Hindawi Publishing Corporation

International Journal of Distributed Sensor Networks

Volume 2015, Article ID 139735, 17 pages
http://dx.doi.org/10.1155/2015/139735

Research Article
Architecting the IoT Paradigm: A Middleware for
Autonomous Distributed Sensor Networks

George Eleftherakis,1 Dimitrios Pappas,2 Thomas Lagkas,1

Konstantinos Rousis,2 and Ognen Paunovski2
1
Computer Science Department, The University of Sheffield International Faculty, CITY College, 54622 Thessaloniki, Greece
2
South-East European Research Centre (SSERC), The University of Sheffield International Faculty, CITY College,
54622 Thessaloniki, Greece

Correspondence should be addressed to Thomas Lagkas; [email protected]

Received 31 May 2015; Revised 6 November 2015; Accepted 17 November 2015

Academic Editor: Davide Brunelli

Copyright © 2015 George Eleftherakis et al. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly
cited.

Actualizing Internet of Things undoubtedly constitutes a major challenge of modern computing and is a promising next step
in realizing the unification of all seamlessly interacting entities, either human users or participating machines, under a shared,
coherent architecture. While it has now become common belief that the related solutions should be based on compatible network
infrastructure employing widely accepted communication schemes, the specifics of the intermediate system that would act as global
interface for all involved “things” are yet to be determined. A rising trend to define such machine-based entities is through cyber-
physical systems, in terms of collaborating elements with physical input and output. Certainly, sensor networks constitute the
most representative realization of such systems. Taking these issues and opportunities under consideration, this work proposes a
bioinspired distributed architecture for an Internet of Things that exhibits self-organization properties to enable efficient interaction
between entities modeled as cyber-physical systems, mainly focusing on sensor networks. Furthermore, a middleware has been
implemented according to the proposed architecture, which serves the role of the backbone of this network as a multiagent and
autonomous distributed system. The evaluation results demonstrate the self-optimization properties of the introduced scheme and
indicate global network convergence.

1. Introduction Modern sensor networks require a flexibility that will

allow them to operate in a decentralized manner and often
The extensive use of the Internet and its phenomenal pen-
without requiring a well-defined infrastructure. There is
etration worldwide, along with the massive trends of social
an increased demand for solutions that work in extremely
networking mobile computing, have created a demand
for unprecedentedly complex applications. Heterogeneous dynamic environments and enable creation of an infrastruc-
platforms and services often require seamless integration ture of a sensor network in a dynamic and autonomous way,
with each other and intuitive sharing of information. Such especially in situations that well-defined and well-designed
demands were undoubtedly a decisive factor on the success infrastructures do not exist or are not preferable. Studies, as
of decentralized architectures during the last decades. The well as simple observation of natural systems, have shown
need to connect everything on the Internet under a com- how self-∗ properties can lead to seamless ways of forming
mon vision of the Internet of Things (IoT), a network of highly robust and dynamic systems. This dictated a research
uniquely identifiable and interacting objects, the introduc- trend over the last decades towards nature-inspired solutions
tion of Cyber-Physical Systems (CPS), and the ambition for a in artificial systems and especially artificial distributed net-
“smart” planet, necessitates architectures that enable scalable, works.
autonomous, and robust solutions which should be capable of This work proposes an architecture for building a self-
exhibiting extremely dynamic behaviors. organizing overlay network of CPS, with a particular focus
2 International Journal of Distributed Sensor Networks

on sensor networks. This architecture is heavily inspired Thus, an asynchronous mode of communication is deemed
by nature, and specifically the fact that innovative design more appropriate for such scenarios.
of individual nodes (microscopic level) can lead to the In a network with the size and interconnection charac-
emergence of desired global properties (macroscopic level). It teristics of an IoT, it is expected that an immense amount
aims to offer an IoT solution of interacting CPS in the form of of information is being generated which should be collected
a middleware that facilitates interaction and interconnection only when this is requested or when a real-world object is
of things in a distributed manner, providing scalability, self- capable of providing that data [7, 8]. This implies that there is
adaptation, and self-organization. The applicability of the no need for processing unless an event is triggered, although
proposed architecture is validated with a realistic design and other solutions could be implemented such as utilizing agents
an implementation solution that could support real-world and agent-based messaging. Equally important to such imple-
scenarios. mentations is how the messages are processed and prioritized;
The following section presents the state of the art on if they are processed in a first-come, first-served manner, that
architectural paradigms of the IoT, different approaches on would simulate an undesired synchronous communication
CPS, and autonomous systems. In Section 3 the proposed model. Apart from an event-driven approach with a properly
architecture is described and justified in different levels structured messaging protocol, the agent-based program-
of abstraction. Section 4 details the design of the pro- ming paradigm has a lot to offer to the IoT. Specifically,
posed architecture and offers a thorough description of the the capacity for intelligence that agents inherently possess
proposed implementation solution. Finally, the preliminary enables the opportunity for achieving autonomy and self-
results and the evaluation strategy are presented in Section 5, management for complex systems such as those that serve as
after which we conclude the paper and present future plans the backbone of an IoT implementation [9–11].
for the middleware. Finding solutions to IoT issues in the agent-based pro-
gramming paradigm hints at the possibility of opportunities
to be found in other popular programming paradigms. In
2. Background and Related Work
particular, the most relevant one may be that of Service-
In this section, we discuss the current status of those Oriented Architecture (SOA). A SOA-compliant middleware
issues that are taken into account for the formulation of would be eligible to provide any valuable information or
the proposed solution. Specifically, the important architec- functionality possessed by the real-world objects to end-users
tural aspects of IoT are provided, key characteristics and in the form of services. The most important property, in terms
approaches for CPS and autonomous systems are presented, of value to the IoT, that can be found in software complying
and we also elaborate on concerns and solutions for modern with a SOA is, unarguably, interoperability. Interconnect-
sensor-based architectures. ing heterogeneous devices and allowing for interoperability
through a SOA are possible through the use of standardized
protocols and semantics [7, 9, 12]. Nonetheless, developing a
2.1. Architectural Issues in the Internet of Things. The vision of SOA for a middleware which aims to realize an IoT is posing
the IoT represents an assortment of interconnected, interact- various research challenges including heterogeneity, mobility,
ing objects, which could be an overlay network that utilizes scalability, adaptiveness, adaptability, awareness, security, and
the Internet itself but is not limited to this solution alone. privacy [13].
Although identifying each network peer can be challenging, it One of the important architectural concerns pertains
is not the primary concern when realizing the IoT. According to the problem of discovery strategies for services in a
to Ashton, the most crucial aspect is the interaction among SOA-compliant middleware. The two fundamental aspects
humans and objects or among objects alone [1]. As the type of service discovery have been shown to correspond to the
of interactions is dependent on the type of network and the following: (a) how services are organized within a system and
needs it was designed to fulfill, aiming for a single architecture (b) how they are located within a system. The former is shown
that fits every scenario is practically impossible and it leads to to be best achieved by storing information about services in
a multitude of standardization issues [2–4]. the form of a localized registry (i.e., a repository on each
The types of interactions within a network may define node), which describes what services can be provided by this
the functionality of each peer, as well as the purpose of the node [13]. Notwithstanding, the crucial issue of locating them
network itself, but the virtual representation of these objects inside the network still remains; to this end, the state of the art
is of utmost importance in realizing the vision of an IoT. dictates two particular approaches: (a) centralized repository
Such representations have an architectural impact on ADS of registries and (b) distributed discovery.
and they also affect the node representation in the network, Centralized approaches are most commonly adopted
as well as the manner in which information flowing between by the industry, including well-established solutions such
humans and devices is being represented [5]. as EPCglobal, Afilias, ID@URI, and the BRIDGE project
An important architectural differentiation is the approach [8]. A centralized implementation necessitates the service
taken towards communication, with the two general cate- consumer to connect to a remote registry, with an interaction
gories being synchronous and asynchronous communication. in the manner of the client-server paradigm and a user
In an IoT implementation, a synchronous communication querying a database. Centralized solutions are undoubtedly
would introduce a large overhead which could defeat its practical but there are several concerns that need to be taken
purpose of guaranteeing delivery in a timely fashion [6, 7]. under consideration. Specifically, the larger the repository
International Journal of Distributed Sensor Networks 3

