TOWER: Topology Optimization for netWork Enhanced Resilience
Enrique de la Hoz
1
, Jose Manuel Gimenez-Guzman
1
, German Lopez-Civera
1
, Ivan Marsa-Maestre
1
and David Orden
2
1
Computer Engineering Department, University of Alcala, Alcala de Henares, Madrid, Spain
2
Physics and Mathematics Department, University of Alcala, Alcala de Henares, Madrid, Spain
Keywords: Network Resilience, Critical Infrastructures, Graph Modelling, Network Reshaping.
Abstract: Nowadays society is more and more dependent on critical infrastructures. Critical network infrastructures
(CNI) are communication networks whose disruption can create a severe impact on other systems including
critical infrastructures. In this work, we propose TOWER, a framework for the provision of adequate
strategies to optimize service provision and system resilience in CNIs. The goal of TOWER is being able to
compute new network topologies for CNIs under the event of malicious attacks. For doing this, TOWER
takes into account a risk analysis of the CNI, the results from a cyber-physical IDS and a multilayer model
of the network, for taking into account all the existing dependences. TOWER analyses the network structure
in order to determine the best strategy for obtaining a network topology, taking into account the existing
dependences and the potential conflicting interests when not all requirements can be met. Finally, we
present some lines for further development of TOWER.
1 INTRODUCTION
Since mid-90s, multiple events caused by natural
disasters or by humans have brought to the fore the
high level of dependency that our society has on
what are known as critical infrastructures. If one of
these infrastructures is compromised, it will have a
serious impact in our life and disrupt the normal
society order as a whole. Therefore, the protection of
critical infrastructures is a top priority in the agendas
of our governments and critical service operators.
Critical infrastructures are characterized by a
high level of interconnection. Many physical, logical
or virtual dependencies are not revealed until a crisis
arises. This high level of interdependencies may lead
to cascading failures. At the same time, small
disruptions can be enough to unleash dramatic
consequences in high complexity systems.
Critical network infrastructures (CNI) are one
subset of this kind of systems. By CNI, we mean
those communication network infrastructures whose
disruption, intentional or accidental, has a high
impact because of either its massive, even global,
deployment (e.g., Internet, cell provider networks)
or its supporting role for other critical infrastructures
(e.g., the network for the internal communications of
a control system in a nuclear facility, or the
communication network for the electrical grid).
When addressing the challenge of offering
robustness to CNI, the concept of network resilience
emerges. We can define network resilience (Smith,
2011) as the ability of a network to defend itself and
keep an acceptable service level in the presence of
challenges such as malicious attacks, hardware
failures, human mistakes (hardware or software
misconfigurations) and large-scale natural disasters
that may threat its normal operation.
In this work, we propose TOWER, a framework
for the provision of adequate strategies to optimize
service provision and system resilience in CNIs. For
achieving this goal, TOWER will use as inputs a risk
analysis and alert reports coming from different
sources and will produce a proposal of a new
network topology. This network topology should be
able to tackle the malicious attack by isolating it and
minimizing its impact in the whole network as the
consequence of the potential cascading failure
induced as a result of the attack. The new topology
will be oriented to offer a robust answer to the
threats that may impact the services at CNI or the
network itself. Therefore, it has to take into account
multiple and heterogeneous input data. We have
identified three main sources: threat analysis, a
Hoz, E., Gimenez-Guzman, J., Lopez-Civera, G., Marsa-Maestre, I. and Orden, D.
TOWER: Topology Optimization for netWork Enhanced Resilience.
DOI: 10.5220/0006017101210128
In Proceedings of the 13th International Joint Conference on e-Business and Telecommunications (ICETE 2016) - Volume 1: DCNET, pages 121-128
ISBN: 978-989-758-196-0
Copyright
c
2016 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
121
multi-layer model of the network and the results
from a cyber-physical intrusion detection system. It
is important to note that, although malicious attacks
are not the only challenge to cope with in CNIs, this
type of attacks use to be the most harmful ones as
they are focused on the blockage of the network-
provided network services. In fact, there are studies
(Albert, 2000) that claim that network resilience
obeys to different structural network properties in
case of malicious attacks, human errors or natural
disasters, due to the fact that these two last ones
have a more random and distributed nature, while
the first ones (malicious attacks) are usually focused
on specific nodes that play an essential role in the
network.
