Content uploaded by Farid Farahmand
Author content
All content in this area was uploaded by Farid Farahmand on Jan 05, 2015
Content may be subject to copyright.
Relay Node Placement in Vehicular Delay-Tolerant
Networks
Farid Farahmand§, Isabella Cerutti†, Ankitkumar N. Patel‡, Qiong Zhang††, and Jason P. Jue‡
§Department of Computer Electronics and Graphics Technology
Central Connecticut State University, New Britain, CT 06050
†Scuola Superiore Sant’Anna, Pisa, Italy
††Photonics Networking Laboratory
Fujitsu Laboratories of America, Inc., Richardson, TX 75082
‡Department of Computer Science
The University of Texas at Dallas, Richardson, TX 75083
email: farahmandfar@ccsu.edu or isabella.cerutti@sssup.it
Abstract— Delay-tolerant networking (DTN) is an architecture to
enable data communications between isolated or remote regions,
where long delays and intermittent connectivity can be tolerated.
An emerging class of DTN, called Vehicular DTNs (VDTN), exploits
transportation systems as the transport layer to transfer data. In
these networks, vehicles (e.g., busses, boats, trains) act as mobile
nodes and carry data messages around. Mobile nodes can exchange
data messages using devices called relay nodes. Relay nodes, placed
in strategic positions along vehicle routes, have the capability to
download, store, and upload the data messages from/to the mobile
nodes.
An important issue in VDTN is the optimal placement of the relay
nodes such that delay-tolerant connectivity in VDTN is ensured at
minimum cost. In this paper we show that the problem of optimal
relay node placement is an NP-hard problem. Other contributions
of this paper are the formulation of the relay node placement
problem using ILP and the proposal of heuristic algorithms solving
the optimization problem. Using simulation results, we compare the
performance of each algorithm under different network constraints,
such as node storage capability and network topology.
Index Terms— Delay-Tolerant Network, Routing, Wireless Net-
works, Transit Networks.
I. INTRODUCTION
Delay/Disruption Tolerant Networks (DTNs) are a class of
architectural solutions that have been proposed to address the
connectivity challenges due to long transmission delays and
intermittent connectivity [1]. The key feature of DTNs lies in their
ability to store the data in the network nodes until the connectivity
required to forward the data becomes available. In general, a DTN
architecture can be characterized by (a) sparse connectivity (i.e.,
end-to-end route between source and destination may not even
exist), (b) long or variable delay, (c) asymmetric data rate, and
(d) high error rate.
A particular class of DTN architecture is the vehicle-based or
vehicular DTN (VDTN). In these networks, also known as transit
networks [2], vehicles (e.g., cars, buses, boats) collect and deliver
data between static nodes. A unique application of VDTN is to
provide asynchronous Internet access to rural and remote regions,
e.g., remote villages, which would be otherwise disconnected
from Internet services.
A number of projects have been dedicated to study and ad-
dress the connectivity challenges in rural regions using a VDTN
architecture [6]. For example, motivated by the Daknet project
[3], the Rural Internet Kiosk project was deployed to provide
a low-cost communication system between rural regions [7].
Another example is the Message Ferry (MF) project that aims
at developing a data delivery system in remote villages using
mobile nodes called message ferries [4]. The DieselNet testbed,
developed at the University of Massachusetts, is another attempt
to implement a VDTN [8]- [9].
In order to provide the requested delay-tolerant connectivity,
each one of the above projects addresses specific network is-
sues and problems under certain service requests and network
assumptions such as topology, node architecture, capabilities,
mobility patterns, and available knowledge about the network.
For example, the Rural Internet Kiosk project focuses on node
hardware and software architecture. In MF, the main idea is to
design the mobility pattern in order to improve the performance.
In the DieselNet project, the developers focus on improving
routing mechanisms.
This paper considers a VDTN architecture consisting of ve-
hicles that act as mobile nodes and carry the data between
terminal nodes located in isolated regions. The data messages
between mobile nodes can only be exchanged via store-and-
forward devices, called relay nodes. Mobile nodes are capable
of dropping and picking up data to/from relay nodes encountered
along their routes.