becomes the more expensive (resource-wise) it is for a Architecturally, CPS are composed of varied subsystems,
single administrator to manage. At the same time, service or parts, each of which can be represented at an abstract level
availability becomes more problematic, since it constitutes a by a component, thereby proving the meaningful application
single point of failure and attack. of component-based software engineering (CBSE) on this
Distributed discovery, on the other hand, is much more particular area. The Dependable Emergent Ensembles of
diverse and a single best approach is lacking. Different Components (DEECo) is an example middleware that has
strategies have been developed to solve different problems evolved into a fully fledged solution for CPS design by
with various trade-offs. Biologically inspired [14], traditional improving interaction among subsystems with minor strain
P2P [8], DNS-based solutions [15], and Web-based solutions and effort on development, all the while being applicable
to a variety of scenarios [23]. Programming Temporally
[16] are few of the most popular approaches.
Integrated Distributed Embedded Systems (PTIDES) model
A recurring theme on most approaches for realizing an is another example, a programming model that succeeds in
IoT architecture is the use of Web services. Inherently, the simplifying and enhancing the development of distributed,
design of Web services has the power to provide unique real-time CPS by resorting heavily to software components
identification to the service providers and exceptional, if and model-based design [24]. To summarize, the literature
not unparalleled, ease of access to the service consumers signifies the capacity of SOA to solve most issues concerning
(i.e., the things in both cases). Their immense popularity the exchange of information among distinct and even hetero-
showcases how they are virtually everywhere in this era of geneous interconnected CPS systems.
cloud computing and attests to their power to contribute to All approaches presented so far require from the system
a different perspective at the vision of the IoT, that of the to operate within its life cycle to its fullest potential with
Web of Things (WoT). Once more, it is essential to stress the minimal to virtually no human intervention at all, which
difference between the IoT and the WoT: the former is about requirement actually matches the concept of autonomy in
interconnection and interaction of the devices and users in computing. It was initially introduced by IBM in 2001 in an
general (which could be the Internet itself or any other type attempt to integrate heterogeneous devices and environments
of network), whereas the latter focuses on achieving these [25]. As a means of evaluating whether a system was exhibit-
goals through the World Wide Web (WWW) and the Web ing autonomy or not, it was decided that the system should
services in particular (which is always an overlay above the possess several properties known as self-∗ : self-configuration,
Internet) [17]. self-healing, self-optimization, and self-protection [26, 27].
Accessing a Web service is fairly uncomplicated as long MAS are often found in the center of discussions about
as its URI is known, along with the type of the requests it can autonomy and particularly for autonomous ADS. This is
respond to. This holds true whether this is a RESTful service attributed to the fact that their interactions have been shown
(adhering to the Representational State Transfer architec- to lead to the emergence of beneficial properties in these
tural constraints) or a Web Services Description Language systems, such as robustness, scalability, and adaptability by
(WSDL) service utilizing the Simple Object Access Protocol using partial views, feedback, and self-evaluation functions
(SOAP), the two predominant approaches to the realization [28].
of Web services [18, 19]. Therefore, the virtual representation Services and a SOA are also capable of providing some
of any device or user comes in the form of either a service degree of autonomy to ADS. Services can handle the task
consumer or a service provider, bringing ever closer the of communication or information sharing, and focusing on
fields of IoT and SOA, along with the formidable advantages lightweight services with descriptions based on standards will
the latter has to offer. In fact, interoperability is achieved definitely contribute to enhancing the awareness aspect of
seamlessly through the Web services and SOA, although there autonomy [29, 30]. In addition to borrowing from the field
still exists the challenge of abstracting the functionality of the of SOA, principles fundamental to the CBSE can be utilized
things as services. for ADS towards autonomy, offering as much as they do to
CPS, if not more.
2.2. Cyber-Physical and Autonomous Distributed Systems. Additionally, architectures such as the Organic Grid,
With the advances in the field of wireless sensor networks where agents attempt to colonize resources they discover in
and the popularity they enjoy nowadays, not discounting a the ADS much like an array of biological organisms would
heavy investment in that field towards encompassing IoT for (humans included), achieve self-organization and hence a
the much-discussed benefits, Cyber-Physical Systems (CPS) satisfying degree of autonomy [31]. On the other end of the
and robotics or sensors have come to be almost synonymous spectrum, a system may exhibit emergent properties that
terms for several practitioners and researchers [20]. On the differ from those envisioned by its designers and that could
other hand, some claim that the term of CPS indicates prove to be harmful to the system. Thus, modeling such
the next step in the evolution of wireless sensor networks solutions requires careful and extensive evaluation of the final
[21]. Despite wireless sensor networks being the prevalent system [32].
CPS family, significant work has begun towards most other
families and generally most devices that have integrated with 2.3. Sensor Network Architectures. It is apparent that archi-
embedded systems [22]. The modern smart car, for example, tectural efforts on sensor networks face the same chal-
is one of the most frequent examples of a CPS. lenges of identification, distribution, and autonomy that were
4 International Journal of Distributed Sensor Networks

previously discussed. The emergence of unified entities with sensor hardware limitations and other significant practical
behavioral patterns arising from the individual participating constraints.
devices has inspired similar architectural ideas for sensor Typically, such implementations employ agent-based
networks. A representative example is BiSNET [33], a sensor middleware solutions to realize communications inside a
network architecture inspired by bee movement patterns, sensor network. One of the most known middleware frame-
which includes a middleware and multiple agents that behave works for wireless sensor networks is Agilla [37]. Agilla
according to biological principles. can be run on TinyOS in order to control mobile agents
From an architectural point of view, the most crucial and over a sensor network. The agents are programmed to act
well-researched part is the middleware as it establishes the towards the achievement of a common goal while their
roles and the connectivity aspects of the whole system. A movement is regulated by the middleware. Another similar
flexible, plug-and-play type of middleware has been proposed middleware for wireless sensor networks is SensorWare [38].
in [34], named Global Sensor Networks (GSN). It specializes Although SensorWare does not employ agents, it provides
in sensors and on how users can access their information. similar capabilities through the use of scripts. AgentScape
The GSN middleware utilizes a directory for interconnecting is an agent-based, multilayer, sensor network architecture
the sensors and users accessing the network, while it employs introduced in [39]. Agents are situated in predefined locations
the IEEE 1451 standard for drawing data from the sensors. and can communicate with external systems. Finally, Deluge
GSN nodes communicate directly with each other in a P2P [40] and Impala [41] are representative examples of platforms
fashion, without any intelligence behind the query processing which allow code distribution over wireless sensor networks.
[34]. Consequently, an implementation through the GSN However, they assume that some code is installed in advance
middleware will not lead to a system that can adapt to changes to all participating nodes as they are not agent-based.
to its expected operational environment. The employed data sampling technique can have a sig-
Another middleware with an interesting architectural nificant impact in the sensor network architecture as well.
approach on interconnecting different types of sensors is Efficiently scheduling the collection of information from
SenaaS, which utilizes the adapter software design pattern sensors is crucial for the effectiveness and the lifetime of
[6]. Examples of middleware that exploit the agent-based the network. A well-designed architecture facilitates the data
paradigm for realizing sensor network architectures include collection process, allowing regular updates while keeping
Flexeo, an intelligent wireless sensor network [35], and Cog- the usage of resources to a minimum. A widely adopted idea
nitive Office, a middleware for intelligent environments [10]. is the use of a portion of the available sensors, since there
Both of them offer a decentralized approach to intelligence are often overlaps and not all nodes are always required.
[11]. Introducing, however, intelligence in the lower layers of The backcasting algorithm proposed in [42] is based on
the considered architecture is highly questionable, as it could this concept. SORA (Self Organizing Resource Allocation)
potentially hinder flexibility. This criticism stems from the [43] is adopting principles of market-oriented programming
fact that a flexible middleware for an IoT model should be to define a penalty scheme for data forwarding, that way
agnostic of what things should or are capable of doing. Rather, reducing resource and energy consumption. A combination
such a middleware should focus on facilitating adaptive of adaptive sampling and effective routing is proposed in
interconnection (encompassing all of its aspects) of machines USAC (Utility-Based Sensing and Communication model)
and users, in the meantime achieving adaptiveness through [44], in an effort to save energy by adapting observations.
intelligent agents in the middle layers. A major concern in such adaptive systems regards the
Undoubtedly, wireless sensor networks have given rise entity that is actually performing the decision making. An
to the usability and effectiveness of agents in distributed efficient and flexible solution should allow individual entities
systems. In particular, the employment of mobile devices to make local decisions on the sampling schedule. In [45],
in such networks has strengthened the arguments for the for example, binary integer programming techniques are
efficacy of mobile software agents in dynamic architectures employed to assign tasks to nodes, allowing them to reach
[36]. This type of agents are autonomous and able to adapt to decisions based on their observations.
the environment. In contrast, a stationary agent provides to
the sensor node it is associated with information originated 2.4. Discussion on Findings. The IoT focuses primarily on
from its adjacent nodes. They can be embedded in the sensing the interconnection of and interactions among the enti-
devices executing their process and then moving between ties connected to it, whether they are human end-users
network sensors. The agent functionality heavily depends or machines. Event-driven, service-oriented approaches to
on the specific architecture. For instance, in hierarchical interaction, along with an overlay network of uniquely
sensor networks, the agents can be created in higher order identifiable entities, are essential to the vision of an IoT.
nodes, whereas in flat topologies typically the sink node is Furthermore, centralized approaches to discovery of services
the one that instantiates the agents. In a standard scenario, are preferred in the industry due to high performance;
the mobile agents traverse the sensor network collecting nonetheless, they have difficulty supporting volatile, ad hoc
monitoring information from each one of the involved networks and overcoming disaster scenarios. The entities
devices and eventually delivers it to the sink. In that manner, in an IoT nowadays are expected to be CPS, such as the
significant portion of the available bandwidth and energy can sensor networks, and designing such systems hints at the
be conserved; however, there are important issues related to need for the following two: (a) properties to be found in
International Journal of Distributed Sensor Networks 5

the field of software components that solve CPS subsystem the need for human configuration and management. EDBO
design issues and (b) a SOA for supporting their interactions. is a continuation of previous work in distributed systems
Meanwhile, infusing autonomy in that distributed system, where the focus was on understanding how relationships
which forms this overlay network actualizing the IoT vision, among network nodes affect resource discovery overall [46].
can be achieved by drawing inspiration from nature for self- Although it is a generic case study which represents dis-
organization and the utilization of the agent-based program- tributed systems, more concrete studies, like file sharing net-
ming paradigm for autonomous operation. works, high performance computing, decentralized schemes
Modern sensor networks have been shown to incorporate of Web services, or sensor and monitoring networks, could
in part several of the aforementioned solutions and program- potentially benefit from the results of this work.
ming paradigms in order to achieve their goals. While the While there is a diverse set of objectives that contempo-
majority of solutions are tailored to a specific scenario, or at rary distributed systems are expected to satisfy, not all of them
least a specific family of them, and they successfully address can be appropriately treated as macroscopic properties which
these issues, they are nonetheless not designed with an eye could emerge from local interactions. The main objectives of
towards a future IoT application with the capacity to support EDBO are to achieve scalability, robustness, and availability
numerous scenarios. These solutions also do not focus on in a distributed system without explicit engineering. Instead,
the need for delivering a system capable of autonomous by following an experiment-driven framework, various prop-
operation, achieved through minimized human intervention erties and behaviors are introduced as different hypotheses
and attributed to self-organization properties. Finally, such of achieving the aforementioned global objectives. Emergent
systems attempt to facilitate effortless, ad hoc connectiv- properties can be utilized in the organization and mainte-
ity of the devices, albeit primarily relying on centralized nance of the network as well as the basis for improving the
approaches for interconnection and resource discovery. discovery of resources within the network.
Addressing the aforementioned needs and exploring The EDBO model is described at length in the following
the alternatives, we propose an architecture that aspires section, along with the biologically inspired concepts and
to use bioinspired concepts to support autonomy through properties that were employed as enabling mechanisms of the
self-adaptation and self-organization in the field of ADS. desired emergent behaviors.
Consequently, we incorporate that model into the design of
a middleware capable of seamlessly interconnecting humans 3.1. Emergent Distributed Bioorganization. The scope of
and heterogeneous machines, which evidently has the capac- EDBO is limited to the top layer of the distributed sys-
ity to facilitate interactions among them. In an attempt to tems paradigm. A generic unstructured distributed system
provide a universal, easily adoptable solution, we ascertain model has been devised which will maintain a high level of
that the middleware possesses properties in its architecture availability, scalability, and robustness, under different opera-
that allow for a substantially high degree of extensibility tional conditions. This is achieved by introducing biologically
and adaptability, when incorporated into an end-system inspired properties in the EDBO model as well as employing
supporting the desired vision of an IoT. agent-based techniques in order to guide emergent phe-
nomena in an ADS. The conceptual structure followed in
3. Proposed Architecture the model is built in terms of a distributed application that
provides resources to external users, rather than a distributed
The Emergent Distributed Bio-Organization (EDBO) is a system that is composed of users (peers).
model that was conceived during long term work on har- EDBO nodes are represented by agents referred to as
nessing emergent phenomena in artificial distributed systems BioBots (Figure 1(b)) which use two-way logical connections
(ADS). Initial work on understanding the micro-macro (relationships) to form an overlay network. Each BioBot has
causal links, along with ADS experience gained through a a limited number of relationships to other BioBots which
simpler version of EDBO, has led to the development of are managed autonomously. In terms of functionality, a
a disciplined framework for engineering emergence, and BioBot serves as a wrapper for a set of resources (abstracting
EDBO as an abstract distributed system model aiming to data, functionality, and services) which are provided to user
engineer emergent properties at the macroscopic level [14]. requests (queries).
EDBO is based on the concept of focusing on properties A BioBot represents the core routing component which
and interactions at the microscopic level in order to allow facilitates the propagation of queries through the network in
desired properties to emerge at the global, or macroscopic, an autonomous manner. BioBot behavior is based on several
level. Previous bioinspired approaches tackle specific prob- bioinspired heuristic mechanisms (elaborated in Section 3.2)
lems with limited scope while EDBO attempts a holistic that guide its decision making. The heuristics rely on the
approach while it addresses multiple operational issues of BioBot’s internal state, the available relationship meta-data,
distributed systems. and the current state of the environment, which in terms of
The main problem that EDBO attempts to solve is the the EDBO paradigm is referred to as BioSpace. This environ-
discovery of resources in an unstructured, fully decentral- ment layer acts as a middleware between the BioBots and the
ized, network, under varying conditions. The rationale of underlying operating system and physical infrastructure (see
introducing emergent properties in a distributed system Figure 1(a)), with different middleware instances running on
is to improve its operational efficiency and to eliminate different servers being interconnected in order to facilitate
6 International Journal of Distributed Sensor Networks

BioBot
BioSpace Attributes
(logical space) (i) Resource set
(ii) Spatial location
(iii) Relationship buffer
(iv) Discovery energy level
Relationship (v) Service energy level
(logical link) Behavioral heuristics
(i) Query routing alg.
BioBot (ii) Movement alg.
(node) (iii) Relationship management alg.
Execution platform (iv) Reproduction alg.
Physical (server)
BioSpace middleware
connections
(a) (b)

Figure 1: (a) Simplified overview of the BioSpace; (b) BioBot’s core attributes and heuristics.

a unified environment and a singular spatial universe. In are dying (i.e., removed from the network), while successful
addition, the BioSpace facilitates several functional opera- agents are rewarded with the ability to reproduce. The latter
tions such as communication, offspring creation, and move- can happen through single-parent replication, which results
ment service. in identical copies, or through binary reproduction in which
both parents’ properties are combined to produce a new
3.2. EDBO: Biological Principles. EDBO utilizes a range mixture of service and discovery characteristics.
of microscopic properties and interactions (at the BioBot
level) in order to achieve global-level properties such as Adaptation through Ad Hoc Selection. This is directly inspired
organization and optimization. by natural selection and fitness-based evolution. BioBot
fitness is expressed through energy levels which clearly
Autonomous Decision-Making. Autonomy is one of the fun- represent how fit or unfit an entity is, at any point during
damental principles of nature; biological entities usually act its life cycle. Death and birth events allow the network to
without explicit central or external control. Autonomous continuously select the most successful BioBots according to
decision making in EDBO is realized through several algo- the current network conditions.
rithms that use local interaction and information to guide
agent behavior.
3.3. EDBO: Operational Emergence. The main challenge in
Death and Birth Events through Energy Maximization. Energy the EDBO model is to devise an effective and efficient
can be seen as a major life force in nature, enabling biological way of discovering resources in an initially unstructured
entities to perform a variety of tasks. In EDBO, the concept and fully decentralized network. Query matching is imple-
of energy is portrayed as a BioBot attribute. More specifically, mented through simple keyword comparison. Intuitively,
the model proposes two different energy types: query routing appears to be the deciding factor for the
success of resource discovery. However, our hypothesis is
(i) Discovery energy is the primary indicator of BioBots’ that providing an appropriate network organization and
usefulness in facilitating the resource discovery pro- resource population size is as important. The rationale behind
cess. The discovery energy is awarded to the BioBots this is that by imposing ad hoc overlay organization and
that facilitate a successful query match. BioBots can adjustment of the number of BioBot/resource instances, the
spend discovery energy while performing various system will be able to maintain acceptable levels of query
activities such as moving, exchanging messages, and match success rate without proactive (well-informed) routing
reproducing. strategies that often impose high communication overheads.
(ii) Service energy denotes how popular or successful is a Towards this direction, the EDBO platform aims to utilize
resource in the system. Service energy is allocated to emergent behaviors in two main aspects of operation: the
each resource individually when the resource matches availability and scalability of resources and the organization
a query. Service energy is consumed each time query and optimization of the network.
matching occurs. The scalability and availability of resources are an impor-
tant aspect in ADS, since the query load imposed by the
In the natural world, biological organisms aspire to users often exhibits extreme fluctuations. Additionally, some
maximize the energy gain and minimize the energy loss as of the resources are in higher demand (popular) than others
a vital principle for survival. In a similar fashion, energy (unpopular). In order to cope with these requirements,
maximization is a core principle of the EDBO model. BioBots EDBO utilizes energy distribution as the main feedback
which fail to maximize their energy and reach a critical low mechanism that regulates the size of the BioBot population.
International Journal of Distributed Sensor Networks 7

Since the amount of energy entering the system is regulated then consume all the provided data from the connected
by the number of successfully resolved queries, in situations CPS without the need of any central control. The abstract
where there is a high query load the amount of available proposed architecture is depicted in Figure 2.
energy in the system will be increased. This allows for a The following section details the design followed towards
larger BioBot reproduction rate which essentially scales up implementing the EDBO for CPS architecture as a viable
the number of available resource providers. Moreover, by solution for real-world wireless sensor networks.
using energy driven selection of parents (see Section 3.2)
popular resources have higher chances of being reproduced.
A decrease in the query load leads to an overall energy
4. Middleware Design
decrease, which in turn increases the BioBot death rate and
In order to fulfill its goal as a middleware, the design of the
reduces the number of birth events. Thus, the size of the
EDBO for CPS implementation allows for several desired
BioBot population scales down.
properties to be present in the final system. To begin with,
Structured overlays are typically performing better than it is essential that an ADS, which can account for all the func-
unstructured ones. In the EDBO this is tackled via an emer- tionality and properties, must be developed as a multiagent
gent overlay organization and connection (relationship) opti- system to enable autonomy and realize the ideas presented by
mization. This is mainly inspired by the supernode architec- the EDBO model (the BioSpace, the BioBot, and their inter-
ture [47] which enables application of discovery mechanisms actions). Meanwhile, the core principles of component-based
to improve discovery performance overall. In the supernode software engineering have to be realized for the middleware
architecture, the overlay structure is logically divided into two to be able to accommodate any future development; those are
layers: the global layer which is composed out of supernodes composability, substitutability, reusability, and extensibility
that enable global connectivity and the local cluster layer [48]. Finally, it was discussed how a SOA is essential to tackle
which facilitates local interconnection. In order to facilitate most of the common architectural and implementation issues
the ad hoc formation of supernode structures through local that are expected of an IoT-enabling middleware.
BioBot decisions, EDBO incorporates flexible relationship
Naturally, a suitable framework for the development of
buffer size, different relationship acquisition strategies, and
such a system is essential, with the following traits being of
energy metadata. The selection of the superbots is based on
paramount importance: offering the ability to design with
the BioBots’ discovery energy level, which in turn determines
SOA principles, encompassing the much-desired properties
the total number of relationships (slots) that a BioBot can
of software components and, of course, providing the tools to
establish. BioBots with high discovery energy and a large
develop a full-fledged MAS. The Jadex Active Components
number of relationships are granted the superbot status. Such
(JAC) framework has been chosen to this end, offering all
BioBots allocate part of their connections to other superbots,
of the above and additionally solving issues such as simu-
while the rest of the connection slots are used to facilitate
lation support, security, as well as offering fast prototyping
the connectivity with the nodes in the local cluster. BioBots
[49]. Finally, it must be mentioned that the nonfunctional
in the local cluster optimize their relationships through
properties expected of the EDBO middleware have been
evaluation of partners’ spatial distance in order to ensure tight
considered during its design. Specifically, software patterns
connectivity in the local cluster.
such as the strategy, factory, observer, and adapters have been
used heavily in order for the software to support a high degree
3.4. Realizing the EDBO Architecture for CPS. The proposed of extensibility and at the same time maintain the highest
architecture aims to serve as a main infrastructure that will possible degree of applicability to assorted scenarios.
enable any authorized consumer to perceive the required The following subsections will detail how the proposed
sensor or other types of data as if connected to the nervous architecture elements depicted in Figure 2 were designed
system of an organism. It provides a solution that enables as a middleware by virtue of the chosen framework for its
utilization of emergent behaviors to achieve availability and development. Firstly, details of the capabilities of the JAC
scalability of resources and the organization and optimization framework are presented, followed by the design details
of the network. The proposed middleware is a realization of the BioSpace and BioBot abstractions as entities in a
of the above-described research prototype network (EDBO) multiagent, distributed system. The section continues with a
that achieves several self∗ properties using the bioinspired presentation of the conceptual architecture of the middleware
solution described above. Thus, it is composed of BioBots and our approach towards solving the IoT issues, eventually
(logical nodes) that are realized in a BioSpace (middleware). concluding with a short presentation of the manner in which
BioBots are capable of communicating with CPS if in range, the various middleware components operate once deployed.
enabling bridging of the desired CPS and the middleware. A package overview of all individual components of the
EDBO for CPS facilitates a plug and play solution which system is illustrated in Figure 3.
allows the addition of such a system to the network at
anytime. By adding a new CPS (provider) to the proposed 4.1. The Jadex Active Components Framework. The JAC
middleware, which is based on the EDBO architecture, a project started out as Jadex, an extension over the JADE
service will be automatically provided and discovery will framework that aimed to utilize the BDI agent model and
be enabled in a fully decentralized manner. Consumers offer an effective and efficient implementation of it over
realized as compatible clients will be able to discover and JADE. Over time, it has grown to become a complete, separate
8 International Journal of Distributed Sensor Networks

Figure 2: An abstract representation of the proposed architecture.

≪Java Package≫ ≪Java Package≫ ≪Java Package≫

endusersystem externalsystem platform

≪Java Package≫ ≪Java Package≫ ≪Java Package≫ ≪Java Package≫

query comm BioBot BioSpace

≪Java Package≫ ≪Java Package≫ ≪Java Package≫ ≪Java Package≫ ≪Java Package≫
local energymanagement birth systemdiscovery
forwarding

≪Java Package≫
≪Java Package≫ ≪Java Package≫ ≪Java Package≫
local
remote migration death

≪Java Package≫
≪Java Package≫ ≪Java Package≫ remote
relationshipmanagement energyallocation

≪Java Package≫
servicediscovery

Figure 3: EDBO middleware package overview. The end-user system running at the nodes consuming services, the external system at the
nodes providing services, and the platform on the nodes forming the EDBO overlay network.
International Journal of Distributed Sensor Networks 9

framework for creating a MAS, or an ADS, that can be BioBot BioSpace

deployed on a number of different devices, provided that they
can run a JVM, including a specialized, lightweight version
of the framework that can be used for deployment of systems
on Android devices. The framework does not offer only BDI Invoke service
agent implementation; in addition, it supports Plain Old Java
Object (POJO) programming for any type of agent (e.g., Return future
model-based, reflex agents).
Add result listener to the future Calculate energies
The main strength of the framework lies in the fact that
it is created with a SOA in mind, with the categorization
of implemented services as either required (an agent cannot
function without this service available somewhere in the
ADS) or provided (functionality that the agent possesses),
thus simplifying the translation of the EDBO model to Notify listener with result
system architecture. Meanwhile, JAC offers asynchronous Extract result from future
agent interaction using the future paradigm (see Figure 4) to
facilitate invocation of agent provided services, which prac-
tically eliminates the need for the utilization of an ACL and
messaging to simulate asynchronous agent interaction. This Figure 4: Sequence diagram depicting the future paradigm at work
interaction is realized in an event-driven manner and hence using the example of the Energy Allocation Service.
covers the needs of an architecture that can support the IoT.
The most integral part of the JAC architecture is the Active
Component (AC), which refers to the agent implementa- the sole purpose of offering the BioSpace services (specified as
tion encapsulated as a software component through several provided). In JAC services also have a scope, which refers to
properties. Furthermore, the additional properties of the AC where they will be available in the ADS (e.g., platform, local
come from the SOA aspect of the framework and essentially network, globally). A platform scope is used to this end so
constitute a separate module that attaches to the agent and that it is only accessible by the BioBots living on the current
is responsible for handling any service-related property or machine.
functionality the AC is qualified for. Finally, apart from An example of this design can be seen in the following
any traditional services offered through an ADS developed scenario: a BioBot is born and the BioSpace has to allocate
with JAC, the framework is capable of exposing any service service and discovery energies to it.
implemented for an AC as either a WSDL or RESTful service. (1) The BioBot AC invokes the Energy Allocation Service
The added value of this capability is the realization of an IoT of the BioSpace AC (assigns a listener to a future
as a WoT, with the latter functioning as either a wrapper of event).
the system or complementing its provided services.
(2) The BioSpace AC receives the invocation (creates the
future event) and determines how much energy to
4.2. The BioSpace. In the model, the BioSpace is described allocate, and when ready it communicates it to the
as a logical, 2D space where the BioBots live and die. An BioBot AC (sets the result of the future and notifies
important distinction that must be made at this point is that the listener).
the model regards it as a single entity offering services to
(3) The BioBot AC retrieves the energy levels sent to it by
the BioBots (e.g., energy allocation and migration), but when
the BioSpace AC (extracts the result wrapped within
translating the model into software the risk of employing
the future event) and begins its life functions.
centralized solutions is inevitable. This is why the software
running on each system has been designed in such a manner Figure 4 provides an illustration of this process involving
that it can still provide BioSpace services to the BioBots on the JAC future paradigm (service-oriented, event-driven
that particular system, without the need to resort to a cen- communication).
tralized mechanism (e.g., a database with BioBot locations).
This is achieved primarily by the abstraction of the migration 4.3. The BioBot. The EDBO model describes the BioBot as a
service to merely provide the functionality of migrating (i.e., very basic, reflexive agent: one that can simply react to queries
changing current BioBot logical space location), while the it receives regarding service discovery (either from users or
BioBots store that location individually instead. other BioBots) and relationship or reproduction requests by
The JAC framework classifies the services in the system other BioBots. Its functionality is hence limited and it is
as either required or provided. A BioBot consuming the clear that the three aforementioned services are specified as
services provided by the BioSpace should have to specify provided services, while the ones provided by the BioSpace
those services as required. But in a system adhering to the AC are required by the BioBot AC to ensure that a BioBot
SOA principles, it is essential that these services are provided cannot be born without a BioSpace to live in. The state of the
by someone; in the case of a JAC the provider is always BioBot, which it monitors as an agent, is distributed among
an AC. Therefore, a very basic AC has been created with the distinct services it provides. Specifically, monitoring of
10 International Journal of Distributed Sensor Networks

Birth Yes

Determine Partner
energy Migrate Reproduce accepted?
levels

Yes
No No
Contact
partner
Migrate? Replicate
Sexually
Asexually

Sexually or
No asexually?

Yes

Death Adjust Manage Reproduction

conditions? No partners relationships
buffer conditions?

Yes

Inform
partners
of death

Death

Figure 5: Flowchart demonstrating the life cycle of the BioBot.

energies is handled by the Energy Management service, relationships before dying and monitoring service levels and
monitoring of location by the migration service, monitoring taking several actions regarding its bioinspired functions
of relationships and metadata on the partners by the Rela- (migration, reproduction, replication, and relationships). As
tionship Management service, and keywords related to the a deviation from the BioBot agent described in the model,
services the BioBot is aware of through the service discovery the BioBot AC is capable of skipping one of its life functions:
service. The observer software pattern has been applied to processing incoming messages and acting on them. This is
these services and the BioBot itself so that monitoring of attributed to the SOA design of the AC in the framework
its state can be carried out elegantly in an event-driven and the event-driven manner of AC interactions thanks to the
manner, in addition to having the capacity to be tailored to future paradigm.
the administrator’s needs. BioBot interactions take place as described above for the
Additionally, the JAC framework specifies three functions BioBot and BioSpace AC, and they are facilitated through
that can be used to abstract the behavior of an AC being cre- two different services specified as provided services in the
ated (the BioBot being born), its main process as expected of design: the Relationship Management service and the Service
an agent (BioBot life cycle), with the last one describing what Discovery service. The former service is used to find a
happens when the AC is being destroyed (BioBot dying). The partner to keep for migration towards possible reproduction
life cycle of the BioBot, how it performs its life functions as and forwarding queries, an action that the latter service is
designed for the middleware, is presented in Figure 5. The responsible for. This Relationship Management service has
BioBot is designed to ask the BioSpace for energy allocation been designed with the strategy software pattern so that it
and introduction to another BioBot when it is born, ending is trivial to configure how BioBots choose their partners
International Journal of Distributed Sensor Networks 11

I IRelationshipService I IEnergyManagementService

+ requestPartnerByChance() + decreaseDiscoveryEnergy()
+ requestPartnerByKSV() + increaseDiscoveryEnergy()
+ requestPartnerByMetaData() + getDiscoveryEnergyLevel() ≪Java interface≫
+ requestPartnerByName() + increaseServiceEnergy() I IEnergyAllocationService
+ requestPartnerIntroduction() + decreaseServiceEnergy() + allocateServiceEnergy()
+ formRelationship() + getServiceEnergyLevel() + allocateDiscoveryEnergy()
+ terminateRelationship()
+ isLookingForPartner()
+ requestPartners() ≪Java class≫ ≪Java class≫
+ partnerDied() C BioBotAgent C BioSpaceServiceProviderAgent
≪Java interface≫
+ partnerAdded() + c BioBotAgent() I IBirthService + c BioSpaceServiceProviderAgent()
+ announceDeath() + agentCreated() + agentCreated()
+ announceMove() + createBioBot() + executeBody()
+ executeBody()
+ manageRelationships() + agentKilled()
+ agentKilled
+ updateMetaData()
+ partnerMoved()

I IServiceDiscoveryService ≪Java interface≫

I IDeathService
+ updateServices()
I IMigrationService + forwardQuery() + killBioBot()
+ Migrate() + returnQuery()
+ getCurrentLocation() + getServices()

Figure 6: The BioBot and BioSpace agents, including their provided and required services.

according to the model: randomly, using keyword similarity Things

or favoring better metadata. This allows for the effortless
integration of better, more optimized strategies in the future. Adapters
BioBots are introduced to each other through either the BioBots
BioSpace or another BioBot, which essentially means that
the BioBot keeps a proxy of its partner so that they can Key
communicate to them directly when needed (framework words
concept of AC External Access). In order to obtain this
proxy, which for newly born BioBots is usually provided by
the BioSpace AC, the design relies on the JAC framework
capacity for awareness of other computer systems. This
awareness mechanism is the core of the system that realizes
it as an ADS, thanks to the various methods it provides
for discovering systems connected to the overlay network: Figure 7: The layered architecture of the EDBO middleware.
Broadcast, Multicast and IP Scanning for local networks, and
Registry, Message and Relay for global networks. Awareness
settings can be enabled and disabled at runtime, and many the mere abstraction of a keyword to represent the func-
can be used together at the same time providing an excellent tionality a thing possesses. Through its layered architecture
solution to ad hoc connectivity and reactivity to disaster (Figure 7) and the exploitation of the mechanisms available
scenarios. An overview of the BioBot agent and the services by the JAC framework, the design of the EDBO middleware
it provides, as well as the one it requires from the BioSpace, is provides the means to leverage the aforementioned issues.
illustrated in Figure 6. Each layer of this architecture will be discussed in this section.
The outermost layer named Things represents the software
4.4. Addressing IoT Concerns. At the very core of the IoT running on the end-point, which can be a single device, an
lie the issues of the unique identification of things, their end-user, or even a complex CPS. Two systems have been
interconnection, and interactions, as well as their virtualiza- developed to support different roles for end-point devices:
tion. The latter solves the problems of device heterogeneity one for “things” offering a service and one for a human end-
due to the fact that the virtual representation of the system user attempting to search for a service. The design is modular
is expected to be uniform. Naturally, the EDBO model has and loosely-coupled; therefore, the appropriate libraries can
not been designed with all these issues in mind, providing be used in any system that aspires to provide a different type
12 International Journal of Distributed Sensor Networks

of user interface with the core of the middleware. This layer a centralized repository of services, rather it aspires to provide
fully supports a WoT implementation because the services, some sort of very primitive semantics to the middleware.
the ones which realize this interface, have been exposed Future work is planned to replace this layer with enhanced
as RESTful services and can thereby be accessed by any semantics, if possible with an extended ontology. The unique
system that is capable of RESTful communication. This can identification of the things connected to the network is
be a simple mobile phone app (the JAC framework provides achieved in part thanks to this layer, but also in part due
extremely lightweight libraries for deployment of the system to how the External Access proxies of platforms connected
on Android devices), a Web application, or a server gathering to the rest of the network are handled by the underlying
data from end-point devices, among many other possibilities. awareness mechanisms of the JAC framework.
The API of this functionality is detailed in the documentation
of the middleware. As such, the ultimate purpose of this layer 4.5. Middleware Operation. The operation of the EDBO
is to serve as the first step in achieving the interaction among middleware in a sample deployment scenario can provide
“things,” one of the core issues IoT implementations ought to an insight into some of its capabilities. The walkthrough
address. presented herein assumes a scenario in the domain of
In the onion-layered architecture of the EDBO middle- health care for elderly patients. In this scenario, the patient
ware, the layer Things communicates directly with the layer possesses a health sensor platform to record some data that
Adapters. Whereas the former aimed to facilitate interaction
that their personal doctor would like to monitor and the
at a higher level, the latter aims to address the problem of the
history of which should be recorded in the hospital server
lower-level interoperability, taking into account the expected
for processing by expert systems. Figure 8 illustrates how
high degree of heterogeneity among “things.” When a user or
these actors and systems are interconnected with the EDBO
a system initiates an action, the information provided by the
middleware deployment.
varied forms of end-point software on the outer layer has to
be communicated to the inner layer of BioBots. Nonetheless, The two application servers run the core of the middle-
the BioBots can only understand a predefined format for ware, each with a BioSpace and BioBots to create the overlay
data communicated to them; therefore, translation is needed. network (link 1, Figure 8) that end-point systems can connect
This second layer provides the means to solve this translation to and offer or consume services. The health sensor platform
problem and achieve interoperability. The adapter software runs software created by the manufacturer; however, they
design pattern has been employed to this end, thus allowing have incorporated the System Discovery libraries from EDBO
developers to extend the provided classes and create their to connect with BioBots (link 2, Figure 8) and they have also
own adapters as needed for translating the information from implemented an adapter to translate the data produced by
their devices or CPS into an EDBO service. Already provided their platform (e.g., the XML adapter) into EDBO services.
are sample adapters that can translate data to and from The android phone possessed by the doctor runs a simple,
Windows INI, XML, and JSON formats. Essentially, the sheer lightweight app that periodically checks for the data of their
flexibility of adapter approach provides the means to achieve patients by using the RESTful API of a specific BioBot (link
the much needed virtualization of a human user or a system 4, Figure 8). The JSON adapter is used to translate the data
in any IoT scenario. from the EDBO service into the format that the app can
Having solved the problem of interoperability and inter- process to produce alerts as needed. Finally, the server runs
action among “things” with the 2 outer layers, the third the already provided EDBO External System EDBO software
layer attempts to solve the interconnection of things. This to connect with the BioBots (link 3, Figure 8) and hence acts
layer is now provided by the core of the EDBO model: the as a service consumer in this case by getting the data from
bioinspired approach that creates the overlay network of the health sensor platform at regular intervals. It implements
the end-point systems and facilitates their communication. a new adapter that parses EDBO service data and maps them
The BioBots are always aware of who their partners are, onto fields for a MySQL database.
what services they can offer, and which partners to contact This scenario depicts only two EDBO middleware sys-
to try and discover a service they cannot offer themselves. tems to serve as the backbone of the overlay network on
The Service Discovery and Relationship services shoulder nodes, but this could be extended to feature any number
the majority of the workload to this end, which is solving of platforms needed to satisfy the area coverage needs of
the problem of interconnection in an IoT implementation, the specific IoT scenario. Whenever a new EDBO middle-
in conjunction with the awareness mechanisms that the ware platform is instantiated on an application server, it
JAC framework provides. The uniqueness in the case of the will attempt to use all of the allowed platform awareness
EDBO middleware is that the contribution of this layer is mechanisms available in order to find an entry point to the
not only limited to the interconnection of devices, rather it network (e.g., local area broadcast and IP message passing
extends to the provision of autonomous operation via self- but not IP scanning). Once it has formed a relationship
organization, not a strict requirement for an IoT application with another BioBot, it can be conversed with during service
but an extremely beneficial property. discovery and it can now offer any services that have been
The innermost layer is that of Keywords, the abstraction published on that application server.
currently used to represent how the BioBots determine if Having showcased the formulation of an IoT with the
the search term supplied by a thing matches a term they two outer layers of the EDBO architecture above (Things
are aware of. This layer does not represent some form of and Adapters, Figure 7), the service discovery capabilities
International Journal of Distributed Sensor Networks 13

≪Device≫
≪Device≫ ≪Device≫
Health sensor platform
Application server Application server
≪2≫

≪Device≫ Platform software

Hospital server
≪3≫ ≪1≫
EDBO.jar EDBO.jar
EDBOExternalSystem.jar ≪Device≫
Android phone
≪4≫

RESTful mobile app

Figure 8: A sample deployment scenario for the EDBO middleware.

Sufficient
service
energy?

Yes

Lose
Keyword No service
found? energy

Yes
No
Gain Have Sufficient Lose
service partners? Yes discovery Yes discovery
energy energy? energy

Provide Forward Randomly,

requested query to similarity strategy,
service partner complex strategy

No
No

Figure 9: Flowchart depicting the decentralized query forwarding mechanism.

of the middleware with the inner layers remain to be seen outlined in Figure 9. The energy level fluctuations presented
(BioBots, Keywords, Figure 7). This process takes place in there trigger the BioBot life cycle decision making process
the case at hand when the mobile phone app is planned (Figure 5) and lead to self-organization.
to update the latest data on the patients being monitored
(e.g., every 10 minutes) and when the server collects data for
processing (e.g., every 1 hour). In the case of the android app, 5. Middleware Evaluation
the BioBot will receive the query on the RESTful wrapper
of the service, process it, and immediately reply with the The EDBO model promises self-adaptivity and self-opti-
requested service data since they are readily available. On mization in an ADS, properties that need to be verified during
the other hand, the hospital server sends the request to the the transition from model to software. In order to ascertain
platform it is connected to but the BioBot receiving that query that the EDBO middleware demonstrates the capabilities
cannot satisfy it itself. Consequently, it forwards the query to offered by the model, extensive testing has taken place
their partner who can provide the requested data. The flow of continuously throughout its iterative development process,
this discovery process as the BioBots perform it in each case is which was in turn followed by an elementary evaluation of
14 International Journal of Distributed Sensor Networks

its capabilities. Both of these tasks have been planned since Table 1: Response accuracy and delay regarding the selected service
the beginning of the development process and have been for the runtime evaluation scenario using the random and complex
facilitated by the functionality offered by the JAC framework, (metadata) strategies.
and specifically through the Jadex Control Center (JCC)
Avg. response Avg. response
software offering a test center and through the integrated Time frame accuracy delay (hops)
logging mechanisms for the ACs.
Random Metadata Random Metadata
The approach towards testing has been carried out in
two ways: (a) unit testing for each component developed After 5 minutes 60.25% 64.09% 2.921 2.896
and (b) integration testing when it was encompassed in the After 20 minutes 57.92% 71.23% 3.057 2.842
rest of the system. The JCC, through its test center, offers After 50 minutes 58.51% 89.56% 3.013 2.439
a frontend for executing JUnit-like, component-based test
cases for each AC that has been developed (in this case
the BioBot and BioSpace agents). As the main guideline (short-term adaptivity) and nearing the end of runtime (long-
for developing these test cases served a formal model of term optimization), in order to determine if the desired
the BioBot implemented as an X-machine, which has been properties emerged in the system. As such, these metrics
developed as part of the EDBO model. The X-machine constitute the Key Performance Indicators (KPIs) for the pur-
represents all possible states and transitions among them poses of this quantitative evaluation: determining efficiency
taking place throughout the lifecycle of the BioBot and offers of the system over the passage of time (totally unstructured
a formal testing methodology. Therefore, it was elementary origins, adaptation, and organized system in the end).
to map these states and transitions onto the JUnit test cases In order to determine how the system can adapt to user
needed to verify and validate BioBot behavior through its behavior, the evaluation scenario required the development
functions and the functions of its services (both provided and of a querying function that sends various keywords to the
required). The BioSpace service provider also had JUnit test BioBots, emphasizing a select keyword at the beginning
cases covering its functionality, although no formal model (to establish a baseline), again after a little time during
existed to guide the process (no life functions to perform). runtime and near the end again. Naturally, this evaluation
The evaluation of the middleware has been focused has been carried out as a demonstration of the system using
on investigating the capabilities of the middleware towards an elementary case study that could simulate the following
realizing a self-adaptive, self-optimizing ADS. The optimal scenario: an ADS of weather sensors. The system is deployed
approach to determining the behavior of an ADS developed over several regions and the end-users can check the weather
with the EDBO middleware was deigned to be the runtime for each region. As an example, after the system has just been
evaluation in the form of a case study comparing results deployed and initialized, a hailstorm affecting crops in one
of a system with no self-organization capabilities to one of the regions prompts users trying to check on the status of
that possesses them. Past results on the evaluation of the their crops to request the data from that region primarily. The
model have shown that the most optimal strategy out of the system thus ought to adapt to this sudden need and try to
three currently formulated and implemented is that of the prepare for future needs. After this initial influx of requests
complex processing of the metadata. This is the case where for this region, users start to check on other regions for a
BioBots tend to form meaningful relationships in the long significant time period to determine if they were also affected.
term (self-optimization) and are capable of handling user After this interval of normalized activity, users decide to
behavior fluctuations by reevaluating current relationships check again on the region suffering from the hailstorm to
and reproducing or replicating (self-adaptive). Consequently, determine its current state. At this point, the system should
the middleware was configured to use two different strategies display its self-organization capabilities by having evolved in
for the same case and compare the results: the complex such a manner that requests towards this “popular” query
strategy and the random strategy. search can be satisfied faster and with higher accuracy.
Evaluation was carried out through the facilities provided The results gathered from the evaluation can be found in
by the JAC framework. Specifically, each AC has an inherent Table 1. Positive results on response accuracy demonstrate
logging mechanism that can be activated and its entries can that the system manages to adapt to user behavior in the
be viewed at real-time during execution via the component short-term, which may become even more significant in the
viewer offered by the JCC. Additionally, the design of the long-term as the system optimizes itself to serve more queries
system enforces the observer pattern and hence a custom similar to the ones it received at the beginning and the
implementation of the abstract observer provided with the first half of its runtime. Delay results did not provide much
middleware offers the means to record all important data information regarding short-term capabilities of the system,
from the state changes and interactions of each BioBot. A which may be attributed to the low initial energy levels of
parser has been developed to read and review the resulting BioBots restricting more partners. Nonetheless, there were
logs, producing information on the data sought after to some changes towards the end of the evaluation hinting at the
determine the properties of the system. The data gathered possibility of self-organization considering that metric, too.
regarded response accuracy (how many queries were satisfied As expected, the random strategy produced no posi-
on average) and response delay (average number of hops tive results whatsoever regarding self-organization. On the
for each query). These values were measured at the very contrary, it appears that random partner selection leads to
beginning of runtime and compared to a few moments later BioBots having “bad” relationships and wasting energy on
International Journal of Distributed Sensor Networks 15

forwarding queries aimlessly. Unlike the evaluation results (the innermost layer of our architecture, Figure 7 earlier).
of the model, no superbots have been observed, which is Discarding the primitive keyword matching functionality,
attributed to the limited resources available for evaluating an the successful integration of semantics will provide a more
ADS at runtime as compared to simulation results. Finally, a refined architecture that allows for better automation of the
criticism of the model, or at least for its initial configurations introduction of new sensor providers to the system. Several
regarding energy levels, stemmed from the fact that several standards for data representation and communication for
BioBots that could not satisfy queries (or forward them) died varied devices are being investigated, in an effort to meet
after some time, which prohibited access to services they this goal and enhance the virtualization aspect, and hence the
offered; in one case, the single BioBot offering a particular interoperability, of the things communicating via the EDBO
service died, effectively losing all access to that service in the middleware. We also aim to work on establishing a security
network. infrastructure that goes beyond the basic security issues
Overall, the system demonstrates the capacity to address already solved in the current realization of our proposed
the issue of load balancing and, moreover, it can do so with- distributed architecture.
out any human intervention, proving the self-organization The issue of BioBots dying and losing access to services
and self-adaptivity capabilities of the system. There is a will be investigated through multiple future evaluations on
differentiation here with the traditional approaches to load- various case studies and assorted size networks. This one
balancing that similar solutions employ, where several factors issue has already sparked research into alternatives to the
are measured over time during network operation and then fundamental biofunction of death the BioBots possess and
settings are adjusted to account for expected load balance its conversion into temporary dormancy. The product of the
depending on network size and connected users, specific evaluation work is expected to produce a policy for optimal
hardware settings of nodes such as remaining energy [50], or energy level configurations for the BioBots and offer further
knowledge of the general structure of the network such as the insight into the scalability in IoT applications. Furthermore,
very efficient SAAS-RWSNs [51]. In such cases this happens the results of these evaluations will be compared to results of
when these values reach the appropriate point that the new similar systems in specific applications in order to develop a
settings have to be applied, and it further demands the exis- better understanding of the overall potential of EDBO and the
tence of a centralized mechanism that keeps all these data of areas that require further optimization, its standing among
all network nodes so that they can be accessed by every node. similar systems, and the possible integration of ideas from
Contrary to this process, the EDBO middleware is capable of those systems with the EDBO ecosystem.
adapting to the changes in its operational environment in a We currently apply the proposed solution in this paper
continuous rather than in a discrete manner, and also without as the core architecture for health monitoring [52], aiming
any knowledge on network structure either. The advantage to evaluate it in a real case study and at the same time
in this case would be the example of 990 users connected at refine it in order to enable seamless, and automate as much
the same time experiencing a slow response because the next as possible, integration of sensors to the system. Research
optimization is to take place when the number goes above is also expanding to the domain of computational acoustic
1000; in the case of the EDBO, the system should have adapted scene analysis, so as to prove the capacity of the system to
gradually to better support these users already. support different and more demanding applications, but also
as an opportunity to investigate the capacity for mapping
BioBot life functions onto real-life actions for connected
6. Conclusions
devices and CPS. One example could be the mapping of
The vision of an Internet of Things is a promising approach BioBot migration onto sensor motor actions towards seeking
to bringing together the two worlds of cyber and physical, a potentially better place for receiving more accurate auditory
especially imagining the gains in accessibility and applica- data. Another is the example of adjusting sensor energy
bility of sensor networks connected to it. By harnessing levels in accordance with BioBot energy levels to fine-tune
the power of component and agent-based services, as well system autonomy through prolonged battery life. Finally, the
as object-oriented approaches, we designed an extensible introduction of new types of energies will be investigated, due
middleware capable of realizing a wide range of scenarios to the research on EDBO that already indicates the potential
with that vision in mind. Its strength lies in the bioinspired emergence of more self-∗ properties that could benefit future
model at the core of its design, one that is capable of offering IoT applications.
autonomous operation to such an artificial distributed system
of interconnected and interacting cyber or physical entities, Conflict of Interests
whether machines or humans. Self-organization features have
emerged in the system thanks to short-term adaptivity to user The authors declare that there is no conflict of interests
behavior, as well as long-term self-optimization for expected regarding the publication of this paper.
user requests.
Future work to the EDBO middleware is planned to add Acknowledgment
features and functionality and to optimize certain aspects
of its operations. As a first step, work is underway on The authors are cofinanced by Greece and the European
developing a data annotation framework to enable the seam- Union (European Social Fund) O.P. “Human Resources
less integration of CPS to the core architecture of BioBots Development,” NSFR 2007–2013.
16 International Journal of Distributed Sensor Networks

References [15] R. Klauck and M. Kirsche, “Bonjour Contiki: a case study of a

DNS-based discovery service for the Internet of Things,” in Ad-
[1] K. Ashton, “That ‘Internet of Things’ thing: in the real world, hoc, Mobile, and Wireless Networks, vol. 7363 of Lecture Notes in
things matter more than ideas,” RFID Journal, vol. 22, pp. 97– Computer Science, pp. 316–329, Springer, Berlin, Germany, 2012.
114, 2009.
[16] D. Guinard, V. Trifa, S. Karnouskos, P. Spiess, and D. Savio,
[2] F. Mattern and C. Floerkemeier, “From the internet of comput- “Interacting with the SOA-based internet of things: discovery,
ers to the Internet of Things,” in From Active Data Management query, selection, and on-demand provisioning of web services,”
to Event-Based Systems and More, vol. 6462 of Lecture Notes IEEE Transactions on Services Computing, vol. 3, no. 3, pp. 223–
in Computer Science, pp. 242–259, Springer, Berlin, Germany, 235, 2010.
2010.
[17] D. Zeng, S. Guo, and Z. Cheng, “The web of things: a survey,”
[3] J. Gubbi, R. Buyya, S. Marusic, and M. Palaniswami, “Internet Journal of Communications, vol. 6, no. 6, pp. 424–438, 2011.
of Things (IoT): a vision, architectural elements, and future
[18] Z. Shelby, “Embedded web services,” IEEE Wireless Communi-
directions,” Future Generation Computer Systems, vol. 29, no. 7,
cations, vol. 17, no. 6, pp. 52–57, 2010.
pp. 1645–1660, 2013.
[19] D. Guinard, I. Ion, and S. Mayer, “In search of an Internet
[4] D. Miorandi, S. Sicari, F. De Pellegrini, and I. Chlamtac, “Inter-
of Things service architecture: REST or WS-∗ ? A developers’
net of things: vision, applications and research challenges,” Ad
perspective,” in Mobile and Ubiquitous Systems: Computing, Net-
Hoc Networks, vol. 10, no. 7, pp. 1497–1516, 2012.
working, and Services, vol. 104 of Lecture Notes of the Institute for
[5] P. Barnaghi, W. Wang, C. Henson, and K. Taylor, “Semantics for Computer Sciences, Social Informatics and Telecommunications
the internet of things: early progress and back to the future,” Engineering, pp. 326–337, Springer, Berlin, Germany, 2012.
International Journal on Semantic Web and Information Systems,
[20] E. A. Lee, “Cyber physical systems: design challenges,” in
vol. 8, no. 1, pp. 1–21, 2012.
Proceedings of the 11th IEEE International Symposium on Object
[6] S. Alam, M. M. R. Chowdhury, and J. Noll, “SenaaS: an event- Oriented Real-Time Distributed Computing (ISORC ’08), pp.
driven sensor virtualization approach for internet of Things 363–369, IEEE, Orlando, Fla, USA, May 2008.
cloud,” in Proceedings of the 1st IEEE International Conference
[21] F.-J. Wu, Y.-F. Kao, and Y.-C. Tseng, “From wireless sensor
on Networked Embedded Systems for Enterprise Applications
networks towards cyber physical systems,” Pervasive and Mobile
(NESEA ’10), pp. 1–6, IEEE, Suzhou, China, November 2010.
Computing, vol. 7, no. 4, pp. 397–413, 2011.
[7] D. Uckelmann, M. Harrison, and F. Michahelles, “An archi-
[22] E. A. