1166 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 14, NO.

4, FOURTH QUARTER 2012

From Delay-Tolerant Networks to

Vehicular Delay-Tolerant Networks
Paulo Rogério Pereira, Member, IEEE, Augusto Casaca, Senior Member, IEEE,
Joel J. P. C. Rodrigues, Senior Member, IEEE, Vasco N. G. J. Soares, Student Member, IEEE,
Joan Triay, Student Member, IEEE, and Cristina Cervelló-Pastor

Script started on Sat Apr 28 11:24:09 2001

Abstract—This paper provides an introductory overview of vegard@gyversalen:~$ /sbin/ifconfig tun0
Vehicular Delay-Tolerant Networks. First, an introduction to tun0 Link encap:Point-to-Point Protocol
Delay-Tolerant Networks and Vehicular Delay-Tolerant Networks inet addr:10.0.3.2 P-t-P:10.0.3.1 Mask:255.255.255.255
UP POINTOPOINT RUNNING NOARP MULTICAST MTU:150 Metric:1
is given. Delay-Tolerant schemes and protocols can help in RX packets:1 errors:0 dropped:0 overruns:0 frame:0
situations where network connectivity is sparse or with large TX packets:2 errors:0 dropped:0 overruns:0 carrier:0 collisions:0
RX bytes:88 (88.0 b) TX bytes:168 (168.0 b)
variations in density, or even when there is no end-to-end vegard@gyversalen:~$ ping -i 900 10.0.3.1
connectivity by providing a communications solution for non real- PING 10.0.3.1 (10.0.3.1): 56 data bytes
64 bytes from 10.0.3.1: icmp_seq=0 ttl=255 time=6165731.1 ms
time applications. Some special issues like routing are addressed 64 bytes from 10.0.3.1: icmp_seq=4 ttl=255 time=3211900.8 ms
in the paper and an introductory description of applications 64 bytes from 10.0.3.1: icmp_seq=2 ttl=255 time=5124922.8 ms
64 bytes from 10.0.3.1: icmp_seq=1 ttl=255 time=6388671.9 ms
and the most important projects is given. Finally, some research --- 10.0.3.1 ping statistics ---
challenges are discussed and conclusions are detailed. 9 packets transmitted, 4 packets received, 55% packet loss
round-trip min/avg/max = 3211900.8/5222806.6/6388671.9 ms
Index Terms—Vehicular Delay-Tolerant Networks, Delay/ vegard@gyversalen:~$ exit
Script done on Sat Apr 28 14:14:28 2001
Disruption-Tolerant Networks, Vehicular Ad Hoc Networks,
Intelligent Transport Systems.
Fig. 1. IP over Avian Carriers (IPoAC) session log, including the output of
a ping command registering round-trip times of about 1.5 hours and 55%
packet loss ratio.
I. I NTRODUCTION

N OWADAYS, all-time and unlimited connectivity to the

Internet seems to be fairly common for a great number of
mobile and fixed devices. However, the truth is that persistent
reception, the paper is scanned. This was implemented by
the Bergen Linux User Group [3] with the results presented
connectivity is not the rule everywhere or even in certain cir-
in Fig. 1. Nine packets were sent over a distance of ap-
cumstances not necessarily mandatory. Thus, further research
proximately 5 Km, each carried by an individual pigeon and
and technical solutions are needed in order to overcome the
containing one ping (ICMP Echo Request) packet. The ses-
lack of connectivity to enable the communications between
sion log shows that four responses were received with largely
nodes and applications in disruptive scenarios. Delay-Tolerant
variable round-trip times, averaging about 1.5 hours, and a
Networks (DTNs) [1] are networks that enable communica-
tion where connectivity issues like sparse and intermittent packet loss ratio of 55%. Naturally, with these unusually high
and largely variable delays and packet loss ratios, the Internet
connectivity, long and variable delay, high latency, high error
protocols would not work properly, resulting in timeouts and
rates, highly asymmetric data rate, and even no end-to-end
cancellations.
connectivity exist.
An example of one of these challenged network scenarios The DTN Research Group (DTNRG) [4], which was char-
is the IP over Avian Carriers (IPoAC) which is a humorously- tered as part of the Internet Research Task Force (IRTF), has
intended proposal to carry Internet Protocol (IP) traffic by proposed an architecture [5] and a communication protocol
birds such as homing pigeons. IP over Avian Carriers was [6] (the Bundle Protocol) for DTNs. In DTNs, a message-
initially described in RFC 1149 [2] released on 1st April oriented overlay layer called “Bundle Layer” is added. The
1990 (April Fools’ Day). According to the RFC, datagrams Bundle Layer exists above the transport (or other) layers
are printed in hexadecimal on a scroll of paper. The paper of the networks it interconnects (Fig. 2). Application data
is wrapped around the pigeon’s leg with duct tape. Upon units are transformed by the Bundle Layer into one or more
protocol data units called “bundles”, which are forwarded
Manuscript received 29 July 2010; revised 24 January 2011 and 18, 30, by DTN nodes according to the Bundle Protocol. The idea
and 31 July 2011.
P. Pereira and A. Casaca are with INESC-ID, Instituto Superior Técnico,
is to “bundle” together all the information required for a
Technical University of Lisbon, Rua Alves Redol, no 9, Lisboa, Portugal (e- transaction, minimizing the number of round-trip exchanges,
mail: {prbp, augusto.casaca}@inesc.pt). which is useful when the round-trip time is very large. To
J. J. P. C. Rodrigues and V. N. G. J. Soares are with Instituto de
Telecomunicaç ões, University of Beira Interior, Covilhã, Portugal (e-mail:
help routing and scheduling decisions, bundles contain an
{joeljr, vasco.g.soares}@ieee.org). originating timestamp, useful life indicator, a class of service
J. Triay and C. Cervelló-Pastor are with the Dept. of Telematics En- assignment and a length indicator.
gineering, Universitat Politècnica de Catalunya (UPC), Esteve Terradas, 7,
Castelldefels, Spain (e-mails: {joan.triay;cristina}@entel.upc.edu). The Bundle Protocol includes a hop-by-hop transfer of
Digital Object Identifier 10.1109/SURV.2011.081611.00102 reliable delivery responsibility, called bundle custody transfer,
1553-877X/12/$31.00 
c 2012 IEEE
PEREIRA et al.: FROM DELAY-TOLERANT NETWORKS TO VEHICULAR DELAY-TOLERANT NETWORKS 1167

Application Application is low, no end-to-end path will be available. So, there is a

dilemma: should the bundle be transferred to this vehicle, or
Bundle end point Bundle end point Bundle end point Bundle end point kept waiting for a better contact opportunity? An alternative
that increases the delivery probability and decreases the delay
Transport Transport Transport Transport is to transfer the bundle and keep a copy. So, there is a bundle
replication that spends transmission and storage resources.
Network Network Network Network This replication can be repeated again and again with the same
costs and possible benefits, that tend to decrease, so at least
an expiration time should exist to delete the bundle copies
Fig. 2. Bundle Protocol in the protocol stack. after some time. When the bundle reaches the destination,
an answer bundle is created and the process starts again to
send it back. The store-and-forward networking paradigm that
evolved to a packet switching paradigm has an alternative that
and an optional end-to-end acknowledgement. When nodes is a store, carry and forward paradigm, where bundles may
accept custody of a bundle, they commit to retain a copy of also be carried by network nodes from a place to another,
the bundle until such responsibility is transferred to another increasing communications efficiency. Unfortunately, carrying
node. Persistent storage may be used in DTN nodes to help is much slower than transmitting. This paradigm allows com-
combat network interruption, storing messages safely until a munication in challenging scenarios if additional delays are
contact opportunity occurs. acceptable. Naturally, the carrying and transferring of bundles
Vehicular Delay-Tolerant Networks (VDTNs) are DTNs creates a number of research challenges, e.g. what information
where vehicles communicate with each other and with fixed should be bundled together? how many bundle copies should
nodes placed along the roads in order to disseminate messages. be created? which bundles should be transferred first? can
Some of the potential applications for these networks are the present or past information about location and contacts be used
following: notification of traffic conditions (unexpected jams), to help improve efficiency? when should bundles be dropped?
road accident warnings, weather reports (ice, snow, fog, wind), can request/replies be cached? if contacts are not long enough,
advertisements (free parking spots, nearby fuel prices, etc.), should bundles be fragmented? can the other nodes be trusted?
cooperative vehicle collision avoidance, web or email access, when should names be resolved?
or even the gathering of information collected by vehicles The remaining of the paper is structured as follows. Section
such as road pavement defects. Vehicular networks have also II presents an overview of DTN. Some additional introductory
been proposed to implement transient networks to benefit tutorials on DTN are available on [4]. Section III provides
developing communities and disaster recovery networks. a detailed description of VDTNs. Section IV presents some
As an example, consider a web-based telematic application VDTN projects and applications. Section V presents some
in a vehicle, where the driver wants to receive relevant in- specific research topics. Section VI concludes the paper.
formation when entering a mountainous region. Is there snow
or other adverse weather conditions? Where is the cheapest II. D ELAY-T OLERANT N ETWORKS
nearby filling station? If there was good cellular network
coverage, the telematic device in the vehicle could send a A. Introduction
request to some server. A typical request would require one The Delay-Tolerant Networking concept was initially pro-
round-trip time (RTT) to resolve a server name to an address, posed as an approach for the InterPlaNetary Internet (IPN)
another RTT to establish a Transmission Control Protocol [7][8]. Deep space communication may suffer from very large
(TCP) connection, another RTT to send an Hypertext Transfer latencies, low data rates, possibly time-disjoint periods of
Protocol (HTTP) request, and when the answer was received reception and transmission, and intermittent scheduled con-
and interpreted, additional requests would be sent to retrieve nectivity.
additional necessary objects requiring several RTTs and some The Internet suite of protocols does not suit to DTN
transfer time. Then the connection could be closed, taking scenarios. The TCP delivers data in transmission order, so
an additional RTT. If the network connectivity is intermittent, any data loss results in at least one RTT delay and decreased
such sequence of protocol interactions may never complete throughput. Establishment of a TCP connection requires at
successfully. A solution might be bundling together a request least one RTT. If the latency exceeds the duration of the
message to resolve the address and get all the parts of the communication opportunity, no data will flow at all. The
answer. This bundle would be sent connectionless, solving slow start mechanism takes a long time to increase the
the RTT problem to a single RTT. But then there is the throughput if the RTT is large, reducing the transmission
problem of finding a route for end-to-end data transfer. If efficiency. In addition, TCP retransmissions are end-to-end,
there is no network infrastructure available, the vehicle has which may be inefficient for high loss ratios. Additionally,
to carry the message until there is a contact opportunity. buffers are required for an RTT at maximum data rate,
These contacts may be with other vehicles or infrastructure which might become sizeable in delay-tolerant scenarios. The
nodes. If one of them has the answer to the initial request, User Datagram Protocol (UDP) leaves the responsibility of
the problem is solved. If it does not, it might be worth data acknowledgement and retransmission to the application
checking if a path can be established through this vehicle or some standard middleware (e.g. Remote Procedure Call,
taking some hops to the destination. But if the vehicle density Remote Method Invocation), resulting in similar problems.
1168 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 14, NO. 4, FOURTH QUARTER 2012