The rest of the paper is organized as follows.
Section 2 presents a review of the state of the art.
Section 3 introduces multilayer networks. In Section
4, the main components and challenges of TOWER
are described. Section 5 discusses network resilience
for CNI. Section 6 presents the main concepts on
network topology adaptation that area applied in
TOWER. Finally, some conclusions and future work
are presented.
2 RELATED WORK
In the last decade, there has been a growing interest
on maximizing the resilience of CNIs. Regarding the
Internet, its vulnerability has been widely
acknowledged (DHS, 2009) (Lin, 2007), and it has
been shown in different global scale incidents, both
accidental (Zmijewski, 2009) and as a result of
malicious activities (Goldberg, 2014). Also, the role
of communication networks in critical
infrastructures has been studied, as there are
interdependences between both (Rinaldi, 2001)
(DHS, 2009). Electrical grids are one of the clearest
examples of this. On the one hand, the
communication networks rely on the power provided
by the electrical grid to work. On the other hand, and
increasingly, the electrical grid relies on the
communication networks for its proper performance,
as it is based on SCADA systems for its
management.
There are some works in different areas that can
be related to the study of resilience in complex
networks. First, mobile ad-hoc networks and
vehicular ad-hoc networks (MANET and VANET).
In these networks, nodes are constantly moving, and,
accordingly, the connectivity must evolve to deal
with this. For these scenarios, it is critical to
guarantee service continuity under this dynamicity
(Landmark, 2015) (Su, 2015) (Dietzel, 2016).
Second, delay tolerant networks (Fan, 2015), where
service interruption after power failures, attacks or
node dispersion are taken into account, has attracted
much interest and its study has been promoted by
DARPA. And last, wireless sensor networks, where
it is necessary to create resilient network topologies
able to provide connectivity even after some sensors
stop contributing as a result of a battery outage
(Younis, 2014) (Yao, 2015).
With this background, there are some works that
face resilience in critical infrastructures by
modelling them as complex networks. The closest
works to our proposal are (Shao, 2015) and
(Berezin, 2015). These works study the robustness
of complex networks under attacks targeted to the
most sensitive points of the network and are more
theoretical than the ones that we propose. While they
deal with generic complex networks, we want to
include more realistic network models. Also, we
propose to use a multi-layer network model, to
capture the features of real-world networks. Finally,
we employ optimization techniques based on
negotiation to keep network resilience in a
distributed and self-organized way.
As research on complex networks has evolved, it
has been clearer the need of going further than
monolayer graph modelling and exploring more
realistic and complex models. Multiplex or multi-
relational networks connect nodes using links that
can express different kinds of relationships (Yagan,
2012). Multilevel networks and meta-networks
enable also hierarchical structures and node and
links of different types (Carley, 2007). Recently,
(Kivelä, 2014) has presented a unified modelling for
multi-layer networks that include these concepts in a
unified manner that takes advantage of the different
mathematical tools available in the state of the art.
In the last years, there have been significant
advances on complex system optimization in
domains that could be modelled as CNIs.
Techniques potentially suited for these domains
include auctions, optimization techniques, and
negotiation protocols. Combinatorial auctions (Xia,
2005) (Sandholm, 2015) can enable large-scale
collective decision-making in nonlinear domains,
but only of limited type (i.e. negotiations consisting
solely of resource/item allocation decisions). Multi-
attribute auctions (Pham, 2015) are also aimed at a
fundamentally limited problem –purchase
negotiations– and require full revelation of
preference information. Constraint-based and other
optimization tools (Chechetka, 2006) (Davin, 2005)
offer good solutions with interdependent issues, but
DCCI 2016 - SPECIAL SESSION ON DATA COMMUNICATION FOR CRITICAL INFRASTRUCTURES
122
have not been applied to contexts with self-
interested parties, thereby ignoring strategic issues
derived from participant’s selfish behaviour. In
particular, the vast majority of these approaches
assume that agreements will be honoured (e.g. after
an auction, the winner will pay the agreed amount to
the auctioneer). In many CNIs domains, however,
this “agreement honouring” assumption cannot be
made (e.g. you can suggest network reshapes to
different network domains, but you cannot force
them to accept them).
The distributed and adaptive nature of CNIs,
along with the need to reach a consensus between
conflictive individual goals to benefit a social goal,
suggests the use of negotiation techniques. However,
most negotiation research has focused on problems
with one issue (typically price) or a few independent
issues (Ren, 2013) and are demonstrably sub-
optimal for negotiations with multiple
interdependent issues (Klein, 2003). A number of
research efforts (Marsa-Maestre, 2009) (Li, 2009)
have attempted to address this challenge, facing
serious limitations in terms of outcome optimality,
strategic stability and scalability. These three
performance indicators are key enablers for the
success of optimization systems in large real-world
CNIs infrastructures, due to their continuous
increase in network size, structural complexity and
dynamicity (Strogatz, 2001). If we want to keep
relying on these exponentially growing
infrastructures for our development as a society, new
distributed optimization mechanisms should be
devised for their management.
3 MULTILAYER NETWORKS
To achieve our purpose, and in order to take
effectively into account the underlying complexity
of CNI, the first step consists in modeling them by
means of multilayer graphs. Graphs have shown to
be a useful tool to model complex systems, as vertex
or nodes can be employed to represent the functional
elements of the system, while edges show the
relationships between them. However, CNIs are
usually complex networks that include intricate
relations between their elements, so a simple graph
can hardly capture these relations. For that reason,
we use multilayer graphs to account for those
relations in a more intuitive and powerful manner.
As described in (Boccaletti, 2006) (Kivelä, 2014),
multilayer networks incorporate multiple levels of
connectivity and provide a natural framework to
describe systems connected through different types
of connections: each channel (relationship, activity
or category) is represented by a layer and the same
node or entity may have different kinds of
interactions (different set of neighbors in each
layer).
As shown in Fig. 1, and for example, a
multilayer graph can be composed by different
layers that express the physical location of the
systems, the communication paths existing between
them and also the functional and management
dependencies. Apart from the more technical
aspects, characteristics like the social or corporate
dependencies present at any kind of critical
infrastructure could be also easily modeled by means
of new layers with their respective inter- and intra-
layer dependencies.
4 TOPOLOGY OPTIMIZATION
FOR NETWORK ENHANCED
RESILIENCE (TOWER)
The main objective of TOWER is to provide
adequate strategies to optimize service provision and
system resilience by taking as input the risk analysis
and alert reports and proposing a new network
topology. This topology will be oriented to offer a
robust answer to the threats that may impact the
services at the critical infrastructure or the network
that supports them. Therefore, it has to take into
account multiple and heterogeneous input data that
can be summarized into three main blocks:
1. Threat analysis: Before the TOWER module can
be deployed a threat analysis and a contingency
plan must be developed. These processes aim to
create an initial status report for the system that
provides enough information to TOWER to start
working. This analysis must include a detailed
analysis of the value, implicit risk and attack
resilience of all the assets that must be protected.
2. A multilayer network model: As described in
section 3, a multilayer network model is needed
to represent the complex system underlying to
the CNI. This transformation will enable to
express hidden dependencies between assets that
would not be covered otherwise. At the same
time, this kind of model provides a wider
perspective over the consequences of an attack as
it can show how the malicious behavior or its
consequences would spread throughout the
network (e.g. a cascading failure). This data
representation could be useful for an intrusion
prevention system to take better decisions about
TOWER: Topology Optimization for netWork Enhanced Resilience
123
Figure 1: Example of three-layer graph.
how to isolate and protect the CNI in the event of
an attack. All this process will be performed
offline, that is, before the system is fully
deployed. The model can be updated lately if
new assets are introduced in the network but a
first model must be created before the online
process begins.
