, placed in the middle of area versus of its broadcasting range and later we show the effectiveness of using the base-station in comparison to

50 100 0.2 0.4 0.6 0.8 1

Tag Buffer Size Message Delivery Rate

50 100 200 400 600 800 1000 1200

Tag Buffer Size Message Delay

• Reg. Grid Based

Random Based

• Reg. Grid Based

Random Based

(a) (b)

• Fig. 10 By initializing the

position of the relay nodes with different techniques, i.e., on a regular grid and randomly chosen locations, there is no significant changes in the performance of the network using probability based

• ptimization placement
• technique. a Message delivery

rate versus tag buffer size, b Message delay versus tag buffer size Wireless Netw (2012) 18:9–31 23

123

SLIDE 16

using multiple small tags introduced in this paper. The results show that using the latter alternative we achieve better performance in terms of power usage. In order to evaluate the performance of using a single base-station placed in the middle of the region we ran an

• experiment. In the experiment we chose same configuration

as scenario 1 in Sect. 4 except Nrelay = 1 and Brelay = 1,600. Our objective function is to maximize q and mini- mize Lavg simultaneously. Conceptually, network ineffi- ciency could be defined vice versa given as Lavg/q. The goal is then to minimize network inefficiency. Figure 12(a) shows that an increased broadcast range of the corresponding base station improves the performance of the network in terms of increasing message delivery rate and decreasing message delay together. As an example, for the RWP mobility model, when the broadcast range of the base station approaches 55m, the performance is almost the same as the GA in Table 3. The covered area by tags is equal to 942.48m2 (16 3

4p52) but the covered area by the

base station is 7127.49m2 (3

4p552); therefore, the required

area to be covered by the base station is 7.5625 times

• larger. If we assume that to cover each unit of area we need

the same amount of energy, then to achieve the same delivery performance using a base-station we need 750%

50 100 0.05 0.1 0.15 0.2 0.25

Tag Buffer Size Delivery Rate

Optimized bf. Regular bf.

50 100 50 60 70 80

Tag Buffer Size Delay

Optimized bf. Regular bf.

(a) (b)

• Fig. 11 By deleting the copies of previously delivered messages the

performance of TBR is increased. We used the following setting: Ntag = 36, Nnode = 10, X/Ymin/max = 1, 000m, d = 200, Bnode = 1000, Brelay = 100, r = 50, Smin/max = [20, 30], k = 0.5, and the mobility model used is random waypoint. Relays are placed on a regular grid. a Delivery rate versus tag buffer size, b Delay versus tag buffer size

(a) (c) (d) (b)

• Fig. 12 Single base station

results and application. a Network inefficiency versusbase station broadcast range, b Packet delivery rate versusbase station broadcast range, c Average end- to-end delay versusbase station broadcast range, d Inhospitable terrain inside the area thatdoesn’t allow deployability

• f any basestation in the middle

24 Wireless Netw (2012) 18:9–31

123

SLIDE 17

more power than using multiple tags. In addition,

• Fig. 12(b, c) show the delivery rate and average end-to-end

delay versus base station range respectively. In some applications, such as military, relying on a single base station may end up with a point of vulnerability which may be attacked and, if destroyed, terminate the network operation. Additionally, in some scenarios such as ring like areas, it is not possible to place a base station in the middle of the area which is unreachable. Alternatively, we can place the tags all over the rings (Fig. 12(d)). 7 Comparisons to existing work 7.1 Comparison to epidemic routing In prior work, Vahdat et al. [56] presented a routing pro- tocol called Epidemic Routing (ER). In ER, every host acts as a carrier to distribute application messages and when- ever hosts meet each other they exchange all the messages together; therefore, after a while all the messages will spread over the network. In this section we will show how we can improve ER performance reported in [56] using TBR considering that we have involved a number of sta- tionary RFID tags in the network. Figure 13(b) shows the CDF of message delay for var- ious transmission ranges in TBR. To have a fair compari- son to ER, we use the same parameters as following: the area size is 1,500 9 300, the average speed of mobile nodes is 10 (Smin = 0, Smax = 20). The message rate is 1 (k = 1) and d = 50 () Nrelays = 217). The buffer size for all the tags and mobile nodes is 1,000 messages. Most of the messages are delivered by time 100, when r C 25. In ER, this phenomenon happens only when the transmission range, i.e., the wireless range between nodes, is more than 100. Furthermore, according to Table 5(a), ER results show that it is not scalable due to its weak perfor- mance for small transmission ranges, where e.g. it takes 44,829.7 s to deliver a message, on average. ER has a better performance when we have a high density network, i.e., 50 mobile nodes with transmission range of higher than 100 in an area 1,500 9 300 m2. If we scale the size of network from 1,500 9 300 to a larger size or reduce the number of participating nodes, using even a larger transmission range, TBR will outper- form the ER approach. In addition, the message delivery rate of TBR will be higher in very sparse networks. We have also investigated the performance of TBR and ER by using different MN/RN buffer sizes. According to Table 5(b), when the transmission range is 50 m, TBR has a higher delivery performance than ER. This case is more significant when we have a limited buffer size. In ER, they also presented the CDF for bounded resource consumption, i.e., bounded buffer sizes. In this case, the amount of buffer space in the nodes is limited. Assuming all the nodes and tags in TBR have the same buffer space and all the parameters are the same as earlier with a transmission range of 50 for all the tags, the results for message delivery rate are plotted in Fig. 13(a). 7.2 Comparison to message ferrying In prior work, Zhao et al. [65] presented a Message Fer- rying (MF) approach for data delivery in sparse MANETs. They described two approaches called NIMF and FIMF. We chose the NIMF (Node-Initiated MF) to be compared with our work based on their results reported in [65]. We ran our simulation based on the same parameters reported in [65] as following: the area is 5, 000 9 5, 000, there is 1 ferry with average speed of 15m/s, Smin = 0, Smax = 5, k = 1.25, d = 50 ()Nrelays = 10,201), Nnode = 40 and r = 20. Note that from a comparison point of view the transmission range of the tags is much less than the transmission range of the mobile nodes in NIMF. Figure 13(c, d) show the effect of node/tag buffer size

• n average message delay and message delivery rate for

different approaches including TBR, NIMF, and ER. The delivery rate in TBR is significantly better than other

Table 5 TBR versus ER Range ER TBR ER TBR r q q Lavg Lavg (a) Different transmission range 250a 100 100 0.2 7.8 100 100 100 12.8 15.5 50 100 100 153.0 42.7 25 100 100 618.9 90.6 10 89.9 100 44829.7 248.9 Buffer size ER TBR ER TBR q q Lavg Lavg (b) Different buffer size 2,000 100 100 147.3 40.6 1,000 100 100 148.7 41.6 500 100 100 149.2 43.8 200 99.6 100 152.0 46 100 95.2 99.5 157.5 41.9 50 79.7 94.1 148.2 41.4 20 50.2 73.1 129.5 35.2 10 29.3 53.2 98.9 34.3

a This network is not sparse by our definition (average number of

connections is more than 20% and our definition requires less than 5%) Wireless Netw (2012) 18:9–31 25

123

SLIDE 18

approaches, because of the added buffer capacity of the tags, which makes it suitable for bounded resource, lossless

• applications. The message delay in TBR seems to be high

for small buffer size but we believe that this higher delay is due to having more messages delivered because there are some messages in the network which are delivered after a long time and in NIMF they are dropped. This phenomenon is intensified in ER which has a low delivery late (according to Fig. 13(c), by increasing buffer size in ER/ NIMF, the delay becomes higher which confirms our belief). In addition, although TBR has a greater delivery rate with buffer size of 800, the message delay is less than MF. Furthermore, our comparisons are based on a regular array of tags. Using either probability based or genetic algorithm based tag placement would further increase the performance of TBR. 8 Conclusion In this paper, we studied the applicability of using point locations as ‘‘mailboxes‘‘ for storage/retrieval of messages to facilitate ubiquitous network connectivity. A practical implementation of our work could use low cost, tiny, unattached-power RFID tags. Our approach is an alterna- tive, for achieving competitively low message delay and high message delivery rates, to existing methods that rely

• n altering/controlling some of the mobile nodes’ move-
• ments. We designed and analyzed a new protocol called

TBR (Tag-based Routing) in which mobile nodes com- municate to each other only via passive relay nodes such as passive RFID tags. Our results show that TBR is effective at expanding the connectivity of DTNs, over a broad range

• f parameter values. Our TBR protocol can provide lower

message delay and higher message delivery rates than existing methods including NIMF and ER. We also showed heuristics methods of placing tags over the geographic region, including relay pruning, a probabilistic distribution based on relay utility and a genetic algorithm approach that attempt to minimize the message delay/rate ratio. The genetic algorithm was shown to be a superior tag place- ment technique. In our future work we will study TBR for intelligent buffer policy management to increase its effectiveness, more advanced relay placement techniques and develop a rigorous analytical framework. DTN definition The performance of the traditional routing protocols for mobile ad hoc networks (MANETs) or mesh networks [24, 42, 43, 48] in terms of packet delivery rate is significantly decreased at a point when the network is getting sparse. To find that point, we ran an experiment shown in the Fig. 14. In Fig. 14 we show a number of well known MANET routing protocols (e.g. AODV [42], DSDV [43], DSR [24], and ADIAN [48]) compared to random gossiping in terms

• f the fraction of delivered packets versus average per-

centage of available contacts. To do this, we used NS2. We

100 200 300 400 20 40 60 80 100

Message Latency Message Delivery Rate

500 1000 20 40 60 80 100

Message Latency Message Delivery Rate

200 400 600 800 2000 4000 6000

Tag/Node Buffer Size Message delay (sec)

200 400 600 800 20 40 60 80 100

Tag/Node Buffer Size Message Delivery Rate

BS=10 BS=20 BS=50 BS=100 BS=200 BS=500 BS=1000 BS=2000 r=10 r=25 r=50 r=100 r=250 TBRP NIMF ER TBRP NIMF ER

(a) (c) (d) (b)

• Fig. 13 TBR’s CDF for

message delivery rate versus message delay, (a) and (b). Note that these CDFs cumulate to the percentage of delivered

• messages. Comparisons to

existing work are shown in (c) and (d). a CDF for message delivery rateas a function of available buffer space for r = 50, b CDF for message delivery rateas a function of tage range, r, c The effect of tag buffer sizeon the average message delivery time, d The effect of tag buffer sizeon the message delivery rate 26 Wireless Netw (2012) 18:9–31

123

SLIDE 19

set a fixed number of mobile nodes, i.e., 20, while increasing the area within which they could move, thus increasing sparseness. The data traffic used in the simula- tion is CBR with a rate of 8 kbps. Mobile nodes move according to the RWP mobility model with a pause time of 0 s and maximum allowed speed of 3 m/s. The simulation time is 1,000 s and radio range of the nodes is 250 m. In the Random Gossip protocol, each node picks a connected node at random and forwards the packet. The maximum number of possible contacts between each pair of nodes can be calculated as follows: Max Contact No: ¼ Nnode Nnode 1 ð Þ=2: According to Fig. 14, the other protocols converge to the performance of the Random Gossip protocol when the total contacts are about 5% of all 190 possible contacts between each pair of nodes. GA placement details In our GA placement approach, the genome is represented as a sequence of relay coordinates: x1; y1; x2; y2; . . .; xn; yn ½

• where xi; yi 2 0::1

½ (0. . .1 is a normalized coordinate). A single population was used of 120 genomes, with each genome initialized by placing relays uniformly at random in the region. The number of elite genomes was set to 10 and the search was run for 100 iterations. We use Genetic Algorithm Solver Toolbox provided by Matlab for our

• simulation. Through some preliminary experiments we

determined some GA parameters values that improved GA performance which is presented in Table 6. Other param- eters are set as the default value in the latter toolbox. We did not do an exhaustive search over the entire parameter

• space. We leave this to future work. The mutation and

crossover functions were customized for our problem. We hypothesized that the structure/topology of the geographic relationships between relays is important in terms of per-

• formance. In order to allow the GA to learn about spatial

characteristics of relay placement, we identified regions of relays using a breadth first tree construction based on the Delaunay graph defined by the genome. Initially, we sim- ply chose relays at random, rather than using a breath first tree approach and the resulting GA performance was comparably poor. We use the Delaunay triangulation operator in Matlab to generate edges between spatially close nodes, creating a

• mesh. The choice of the Delaunay triangulation is arbitrary

and unimportant; there are many different ways to generate these edges. We then select a node at random in the mesh and construct a tree using a breadth search search from that node, with a target number of nodes to be included in the tree. Figure 15(a, b), show two genomes, each consisting of 50 relays. The Delaunay graph of the relays is shown using light lines. A random breadth first tree is constructed by choosing a relay at random and forming a breadth first tree that consists of the required number of relays. An example set of relays in such a tree is shown using solid dots for each genome. The child shown in Fig. 15(c) is constructed from the selected trees in genome 1 and genome 2. In practice, for

• ur crossover function, the number of trees and the number

10% 20% 30% 40% 50% 10 20 30 40 50 60 70 80 90 100

Contact_Fraction Percentage Packet Delivery Ratio

ADIAN DSR AODV DSDV RndGossip

• Fig. 14 Packet delivery rate versus fraction of available contacts to

all potential contacts between each pair of nodes Table 6 GA parameter settings Parameter name Description Parameter value Population size number of individuals 120 Elite count Number of best individuals that survive to next generation without any change 10 Crossover fraction The fraction of genes swapped between individuals 0.8 Mutation probability The probability that how often will be parts of chromosome mutated 0.05 Generations Maximum number of generations allowed 100 Migration interval The number of generations between the migration of the fittest individuals to other sub-populations 5 Migration fraction Fraction of those individuals scoring the best that will migrate 0.2 Wireless Netw (2012) 18:9–31 27

123

SLIDE 20

• f relays in each tree selected from genome 1 is random-

ized, i.e., we choose a small number of random trees from genome 1. The resulting number of relays (and number of trees) selected from genome 2 is constrained so that the total number of relays in the child equals Nrelay. If genome 1 and 2 have identical selected relays (as in the case when they have common ancestors) then one of the identical relays is replaced with a point selected at random in the region. As an example representing crossover operator, assume the following two genomes: g1 ¼ x1;1; y1;1; x1;2; y1;2; :::; x1;n; y1;n

• ;

g2 ¼ x2;1; y2;1; x2;2; y2;2; :::; x2;n; y2;n

• :

The crossover function then takes a subset of points from g1, and the remaining points from g2. Therefore, the result

• f crossover operator would be a new genome as follows:

g3 ¼ g1 g2 ¼ x3;1; y3;1; x3;2; y3;2; . . .; x3;n; y3;n

• Our mutation operation similarly selects a random

breadth first tree (in practice consisting of up to 5% of the relays). For mutation, the selected relays are replaced with relays chosen at random in the region. Figure 15(d) is a mutation of genome 2. Note that a ‘‘hole‘‘ appears where the selected relays were, since the new relays are less likely to appear in the selected region (for a small number of selected relays and hence a small selected region). References

• 1. Akyildiz, I. F., Pompili, D., & Melodia, T. (2005). Underwater

acoustic sensor networks: Research challenges. Ad Hoc Networks 3(3), 257–279.

• 2. Al Hanbali, A., Ibrahim, M., Simon, V., Varga, E., & Carreras, I.

(2008). A survey of message diffusion protocols in mobile ad hoc

• networks. In Proceedings of the 3rd international conference on

performance evaluation methodologies and tools (ICST, Brussels, Belgium, Belgium), ValueTools ’08 (pp. 82:1–82:16). ICST (Institute for Computer Sciences, Social-Informatics and Tele- communications Engineering).

• 3. Baghaei-Nejad, M., Mendoza, D. S., Zou, Z., Radiom, S., Gielen,

G., Zheng, L. -R., & Tenhunen, H. (2009). A remote-powered rfid tag with 10mb/s uwb uplink and -18.5 dbm sensitivity uhf downlink in 0.18 lm cmos. In Solid–state circuits conference— digest of technical papers, 2009. ISSCC 2009. IEEE International (pp. 198–199,199a).

• 4. Balasubramanian, A., Levine, B., & Venkataramani, A. (2007).

Dtn routing as a resource allocation problem. SIGCOMM Com- puter Communication Review, 37(4), 373–384.

• 5. Banerjee, N., Corner, M. D., & Levine, B. N. (2007). An energy-

efficient architecture for dtn throwboxes. INFOCOM 2007. 26th IEEE international conference on computer communications. IEEE (pp. 776–784).

• 6. Banerjee, N., Corner, M. D., Towsley, D., & Levine, B. N.

(2008). Relays, base stations, and meshes: Enhancing mobile networks with infrastructure. In MobiCom ’08: Proceedings of the 14th ACM international conference on mobile computing and networking (pp. 81–91). New York, NY, USA: ACM.

• 7. Bettstetter, C. (2001). Smooth is better than sharp: A random

mobility model for simulation of wireless networks. In MSWIM ’01: Proceedings of the 4th ACM international workshop on

0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1

(a) (c) (d) (b)

• Fig. 15 Example genetic

algorithm crossover and

• mutation. a Genome 1,

b Genome 2, c Crossover of genome 1 and 2, d Mutation of genome 2 28 Wireless Netw (2012) 18:9–31

123

SLIDE 21

modeling, analysis and simulation of wireless and mobile systems (pp. 19–27). NY, USA: ACM.

• 8. Bettstetter, C., Hartenstein, H., & Perez-Costa, X. (2002). Sto-

chastic properties of the random waypoint mobility model: epoch length, direction distribution, and cell change rate. In MSWiM ’02: Proceedings of the 5th ACM international workshop on modeling analysis and simulation of wireless and mobile systems (pp. 7–14). NY, USA: ACM.

• 9. Tariq, M. M. B., Ammar, M., & Zegura, E. (2006). Message ferry

route design for sparse ad hoc networks with mobile nodes. In MobiHoc ’06: Proceedings of the 7th ACM international sym- posium on Mobile ad hoc networking and computing (pp. 37–48). New York, NY, USA: ACM.

• 10. Burgess, J., Gallagher, B., Jensen, D., & Levine, B. N. (2006).

Maxprop: Routing for vehicle-based disruption-tolerant net-

• works. INFOCOM 2006. 25th IEEE international conference on

computer communications. Proceedings (pp. 1–11).

• 11. Camp J. B., & Davies, V. (2002). A survey of mobility models for

ad hoc network research. Wireless Communication and Mobile Computing (WCMC), 2(5), 483–502.

• 12. Chen, W., Guha, R. K., Kwon, T. J., Lee, J., & Yuan-Ying, H.

(2009). A survey and challenges in routing and data dissemina- tion in vehicular ad hoc networks. Wireless Communications and Mobile Computing.

• 13. Daly, E. M., & Haahr, M. (2007). Social network analysis for

routing in disconnected delay-tolerant manets. In Proceedings of the 8th ACM international symposium on mobile ad hoc net- working and computing, MobiHoc ’07 (pp. 32–40). New York, NY, USA: ACM.

• 14. de Oliveira, E. C. R., & de Albuquerque, C. V. N. (2009). Nectar:

A dtn routing protocol based on neighborhood contact history. Proceedings of the 2009 ACM symposium on applied computing, SAC ’09 (pp. 40–46). New York, NY, USA: ACM.

• 15. Fall, K. (2003). A delay-tolerant network architecture for chal-

lenged internets. In SIGCOMM ’03: Proceedings of the 2003 conference on applications, technologies, architectures, and protocols for computer communications (pp. 27–34). New York, NY, USA: ACM.

• 16. Farahmand, F., Cerutti, R. I., Patel, A. N., Jue, J. P., & Rodrigues,
• J. J. P. C. (2009). Performance of vehicular delay-tolerant net-

works with relay nodes. Wireless Communications and Mobile Computing.

• 17. Gao, W., Li, Q., Zhao, B., & Cao, G. (2009). Multicasting in

delay tolerant networks: A social network perspective. In Pro- ceedings of the tenth ACM international symposium on Mobile ad hoc networking and computing, MobiHoc ’09 (pp. 299–308). New York, NY, USA: ACM.

• 18. Goldenberg, D. K., Lin, J., Morse, A. S., Rosen, B. E., & Richard
• Y. Y. (2004). Towards mobility as a network control primitive. In

MobiHoc ’04: Proceedings of the 5th ACM international sym- posium on Mobile ad hoc networking and computing (pp. 163–174). NY, USA: ACM.

• 19. Han, J. K., Park, B. S., Choi, Y. S., & Park, H. K. (2001). Genetic

approach with a new representation for base station placement in mobile communications. In Vehicular technology conference,

• 2001. VTC 2001 Fall. IEEE VTS 54th (Vol. 4, pp. 2703 –2707).
• 20. Hong, X., Gerla, M., Pei, G., & Chiang, C.-C. (1999). A group

mobility model for ad hoc wireless networks. In MSWiM ’99: Proceedings of the 2nd ACM international workshop on model- ing, analysis and simulation of wireless and mobile systems (pp. 53–60). NY, USA: ACM.

• 21. Hui, P., Crowcroft, J., & Yoneki, E. (2008). Bubble rap: Social-

based forwarding in delay tolerant networks. In Proceedings of the 9th ACM international symposium on mobile ad hoc net- working and computing, MobiHoc ’08 (pp. 241–250). New York, NY, USA: ACM.

• 22. Ibrahim, M., Al Hanbali, A., & Nain, P. (2007). Delay and

resource analysis in manets in presence of throwboxes. Perfor- mance Evalution, 64(9–12), 933–947.

• 23. Jain, S., Fall, K., & Patra, R. (2004). Routing in a delay tolerant
• network. In SIGCOMM ’04: Proceedings of the 2004 conference on

applications, technologies, architectures, and protocols for com- puter communications (pp. 145–158). New York, NY, USA: ACM.

• 24. Johnson, D. B., & Maltz, D. A. (1996). Dynamic source routing

in ad hoc wireless networks. In: T. Imielinski & Korth, H. (Eds.), Mobile computing (pp. 153–181). Dordrecht: Kluwer.

• 25. Juang, P., Oki, H., Wang, Y., Martonosi, M., Peh, L. S., &

Rubenstein, D. (2002). Energy-efficient computing for wildlife tracking: Design tradeoffs and early experiences with zebranet. SIGPLAN, 37, 96–107.

• 26. Dheeraj, K., Kwan-Wu, C., & Raad, R. (2009). On the energy

consumption of pure and slotted aloha based rfid anti-collision

• protocols. Computer Communications, 32, 961–973.
• 27. Lahde, S., Doering, M., Pttner, W.-B., Lammert, G., & Wolf, L.

(2007). A practical analysis of communication characteristics for mobile and distributed pollution measurements on the road. Wireless Communications and Mobile Computing, 7(10), 1209–1218.

• 28. LeBrun, J., Chuah, C.-N., Ghosal, D., & Zhang, M. (2005).

Knowledge-based opportunistic forwarding in vehicular wireless ad hoc networks. In Vehicular technology conference, 2005. VTC 2005-Spring. 2005 IEEE 61st (Vol. 4, pp. 2289–2293).

• 29. Leguay, J., Friedman, T., & Conan, V. (2006). Evaluating

mobility pattern space routing for dtns, INFOCOM 2006. In 25th IEEE international conference on computer communications. Proceedings (pp. 1 –10).

• 30. Li, Q., & Rus, D. (2000). Sending messages to mobile users in

disconnected ad-hoc wireless networks. In Proceedings of the 6th annual international conference on mobile computing and net- working (pp. 44–55). New York, NY, USA: MobiCom ’00, ACM.

• 31. Lin, X., & Chen, H.-H. (2010). A secure and efficient rsu-aided

bundle forwarding protocol for vehicular delay tolerant networks. In Wireless communications and mobile computing.

• 32. Lindgren, A., Doria, A., & Schele

´n, O. (2004). Probabilistic routing in intermittently connected networks. Lecture Notes in Computer Science 3126, 239–254.

• 33. Ma, Y., & Jamalipour, A. (2009). Optimized message delivery

framework using fuzzy logic for intermittently connected mobile ad hoc networks. Wireless Communications and Mobile Com- puting, 9(4), 501–512.

• 34. Maral, G., Bousquet, M., & Sun, Z. (2009). Satellite communi-

cations systems: Systems, techniques and technology, 5th edn., communication and distributed systems. New Jersey: Wiley.

• 35. Meunier, H., Talbi, E.-G. & Reininger P. (2000). A multiobjec-

tive genetic algorithm for radio network optimization, evolu- tionary computation, 2000. In Proceedings of the 2000 congress

• n (Vol. 1, pp. 317 –324).
• 36. Melanie, M. (1998). An introduction to genetic algorithms.

Cambridge, MA, USA: MIT Press.

• 37. Musolesi, M., Hailes, S., & Mascolo, C. (2002). Adaptive routing for

intermittently connected mobile ad hoc networks. In World of wireless mobile and multimedia networks, 2005. WoWMoM 2005. SixthIEEEInternationalSymposiumona,June2005(pp.183–189).

• 38. Ott J., & Kutscher D.(2005). A disconnection-tolerant transport

for drive-thru internet environments, INFOCOM 2005. 24th annual joint conference of the ieee computer and communications

• societies. Proceedings IEEE (Vol. 3, pp. 1849–1862).
• 39. Pabst, R., Walke, B. H., Schultz, D. C., Herhold, P., Yaniko-

meroglu, H., Mukherjee, S., Viswanathan, H., Lott, M., Zirwas, W.,Dohler, M., Aghvami, H., Falconer, D. D. & Fettweis, G. P. (2004). Relay-based deployment concepts for wireless and mobile broadband radio. Communications magazine, IEEE (Vol. 42, pp. 80–89. Wireless Netw (2012) 18:9–31 29

123

SLIDE 22

• 40. Partan, J., Kurose, J., & Levine B. N. (2007). A survey of prac-

tical issues in underwater networks. SIGMOBILE Mobile Com- puter Communication Review, 11, 23–33.

• 41. Pei, G., Gerla, M., & Hong, X. (2000). Lanmar: Landmark

routing for large scale wireless ad hoc networks with group

• mobility. In Proceedings of the 1st ACM international symposium
• n mobile ad hoc networking & computing (pp. 11–18), Piscat-

away, NJ, USA: MobiHoc ’00, IEEE Press.

• 42. Perkins, C. E., & Royer, E. M. (1999). Ad-hoc on-demand dis-

tance vector routing. In Mobile computing systems and applica- tions 1999. Proceedings. WMCSA ’99. Second IEEE Workshop

• n (pp. 90 –100).
• 43. Perkins, C. E., & Bhagwat, P. (1994). Highly dynamic destina-

tion-sequenced distance-vector routing (dsdv) for mobile com-

• puters. SIGCOMM Computer Communication Review, 24(4),

234–244.

• 44. Perur, S., & Iyer, S. (2006). Characterization of a connectivity

measure for sparse wireless multi-hop networks. Distributed computing systems workshops. International conference on (p. 80).

• 45. Ristic, B., Arulampalam, S., & Gordon, N. (2004). Beyond the

Kalman filter: Particle filters for tracking applications. Boston: Artech House.

• 46. Santi, P., & Blough, D. M. (2003). The critical transmitting range

for connectivity in sparse wireless ad hoc networks. Mobile Computing, IEEE Transactions on, 2(1), 25–39.

• 47. Seth, A., Kroeker, D., Zaharia, M., Guo, S., & Keshav, S. (2006).

Low-cost communication for rural internet kiosks using mechanical backhaul. In MobiCom ’06: Proceedings of the 12th annual international conference on Mobile computing and net- working (pp. 334–345). NY, USA: ACM.

• 48. Shahbazi, S., Ghassem-Sani, G., Rabiee, H., Ghanbari, M., &

Dehghan, M. (2006). Adian: A distributed intelligent ad-hoc

• network. In Distributed computing and networking, lecture notes

in computer science (Vol. 4308, pp. 27–39). Berlin, Heidelberg: Springer.

• 49. Shahbazi, S., Harwood, A., & Karunasekera, S. (2008). Achiev-

ing ubiquitous network connectivity using an rfid tag-based routing protocol. In ICPADS ’08: Proceedings of the 2008 14th IEEE international conference on parallel and distributed sys- tems (pp. 391–398).

• 50. Shahbazi, S., Harwood, A., & Karunasekera, S. (2009). An ana-

lytical model for performance evaluation in sparse mobile ad hoc

• networks. In WD’09: Proceedings of the 2nd IFIP conference on

wireless days (pp. 236–241). Piscataway, NJ, USA: WD’09, IEEE Press.

• 51. Shahbazi, S., Harwood, A., & Karunasekera, S. (2011). On

placement of passive stationary relay points in delay tolerant

• networking. In AINA 2011: Advanced information networking

and applications, international conference on (pp. 764–771). Los Alamitos, CA, USA: IEEE Computer Society.

• 52. Spyropoulos, T., Psounis, K., & Raghavendra, C. S. (2005). Spray

and wait: an efficient routing scheme for intermittently connected mobile networks. In WDTN ’05: Proceedings of the 2005 ACM SIGCOMM workshop

• n

delay-tolerant networking (pp. 252–259). NY, USA: ACM.

• 53. Spyropoulos, T., Psounis, K., & Raghavendra, C. S. (2007). Spray

and focus: Efficient mobility-assisted routing for heterogeneous and correlated mobility. In Pervasive computing and communi- cations workshops, 2007. PerCom workshops ’07. Fifth annual IEEE international conference on (pp. 79–85).

• 54. Ting, C.-K., Lee, C.-N., Chang, H.-C., & Wu, J.-S. (2009). Wireless

heterogeneous transmitter placement using multiobjective vari- able-length genetic algorithm. Systems, man, and cybernetics, part B: Cybernetics. IEEE Transactions on, 39, 945 –958.

• 55. Tournoux, P.-U., Leguay, J., Benbadis, F., Conan, V., de Amo-

rim, M.D., & Whitbeck J. (2009). The accordion phenomenon: Analysis, characterization, and impact on dtn routing. INFO- COM, IEEE (pp. 1116 –1124).

• 56. Vahdat, A., & Becker, D. (2000). Epidemic routing for partially-

connected ad hoc networks. Tech. report, Duke University CS- 2000-06.

• 57. Wang Y., Dang H., & Hongyi, W. (2007). A survey on analytic

studies of delay-tolerant mobile sensor networks. Wireless Communications and Mobile Computing, 7(10), 1197–1208.

• 58. Whitehouse, K., Woo, A., Jiang, F., Polastre, J., & Culler, D.

(2005). Exploiting the capture effect for collision detection and recovery, Embedded Networked Sensors, 2005. In EmNetS-II. The second IEEE workshop on (pp. 45–52).

• 59. Wu, J., Yang, S., & Dai, F. (2007). Logarithmic store-carry-

forward routing in mobile ad hoc networks. Parallel and Dis- tributed Systems, IEEE Transactions on, 18(6), 735–748.

• 60. Yang, J., Chen, Y., Ammar, M., & Lee, C. (2005). Ferry

replacement protocols in sparse manet message ferrying systems. In Wireless communications and networking conference, 2005 IEEE (Vol. 4, pp. 2038–2044).

• 61. Yuan, Q., Cardei, I., & Wu, J. (2009). Predict and relay: an

efficient routing in disruption-tolerant networks.In E.

• W. Knightly, C.-F. Chiasserini, & X. Lin, (Eds.), MobiHoc (pp.

95–104). ACM.

• 62. Zhang, Y. P., & Hwang, Y. (1998). Characterization of uhf radio

propagation channels in tunnel environments for microcellular and personal communications. Vehicular Technology, IEEE Transactions on, 47, 283–296.

• 63. Zhang Z., & Qian Z. (2007). Delay/disruption tolerant mobile ad

hoc networks: Latest developments. Wireless Communications and Mobile Computing, 7(10), 1219–1232.

• 64. Zhang, Z., Lu, Z., Pang, Z., Yan, X., Chen, Q., & Zheng, L.-R.

(2010). A low delay multiple reader passive rfid system using

• rthogonal th-ppm ir-uwb. In Computer communications and

networks (ICCCN), 2010 proceedings of 19th international con- ference on (pp. 1–6).

• 65. Zhao, W., Ammar, M., & Zegura, E. (2004). A message ferrying

approach for data delivery in sparse mobile ad hoc networks. In Proceedings of the 5th ACM international symposium on Mobile ad hoc networking and computing (pp. 187–198). New York, NY, USA: MobiHoc ’04, ACM.

• 66. Zhao, W., Chen, Y., Ammar, M., Corner, M., Levine, B., &

Zegura, E. (2006). Capacity enhancement using throwboxes in

• dtns. In: Mobile Adhoc and Sensor Systems (MASS), 2006 IEEE

International Conference on (pp. 31–40).

Author Biographies

Saeed Shahbazi received the BEng degree in computer soft- ware engineering from Iran University of Science and Tech- nology in 2002 and the M.Eng. degree in Artificial Intelligence in Sharif University of Technol-

• gy in 2005. He recieved his

Ph.D. degree in computer sce- ince and software engineering at the University of Melbourne in

• Aug. 2011. He is currently a

research engineer in National ICT of Australia. Previously, he was with Iran Telecommunica- tion Research Center and Nokia. His research interests include routing in mobile ad hoc networks, cross-layer network protocol design for 30 Wireless Netw (2012) 18:9–31

123

SLIDE 23

wireless networks, distributed systems, distributed artificial intelli- gence, and fuzzy systems. Shanika Karunasekera received the B.Sc. (Honours) degree in electronics and telecommunica- tions engineering from the Uni- versity of Moratuwa, Sri Lanka, in 1990 and the Ph.D. degree in electrical engineering from the University of Cambridge, UK, in

• 1995. From 1995 to 2002, she

was a Software Engineer and a Distinguished Member of Tech- nical Staff at Lucent Technolo- gies, Bell Labs Innovations,

• USA. Since January 2003, she

has been a Senior Lecturer at the Department of Computer Science and Software Engineering, Univer- sity of Melbourne. Her current research interests are distributed computing, software engineering and peer-to-peer computing. Dr. Karunasekera is a member of the ACM. Aaron Harwood received the BIT, B.Eng. (M.E.), and Ph.D. degrees from Griffith University, Brisbane, Australia, in 1998 and 2002, respectively. He is cur- rently a senior lecturer in the Department of Computer Sci- ence and Software Engineering, University of Melbourne, Mel-

• bourne. He is a founding member
• f the Peer-to-Peer Networks and

Applications research group. His research interests are in the per- formance of large-scale, decen- tralized, autonomous systems. He has been a member of the ACM, the IEEE, and the IEEE Computer Society since 2003. Wireless Netw (2012) 18:9–31 31

123