The main contribution of this paper is the optimization of
the VDTN design. The optimal VDTN design problem aims at
finding the relay node placement that minimizes the network cost,
while providing full connectivity between users located in remote
regions. Network cost is considered proportional to the number
of relay nodes that need to be installed to offer delay-tolerant
connections. To the best of our knowledge, this is the first time
such a study has been conducted in the context of VDTN.
The remainder of this paper is structured as follow. In Sec-
tion II, the network model and the assumptions are described.
In Section III, the relay node placement problem is formulated
using an integer linear programming formulation. In Section IV,
heuristic algorithms solving the relay node placement problem
are presented. Section V shows the performance comparison
among the proposed heuristic algorithms under different network
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.
978-1-4244-2324-8/08/$25.00 © 2008 IEEE.
Fig. 1. An example of a VDTN and its components. Isolated regions are circled.
A candidate node can be any of the intersecting points or terminal nodes.
constraints, followed by concluding remarks in Section VI.
II. NETWORK MODEL AND ASSUMPTIONS
The proposed VDTN network utilizes a transportation system
as its transport layer to deliver data. Hence, VDTN consists of
vehicles (e.g., trains, buses) along predetermined routes, as shown
in Fig. 1. Each route passes along multiple terminal stations of
the transportation system. Two or more routes may intersect at
intersection or terminal station.
Abundle layer [1] is used as a new layer over the transportation
system to aggregate incoming IP data packets into bundle mes-
sages, or messages, and to provide end-to-end message delivery.
Due to the intermittent connectivity and the discrete nature of
transfer opportunities in the transportation system, routing large
size messages is more suitable than small size IP packets. The
bundle layer allows uploading, downloading, and exchanging
messages from/to the terminal stations. These operations can take
place only when the vehicles approach the terminal nodes. We
assume that messages are based on the IP paradigm and the access
network connected to the VDTN is an IP network.
Three different types of nodes, namely (a) terminal nodes, (b)
relay nodes, and (c) mobile nodes, are deployed in the proposed
VDTN architecture.
Terminal nodes are access points to VDTN and they are
strategically located at terminal stations to support their surround-
ing users. Individual end-users are connected to terminal nodes
through low-range low-power radio frequency signals, such as
the IEEE 802.11 wireless protocol. A key function of terminal
nodes is to aggregate the incoming IP data packets and to create
messages. Outgoing IP data packets between a terminal node and
its surrounding end-users can be routed using commonly known
protocols, such as dynamic source routing (DSR) [5].
Mobile nodes are mounted on vehicles (e.g., buses) and act as
store-carry-forward devices. It is assumed that message exchange
between mobile nodes is not possible due to the short and infre-
quent contact times among mobile nodes (i.e., the time intervals
during which the mobile nodes are within communication range)
and the low data-rate that would be reached. In order to store
the incoming data messages until successful forwarding, mobile
nodes are provided with considerable storage capacity.
Relay nodes are also store-and-forward devices which receive
messages from mobile nodes and store them until they can be
uploaded on another mobile node. The main function of these
nodes is to allow message exchanges between various mobile
nodes. Relay nodes are typically stationary and located at terminal
nodes shared by different routes, as shown in Fig. 1. Any route
intersection is a candidate location for a relay node and hence,
it is referred to as candidate node. Generally, relay nodes have
large storage requirements. Depending on the traffic volume, relay
nodes may be equipped with hard drives having several thousand
gigabytes of capacity.
The wireless technology, that should be selected to support
message exchanges between terminal nodes, relay nodes, and
mobile nodes, depends on convergence and economical consider-
ations. For instance, low-powered radio communications based on
IEEE 802.11 a/b/g or ultra-wideband (UWB) radio are promising
technological solutions for VDTN implementation.
In the next section, the problem of selecting which candidate
nodes should act as relay nodes is formalized.
III. DESCRIPTION OF THE PROBLEM
A VDTN network can be represented as a graph G(V, E ).
The set of nodes Vincludes the set of terminal nodes where
the traffic is originated, Vt, and the set of terminal nodes where
routes intersect (i.e., candidate nodes), Vc. Thus, V⊆Vt∪Vc. A
directed link iexists from node sto node d, when mobile nodes
moving along route iconnect sto d, i.e., vehicle route ipasses
both nodes in sequence. Fig. 2 shows the graph derived for the
VDTN network in Fig. 1.
The relay nodes can be placed in any of the candidate nodes
in Vc. The relay node placement problem can be defined as
follows. Given a VDTN graph and the requested traffic rate (e.g.,
message rate) between source and destination nodes, the relay
node placement problem aims at finding the placement of the
minimum number of relay nodes, in order to support the requested
traffic. The relay placement problem can be described with an
integer linear programming formulation, as presented next.
Fig. 2. DTN network graph derived for the transportation system in Fig. 1.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.
978-1-4244-2324-8/08/$25.00 © 2008 IEEE.
A. Problem Formulation
Given
•G(V, E ): VDTN network graph as defined above;
•Λ: the matrix of the traffic requests, Λsd, expressed in terms
of traffic rates, from source node s∈Vtto destination node
d∈Vt, i.e., Λ∪s,d Λsd;
•W: the set of routes in the transportation system;
•πi:ith route (i.e., sequence of nodes in V);
•Cb: maximum data transportation rate along a route;
•B: maximum transmission bandwidth at a terminal node;
•Cr: maximum storage capacity of a relay node;
•T: large constant, e.g., T=sd Λsd.
Variables:
•Rmdenotes whether node mis a relay node.
Rm=1if candidate node m∈Vcis a relay node
0if m∈Vtor if m∈Vcis not a relay node;
(1)
•λsd,i
mn denotes the traffic rate from sto dpassing on directed
link (m,n) of G, using route i. The end-to-end traffic rate
Λsd may be carried using one route ior multiple routes
in parallel or in sequence thanks to the store-and-forward
capabilities of relaying nodes.
Objective Function:
Minimize: m∈VcRm
Constraints:
i∈W
m:(s,m)∈πi
λsd,i
sm = Λsd ∀s, d (2)
i∈W
m:(m,s)∈πi
λsd,i
ms = 0 ∀s, d (3)
i∈W
m:(m,d)∈πi
λsd,i
md = Λsd ∀s, d (4)
i∈W
m:(d,m)∈πi
λsd,i
dm = 0 ∀s, d (5)
i∈W
m:(m,k)∈πi
λsd,i
mk =
i∈W
n:(k,n)∈πi
λsd,i
kn ∀s, d ∀k∈V(6)
m:(m,k)∈πi
λsd,i
mk −
n:(k,n)∈πi
λsd,i
kn ≤RkT∀s, d ∀k∈V(7)
Eq. 2-6 are flow conservation constraints, i.e., the requested
traffic rate should be routed in the network from sto d, using
one or multiple routes and passing through intermediate relay
nodes.
Eq. 7 forces to route the traffic on the same route iif node
klacks relaying capabilities. In other words, the combination of
Eq. 6 and 7 allows the relaying of traffic on a different routes at
relay nodes, while it forces the routing of the traffic on the same
route in the intermediate nodes without relaying capabilities.
B. Additional Constraints
The relay placement problem described above can be subject
to one or more of the following constraints:
•Constraint on relay node storage capacity. The amount of
data stored by relay node mshould not be greater than Cr,
i.e., T=Cr.
•Constraint on route capacity. The amount of data carried by
mobile nodes along route ifrom mto nshould not be greater
than Cb, i.e.,
sd
λsd,i
mn ≤Cb∀(m, n)∀i. (8)
•Constraint on transmission bandwidth:
i∈W
m:(m,k)∈πi
λsd,i
mk ∀m6=d∈V. (9)
In absence of constraints (e.g., when the capacity is large)
and when the set of terminal nodes coincides with the set of
candidate nodes, the solution of the connected dominating set
(CDS) problem [10]–[12] identifies a set of relay locations that
assure a feasible route between the source-destination pair of the
traffic requests. Finding the smallest CDS on a general graph is
NP-hard and thus the complexity of the optimal relay placement
problem in the general case is also NP-hard.
When solved, the above presented ILP formulation finds the
placement of the minimum number of relay nodes in the DTN.
Using this solution, traffic request routing can be further op-
timized by routing traffic along the minimum latency routes,
while meeting the capacity constraints. Under the assumption that
message storage time at relay nodes is negligible compared to
the transport time of the messages along the routes, the routing
can be optimized using a well-known multi-commodity flow
formulation that aims at minimizing the average message delivery
time (AMDT), i.e., total time spent by a message along the routes.
Next, heuristic algorithms that aim at optimizing the relay
node placement and then the average message delivery time are
presented.
IV. HEURISTIC ALGORITHMS
This section presents two heuristic algorithms that attempt to
minimize the number of required relay nodes in the network,
while guaranteeing end-to-end connectivity between the terminal
nodes. Each algorithm aims at minimizing one of the following
cost functions subject to minimum number of new relay nodes: (a)
minimizing the number of hop counts between source-destination
terminal nodes, (b) minimizing the average message delivery
time (AMDT) to the destination. We also introduce a heuristic
algorithm, which provides an upper bound solution to the relay
node placement problem.
A. Minimizing Relay Node and Hop Count (MRH) Algorithm
Minimizing Relay Node and Hop Count (MRH) algorithm
maximizes the relay node usage in order to reduce the number
of required relay nodes. Hence, no new relay nodes are added
unless all existing ones have been fully utilized. The description
of the MRH algorithm is given below:
Step 1 : Initialization
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.
978-1-4244-2324-8/08/$25.00 © 2008 IEEE.
a) Given the traffic matrix Λsd ∈Λwith traffic rates
between each s−dnode pair, make a copy Q= Λ.
b) Set ˇ
Ci
r=Cr∀i∈Vc, where ˇ
Ci
ris the available
average storage capacity of relay node i.
c) Select the largest traffic request, i.e., qmax
sd =
maxij Qij .
Step 2 : Find the route (i.e., routing on G(V, E )) for qmax
sd with
the fewest existing relay nodes.
a) If there are more than one such route use the one
with the smallest hop count.
b) If no such route exists, convert another candidate
node to a relay node such that a route with the lowest
hop count is found.
Step 3 : Verify that the constraint on route capacity is met, i.e.,
ˇ
Cj
bshould not be greater than its maximum capacity Cb.
Verify that the constraint on relay node storage capacity
is met, i.e., the total traffic stored by each relay node,
ˇ
Ci
r, should not be greater than its maximum capacity,
Cr.
Step 4 : Set qmax
sd = 0, update ˇ
Cj
band ˇ
Ci
rfor all relay nodes
and routes utilized along the path. Then return to Step
2until all nonzero traffic demands are satisfied.
Step 5 : Count the number of relay nodes carrying traffic.
B. Minimizing Relay Node and Delivery Time (MRD) Algorithm
Minimizing Relay Node and Delivery Time (MRD) algorithm
minimizes AMDT while adding the fewest number of new relay
nodes. The basic difference between this algorithm and MRH
is that, rather than focusing on minimizing the number of relay
nodes, messages are routed on the route with smallest AMDT.
All steps in MRD algorithm are similar to MRH except Step 2,
which must be modified as follow:
Step 2 : Find the path on G(V, E )for qmax
sd requiring the
minimum number of new relay nodes.
a) If there are more than one such path, use the one
with the smallest hop count.
b) If no such path exists, convert another candidate node
to a relay node such that a path with the smallest AMDT
is obtained.
C. Minimizing Delivery Time (MDT) Algorithm
An upper bound solution to the relay node placement prob-
lem can be obtained by routing all requests on the route with
the smallest AMDT. Hence, Minimizing Delivery Time (MDT)
algorithm assumes that all candidate nodes already act as relay
nodes. Therefore, routing in MDT results in finding the shortest
route between node pairs. If there are more than one such route,
the route with the smallest hop count is selected. Once all traffic
demands are routed, relay nodes which are not carrying any traffic
are removed. We note that MDT provides the lower bound to the
average end-to-end message delivery time.
V. PERFORMANCE ANALYSIS
This section compares the performance of the aforementioned
algorithms. Network graphs, as defined in Section III, are ran-
domly generated by uniformly distributing Nnodes in a 20 ×
20km2geographical area. The probability of establishing a link
between a node pair is uniform. The network connectivity is
given by connectivity factor (K), indicating the percentage ratio
of established links over the number of links in a fully connected
network (i.e., N(N−1)/2). The number of generated network
topologies is sufficient to guarantee that the mean results a
confidence interval of 15% or better at 90% confidence level.
For each network topology, a different traffic matrix is gen-
erated. Each element in the matrix represents the traffic rate
(total number of messages generated in the network per unit
time) between a given node pair. Traffic requests are uniformly
distributed among all the node pairs. The network load is defined
as the total traffic rate offered to the network.
It is assumed that mobile nodes (e.g., buses) move at a speed
of 40 km/hr and have a transmission range of 10 m. Also, the
average time spent by a message along each given link (including
the waiting time and transport time) is known in advance. Unless
otherwise indicated, the following assumptions are made: N= 15,
K= 14%, unlimited storage capacity at relay nodes and mobile
nodes, and unlimited transmission bandwidth.
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
10
20
30
40
50
60
70
80
Network Load
% Ratio of Relay Node to Candidate Node
MRH
MDT
MRD
ILP−Results
(a) PRC vs. network load.
10% 14% 28%
0
5
10
15
20
25
30
35
Percentage Connectivity (K)
% Ratio of Relay Node to Candidate Node
MRH
MRD
(b) Ratio of relay nodes to candidate
nodes for K= 10%, 14%, and 28%, when
network load equals 0.8.
Fig. 3. PRC and ratio of relay nodes to candidate nodes.
Fig. 3(a) shows the percentage ratio of relay nodes to candidate
nodes (PRC) for each heuristic algorithm versus the network load.
When ratio of relay nodes to candidate nodes reaches 100%, each
candidate node is selected to be a relay node. This figure suggests
that MRH consistently performs better than MRD. This is mainly
due to the fact that MRH attempts to find the route with the lowest
hop count, resulting in utilizing fewer relay nodes to satisfy all
demands. The figure also shows that the results obtained by the
ILP and the MDT algorithm represent the lower and upper bounds
on the number of relay nodes, respectively.
Fig. 3(b) shows the ratio of relay nodes to candidate nodes for
different values of connectivity factor. As K increases (from 10%
to 28%), both MRH and MRD algorithms require smaller ratio
of relay nodes to candidate nodes. The reason is that the network
is more connected and fewer relay nodes are required. The figure
suggests that MRH consistently outperforms MRD, regardless of
the value of K.
Fig. 4 show the performance of the proposed algorithms in
terms of average message delivery time (AMDT) versus the
network load and network size, respectively. Note that AMDT
also provides an insight into the network throughput, i.e., the
amount of messages delivered between nodes in a give time unit.
In fact, AMDT is inversely proportional to network throughput.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.
978-1-4244-2324-8/08/$25.00 © 2008 IEEE.
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
18
20
22
24
26
28
30
32
34
Network Load
Average Message Delivery Time (Min.)
MRH
MDT
MRD
(a) AMDT vs. network load.
15 20 25
0
5
10
15
20
25
30
35
40
45
50
Network Size
Average Message Delivery Time (Min.)
MDT
MRD
MRH
(b) AMDT vs. network size; Nchang-
ing between 15, 20, and 25.
Fig. 4. AMDT as network load and size varies.
Fig. 4(a) suggests that, in general, as the network load changes,
AMDT varies slightly. This is due to the fact that transmission
bandwidth is assumed to be infinity. The figure suggests that
MRD outperforms MRH in terms of AMDT, while MDT provides
the lower bound for AMDT. When the network load reaches
its mid-range values (0.5-0.7) AMDT decreases. This counter-
intuitive result is due to the fact that as the network load increases,
more relay nodes are required to be placed in the network. Once
sufficient number of relay nodes are placed, more route options
are available, resulting in lower AMDT.
Fig. 4(b) shows AMDT obtained by the algorithms when
the total number of nodes in the network increases. The figure
suggests that MRD outperforms MRH regardless of the network
size.
Fig. 5(a) shows the expected maximum storage capacity re-
quired by relay nodes versus the network load. Note that as the
network load increases, MRH requires larger buffer capacity. This
is mainly due to the fact that the total number of relay nodes
placed by MRH algorithm is lower compared to MRD, resulting
in large amount of relayed traffic in MRH.
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
20
40
60
80
100
120
140
160
180
200
Network Load
Average Required Buffer Per Relay (MB)
MRH
MDT
MRD
(a) Average maximum buffer size vs.
network load.
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
50
100
150
200
250
300
350
400
450
500
Network Load
Average Required Buffer Per Relay in MB
MRH (Long Waiting Time in Relay)
MRH
MRD (Long Waiting Time in Relay)
MRD
(b) Expected maximum buffer capacity
vs. network load.
Fig. 5. Average maximum buffer size and expected maximum buffer capacity
as network load varies.
In general, as the inter-contact time (frequency of vehicles
visiting relay nodes) reduces, messages tend to experience longer
waiting time at intermediate relay nodes. Fig. 5(b) shows that
when inter-contact time becomes longer, MRD outperforms MRH
in terms of expected maximum storage capacity at relay nodes;
the difference becomes more significant as the network load
increases. This is due to the fact that MRH chooses the routes with
fewer hops; therefore, messages may experience longer waiting
time at relay nodes before being transferred to the mobile nodes.
In summary, MRD and MRH offer different advantages in
terms of hardware resources (i.e., storage capacity and number
of relay nodes) and message delivery time. While, MRD results
in faster message delivery, MRH is more resource efficient.
VI. CONCLUSION
This paper considered the problem of the optimal relay node
placement in vehicular delay tolerant networks (VDTN). The
main motivation for this work is the reduction of the network
cost in terms of the number of relay nodes required to be
installed throughout the network. Two heuristic algorithms aiming
at minimizing the number of relay nodes, namely MRH and
MRD, were proposed to solve the relay node placement problem.
We compared the results obtained from each algorithm and
the optimal solution obtained by ILP formulation. The upper
bound solution was evaluated using a heuristic that minimizes the
delivery time, i.e., MDT. Results obtained on randomly generated
VDTN graphs evaluate the algorithm performance in terms of
hardware resources (i.e., number and storage capacity of relay
nodes) and message delivery time. These results indicate that
while MRD results in faster message delivery, MRH is more
resource efficient.
In this work we ignored bus storage and transmission band-
width limitations. Further studies are required to fully understand
the benefits of emerging VDTN, in terms of cost effectiveness for
the offered performance.
REFERENCES
[1] V. Cerf et al., “Delay-Tolerant Networking Architecture,” RFC 4838, April
2007.
[2] LeBrun, J. Chen-Nee Chuah, D. Ghosal, M. hang, “Knowledge-based
opportunistic forwarding in vehicular wireless ad hoc networks,” Vehicular
Technology Conference, 2005.
[3] A. Pentland, R. Fletcher, and A. Hasson, “Daknet: Rethinking connectivity
in developing nations,” IEEE Computer, Vol. 37, No. 1, pp. 7883, 2004.
[4] W. Zhao and M. H. Ammar, “Message ferrying: proactive routing in highly-
partitioned wireless ad hoc networks,” In proceedings of the IEEE Workshop
on Future Trends in Distributed Computing Systems, Puerto Rico, May 2003.
[5] D. B. Johnson and D. A. Maltz, “Dynamic source routing in ad hoc wireless
networks,” Mobile Computing, Kluwer Academic Publisher, Vol. 353, 1996.
[6] Z. Zhang, “Routing in intermittently connected mobile ad hoc networks and
delay tolerant networks: overview and challenges,” IEEE Communications
Surveys and Tutorials, Vol. 8, No. 1, pp. 24-37, 2006.
[7] A. Seth, D. Kroeker, M. Zaharia, S. Guo, S. Keshav, “Low-cost communi-
cation for rural internet kiosks using mechanical backhaul,” Proceedings of
the 12th annual international conference, 2006.
[8] J. Burgess, B. Gallagher, D. Jensen, and B. N. Levine, “MaxProp: Routing
for Vehicle-Based Disruption- Tolerant Networks,” In Proc. IEEE INFO-
COM, April 2006.
[9] A. Balasubramanian, B. N. Levine, and A. Venkataramani, “DTN Routing
as a Resource Allocation Problem,” In Proc. ACM SIGCOMM, 2007.
[10] M. S. Savasini, P. Monti, M. Tacca, A. Fumagalli, and H. Waldman,
“Regenerator placement with guaranteed connectivity in optical networks,”
ONDM 2007.
[11] T. Carpenter, D. Shallcross, J. Gannett, J. Jackel, A. Von Lehmen, “Maxi-
mizing the transparency advantage in optical networks,” OFC 2004.
[12] S. Guha and S. Khuller, “Approximation Algorithms for Connected Domi-
nating Sets,” Algorithmica, vol. 20, no. 4, April 1998.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.
978-1-4244-2324-8/08/$25.00 © 2008 IEEE.