3. IDS alerts: These alerts contain information
about the type of attack, its location and the
assets involved. Based on this data, TOWER will
be able to evaluate the overall status of the
network and respond accordingly. Therefore, a
consistent and meaningful communication
channel must be developed to link the cyber-
physical IDS and the recovery module.
After analyzing all this data, TOWER will
compute a network topology intended to isolate the
attack and minimize its impact. This topology will
consider the required paths that must be always
linked and make use of backup links or modified
routing paths to create new ways to reach each point
of the network. TOWER will offer a ranked list of
different topologies, ordered by a score metric. This
ranked list will provide different options for the rest
of the recovery module. Therefore, if a particular
topology cannot be deployed in the network, other
options that will also reduce the impact of an attack
should be available.
The purpose of the topology computation engine
inside TOWER is to prevent not only the direct
consequences of a loss of connectivity or a system
failure (produced by an attacker or not), but also to
avoid the cascading effects that it may produce.
Consequently, TOWER will analyze the
interdependencies that may emerge from the
network elements and the risk analysis performed
over them. All this information has to be translated
into graph elements and this is when multilayer
network modelling becomes useful for modelling all
this complexity into multiple graphs for the
representation of the existing different kinds of
dependencies between network nodes. This
relationship between the multiple layers of the
network model provides an easier way to create a
data structure for TOWER to find hidden
consequences that cannot be directly traced to a
failure in a network node.
The kind of network topology that TOWER will
compute will be highly dependent on the type of
attack detected. Different kinds of attacks will imply
a particular type of topology modification, not only
due to its location but also because of its inner
characteristics. This implies that TOWER must take
into account how the threat could spread through the
network. For example, if the infrastructure is
suffering a DDoS attack, TOWER may modify the
network by cutting the paths under attack and
reroute traffic through other alternative paths. On the
other side, if a server gets compromised, TOWER
will isolate the node and limit the communications
made by all the computers that could have been
reached from that server. Therefore, pivoting attacks
that employ accessible servers to reach the internal
network, that could not be attacked otherwise, are
also covered by the TOWER approach.
One of the main challenges TOWER will face is
the situation where updating the topology will
protect part of the network but may introduce new
threats or attack surface. Unfortunately, this is the
most common case as computer networks tend to be
highly coupled with some central nodes that manage
a large amount of traffic. The existence of these
nodes may create central points of failure which can
originate a cascading effect if they are compromised.
The purpose of the initial network topology
deployment and all modifications made to it through
TOWER is to minimize these interdependencies and
maintain the network in a resilient state.
DCCI 2016 - SPECIAL SESSION ON DATA COMMUNICATION FOR CRITICAL INFRASTRUCTURES
124
To achieve this purpose, different approaches are
being studied, but the main efforts are focused on
employing graph theory techniques to analyze the
status of the network and compute how to react
against an attack or a system malfunction. Apart
from this data representation, the progress made in
TOWER module is focused on finding relevant
graph metrics that helps to characterize the network
they model. These metrics are different than the
characteristics modelled through the use of multiple
layer networks. The layers are employed to show the
dependencies present in the network at different
levels, while these metrics describe the behavior of
the dependencies. For example, if a layer of the
network model is composed by the existing
communications flows between the nodes, a metric,
e.g. node degree, can help identifying the nodes that
are the origin or destination of the majority of the
traffic. One of the long-term purposes of our work is
to study how the network structural properties of a
problem influence the performance of optimization
and the choice of the negotiation approaches best
suited to it. To this end, we have selected a number
of graph metrics from the literature, being the
following:
1. Graph order: the number of nodes in the graph.
2. Graph diameter: the longest distance (number
of traversed edges) between any pair of nodes in
the graph (Newman, 2010).
3. Wiener index: gives a measure of graph
complexity from the distances in the graph. It is
computed as,
where |N| are the number of vertices of the graph
and d(n
i
, n
j
) is the shortest distance between
nodes (Wiener, 1947).
4. Graph density: the ratio between the number of
edges in the graph and the maximum possible
number of edges (that is, for a fully-connected
graph).
5. Clustering coefficient: a measure of the degree
to which nodes in a graph tend to cluster
together. The cluster coefficient of a graph is
computed as the average of the local clustering
coefficient of its nodes, which is the ratio
between the number of links between a node’s
neighbors and the maximum possible number of
links between them (that is, if they were fully
connected).
Another relevant type of metric are the centrality
metrics. These metrics allow to identify the most
important nodes inside a network, which can be
critical to know the most vulnerable points at each
ground station. There are a number of centrality
metrics in the literature, such as degree centrality
(Freeman, 1979), hub and authority centrality
(Chakrabarti, 1999), PageRank centrality (Brin,
1998), Katz centrality (Katz, 1953) or betweenness
centrality (Freeman, 1979) (Koschützki, 2005). In
particular, betweenness centrality of a node is the
ratio of shortest paths in the graph that traverse the
node.
In particular, due to the need for TOWER to
respond in a timely fashion to incidents, we are
testing efficient methods to approximate these
metrics, such as the ones in (Chierichetti, 2015)
(Ohara, 2014) (Kimura, 2016).
We are studying the relevance of these and other
metrics on the resilience of different kinds of
networks, and we expect to be able to derive novel
metrics and methods to detect the cascading effects
of an attack and be able to react proactively to
network intrusions. This way the system will not
only react when an attack is made but also it will
prevent the actual attacks strengthening the security
measures or creating new backup paths if needed.
Finally, another set of techniques that allow to
split a graph into multiple pieces are being studied
during the design of TOWER module. These
techniques allow to distribute the computational cost
of running complex algorithms against big scale
graphs. Clustering algorithms may be employed to
detect sets of vertex that are more tied and therefore,
should not be split when dividing the network.
Particular cautions are needed here, since this kind
of division may hide dependencies that could be
relevant when deciding how to protect the network.
Consequently, distributed computation algorithms
will be evaluated against different use cases to
explore how they impact the computation of new
topologies compared to centralized algorithms.
5 NETWORK RESILIENCE
One of the main objectives of TOWER is being able
to improve network resilience for CNI. For this to
succeed, we need a way to identify and measure
resilience in a network. In this section we will point
out the most important points that should be taken
into account.
To characterize a resilient network three levels of
functional dependencies can be identified:
1. System properties: these entities represent the
functions of the system that have to be protected
and must be working permanently.
TOWER: Topology Optimization for netWork Enhanced Resilience
125
2. System attributes: they show the characteristics
that tells us if a system property is robust enough
to resist an attack. Some examples are the
following:
1. Diversity: It tells how much different kinds
of elements are used in the network. The
more diverse the system is, the more
complex is to attack it as the adversary has
to find vulnerabilities on multiple kind of
systems. This attribute has to reach a point
where the redundancy of the system is not
compromised since redundant systems tend
to be built using the same infrastructure
employed for the replicated element.
2. Fault tolerance: Is the system able to
guarantee correct operation if the presence
of a failure? This attribute is applied to
distributed system where if a node fails, the
remaining nodes must be able to continue
working.
3. Deceptiveness: It measures how much
information can be manipulated about the
infrastructure to fool the attacker.
4. Velocity/Fluidity: this concept expresses the
level of uncertainty offered to an attacker
when he tries to attack the infrastructure.
For example, the IP address of a system can
change over time or there can be different
ways to obtain the same information from
different locations, making it harder to
identify patterns or security measures.
5. Self-stability: No matter how the system is
compromised, the reaction performed has to
lead to a stable state where the normal
operation of the infrastructure is
guaranteed.
3. Metrics: All the attributes described above
cannot be measured directly (i.e. it does not exist
a fault tolerance indicator on our systems).
Therefore, specific observations must be
performed on the infrastructure to determine the
level of fulfillment of all these attributes. For
example, to measure the diversity of the network,
we may need a system inventory where the
number of different operating systems and
versions are described.
The metrics have to be carefully chosen as they
generate the vision the system will have of the actual
status of the network. They can also be modeled as
part of the graph representation, showing the
dependencies that exist between different attributes
and metrics and the function they cover in the
infrastructure.
6 ADAPTATION OF NETWORK
TOPOLOGY
Up to this point, we have presented the main
underlying concepts and technologies to TOWER
for identifying and modelling the dependencies of a
CNI. Using this information, and taking into account
the threat model, TOWER will produce a proposal
of new network topologies. These network
topologies will be oriented to offer a robust answer
to the threats that may impact the services at CNI or
the network itself. In practice, this means that
TOWER should be able to lead the CNI from a
potentially harmful state, e.g. the preliminary steps
of an attack or early signs of a system being
compromised, to a new good state, ensuring or
minimizing the impact for the whole network. It is
when evaluating this impact, that the previous
models reveal its importance: we ensure that no
important dependence is missing. It can also be the
case that it is not possible to ensure the safe
operation of the whole network. For agreeing on the
best outcomes for all the interesting parties,
negotiation techniques, such as the one pointed out
in section 2, can be used.
We also need to know how to reach a specific
state. To do so, we need to model the viable
transitions that exist between states, describing
which methods or procedures have to be employed
to get to that state. These transitions can also be
modeled as a graph where the nodes represent the
states the system can be and the edges show how to
go from one state to another. To decide which path
to follow inside this transition graph a decision
system has to be developed. This system has to
come up with the decision of which the best way is
for protecting the network and the actions that have
to be taken to reach a stable state that can guarantee
the normal operation of the system.
To do so, first we need to perform an exhaustive
enumeration of the states the system could be at.
Since the computational cost of this enumeration
process can exceed the time requirements of the
platform, clustering techniques may be applied
during offline pre-computation to reduce the number
of nodes in the graph. Apart from the normal or
working states of the system, we need also to model
the states where the system is compromised.
Therefore, we need to create also “bad” states that
account for an attacker getting inside the network,
with different degrees of success. These states must
be based on the risk analysis used as input to the
TOWER module. Using these compromised states,
the system would be able to predict a malicious
DCCI 2016 - SPECIAL SESSION ON DATA COMMUNICATION FOR CRITICAL INFRASTRUCTURES
126
behavior based on the preliminary states that can
lead to the actual attack. For example, typically
reconnaissance techniques are performed before
launching an attack, thus if a port scan alert is
received it can be mapped to a prior-attack state that
can trigger preemptive security measures. Moreover,
post-attack states will be also included in the
enumeration as they can show how an attack could
have cascading effects on multiple elements of the
network. This analysis will help to choose the
appropriate reaction method and also to determine
the coverage of the response action.
Once these states are modeled, the next step
consists on determining the current status of the
system. In order to do so, a method that maps the
information collected from the alerts to one of the
states generated before has to be developed. To
achieve this purpose, clustering techniques can also
be employed, as they can compute the nearest or
more similar state.
Once we know the condition of the system, we
have translated it to an actual state node in our graph
and we know the transitions we can employ to travel
the graph from that node; we are ready to continue.
The final step is to compute the path we have to
follow to a better state. Distributed agent-based
decision systems are being studied to solve this
problem. This kind of systems allow to get a
response even in the case where a particular action
may only benefit part of the infrastructure. This way
the overall operation of the infrastructure is
guaranteed even if part of it has to be partially
harmed during the transition.
Game theory (Myerson, 2013) techniques are
also being applied to model the behavior of an
adversarial attack and predict how he can breach
into the system, taking additional security measures
at weak points. Therefore, we can also model the
steps an attacker can take to compromise the
systems and identify if the infrastructure actual state
is leading to one of them, preventing further
consequences. These techniques help also to reduce
the amount of information that an attacker can
obtain from the system, as this information is the
input for the algorithm.
7 CONCLUSIONS
This paper presents TOWER (Topology
Optimization for netWork Enhanced Resilience), a
decision support system intended to provide
topology alternatives in Critical Network
Infrastructures (CNI), so that the impact of an attack
on data communication is mitigated. The system
operates from a multilayer network model of the
asset inventory of the CNI, which specifies the
dependencies between nodes and the associated
risks, and is able to react to network incidents
(modeled as perturbations in the network) by
providing a ranked set of alternative topologies
maximizing network resiliency. We discuss the most
important design principles behind TOWER and
finally, point out some research lines and ideas that
will guide future TOWER development.
ACKNOWLEDGEMENTS
This work was partially supported by SCOUT, a
research project supported by the European
Commission under its 7th Framework Program
(contract-no. 607019). The views and conclusions
contained herein are those of the authors and should
not be interpreted as necessarily representing the
official policies or endorsements, either expressed or
implied, of the SCOUT project or the European
Commission.
REFERENCES
Albert, R. Error and attack tolerance of complex
networks. Nature, 406(6794), 378-382.
Berezin, Y. “Localized attacks on spatially embedded
networks with dependencies,” Scientific reports 5.
Boccaletti, S., Latora, V., Moreno, Y., Chavez, M. and
Hwang, D. U. Complex networks: Structure and
dynamics. Physics reports, 424(4), 175-308. (2006).
Brin, S. and Page, L.: The anatomy of a large-scale
hypertextual web search engine. Computer Networks
and ISDN Systems 30, 107–117 (1998).
Carley, K.M. "Toward an interoperable dynamic network
analysis toolkit." Decis. Support Syst., 43, 1324–1347.
(2007).
Chakrabarti, S., Dom, B., Kumar, R., Raghavan, P.,
Rajagopalan, S., Tomkins, A., Gibson, and D.,
Kleinberg, J.: Mining the web’s link structure. IEEE
Computer 32, 60–67 (1999).
Chechetka, A. and K. Sycara, No-commitment branch and
bound search for distributed constraint optimization.
AAMAS International Conference. Hakodate, Japan.
(2006).
Chierichetti, F., Epasto, A., Kumar, R., Lattanzi, and S.,
Mirrokni, V.: Efficient algorithms for public-private
social networks. In: Proceedings of the 21st
ACMSIGKDD International Conference on
Knowledge Discovery and Data Mining (KDD’15).
pp. 139–148 (2015).
TOWER: Topology Optimization for netWork Enhanced Resilience
127
Davin, J. and Modi, P.J. Impact of problem centralization
in distributed constraint optimization algorithms.
AAMAS International Conference. (2005).
DHS, A Roadmap for Cybersecurity Research, Technical
Report, Department of Homeland Security (DHS).
(2009).
Davin, K. Impact of problem centralization in distributed
constraint optimization algorithms. In Proceedings of
The 4th International Conference on Autonomous
Agents and Multiagent Systems AAMAS. (2005).
Dietzel, S. A resilient in-network aggregation mechanism
for VANETs based on dissemination redundancy, Ad
Hoc Networks 37, 101-109. (2016).
Fan, R. RobustGeo: A Disruption-Tolerant Geo-Routing
Protocol, 24th International Conference on Computer
Communication and Networks (ICCCN). (2015).
Freeman, L.: Centrality in social networks: Conceptual
clarification. Social Networks 1, 215–239 (1979).
Goldberg, S. Why is it taking so long to secure internet
routing? Communications of the ACM, 57(10), 56-63.
(2014).
Katz, L. A New Status Index Derived from Sociometric
Index. Psychometrika, 39-43. (1953).
Kimura, M., Saito, K., Ohara, K., and Motoda, H.:
Speeding-up node influence computation for huge
social networks. International Journal of Data Science
and Analytics 1, 1–14 (2016).
Kivelä, M. Multilayer networks, Journal of Complex
Networks, 2(3), 203-271. (2014).
Klein, M., P. Faratin, H. Sayama and Y. Bar-Yam.
Negotiating Complex Contracts. Group Decision and
Negotiation 12(2), 111 - 125. (2003).
Koschützki, D., Lehmann, K.A, Peeters, L., Richter, S.
Tenfelde-Podehl, D. and Zlotowski, O.. Centrality
indices. Network analysis. Lecture Notes in Computer
Science. 3418:16–61, (2005).
Landmark, L. Resilient internetwork routing over
heterogeneous mobile military networks, IEEE
Military Communications Conference (MILCOM),
388-394. (2015).
Li, M., Q. B. Vo and R. Kowalczyk Searching for fair
joint gains in agent-based negotiation. Autonomous
Agents and Multi-agent Systems (AAMAS-09). (2009).
Lin, H. S. and Goodman S. E. Toward a Safer and More
Secure Cyberspace, National Academies Press (2007).
Marsa-Maestre, I., Lopez-Carmona, M. A., Velasco, J. R.,
and de la Hoz, E. Effective bidding and deal
identification for negotiations in highly nonlinear
scenarios. In Proceedings of The 8th International
Conference on Autonomous Agents and Multiagent
Systems 2, 1057-1064. (2009).
Myerson, Roger B. Game theory. Harvard university
press, (2013).
Newman, M. Networks: an introduction. Oxford
University Press, (2010).
Ohara, K., Saito, K., Kimura, and M., Motoda, H.:
Resampling-based framework for estimating node
centrality of large social network. In: Proceedings of
the 17th International Conference on Discovery
Science (DS’14). pp. 228–239. LNAI 8777 (2014).
Pham, L., Teich, J., Wallenius, H., and Wallenius, J.
Multi-attribute online reverse auctions: Recent
research trends. European Journal of Operational
Research, 242(1), 1-9. (2015).
Ren, F., and Zhang, M. Bilateral single-issue negotiation
model considering nonlinear utility and time
constraint. Decision Support Systems. 60, 29-38.
(2013).
Rinaldi, S.M. Identifying, understanding and analyzing
critical infrastructures interdependencies, IEEE
Control Systems Magazine, 21(6), 11-25. (2001).
Sandholm, T., and Likhodedov, A. Automated design of
revenue-maximizing combinatorial auctions.
Operations Research, 63(5), 1000-1025. (2015).
Shao, S. Percolation of localized attack on complex
networks, New Journal of Physics, 17, 023049.
(2015).
Smith, P. Network resilience: a systematic approach, IEEE
Communications Magazine, 49(7), 88-97. (2011).
Strogatz, S. H. Exploring complex networks. Nature,
410(6825), 268-276. (2001).
Su, M.Y. A resilient routing approach for Mobile Ad Hoc
Networks, International Conference on High
Performance Computing & Simulation (HPCS).
(2015).
Wiener, H. Structural determination of paraffin boiling
points. Journal of the American Chemical Society,
69(1):17–20, (1947).
Xia, M., Stallaert, J. and A. B. Whinston. Solving the
combinatorial double auction problem. European
Journal of Operational Research 164(1), 239-251.
(2005).
Yağan, O. and Gligor, V. Analysis of complex contagions
in random multiplex networks. Physical Review E,
86(3), 036103. (2012).
Yao,Y. EDAL: An Energy-Efficient, Delay-Aware, and
Lifetime-Balancing Data Collection Protocol for
Heterogeneous Wireless Sensor Networks, in
IEEE/ACM Transactions on Networking, 23(3), 810-
823. (2015).
Younis, M. Topology management techniques for
tolerating node failures in wireless sensor networks: A
survey, Computer Networks, 58, 254-283. (2014).
Zmijewski, Reckless Driving on the Internet,
http://research.dyn.com/2009/02/the-flap-heard-around
-the-world/ .(2009).
DCCI 2016 - SPECIAL SESSION ON DATA COMMUNICATION FOR CRITICAL INFRASTRUCTURES
128