Routing protocols are also affected as they rely on timely nodeidentifier/applicationtag is used, where node identifier
updates, which may lead to wrong interpretations of what is is a DNS-style hostname string and application tag is any
happening. In addition, most routing protocols assume that string of URI-valid characters that identifies the DTN applica-
the network is not partitioned - a situation typical for DTN tion addressed. The mapping of EIDs to lower layer addresses
scenarios. BGP (Border Gateway Protocol), responsible for is done when deciding to forward bundles. This requires a
routing between IP autonomous systems, is built on TCP, translation mechanism, which is based on the late-binding
being limited by the TCP operational issues discussed above. principle in the DTN architecture. This principle establishes
While several optimizations were developed to improve the that the translation into a lower layer address should occur as
performance of TCP transmissions over satellite [9][10], space close to the final destination as possible. In a network with
communications present a challenge that requires the use of frequent disruptions, late-binding is advantageous because the
specifically tailored protocols. The Consultative Committee for transit time of a message may exceed the validity time of a
Space Data Systems (CCSDS) [11] has been addressing this binding, making the binding at the source impossible or invalid
issue by standardizing protocols for these scenarios. [5].
The pushing into the market of a broad variety of mesh, The DTN bundle protocol assumes that contact between
ad hoc and wireless sensor network technologies that can nodes is not always available. The communication is only
rapidly vary the network connectivity presents new challeng- possible at certain times, which is different from the as-
ing opportunities to DTN. Some solutions have been found sumption done in the Internet. During the contact, a certain
for these cases, but they were not generic enough for a broad number of bits can be transmitted, which defines the contact’s
application. DTN research has the main objective of providing volume. If the data to be transferred exceeds the volume of
a generic network architecture, which can apply in cases where any known contact, the bundle has to be fragmented. Two
long end-to-end delays, frequent network disruptions or high types of fragmentation are foreseen: proactive and reactive.
bit error probability occurs. In proactive fragmentation, a DTN node may divide a block
of application data into multiple smaller blocks and transmit
each block as an independent bundle. This approach is called
B. Architecture and Bundle Protocol proactive fragmentation because it is used primarily when
The bundle layer stores and forwards bundles (also called contact volumes are known (or predicted) in advance. In
messages) between nodes. A single bundle layer protocol is the reactive fragmentation, the DTN nodes in communication
used across all the DTN. Instead of working end-to-end as the may fragment a bundle cooperatively when a bundle is only
Internet transport protocols, the bundle layer forms an overlay partially transferred; the receiving bundle layer modifies the
that employs a store, carry and forward message switching incoming bundle to indicate the data received is a fragment and
paradigm. Messages or fragments of such messages are moved forwards it normally; the sender node truncates the success-
(forwarded) from a storage place on one node to a storage fully transferred part from the original bundle and keeps the
place on another node, along a path that eventually reaches the undelivered fragment for a subsequent contact. This approach
destination. The two intermediary bundle protocol instances is called reactive fragmentation because the fragmentation
in Fig. 2 represent gateways between the two networks. Of process occurs after an attempted transmission has taken place
course, this can be generalized for any number of gateways. [5].
The protocols which are used in the layers below the bundle The contact type depends on the underlying network.
layer might be diverse and are chosen according to the They can extend from persistent, which are Internet-like, to
communication environment of each region. The underlying opportunistic, which are typical of ad-hoc networks. Pre-
networks might be of any kind, e.g. terrestrial Internet, space dicted contacts require analyzing previous observed contacts
links, ad-hoc networks or wireless sensor networks. to predict the future opportunities to transmit data. There can
The applications are running on top of the bundle protocol also be scheduled contacts, e.g. for the space communication
and communicate intermittently. Messages should be as much environment. In this latter contact type, the sender and the
as possible self-contained atomic units of work. Messages receiver must be synchronized which requires for both sides,
should be sent asynchronously, not wait for the next response at least, to have loosely synchronized clocks. This feature is
before sending the next message. For instance, a DTN file presently assumed by the bundle protocol. It is questionable
transfer application would not initiate a dialog as in the File if this requirement for synchronization should be compulsory
Transfer Protocol (FTP). It would bundle the requested file in the future for the bundle protocol, except for scheduled
names with the username, password and transfer mode in a contacts. Present research goes into the direction of removing
single message to be sent and then wait for a single message the requirement for time synchronization in the bundle proto-
with the requested data. This would make the communication col, introducing a scheme in which bundles expire based on
more efficient in the presence of large delays or connection identifiers and hop counting.
disruptions. As the protocols below the bundle layer may provide dif-
A DTN node is an entity able to run an instance of the ferent semantics, a collection of protocol-specific convergence
bundle protocol. Nodes are identified by End Point Identifiers layer adapters provides the functions necessary to carry the
(EID). Each EID refers to one or more bundle nodes. When bundles on each of the corresponding protocols, as shown in
the EID refers to more than one node, we are in a multicast Fig. 3 [1]. In this conceptual implementation architecture, a
situation. An EID is a Uniform Resource Identifier (URI) from central forwarder is responsible for moving bundles between
the syntactic point of view. Usually, the canonical form: dtn:// applications, convergence layer adapters and storage, accord-
PEREIRA et al.: FROM DELAY-TOLERANT NETWORKS TO VEHICULAR DELAY-TOLERANT NETWORKS 1169

D. Applications
As previsously stated, the DTN concept was initially de-
signed for communicating with spacecraft, to compensate for
disconnections over interplanetary distances. However, over
the years, researchers have identified numerous terrestrial
environments where DTN concepts may be employed.
For example, underwater networks make use of the DTN
paradigm to enable applications for oceanographic data col-
lection, pollution monitoring, offshore exploration, disaster
prevention, assisted navigation and tactical surveillance ap-
plications [17]. Wildlife tracking networks, which are de-
signed for biology research, allow monitoring the long-term
Fig. 3. An example implementation architecture shows how a bundle
forwarder interacts with storage, routing decisions and convergence layer behaviour of wild animals sparsely distributed over a large
adapters to utilize various protocols for delivery. area. Examples of projects in this area are ZebraNet [18] for
zebra tracking, SWIM [19] for whale tracking, and TurtleNet
for turtle tracking [20].
ing to decisions made by routing algorithms. Arrows indicate Sparse wireless sensor networks (e.g. space, terrestrial,
interfaces, which may carry either bundles or directives. and airborne) can also apply DTN technology [21]. These
networks are usually employed to monitor science and hazard
events, like earthquakes, volcanos, floodings, forest fires, sea
C. Other Protocols ice formation and breakup, lake freezing and thawing, and en-
vironmental monitoring. The problem of providing data com-
If we look at the lower layers and try to deal with delay munications to remote and underdeveloped rural communities
tolerance and disruption in a point-to-point environment, we in developing countries has been addressed by several projects
will also be led to a different type of protocols . The most well with approaches that focus on asynchronous (disconnected)
known protocol for operation on a point-to-point link subject messaging by transportation systems [26]. Examples include
to extremely long delay is the Licklider Transmission Protocol DakNet [24], Saami Network Connectivity (SNC) [25], Wizzy
(LTP) [12]. LTP is named in honor of JCR Licklider, one of Digital Courier [23], Networking for Communications Chal-
the pioneers of ARPANET, who envisioned there would be lenged Communities (N4C) [22], First Mile Solutions [27],
interplanetary links in the future. LTP was conceived for single and KioskNet [28].
hop deep space communication links, although it can also be People networks, also called pocket switched networks and
useful for terrestrial networks where disruption might occur. social networks explore transfer opportunities between mobile
LTP can isolate the complexity of a disruptive environment wireless devices carried by humans to transfer data [29].
from the application layer. Moreover, thanks to its design, LTP These networks enable applications that forward data based on
is a good convergence layer to support the bundle protocol. people’s social interest (e.g., news, music, movies, arts). Inte-
In LTP, all the parameters required for interoperability have grating DTN concepts to military tactical networks can ease
to be agreed before a contact occurs, as in a disruptive communications in hostile environments (battlefields) where
environment there is no chance of negotiating parameters. a network infrastructure is unavailable [30]. DTN principles
For this reason, LTP requires information about previous and can also be considered for disaster recovery networks. These
upcoming contacts. networks can support communications in catastrophe hit areas
There are alternative DTN protocols to the bundle/LTP lacking a functioning communication infrastructure [31].
approach. Although the latter is the preferable approach nowa-
days, as DTN is a relatively recent research area in networking,
some new protocols might appear that prove to be more E. Implementations and Simulators
favorable. A survey on protocols for space communication is Several implementations of DTN protocols and simulators
provided in [13]. Of the existing proposals, it is worth referring are listed in [32]. Only the most important are described here.
the LTP-T protocol [14], TP-Planet [15] and DS-TP [16]. LTP- The Bundle Protocol reference implementation is called
T is an extension of LTP to provide a transport layer protocol, DTN2. Besides the Bundle Protocol, DTN2 also supports
which can be useful for cases in which LTP is used in all optionally the Bundle Security Protocol to provide authen-
the links of a path. In this case, having the LTP-T transport tication and/or integrity protection for transmitted bundles if
protocol can be proved to be more efficient than running required by the application. Bundles can be transmitted over
the bundle protocol over LTP. TP-Planet is another proposal either IP transport layers or various link layers including
for a transport protocol, whose aim is to create a transport Ethernet and Bluetooth. Support for LTP is provided through
protocol running directly over IP able to handle high latency LTPlib that can be compiled into DTN2. DTN2 implements
links. Finally, the Deep-Space Transport Protocol (DS-TP) can a number of convergence layers that interface between the
become two times faster than conventional protocols by using bundle and transport protocols. DTN2 also provides a number
techniques such as proactive transmission and retransmission of routing mechanisms to direct the forwarding of bundles to
scheduling rules in order to deal with the unique characteristics their intended destinations, including a static routing scheme
of the deep-space networking environment. based on pre-configured routes, epidemic routing which floods
1170 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 14, NO. 4, FOURTH QUARTER 2012

bundles to any node it encounters and PRoPHET which is is the best when it comes to simulating VDTN scenarios
described in section V.A. Finally, DTN2 provides some exam- However, this simulator does not support ITS protocols or
ple applications such as dtnping to verify network operation, microscopic mobility models as the NCTUns simulator does.
dtnsend and dtnrecv to send and receive bundles, and dtncp On the other hand, the NCTUns simulator lacks support for
and dtncpd to send and receive files. DTN protocols. If ITS functionalities are really required, either
DTN2 also has its own simulation framework still in a pre- they are implemented in the ONE simulator or DTN protocols
release state. TCL (Tool Command Language) scripts allow are implemented in NCTUns. A more complete survey of
creating nodes and connections, generating traffic, configuring VANET simulators is available in [39].
the simulation and retrieving statistics.
The Interplanetary Overlay Network (ION) [33] is an im- III. V EHICULAR D ELAY-T OLERANT N ETWORKS
plementation of the Bundle Protocol, LTP, and the CCSDS
File Delivery Protocol (CFDP) and Asynchronous Message Vehicular Ad hoc Networks (VANETs) have been an impor-
Service (AMS). CFDP is an application-layer service that per- tant research topic for many years. It is an extension of Mobile
forms the segmentation, transmission, reception, reassembly, Ad hoc Networks (MANETs) to vehicle systems, spanning to
and delivery of files in a delay-tolerant manner. AMS is an planes, trains, boats, automobiles and robots.
application-layer service for message distribution based on the MANETs have a set of attributes and requirements [40]:
publish/subscribe or the client/server model. CFDP and AMS 1) Self-organization: a MANET does not depend on a pre-
are not part of the DTN architecture but utilize underlying existing infrastructure but, rather creates one within the wire-
DTN protocols. ION is designed to enable inexpensive in- less network itself; the nodes are both router and terminal; 2)
sertion of DTN functionality into embedded systems such as Mobility: nodes move and protocols have to adapt to this; 3)
robotic spacecraft. ION implements several convergence-layer Multihopping: certain nodes can be reached only by hopping
adapters: TCP (interoperable with DTN2), UDP (likewise over other nodes; 4) Energy conservation: nodes are typically
interoperable with DTN2), Bundle Relay Service, simplified small devices with a limited power supply; 5) Scalability:
TCP, and LTP. applications can grow at any moment, increasing complexity;
The Opportunistic Networking Environment (ONE) simu- and 6) Security: due to their wireless nature, security is
lator [34][35] is a Java-based simulator specifically designed complex and a major issue.
for evaluating DTN routing and application protocols. The VANETs have special characteristics [40]: 1) Predictable
authors demonstrate the tremendous capabilities supported mobility: movements are not random, since vehicles have to
by the simulator critical to implementing various scenarios stay on the road, for example; 2) High mobility: the network
in DTN, including mobility and event generation, message topology changes rapidly because of vehicle speed; 3) Variable
exchange, a basic notion of energy consumption, visualization topology in time and place: the network topology evolves
and analysis, interfaces for importing and exporting mobil- depending on time (e.g., traffic jams) and location (urban,
ity traces, events and entire messages. The following DTN rural); 4) Large scale: all vehicles are potential nodes; 5)
routing protocols are supported: 1) Direct Delivery, 2) First Partitioned networks: the hop range in a wireless car-to-car
Contact, 3) Spray-and-Wait, 4) Epidemic, 5) PRoPHET, and network is about 1000 m, limiting the communication range of
6) MaxProp. These protocols are described in section V.A. vehicles; 6) No significant power of computation constraints:
Mobility models define the algorithms and rules that generate a vehicle can generate sufficient power. An exception is for
the node movement paths. Three types of synthetic movement stationary nodes, which may be battery operated.
models are included in the simulator: 1) random movement, 2) The main difference between VANETs and VDTNs is that
map-constrained random movement, and 3) human behavior VANETs assume that end-to-end connectivity exists through
based movement. An extensive survey on mobility models is some path, while VDTNs do not [41][42][43]. So, VANETs
provided in [36]. concepts are more appropriate for dense networks, while
The National Chiao Tung University Network Simulator VDTNs accept also sparse networks through its store-carry-
(NCTUns) [37][38] does not have specific support for DTN forward paradigm.
protocols. However, it deserves a reference as it supports VDTNs extend VANETs with DTN capabilities to support
protocols important for Vehicular Ad hoc Networks (VANETs) long disruptions in network connectivity. The DTN concepts
such as the IEEE 802.11p/1609 WAVE (Wireless in Vehicular are useful as vehicular networks are characterized by scarce
Networks) and Intelligent Transport Systems (ITS), including transmission opportunities and intermittent connectivity, par-
vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I). ticularly in rural or mountainous areas. A recent study [44]
NCTUns is a C++ based open system network simulator and shows that the duration of contacts between cars using IEEE
emulator designed to work under Red Hat’s Fedora 12 Linux. 802.11g crossing at 20 Km/h is about 40 s, at 40 Km/h is
A Graphical User Interface (GUI) allows creating Tcl files about 15 s and at 60 Km/h is about 11 s. If TCP is used at
to describe the nodes’ protocol stack, connectivity among 60 Km/h, the goodput is very low (average of 80 KB) and in
nodes and their configuration. Additionally, the GUI provides 4 out of 10 experiments no data was transferred at all. UDP
visual observation of the simulated network, including node gives better results, with about 2 MB transferred in a contact
movements. NCTUns supports microscopic mobility models at 60 Km/h.
including car following and lane changing. Most of the problems in vehicular networks arise from
For the deployment of VDTN applications, the best choice the mobility and speed of vehicles that are responsible for a
is to use the DTN2 reference implementation, while ONE highly dynamic network topology and short contact durations.
PEREIRA et al.: FROM DELAY-TOLERANT NETWORKS TO VEHICULAR DELAY-TOLERANT NETWORKS 1171

Limited transmission ranges, radio obstacles due to physical relay node on the base of a mountain storing ice and fog
factors (e.g., buildings, tunnels, terrain and vegetation), and warnings from cars coming down the mountain to provide
interferences (i.e., high congestion channels caused by high them to cars going up the mountain.
density of nodes), lead to disruption, intermittent connectivity, On the contrary, for class 2 services, delayed information
and significant loss rates. All these conditions make vehicular may compromise safety. Examples are context-specific safety
networks subject to frequent fragmentation/partition (i.e., end- alerts related to potential dangers, such as surface conditions
to-end connectivity may not exist), resulting in small effective or abnormal vehicle behavior due to an accident or some other
network diameter. Furthermore, vehicular networks have the failure. As the tolerable delays are very short, DTN cannot be
potential to grow to a large-scale, and its node density, which used effectively.
is affected by location and time, can be highly variable. For Finally, class 3-4 services require real-time communication
example, a vehicular network can be categorized as being and are typically sensitive to delayed or loss of data. These
dense in a traffic jam, whereas in suburban traffic it can be applications may be used to control the vehicles throttle and
sparse. In fact, in rural areas, the network can be extremely brakes on a real-time basis. An example is vehicle collision
sparse. For all these scenarios, DTN mechanisms provide a avoidance. Again, DTN may not improve the quality of this
significant advantage. kind of services, as permanent connectivity with low delay is
Some interesting new types of nodes in VDTNs are data required.
MULEs (Mobile Ubiquitous LAN Extensions) [45] and sta- Table I lists the main characteristics required by VDTN
tionary relay nodes [46]. Mules are mobile nodes that pick applications, their definition and the corresponding protocol
data in one place and drop it in another place. Mules are supporting mechanisms. Usually, there is a compromise be-
added to the network to extend coverage and/or the number of tween latency and delivery ratio. If more time is allowed
communication opportunities. Stationary relay nodes are fixed to deliver messages, more messages will be delivered suc-
devices with store-and-forward capabilities that are located cessfully. Naturally, the resulting network resources spent
at road intersections. Mobile nodes use them to deposit and can increase, reducing scalability. Several mechanisms may
pickup data. Relay nodes increase the number of communica- be used to reduce network resource consumption. A routing
tion opportunities in scenarios with low node density. Hence, protocol can be used to select the most appropriate next hop(s)
MULEs and relay nodes contribute to increase the bundles for a message. Scheduling and prioritization mechanisms may
delivery ratio, and decrease their delivery delay. The best be used to select which messages to be forwarded first in a
placement for relay nodes and how to manage the available limited duration contact between nodes. Cache management
limited storage for bundles in the different types of nodes are mechanisms and acknowledgements may be used to clean
still research challenges. older messages when storage space is scarce. The DTN’s
Some examples of potential VDTN applications are: im- custody transfer mechanism can be used to ensure delivery of
proving road safety (e.g., cooperative collision avoidance, messages at the expense of protocol complexity and storage
emergency break warning, road hazard notification); optimiz- space. Security and delivery of messages to an appropriate
ing the traffic flow and road capacity (e.g., road congestion set of destinations are also important protocol characteristics.
prevention, traffic condition monitoring); monitoring networks Most of the protocol design issues that address these appli-
for sensor data collection (e.g., weather conditions, pollution cation requirements are still research challenges that will be
measurements, road surface conditions); commercial applica- addressed in more detail in section V. A generic discussion
tions (e.g., commercial advertisements, marketing data, travel, of inter-vehicle protocol characteristics without special focus
tourist and leisure information and parking space availability); in VDTNs is given in [47].
entertainment applications (e.g., Internet access and multime- The following section gives further details about VDTN,
dia content sharing); provide connectivity to remote rural com- summarizing some of its applications and present projects.
munities and regions (e.g., file-transfer, electronic mail, cached
Web access, and telemedicine); assist communication between IV. VDTN P ROJECTS AND A PPLICATIONS
rescue teams and other emergency services in catastrophe hit
areas lacking a conventional communication infrastructure. A. KioskNet Project
A survey of inter-vehicle communication protocols and The KioskNet project [48] provides low-cost Internet kiosks
their applications is given in [47]. This survey classifies in rural areas with some services, such as email, web browsing,
applications in four classes: 1) General information services; telemedicine, crop prices information and taxpaying. As the
2) Information services for vehicle safety; 3) Individual motion kiosks have no permanent Internet connection, a bus and DTN
control using inter-vehicle communication; and 4) Group protocols offer the gateway between the kiosks and the Internet
motion control using inter-vehicle communication. at a neighboring town as shown in Fig. 4.
For class 1 services, delayed or lost information does The user applications contact the DTN protocols, generating
not compromise safety or render the application useless. bundles that are stored in persistent storage until the bus
Examples are information queries (e.g. weather reports, web passes by the village. Then, the bundles are transferred to the
browsing, business services, road conditions, traffic volume) DTN agent in the bus and carried to a town to be delivered
and context-specific broadcasts (e.g. advertising, entertainment to an Internet gateway. These procedures are illustrated in
feeds). Thus, DTN capabilities can improve the quality of Fig. 5 for the case of the email service. An application pro-
class 1 services by increasing the message delivery ratios and tocol, named Opportunistic Connection Management Protocol
reducing delays in challenging scenarios. An example is a (OCMP), runs on top of the bundle layer. OCMP provides the
1172 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 14, NO. 4, FOURTH QUARTER 2012

TABLE I Village
VDTN P ROTOCOL C HARACTERISTICS .
Hub (Internet
Characteristic Description Supporting Mechanisms
access point) Village
End-to-end packet Routing protocols, Kiosk
Latency delivery time scheduling mechanisms
Routing protocols,
Delivery ratio Ratio of delivered custody transfer, mobile access
messages over Kiosk
end-to-end point
transmitted messages acknowledgement,
caching mechanisms Town Village
Support a large number
Scalability of nodes, a large number Routing protocols,
of hops and varying node prioritization
densities Kiosk
Transmission to a single,
Routing multiple, all, any, all Routing protocols, Fig. 4. A route bus and DTN protocols offer the gateway between the kiosks
scheme within a region membership services, and the Internet at a neighboring town.
destinations: unicast, locating services
multicast, broadcast,
anycast, geocast
Prevent unauthorized a throwbox will often remain stationary for several hours or
Security access, misuse, Authentication,
encryption protocols, days, their nomadic features enable them to be flexibly placed
modification or denial of
use of network resources security policies across the DOME testbed.
The mesh network consists of 26 Wi-Fi access points
mounted on different buildings and light poles of the town.
interface between the email client and the DTN agent at the The former provide direct connection to the local fiber infras-
kiosk and between the DTN agent at the gateway and an email tructure, while the latter provide a connection at no more than
server. three hops from the wired network.
The system is rapidly deployable, low-power (can be pow- The testbed covers a large geographic region, from urban
ered by a solar panel) and uses free software allowing low-cost areas with dense connectivity to rural areas with sparse con-
Internet access even in remote areas. Naturally, if either dialup, nectivity. It provides a rich environment for research of DTN
long-range wireless or cellular phone service is available, the concepts such as routing, power management, system design,
kiosk controller can be configured to use these communication and application design. Traces of bus movements and contacts
links in conjunction with the longer delay access provided by are provided for the research community to do experiments
the bus to get lower Internet access delays. The system also based on a real testbed.
includes a user management interface for local administration An interesting application is enhanced web browsing for bus
of the kiosk, remote management, log collection and extensive users [50]. The use of aggressive prefetching of the links in a
security support. given page can convert an interactive web session into a one-
The KioskNet project has developed a complete system shot request/response adequate for an opportunistic contact
proving that DTN protocols can exploit periodic bus move- network. The use of mobile-to-mobile routing and caching can
ments to move data cost-effectively between isolated places, improve the number of relevant responses for sparse networks
offering valuable Internet services to the poorest sections of in rural areas by 58%, but the mean delay is significantly
society. high at 6.7 minutes, calling into question its practicality for
interactive applications.
B. DieselNet Project A miniature version of the bus-based technology coupled
The Diverse Outdoor Mobile Environment (DOME) [49] with some intelligent power management is used to monitor
is a testbed for large-scale mobile experimentation consisting endangered wood turtles in the Amherst area for ecologic and
of three major hardware components: the DieselNet vehicular biodiversity conservation purposes.
network, a set of nomadic throwboxes and an outdoor mesh The main contributions of the DieselNet project have been
network. several new routing protocols and applications for DTNs,
DieselNet is comprised of 40 transit buses, covering an and the longest-running large-scale, energy-efficient, highly
area of 150 square miles around Amherst, each equipped diverse mobile systems testbed, which constitutes a significant
with an embedded PC with GPS (Global Positioning System), engineering challenge.
802.11abg card, 802.11g wireless access point, wireless 3G
USB modems and a 900MHz USB RF modem. The access
C. VDTN Project
point allows other buses, or bus riders, to establish 802.11
connections, giving them access to the Internet via one of The Vehicular-Delay Tolerant Networks (VDTN) project
the external radio interfaces. The Wi-Fi interface is used to [51] proposes a layered architecture for VDTNs, where the
connect to foreign access points, including the access points bundle layer is placed below the network layer instead of
from other buses. above the transport layer. The objective is to route large size
Throwboxes are wireless nodes that can act as relays, cre- messages instead of small size IP packets. This results in fewer
ating additional contact opportunities among DieselNet buses. packet processings and routing decisions, which can result into
Throwboxes use batteries recharged by solar cells. Although less complexity, lower cost and energy savings.
PEREIRA et al.: FROM DELAY-TOLERANT NETWORKS TO VEHICULAR DELAY-TOLERANT NETWORKS 1173

Email client Internet

running on
terminal Legacy
bus Gateway email
Kiosk controller server
Proxy
DTN agent DTN agent
Email client OCMP SMTP plugin on bus at gateway OCMP SMTP
plugin
HELO OCMP deamon
OCMP deamon
FROM: src DTN agent DTN agent
TO: dst Create
DTN
AppData SMTP
bundles
Data plugin
AppData: Bulk
QUIT SMTP, Data
src, dst
OK Reassemble
Bulk Data Delivery
Acks

Persistent
storage

Persistent
storage
HELO

DTN
bundles
Delivery QUIT
Delivery Delivery OK
Acks Ack
Acks

Fig. 5. The opportunistic connection management protocol (OCMP) runs on top of the DTN bundle layer to allow email transfers.

Application Layer of messages. Relay nodes are simpler, as they only need to
implement the three lower layers of the protocol stack.
Transport Layer
A prototype testbed was implemented [53] using Lego
Persistent Network Layer Mindstorm NXT robotic cars equipped with a Personal Digital
Storage Assistant (PDA) device coupled on it to emulate mobile nodes
BAD Layer BSC Layer (e.g., vehicles). Terminal nodes and relay nodes are emulated
using desktop and laptop computers. A Bluetooth connection
MAC Layer MAC Layer
is always active to process out-of-band control information.
PHY Layer PHY Layer When necessary, a Wi-Fi connection is activated to exchange
data bundles. This contributes to saving energy, which is
Data Plane Control Plane very important for energy-constrained network nodes such as
Fig. 6. IP-over-VDTN layered architecture.
stationary relay nodes.
The prototype allows the study and evaluation of the differ-
ent nodes behavior and their corresponding caching, carrying
and forward/routing mechanisms. The testbed also allows the
The architecture uses out-of-band signaling, based on the development of new protocol services and the comparison with
separation of the control plane and data plane (Fig. 6). results obtained by simulation.
The Bundle Aggregation and De-aggregation (BAD) layer The VDTN project has shown that a separate control
aggregates incoming IP messages into bundle messages that plane can optimize the use of the data plane resources (e.g.,
are transferred in the data plane and de-aggregated at the des- storage and bandwidth) and save energy. This project has
tination. The Bundle Signaling Control (BSC) layer provides also explored the use of scheduling and dropping policies,
a signaling protocol for use at the connection setup phase. traffic differentiation, node localization, stationary relay nodes,
The nodes exchange control information to discover each geographic routing, and caching mechanisms to increase the
other’s characteristics and prepare the data transfer to occur efficiency of communication.
in the data plane. This layer also includes routing algorithms.
The separation of the control and data planes is conceptually
similar to Optical Burst Switching [52]. D. CarTel Project
Three types of nodes are considered to exist: terminal CarTel [54][55] offers two DTN networking abstractions:
nodes provide the connection to end-users; mobile nodes (e.g., dPipe and CafNet (“carry and forward network”). dPipe is a
vehicles) carry messages between terminal nodes; and relay conceptual extension to the UNIX pipe abstraction that allows
nodes are fixed nodes located at crossroads to improve delivery processes on separate hosts to communicate via a reliable,
1174 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 14, NO. 4, FOURTH QUARTER 2012

Web Drive-thru Drive-thru proxy Web

Application 1 Application N browser client server
HTTP HTTP+X HTTP+X HTTP HTTP
Transport Layer
PCMP PCMP
Network Layer
TCP TCP TCP TCP TCP
Mule Adaptation Layer IP IP IP IP IP
802.11 Internet
Device Driver Device Driver Mobile Node radio Drive-thru proxy backbone Server

Fig. 8. Drive-thru Internet architecture.

Fig. 7. CafNet protocol stack.

server via the WAN. In addition, control messages for traffic

delay-tolerant data stream. It is implemented using several file management devices or information displays are forwarded
based buffers for storage, and, when connectivity is present, from the WAN to the DTN. Smart display panels show
uses TCP sockets to send buffered data and application-level information on current pollutant concentrations. The measure-
acknowledgments to ensure that all data gets written to disk. ment results are gathered from passing vehicles and computed
CafNet is a delay-tolerant network stack, as illustrated in autonomously by the panel. The display may also receive
Fig. 7, which delivers data in intermittently connected envi- messages from a control center (via the DTN) in order to
ronments possibly through mule nodes. The Mule Adaptation show, for example, traffic information. Displays also act as
Layer hides details of the communication medium from the relay nodes, speeding up the bundle distribution process.
higher layers. Unlike the traditional sockets interface, the The EMMA project has shown that a public bus trans-
CafNet interface uses callbacks across all its layers. By issuing portation network can provide a cost-effective environmental
callbacks whenever network conditions change, CafNet makes monitoring, traffic management and information services so-
it possible for the sender application to dynamically prioritize lution. Although EMMA is similar to DieselNet in several
data. At the same time, CafNet’s network layer provides some aspects, DieselNet is not a distributed information system
buffering to achieve high utilization when network connectiv- where measured data is spread through the network as EMMA
ity is fleeting (e.g., a few seconds), a common situation at is.
vehicular speeds.
An application example is the detection of road pavement
defects. A three-axis acceleration sensor and GPS device F. Drive-Thru Internet Project
is deployed on an embedded computer in the car. Using a The Drive-Thru Internet project [57] aims to provide Inter-
simple machine-learning approach, it is possible to identify net access for vehicles, by exploiting intermittent connectivity
potholes and other severe road surface anomalies from the to wireless access points along the road. The concept of
accelerometer data. The detections are uploaded to a central Performance Enhancing Proxies (PEP) [58] is used to hide
server using opportunistic Wi-Fi connections provided by the effects of intermittent connectivity.
participating open Wi-Fi access points, or using a cellular data Fig. 8 shows the system architecture. Mobile nodes are
service, where available. equipped with client proxies that relay transport and appli-
The CarTel project has a vision that the existing hundreds cation layer protocol interactions to the corresponding Drive-
of millions of cars and several billion mobile phones can be thru proxy in a nearby wireless access point. These Drive-
the carriers of the world’s largest and most dynamic sensor thru proxies have a direct connection to the Internet backbone.
networks. CarTel provides software to collect, process, deliver, The session persistence across disconnections is provided by
and visualize data from sensors located on mobile devices to the Persistent Connection Management Protocol (PCMP), a
a portal. Application data requests are executed using a delay- session layer protocol that operates based on TCP connections.
tolerant continuous query processor on the remote nodes. Drive-thru proxies can perform application layer functions
such as web- and e-mail prefetching to improve performance.
E. EMMA Project This is why the HTTP layer in Fig. 8 is extended while using
the proxies. A possible drawback of this architecture is the use
The Environmental Monitoring in Metropolitan Areas of TCP for short-lived contacts which may cause an excessive
(EMMA) project [56] uses a public bus transportation network overhead as discussed before.
to monitor pollution. The buses have a GPS and a number The Drive-Thru Internet project has shown that a session
of pollution detection sensors that continuously monitor the layer above the transport layer can maintain a persistent ap-
environment. The collected data is transmitted through a DTN plication session throughout a period of several disconnections
stack to a central server where it is analyzed. and new network attachments.
EMMA considers two types of stationary DTN nodes:
gateways and smart display panels. The former provide an
interface between the buses and the traffic management Wide G. CONDOR Project
Area Network (WAN), and when data is available they forward Military communication systems have to adapt to situations
the bundles with the measurement results to the evaluation that offer several challenges: lack of fixed infrastructure, lim-
PEREIRA et al.: FROM DELAY-TOLERANT NETWORKS TO VEHICULAR DELAY-TOLERANT NETWORKS 1175

beyond
CONDOR
PoP-V line of Gateway
CONDOR PoP-V sight
Gateway beyond
line of
sight

CONDOR
Jump C2

Displacing Division Commander

Fig. 9. CONDOR network.

Fig. 10. Car2Car reference architecture.

ited spectrum availability, difficult propagation environments,

and rapidly fluctuating information demands by end users. H. Non-DTN Projects
The Marine Corps is already using DTN concepts in its Although not currently using DTN protocols, a few projects
Command and Control, On-the-Move, Network, Digital, Over- are worth referencing due to their relevance in vehicular
the-horizon Relay (CONDOR) project [59]. There are three networks, either because of their standardization effort or the
types of CONDOR vehicles, as shown in Fig. 9: Gateway, availability of commercial services. A more complete list can
Point of Presence and Jump C2 (Command and Control) be found in [60].
vehicle. The Gateway vehicle is used to extend communica- The Car 2 Car Communication Consortium (C2C-CC) [61]
tions beyond line of sight, by bridging the Enhanced Position aims to standardize interfaces and protocols of wireless com-
and Location Reporting System (EPLRS) networks through a munications between vehicles and their environment in order
satellite as necessary. The Point of Presence vehicle (PoP-V) to make the vehicles of different manufacturers interoperable,
connects older technology radios to the EPLRS or satellite not only among them but also with road-side units. The draft
systems by acting as a translator and repeater. The Jump C2 reference architecture of the C2C Communication System is
vehicle acts as a mobile command post by keeping continuous shown in Fig. 10. It comprises three distinct domains: in-
satellite communications and connecting nearby command vehicle, ad hoc, and infrastructure domain. The in-vehicle
vehicles using wireless technology. domain refers to a network logically composed of an on-board
To avoid rebuilding applications for DTN support, unit (OBU) and (potentially multiple) application units (AUs).
application-layer proxies are used in the CONDOR project An AU is typically a dedicated device that executes a single
to convert the application data into bundles to transmit over or a set of applications and utilizes the OBU’s communication
the DTN network. So, an application-layer proxy for SMTP capabilities. An AU can be an integrated part of a vehicle or a
behaves like an SMTP server (to SMTP clients), responding portable device such as a laptop, PDA or game pad. The ad hoc
with the appropriate SMTP protocol responses, gathering domain, or VANET, is composed of vehicles equipped with
all information related to a mail message and building the OBUs and stationary units along the road, termed road-side
corresponding bundle to be transmitted through the DTN. This units (RSUs). An OBU is at least equipped with a (short range)
is very similar to what is illustrated for the Kiosknet project wireless communication device dedicated for road safety, and
in Fig. 5. potentially with other optional communication devices. The
A DTN/http proxy was also built based on the World Wide primary role of an RSU is the improvement of road safety.
Web Offline Explorer (WWWOFFLE). WWWOFFLE was An RSU can be attached to an infrastructure network, which
designed to be a web cache/offline viewer for users with a in turn can be connected to the Internet. As a result, RSUs
dialup or other intermittent connection to the Internet. The may allow OBUs to access the infrastructure. An OBU may
cache receives http requests from standard web browsers and, also communicate with Internet nodes or servers via public,
if not connected to the Internet, stores them in the local file commercial, or private hot spots (HS). In case that neither
system for later retrieval when online. The WWWOFFLE RSUs nor hot spots provide Internet access, OBUs can also
proxy was split in two, to form a client and a server side. utilize communication capabilities of cellular radio networks
The client side lives on the challenged or tactical piece of if they are integrated in the OBU, in particular for non-safety
the network, and uses DTN bundles to communicate with the applications.
server side. The server side is assumed to be permanently A simplified version of the protocol stack is shown in
connected to the Internet, so that when it receives requests Fig. 11. This stack is being normalized by the Intelligent
from clients, it can retrieve the requested web pages and Transport Systems (ITS) workgroup [62]. The protocol stack
include them in a bundle to be sent to the client side. is similar to an Open Systems Interconnection (OSI) stack.
The CONDOR project has deployed DTN protocols in The “access” layer represents layers 1 and 2, providing access
military environments showing they efficiently deliver data, to a physical network. The “networking & transport” layer
where simple IP protocols do not. Besides the improvement represents layers 3 and 4, providing routing and end-to-end
in packet delivery ratio, latency is reduced and the link load delivery of data. The “facilities” layer represents layers 5, 6
is also reduced as there are fewer retransmissions. and 7, providing session management to hide communication
1176 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 14, NO. 4, FOURTH QUARTER 2012

technology is a cellular floating phone data (CFCD) system ex-

ploiting signaling data from the GSM operator network, which
is enhanced by GPS-based probe data (both anonymously
gathered) along with conventional data locally detected by
third parties like local authorities and loop systems on the
road. The CFCD is based on changes of the Timing Advance
measurement values while a GSM handset is in an active call.
Timing Advance is a measurement of the distance between
the cell phone and the serving base station, which is important
to synchronize phone calls. This feature allows triangulating
the position of the cellular from the base station location and
the underlying road network. A data fusion engine merges the
information received from the different sources with historical
information, filtering out anomalous readings.
It is expected that, in the future, cars will have communi-
cation capabilities that enable a rich set of applications that
have already started to emerge into the market. The challenges
cover all protocol layers from the radio physical layer up to
the application layer. As the nodes are highly mobile, the use
of delay-tolerant mechanisms in the protocols is an important
aspect to be considered.

I. Summary
Fig. 11. Intelligent Transport Systems’ simplified protocol stack. Table II summarizes the main characteristics of the projects
presented. For each project, the main application area is indi-
cated. Most projects focus on Internet applications such as web
disruptions and delays, standardized data formatting and ap- and e-mail, while few have vehicle specific applications. The
plication support. The session layer provides functions similar protocol stack and the routing protocols used are also listed.
to the bundle layer in the DTN stack, although the work has Most projects use a standard DTN stack, while others have
been done in both areas independently. minor modifications on this stack. Most projects use Epidemic
A set of applications are foreseen by the C2C-CC [61] routing, which is one of the simplest to implement. The routing
to 1) improve driving safety, 2) improve efficiency of the protocols will be further discussed in the following section.
traffic network and 3) provide information or entertainment Unluckily, most VANET projects do not consider the use
to the vehicle passengers. Example use cases are: coopera- of DTN concepts, as they would certainly increase the com-
tive collision warning, providing assistance to the driver to munication efficiency [64][65][66]. However, as the difference
avoid collisions; pre-crash sensing, to activate air bags before is only at the software level, it is expected that in the future
crashing; hazardous location notification, to share information more projects will include DTN capabilities to optimize the
about dangerous locations such as slippery roads or potholes; use of the limited contact opportunities between vehicles and
enhanced route guidance and navigation, giving information to reduce the use of infrastructure communications.
the driver about expected delays or better routes; green light The DTN architecture and protocols [1][4] have proven to
optimal speed advisory, to inform the driver of the optimum be useful technologies for VDTNs. The use of MULE [45]
speed to reach a signalized intersection with the green light nodes, that move data from one place to another, and relay
on; V2V merging assistance, to help the driver to merge nodes [46], that store data for later delivery to passing by
into the traffic flowing on a road he/she is about to enter; vehicles, can increase the efficiency of inter-vehicle non real-
Internet access in the vehicle, to allow all kinds of IP-based time communications [47]. Projects such as KioskNet [48]
services in the vehicle; point of interest notification, to provide were pioneers in showing that the theoretical DTN concepts
information about nearby businesses, tourist attractions or can be effectively useful in practice, as the movement of vehi-
other points of interest; or remote diagnostics to speed up cles can be exploited to carry data to other locations. Routing
appointments to the service garage. protocols [67] play an important role in efficient delivery
TomTom’s HD Traffic [63] navigation and traffic informa- of data to the intended destination nodes. Standardization of
tion service is already available in several European countries. vehicle communications is taking place within C2C-CC [61],
Although no DTN protocols are used, it is an example of but still not including DTN mechanisms.
an almost real-time commercial vehicular information system.
TomTom uses a bi-directional GPRS communication channel
V. S PECIFIC R ESEARCH T OPICS
to deliver traffic information and other relevant messages to
the device in the vehicle every three minutes. This information VDTNs face various network issues and challenges such as
gives an accurate estimate of travel times that can be used architecture, node design and typology, interactions and coop-
to select the fastest route. The core traffic data collection eration, network topology, mobility pattern, packet scheduling,
PEREIRA et al.: FROM DELAY-TOLERANT NETWORKS TO VEHICULAR DELAY-TOLERANT NETWORKS 1177

TABLE II
VDTN P ROJECT ’ S CHARACTERISTICS . Spray-and-Wait [69] generates n copies of a message. In
normal mode, a node gives one copy to each contact; in binary
Project Applications Protocol Stack Routing mode, half of the copies are forwarded to a contact. Once
Protocol
only a single copy is left, it is forwarded only to the final
KioskNet Internet access for
[48] rural sites DTN standard stack Epidemic recipient. Spray-and-Wait is another example of protocol that
MaxProp, limits message replication as compared with Epidemic routing.
DieselNet Internet access for DTN standard stack RAPID, or
[49] buses The PRoPHET (Probabilistic Routing Protocol using His-
others
Bundle layer below Epidemic, tory of Encounters and Transitivity) [72] protocol transfers the
VDTN Internet access for network layer. Spray-and- message to a neighbor if it estimates the neighbor has a higher
[51] vehicles Separate data and wait or “likelihood” of being able to deliver the message to the final
control planes others
Mule adaptation
destination based on past node encounter history.
CarTel Detection of road layer below Static, MaxProp [65] floods the messages but explicitly clears them
[55] pavement defects network layer Epidemic
once a copy gets delivered to the destination. In addition,
Pollution
EMMA measurements, DTN standard stack Epidemic MaxProp sends messages to other hosts in a specific order that
[56] Traffic information takes into account message hop counts and message delivery
Drive- probabilities based on previous encounters.
Thru E-mail, Session layer above Through in-
Internet Web browsing the transport layer frastructure RAPID (Resource Allocation Protocol for Intentional DTN)
[57] [73] can optimize a specific routing metric such as worst-case
Email, Web delivery delay or the fraction of packets that are delivered
CONDOR browsing, IRC, DTN standard stack Static
[59] within a deadline. The key insight is to treat DTN routing as
Voice mail
Safety, Traffic a resource allocation problem that translates the routing metric
C2C-CC Efficiency, Standard OSI stack Geographic into per-packet utilities which determine at every transfer
[61] Infotainment routing
opportunity if the marginal utility of replicating a packet
TomTom Navigation, Traffic Through in-
[63] information GSM frastructure justifies the resources used.
Fairly similar to RAPID, MORA (Multi-Objective Robotic
Assistance) [74] learns from the structure of the node move-
traffic type and prioritization, convergence layer adapters, rout- ment patterns and uses this information to improve the
ing protocols, bundle format, caching mechanisms, security, message routing. Moreover, to further increase the delivery
and supported applications. The most relevant are described ratio, MORA introduces autonomous agents that adapt their
in greater detail in the remaining of this section. movement based on the variations in network capacity and
demand.
MaxProp and RAPID were developed under the DieselNet
A. Routing project. The DieselNet and VDTN projects have used several
The bundle protocol does not describe how to set up routing protocols. The remaining VDTN projects described in
routes between the nodes. It deals only with the forwarding section IV use Epidemic routing. The C2C-CC foresees the
plane. The control plane issues remain open. However, a use of geographical routing in which the geographic position
number of studies exist for applicable routing protocols based of nodes is known and used for routing purposes. The basic
on different schemes, such as oracle schemes, model-based algorithm used is greedy forwarding, e.g. with the maximum
schemes, epidemic schemes and estimation schemes. A survey progress within radius policy.
of routing algorithms is provided in [67]. An update of this There are many routing schemes for the ITS. However, only
survey is provided in [68]. Some typical routing schemes are a few deal with delay-tolerant scenarios [75]. From these,
summarized in the remaining of this section. some try to choose paths through denser areas, which may
A very simple protocol is Direct Delivery, in which the cause congestion. Others store data in fixed relay nodes until
node originating a message carries it until it meets its final a vehicle going to an adequate destination passes by, which
destination. may take some time. Others try to forward the data along
In First Contact routing, the nodes forward messages to the the direction to the destination, which may also take some
first node they encounter, which results in a “random walk” time. Finally, others combine trajectory information and traffic
search for the destination node. statistics to find the best path, which may be complex.
Epidemic routing [70] replicates messages to all encoun- As VDTNs are characterised by generally short contacts
tered peers that still do not have them. If message storage between nodes and a highly dynamic network topology, rout-
space is unlimited and contacts between nodes are long ing is a particularly challenging problem. Routing protocols
enough, epidemic minimizes the delivery delay and maximizes that need to exchange control information during contacts to
the delivery ratio. However, since those resources are usually update routing tables or other information databases have less
limited, epidemic wastes storage and bandwidth in comparison time to transfer data bundles. For instance, PRoPHET requires
with other protocols. For instance, Surround routing [71] tries some overhead for maintaining the estimates of meeting
to minimize the storage consumption and overhead by also probabilities. On the other hand, routing protocols that do
sending messages to all the nodes, but only the nodes that not maintain such control information generally have to create
surround the final recipient will keep the copies longer than more bundle copies to achieve the same delivery performance.
others. This represents an efficiency compromise, as more copies
1178 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 14, NO. 4, FOURTH QUARTER 2012

spend more storage and transmission resources, contributing to And finally, geographically-scoped anycast is similar to the
congestion. As the network topology is highly dynamic, nodes previous method, but the message is not forwarded further
have to take into account that any information maintained may after reaching the destination area. Geocast is of significant
be outdated soon. So, there is also a compromise between the importance to VDTNs, as frequently the information to be
value of information exchanged and the cost of keeping it disseminated is only locally useful. In addition, it allows
updated. restricting resource consumption.
Three other research challenges related to routing are the
optimal placement of relay nodes [76], traffic differentiation C. Applications and Scalability
[77] and congestion control [1]. The problem of optimal The scalability of the proposed DTN protocols has yet to be
relay node placement is an NP-hard problem. So, heuristic proved. There are no large scale deployments of DTNs running
algorithms may be used to solve the optimization problem the proposed protocols, and all studies for a large number of
and select placements for relay nodes for improving deliver nodes are currently based on simulations. Scalability of the
probabilities. Traffic differentiation is a scheduling problem network includes also the issues of how to effectively manage
for which a decision has to be made on the bundles to the network and how to deal with interoperability problems
be transmitted first when a limited time contact between when several network operators are involved.
nodes occurs and also which bundles should be discarded first The problem of aggregating application data into bundles
if storage is exhausted. Congestion occurs when too much is called bundling. Bundles should be as large as possible to
bundle data is circulating in the network and new data cannot minimize RTTs in the presence of link disruptions. However,
be properly handled. Several mechanisms can be used to large bundles may require more transmission time than avail-
reduce congestion, such as scheduling and dropping policies, able in short vehicle contacts. This may require fragmentation,
dynamic control of bundles replication, the use of explicit which causes additional overhead. So, a compromise in bundle
delivery acknowledgments to notify intermediate nodes about size may prove more efficient.
delivered bundles and therefore delete these bundle copies If the application allows for caching of data in the network,
[65], and selection of less congested paths at a routing level. as for instance a web browsing application, the use of bundle
Some of these mechanisms require additional traffic, so there caching may improve performance [82][83]. So, the additional
is again an efficiency compromise. bundle copies created by routing protocols in different loca-
It is worth mentioning that the performance of most of the tions may result in an additional benefit for the applications
introduced routing protocols highly depends on the level of if content storage and retrieval methods are implemented.
cooperation and autonomy of the nodes. By default, most of
the protocols assume full node cooperation and little attention D. Convergence Layer Adapter
has been devoted to study the effect of reduced levels of coop- A survey of inter-vehicle communication systems, covering
eration. In fact, by applying and fine-tuning simple knowledge- the lower layers is presented in [84]. The IEEE 802.11p stan-
based cooperation mechanisms, the routing performance can dards for the MAC and Physical layer of vehicular networks
be considerably improved [78]. are finished [85]. They provide an optimized version of the
Wi-Fi protocols for vehicle communications. The range is
extended up to 1 Km. The data rates range from 3Mbit/s to
B. Anycast and Multicast
27Mbit/s. A faster association is supported at the MAC layer.
The DTN anycast and multicast routing topic is still little IEEE 1609.4 further enhances the MAC layer by offering
studied. Anycast refers to delivering a message to any of 7 different channels and Quality of Service (QoS) support
the group member destinations, whereas multicast does the with 4 priority levels. One of the channels is reserved for
same but to all destinations within the group. When the group broadcasting high-priority safety information.
of destinations is defined by their geographic locations, it is While the operation of a convergence layer adapter over
called geocast. IEEE 802.11p seams natural in the near future, the respective
An overview of multicast models for DTNs is presented interactions with other bundle protocol modules is still a
in [79], but no specific protocol is proposed. A performance research area of interest for VDTNs. For instance, using a
comparison of different multicast routing strategies for DTNs received signal strength indicator (RSSI) and location infor-
is given in [80]. As an illustration, a multicast extension of mation provided by GPS may help estimating the duration
Prophet is described in [81]: a heuristic is used to select the of contacts. If known, the duration of a contact may help
best contacts where to forward bundles to build a pseudo bundle scheduling and fragmentation decisions. In addition,
multicast tree and reach the destination group. As no end- if multiple nodes are within reach, a decision may be made
to-end path is assumed, a complete multicast tree does not about which contacts should be established first or if multiple
exist either. simultaneous contacts can be established.
None of the VDTN projects described in section IV deals Several shortcomings of UDP and TCP were identified
with anycast or multicast. The C2C-CC foresees three any- in section II. To address these shortcomings, [84] suggests
cast/multicast methods [61]. Topologically-scoped broadcast is modifying TCP or developing new transport protocols specific
used to transmit data to nodes within a given distance from the for vehicular communications. An alternative is to address
source node, for instance a 2-hop range. In the geographically- these shortcomings in the bundle layer, above the transport
scoped broadcast the data is transmitted to all nodes within protocol, as shown in Fig. 2. This even allows for different
a defined geographic area such as a rectangle or a circle. transport protocols for different path segments.
PEREIRA et al.: FROM DELAY-TOLERANT NETWORKS TO VEHICULAR DELAY-TOLERANT NETWORKS 1179

E. Security are still in a very active research phase. The DTN architecture
Security is a concern in DTNs. This issue dates from the and bundle protocol have been standardized. Although some
IPN origins, in which the exposure of space communications aspects are a lively research area, its main concepts have
to a similar level of threat experienced in the Internet was already been successfully used in vehicular networks and other
undesirable. When the IPN concept was extended to terrestrial applications.
networks to originate DTNs, the concern remained because it DTNs introduce a store-carry-forward paradigm that per-
is difficult to do a manual reset and reconfiguration of DTN forms better and uses fewer resources than end-to-end proto-
nodes after a successful attack. Most network security methods cols, as each hop is optimized individually. The bundle layer
attempt to authenticate user identities and the integrity of hides internal network details and improves message delivery
messages, but they do not attempt to authenticate the routers ratios in challenging environments.
involved in the communication process. In DTNs, the routers The DTN area has experienced an increased interest over
and the gateways should also be authenticated, and sender in- the past years and it is expected that the use of its concepts for
formation should be authenticated by forwarding nodes. In this vehicular networks also increases in the near future. VDTN
way, the transport of prohibited traffic would be prevented at is thus an emerging area of research with a large number of
the earliest opportunity. A rather detailed discussion on DTN practical applications. By providing DTN capabilities to vehic-
security issues can be found in [86], including a discussion on ular networks, new challenging situations typical of vehicular
the suitability of delay tolerant key management schemes for networks may be overcome, such as sparse and intermittent
DTNs, which is still one of the open points in DTN security. connectivity, variable delays, high error rates and nonexistence
Security in DTN presents also some research challenges. of an end-to-end path.
The requirement for out-of-band contacts, for example, to Several VDTN projects were presented. These projects have
check that public keys are not in blacklist in the Internet proven that the movement of vehicles can be exploited to carry
is problematic in DTNs. Another issue is the lack of a data from one place to another. This carry paradigm is much
delay tolerant method for key management. Also methods for slower than wired or wireless transmissions, but combined
protection of the network against traffic analysis are missing, with delay-tolerance can provide cost-effective support for
as well as a policy for introducing new nodes in the network non real-time applications. The projects presented identified
without impacting the security level. a number of research challenges that were described. The use
From the VDTN projects described in section IV, only of advanced routing mechanisms, scheduling and dropping
Kiosknet deals with security issues. Kiosknet uses a Public policies, traffic differentiation, node localization, contact time
Key Infrastructure (PKI) to sign and encrypt transmitted estimation, stationary relay nodes, and caching mechanisms
data, as well as encrypted virtual disk volumes to prevent may provide significant performance advantages to VDTNs.
unauthorized access to stored data. The C2C-CC also foresees
the use of a PKI certification infrastructure for securing ACKNOWLEDGMENT
communication between nodes. This work was supported in part by the Instituto de
Telecomunicações, Next Generation Networks and Applica-
F. Cooperation tions Group (NetGNA), Portugal, in the framework of the
VDTN@Lab Project, by the Euro-NF Network of Excel-
The effective operation of VDTN networks relies on the lence of the Seventh Framework Programme of EU, in the
cooperation of nodes to store-carry-and-forward data. In fact, framework of the VDTN Project, by the MPSat project (FCT
most of the research on these networks assumes that nodes Ref. PTDC/EEA-TEL/099074/2008), as well as by national
are fully cooperative, which means that they collaborate with funding from FCT - Fundação para a Ciência e a Tecnolo-
each other, storing and distributing bundles not only in their gia (INESC-ID multiannual funding) through the PIDDAC
own interest, but also in the interest of the other nodes. program funds and through the PEst-OE/EEI/LA0008/2011
Such a behaviour, in conjunction with a multiple-copy routing project.
scheme, increases the number of possible transmission paths,
thus improving the robustness to failure of individual nodes. R EFERENCES
However, this assumption may not be realistic. Network
[1] K. Fall and S. Farrell, “DTN: An Architectural Retrospective”, IEEE
nodes may exhibit a selfish behaviour caused by several rea- J. Sel. Areas Commun., vol. 26, no. 5, pp. 828-836, June 2008.
sons, such as resource limitations (e.g., storage and energy) or [2] D. Waitzman, “A Standard for the Transmission of IP Datagrams on
rogue operation (i.e., malicious behaviour). This can severely Avian Carriers”, IETF, RFC 1149, April 1, 1990.
[3] IP over Avian Carriers, [Online]. Available: http://en.wikipedia.org/wiki/
affect the network performance. IP over Avian Carriers
A detailed discussion on VDTN cooperation issues can be [4] Delay Tolerant Networking Research Group. [Online]. Available: http:
found in [87], including a review of the state of the art of //www.dtnrg.org/wiki
[5] V. Cerf et al., “Delay Tolerant Network Architecture”, IETF, RFC 4838,
this field. This work reinforces the need for developing an April 2007.
incentive mechanism that can act as a basis for achieving [6] K. Scott and S. Burleigh, “Bundle Protocol Specification”, IETF, RFC
cooperation on these networks. 5050, November 2007.
[7] S. Burleigh, A. Hooke, L. Torgerson, K. Fall, V. Cerf, B. Durst, K.
Scott, and H. Weiss, “Delay-Tolerant Networking: An Approach to
VI. C ONCLUSION Interplanetary Internet”, IEEE Commun. Mag., vol. 41, no. 6, pp. 128-
136, June 2003.
Standards for MAC and physical layers of vehicular ad hoc [8] InterPlaNetary (IPN) Internet project. [Online]. Available: http://www.
networks are finished. However, the network layer and above ipnsig.org/
1180 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 14, NO. 4, FOURTH QUARTER 2012

[9] M. Allman et al., “Enhancing TCP Over Satellite Channels using [34] A. Keränen, T. Kärkkäinen and J. Ott, “Simulating Mobility and DTNs
Standard Mechanisms”, IETF, RFC 2488, January 1999. with the ONE”, Journal of Communications, vol. 5, no. 2, pp. 92-105,
[10] M. Allman et al., “Enhancing TCP Over Satellite Channels using Feb. 2010.
Standard Mechanisms”, IETF, RFC 2760, February 2000. [35] The ONE. [Online]. Available: http://www.netlab.tkk.fi/tutkimus/dtn/
[11] Consultative Committee for Space Data Systems. [Online]. Available: theone/
http://public.ccsds.org/ [36] J. Härri, F. Filali and C. Bonnet, “Mobility Models for Vehicular Ad
[12] M. Ramadas, S. Burleigh and S. Farrell, “Licklider Transmission Hoc Networks: A Survey and Taxonomy”, IEEE Commun. Surveys
Protocol - Specification”, IETF RFC 5326, September 2008. Tutorials, vol. 11, no. 4, pp. 19-41, 2009.
[13] R. Wang, T. Taleb, A. Jamalipour, B. Sun, “Protocols for Reliable Data [37] S. Y. Wang and C. L. Chou, “NCTUns Tool for Wireless Vehicular
Transport in Space Internet”, IEEE Commun. Surveys Tutorials, vol. 11, Communication Network Researches”, Simulation Modelling Practice
no. 2, pp. 21-32, 2009. and Theory, vol. 17, no. 7, pp. 1211-1226, Aug. 2009.
[14] S. Farrell and V. Cahill, “Evaluating LTP-T: A DTN-Friendly Transport [38] NCTUns. [Online]. Available: http://nsl.csie.nctu.edu.tw/nctuns.html
Protocol”, Int. Workshop on Satellite and Space Communications, [39] F. Martinez, C. Toh, J. Cano, C. Calafate, P. Manzoni, “A survey
(IWSSC ’07), pp. 178-181, Salzburg, Austria, Sept. 2007. and comparative study of simulators for vehicular ad hoc networks
[15] O. Akan, J. Fang and I. Akyildiz, “TP-Planet: A New Transport Protocol (VANETs)”, Wireless Communications and Mobile Computing, 2009
for InterPlanetary Internet”, IEEE J. Sel. Areas Commun., vol. 22, no. [40] F. Gil-Castiñeira, F. J. González-Castaño and L. Franck, “Extending
2, pp. 348-361, Feb. 2004. Vehicular CAN Fieldbuses With Delay-Tolerant Networks”, IEEE Trans.
[16] G. Papastergiou, I. Psaras, V. Tsaoussidis, “Deep-Space Transport Ind. Electron., vol. 55, no. 9, pp. 3307-3314, Sep. 2009.
Protocol: A novel transport scheme for Space DTNs”, Computer [41] M. Zhang and R. S. Wolff, “Routing Protocols for Vehicular Ad Hoc
Communications, vol. 32, no. 16, pp. 1757-1767, Oct. 2009. Networks in Rural Areas”, IEEE Commun. Mag., vol. 46, no. 11, pp.
[17] J. Partan, J. Kurose, and B. N. Levine, “A Survey of Practical Issues 126-131, Nov. 2008.
in Underwater Networks”, in 1st ACM International Workshop on [42] M. Zhang and R. S. Wolff, “A Border Node Based Routing Protocol
Underwater Networks, in conjunction with ACM MobiCom 2006, Los for Partially Connected Vehicular Ad Hoc Networks”, Journal of Com-
Angeles, California, USA, Sep. 25, 2006, pp. 17 - 24. munications, Academy Publisher, vol. 5, no. 2, pp. 130-143, February
[18] P. Juang, H. Oki, Y. Wang, M. Martonosi, L. S. Peh, and D. Rubenstein, 2010.
“Energy-Efficient Computing for Wildlife Tracking: Design Tradeoffs [43] N. Wisitpongphan, F. Bai, P. Mudalige, V. Sadekar and O. Tonguz,
and Early Experiences with ZebraNet”, ACM SIGOPS Operating “Routing in Sparse Vehicular Ad Hoc Wireless Networks”, IEEE J.
Systems Review, vol. 36, no. 5, pp. 96-107, 2002. Sel. Areas Commun., vol. 25, no. 8, pp. 1538-1556, October 2007.
[19] T. Small and Z. J. Haas, “The Shared Wireless Infostation Model - A [44] M. Rubinstein et al, “Measuring the Capacity of In-Car to In-Car
New Ad Hoc Networking Paradigm (or Where there is a Whale, there Vehicular Networks”, IEEE Commun. Mag., vol. 47, no. 11, pp. 128-
is a Way)”, in 4th ACM International Symposium on Mobile Ad Hoc 136, Nov. 2009.
Networking and Computing (MobiHoc 2003), Annapolis, MD, USA, [45] R. C. Shah, S. Roy, S. Jain, W. Brunette, “Data MULEs: modeling
June 1-3, 2003. a three-tier architecture for sparse sensor networks”, 1st IEEE Inter-
national Workshop on Sensor Network Protocols and Applications, pp.
[20] UMass Diverse Outdoor Mobile Environment (DOME) project,
30-41, Anchorage, AK, USA, May. 2003.
“UMass TurtleNet”, [Online]. Available: http://prisms.cs.umass.edu/
dome/turtlenet. [46] V. Soares, F. Farahmand and J. Rodrigues, “Improving Vehicular
Delay-Tolerant Network Performance with Relay Nodes”, 5th EuroNGI
[21] R. Sherwood and S. Chien, “Sensor Webs for Science: New Directions
Conference on Next Generation Internet Networks (NGI’2009), Aveiro,
for the Future”, in AIAA Infotech@Aerospace 2007, Rohnert Park, CA,
Portugal, July 1-3, Jul. 2009.
May 7-10, 2007.
[47] T. Willke, P. Tientrakool, N. Maxemchuk. “A Survey of Inter-Vehicle
[22] N4C and eINCLUSION, “Networking for Communications Challenged
Communication Protocols and Their Applications”, IEEE Commun.
Communities: Architecture, Test Beds and Innovative Alliances”, [On-
Surveys Tutorials, vol. 11, no. 2, pp. 3-20, 2009.
line]. Available: http://www.n4c.eu/.
[48] S. Guo, M. H. Falaki, E. A. Oliver, S. Rahman, A. Seth, M. A. Zaharia
[23] Wizzy Digital Courier, “Wizzy Digital Courier - leveraging locality”, and S. Keshav, “Very Low-Cost Internet Access Using KioskNet”, ACM
[Online]. Available: http://www.wizzy.org.za/. Computer Communication Review, Oct. 2007.
[24] A. Pentland, R. Fletcher, and A. Hasson, “DakNet: Rethinking Connec- [49] H. Soroush, N. Banerjee, A. Balasubramanian, M. D. Corner, B. N.
tivity in Developing Nations”, in IEEE Computer, vol. 37, 2004, pp. Levine and B. Lynn, “DOME: A Diverse Outdoor Mobile Testbed”,
78-83. Proc. ACM Intl. Workshop on Hot Topics of Planet-Scale Mobility
[25] A. Doria, M. Uden, and D. P. Pandey, “Providing Connectivity to Measurements (HotPlanet), June 2009.
the Saami Nomadic Community”, in 2nd International Conference [50] A. Balasubramanian, B. N. Levine and A. Venkataramani, “Enabling In-
on Open Collaborative Design for Sustainable Innovation, Bangalore, teractive Web Applications in Hybrid Networks”, Proc. ACM Mobicom,
India, December, 2002. pp. 70-80, Sep. 2008.
[26] A. Seth, D. Kroeker, M. Zaharia, S. Guo, and S. Keshav, “Low-cost [51] V. Soares, F. Farahmand and J. Rodrigues, “A Layered Architecture for
Communication for Rural Internet Kiosks Using Mechanical Backhaul”, Vehicular Delay-Tolerant Network”, IEEE Symposium on Computers
in 12th ACM International Conference on Mobile Computing and and Communications (ISCC 2009), Sousse, Tunisia, July 5-8.
Networking (MobiCom 2006), Los Angeles, CA, USA, September 24- [52] Y. Chen, C. Qiao and X. Yu, “Optical Burst Switching: A New Area
29, 2006, pp. 334-345. in Optical Networking Research”, IEEE Network, vol. 18. No. 3, pp.
[27] United Villages, Inc., “First Mile Solutions”, [Online]. Available: http: 16-23, May 2004.
//www.firstmilesolutions.com/. [53] J. Dias, J. Isento, B. Silva, V. Soares, P. Salvador, A. Nogueira and
[28] Tetherless Computing Lab - University of Waterloo, “The KioskNet J. Rodrigues, “Creation of a Vehicular Delay-Tolerant Network Proto-
Project”, [Online]. Available: http://blizzard.cs.uwaterloo.ca/tetherless/ type”, Engenharia’2009 - Inovação e Desenvolvimento, UBI, Covilhã,
index.php/KioskNet. Portugal, Nov. 2009.
[29] N. Glance, D. Snowdon, and J.-L. Meunier, “Pollen: Using People as [54] CarTel Project. [Online]. Available: http://cartel.csail.mit.edu/
a Communication Medium”, Computer Networks: The International [55] B. Hull, V. Bychkovsky, K. Chen, M. Goraczko, A. Miu, E. Shih, Y.
Journal of Computer and Telecommunications Networking, vol. 35, no. Zhang, H. Balakrishnan and S. Madden, “CarTel: A Distributed Mobile
4, pp. 429-442, March 2001. Sensor Computing System”, in Proc. Fourth Int. Conf. Embedded Netw.
[30] J. Jormakka, H. Jormakka, and J. Väre, “A Lightweight Management Sens. Systems (SenSys’06), pp. 125-138, Oct. 2006.
System for a Military Ad Hoc Network”, in Lecture Notes in Computer [56] S. Lahde, M. Doering, W. Pöttner, G. Lammert and L. Wolf,“A practical
Science, vol. 5200/2008: Springer Berlin / Heidelberg, 2008, pp. 533- analysis of communication characteristics for mobile and distributed
543. pollution measurements on the road”, Wireless Communications and
[31] M. Asplund, S. Nadjm-Tehrani, and J. Sigholm, “Emerging Information Mobile Computing, vol. 7, no. 10, pp. 1209-1218, Dec. 2007.
Infrastructures: Cooperation in Disasters”, in Lecture Notes in Computer [57] Jörg Ott, Dirk Kutscher, “From Drive-thru Internet to Delay-tolerant
Science, Critical Information Infrastructure Security, vol. 5508/2009: Ad-hoc Networking”, Book chapter in “Mobile Ad-hoc Networks: From
Springer Berlin / Heidelberg, 2009, pp. 258-270. Theory to Reality”, Nova Science Publishers, Inc., 2007.
[32] DTN Code. [Online]. http://www.dtnrg.org/wiki/Code [58] J. Border, M. Kojo, J. Griner, G. Montenegro, Z. Shelby, “Performance
[33] ION (Interplanetary Overlay Network). [Online]. Available: https://ion. Enhancing Proxies Intended to Mitigate Link-Related Degradations”,
ocp.ohiou.edu/ IETF RFC 3135, June 2001.
PEREIRA et al.: FROM DELAY-TOLERANT NETWORKS TO VEHICULAR DELAY-TOLERANT NETWORKS 1181

[59] K. Scott, “Disruption tolerant networking proxies for on-the-move [83] J. Ott and M. Pitkänen, “DTN-based Content Storage and Retrieval”,
tactical networks”, Military Communications Conference, MILCOM IEEE International Symposium on a World of Wireless, Mobile and
2005, vol. 5, pp. 3226-3231. Multimedia Networks (WoWMoM 2007) - First IEEE Workshop on
[60] Y. Toor, P. Mühlethaler, A. Laouiti, A. de la Fortelle, “Vehicle Ad Hoc Autonomic and Opportunistic Communications (AOC 2007), Helsinki,
Networks: Applications and Related Technical Issues”, IEEE Commun. Finland, June 18-21, 2007.
Surveys Tutorials, vol. 10, no. 3, pp. 74-88, 2008. [84] M. Sichitiu, M. Kihl, “Inter-Vehicle Communication Systems: A
[61] Car 2 Car Communication Consortium Manifesto. Overview of the C2C- Survey”, IEEE Commun. Surveys Tutorials, vol. 10, no. 2, pp. 88-105,
CC System, 2007. [Online]. Available: http://www.car-2-car.org/ 2008.
[62] Intelligent Transport Systems (ITS); Communications Architecture; [85] F. Qu, F. Wang, L. Yang, “Intelligent Transportation Spaces: Vehicles,
ETSI EN 302 665, v1.1.1, September 2010. Traffic, Communications, and Beyond”, IEEE Commun. Mag., vol. 48,
[63] White paper, “How TomTom’s HD Traffic and IQ Routes data provides no. 11, pp. 136-142, Nov. 2010.
the very best routing”, [Online]. Available: http://www.tomtom.com/ [86] S. Farrell, S. Symington, H. Weiss, P. Lovell, “Delay-Tolerant Network-
[64] L. Franck and F. Gil-Castiñeira, “Using Delay Tolerant Networks ing Security Overview”, IRTF, draft-irtf-dtnrg-sec-overview-06, March
for Car2Car Communications”, IEEE International Symposium on 2009.
Industrial Electronics, ISIE 2007, pp. 2573-2578, Vigo, Spain, June [87] V. Soares and J. Rodrigues, “Chapter 7: Cooperation in DTN-Based
2007. Network Architectures”, in Cooperative Networking, S. Misra and M.
[65] J. Burgess, B. Gallagher, D. Jensen, and B.N. Levine, “MaxProp: routing Obaidat (Ed.s), pp. 101-115, ISBN: 978-0-470-74915-9, Wiley, 2011.
for vehicle-based disruption-tolerant network”, in 25th IEEE INFOCOM
Conference, pp. 1688-98, Barcelona, Spain, Apr. 2006.
[66] V. Cabrera, F. J. Ros and P. M. Ruiz, “Simulation-based Study of
Common Issues in VANET Routing Protocols”, IEEE 69th Vehicular
Technology Conference, VTC2009-Spring, Barcelona, Spain, April 2009. Paulo Rogério B. A. Pereira received his gradua-
[67] Z. Zhang, “Routing in Intermittently Connected Mobile Ad Hoc tion, M.Sc. and Ph.D. degrees in Electrical and Com-
Networks and Delay Tolerant Networks: Overview and Challenges”, puter Science Engineering from Instituto Superior
IEEE Commun. Surveys Tutorials, vol. 8, no. 1, pp. 24-37, 2006. Técnico, Technical University of Lisbon (IST/UTL),
[68] M. Khabbaz, C. Assi, W. Fawaz, “Disruption-Tolerant Networking: A Portugal, in 1991, 1994 and 2003, respectively.
Comprehensive Survey on Recent Developments and Persisting Chal- He is an assistant professor of computer networks
lenges”, Accepted for publication in IEEE Commun. Surveys Tutorials. related subjects at IST/UTL, and a researcher at
[69] F. Tchakountio and R. Ramanathan, “Tracking Highly Mobile Ed- the Communications and Mobility Networks lab-
points”, ACM Workshop Wireless Mobile Multimedia (WoWMoM) , oratory of INESC-ID Lisbon, since 1991. He has
Rome, Italy, July 2004. participated in the IST European projects EuroNGI,
[70] A. Vahdat and D. Becker, “Epidemic Routing for Partially Connected EuroFGI, EuroNF, UbiSec&Sens and WSAN4CIP.
Ad Hoc Networks”, Tech. Rep. CS-200006, Department of Computer His research interests include IP quality of service, network management,
Science, Duke University, Durham, NC, 2000. wireless sensor networks and delay-tolerant networking.
[71] Q. He, Y. Li, and X. Fan, “A Study on Buffer Efficiency and
Surround Routing Strategy in Delay Tolerant Network”, in Proc.
2009 International Conference on Dependable, Autonomic and Secure
Computing (DASC 2009), pp. 566-570, Chengdu, China, Dec. 2009
Augusto Casaca graduated in Electrical Engineer-
[72] A. Lindgren, A. Doria and O. Schelen, “Probabilistic Routing in
ing at the Instituto Superior Técnico (IST), Technical
Intermittently Connected Networks”, in Proc. First Int. Workshop on
University of Lisbon, Portugal. He got the M.Sc.
Service Assurance with Partial and Intermittent Resources, pp. 239-254,
degree in Digital Electronics at UMIST, Manch-
Fortaleza, Brazil, Sep. 2004.
ester, UK, and the Ph.D. in Computer Science
[73] A. Balasubramanian, B.N. Levine, and A. Venkataramani, “DTN routing at the University of Manchester, Manchester, UK.
as a resource allocation problem”, ACM Computer Communication Presently he is Full Professor at IST and Leader of
Review, vol. 37, no. 4, pp. 373-384, Oct. 2007. the Action Line on “Communications and Mobility
[74] B. Burns, O. Brock, and B.N. Levine, “MORA Routing and Capacity Networks” at INESC-ID. He has been involved in
Building in Disruption-Tolerant Networks”, Ad Hoc Networks, Elsevier, many research and development activities in the
vol. 6, no. 4, pp. 600-620, June 2008. area of Networking and has participated in several
[75] G. Karagiannis, O. Altintas, E. Ekici, G. Heijenk, B. Jarupan, K. Lin, EURESCOM, RACE, ACTS and IST projects in the area of Broadband Com-
T. Weil, “Vehicular Networking: A Survey and Tutorial on Require- munications and Networking. He has actively participated in standardization
ments, Architectures, Challenges, Standards and Solutions”, Accepted activities at ITU-T and ETSI, has published a significant number of scientific
for publication in IEEE Commun. Surveys Tutorials. papers in journals and conferences, and has chaired or co-chaired five IEEE
[76] F. Farahmand, I. Cerutti, A. N. Patel, J. P. Jue and J. Rodrigues, and IFIP conferences on Networking. He was Chairman of IFIP Technical
“Performance of vehicular delay-tolerant networks with relay nodes”, Committee 6 (Communication Systems) from 1998 to 2004 and is a Senior
Wireless Communications and Mobile Computing Journal, 2009; 9:1- Member of IEEE. His present research interests are in the areas of DTN and
11. WSN.
[77] V. Soares, F. Farahmand and J. Rodrigues, “Scheduling and Drop Poli-
cies for Traffic Differentiation on Vehicular Delay-Tolerant Networks”,
17th International Conference on Software, Telecommunications and
Computer Networks (SoftCOM 2009), Croatia, September 24-26, 2009.
[78] A. Panagakis, A. Vaios, and I. Stavrakakis, “On the Effects of
Cooperation in DTNs”, in Proc. 2007 2nd Int. Conf. on Communication
System Software, Middleware and Workshops, pp. 749-754, Bangalore,
India, Jan. 2007.
[79] W. Zhao, M. Ammar, E. Zegura, “Multicasting in Delay Tolerant
Networks: Semantic Models and Routing Algorithms”, Proc. SIG-
COMM’05, Aug. 2005.
[80] Q. Ye, L. Cheng, M. C. Chuah, B. Davison, “Performance comparison
of different multicast routing strategies in disruption tolerant networks”,
Computer Communications, vol. 32, pp.1731-1741, Elsevier, 2009.
[81] J. Santiago, A. Casaca, P. R. Pereira, “Multicast in Delay Tolerant
Networks using Probabilities and Mobility Information”, Ad Hoc &
Sensor Wireless Networks, An International Journal, vol. 7, no. 1-2, pp.
51-68, 2009.
[82] B. Silva, F. Farahmand, J. Rodrigues, “Performance Assessment
of Caching and Forwarding Algorithms for Vehicular Delay Tolerant
Networks”, Proceedings IEEE Globecom’2010, Dec. 2010.
1182 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 14, NO. 4, FOURTH QUARTER 2012

Joel J. P. C. Rodrigues (S’01 - M’06 - SM’06), is Joan Triay received a BEng and a MEng
a Professor in the Department of Informatics of the in Telecommunications Engineering, and a MSc
University of Beira Interior, Covilhã, Portugal, and a in Telematics Engineering at the Universitat
Researcher at the Instituto de Telecomunicaç ões, as- Politècnica de Catalunya (UPC), Spain, in 2004,
sociated Lab, Portugal. He received a five-year B.S. 2006 and 2007, respectively. In 2007, he was
degree (licentiate) in Informatics Engineering from awarded with a 4-year pre-doctoral scholarship
the University of Coimbra, Portugal and the M.Sc. through the Research Training Program (FI) from
degree and Ph.D. degree in Informatics Engineer- the Government of Catalonia, and since then he
ing from the University of Beira Interior, Portugal. is a PhD candidate and research assistant in the
His main research interests include delay-tolerant Department of Telematics Engineering at UPC. He
networks, sensor networks, high-speed networks, e- was a visiting fellow at University of Essex, United
Learning, e-Health, and mobile and ubiquitous computing. He is the leader Kingdom, from June 2009 to February 2010 thanks to a BE-DGR fellowship.
of the NetGNA Research Group (http://netgna.it.ubi.pt) and the founder and His research interests include, but not limited to, future optical network
leader of the IEEE ComSoc CSIM Special Interest Group on modeling architectures and the provisioning of multi-service capabilities on high-speed
and simulation tools (http://mst.it.ubi.pt). He is the Secretary of the IEEE networks.
ComSoc Technical Committee on eHealth and the Vice-Chair of the IEEE
ComSoc Technical Committee on Communications Software. He has authored
or coauthored over 170 technical papers in refereed international journals and
conferences, book chapters, a book, and a patent. He is the editor-in-chief Cristina Cervelló-Pastor received her MSc degree
of the International Journal on E-Health and Medical Communications and in Telecom Engineering in 1989 and Ph.D degree in
served several Special Issues as a Guest Editor (IEEE Trans. Multimedia, El- Telecommunication Engineering in 1998, both from
sevier Journal of Network and Computer Applications, IET Communications, the Escola Tècnica Superior d’Enginyers de Tele-
Journal of Communications, etc.). He has served as General Chair, Technical comunicació, Universitat Politècnica de Catalunya
Program Committee Chair, and symposium Chair for many international (UPC), Barcelona, Spain. She is currently an As-
conferences, including IEEE ICC/GLOBECOMs, CAMAD, MAN, ITST, sociated Professor in the Department of Telematics
ICNC, SoftCOM, among others. He participated in tens of international Engineering at UPC, which she joined in 1989,
TPCs and several editorial review boards (including IEEE Commun. Mag., and leader of the optical networks research group
International Journal of Communications Systems, etc.). He is a licensed within the BAMPLA research line. Her research
professional engineer (as a senior member), and he is member of ACM trajectory has been centered on the field of routing
SIGCOMM, a member of the Internet Society, an IARIA fellow, and a senior in high speed networks and the development of new protocols and services in
member of the IEEE Computer Society, IEEE Communications Society, and OBS/OPS, taking part in diverse national and European projects (FEDERICA,
IEEE Education Society. ATDMA, A@DAN, Euro-NGI, Euro-FGI, EURO-NF) and being responsible
of various public and private funding R&D projects, some of them with the
i2CAT Foundation. In parallel she has presented several patent proposals about
OBS networks.

Vasco N. G. J. Soares received his five-year B.S. de-

gree (licentiate) in Informatics Engineering from the
University of Coimbra, Portugal, in 2002. He is an
Assistant Professor in the Technical-Scientific Unit
of Informatics at the Superior School of Technology
of the Polytechnic Institute of Castelo Branco, Por-
tugal. He is currently a Ph.D. student in Informat-
ics Engineering at the University of Beira Interior,
Covilhã, Portugal, and a PhD student member of
the Instituto de Telecomunicaç ões, Portugal. He has
authored or coauthored over 20 technical papers
in refereed international journals, conferences, and book chapters. He has
served as a Technical Program Committee Member and Web Chair of several
international conferences and has participated in the European project Euro-
NF. His main research interests include delay-tolerant networks and vehicular
networks.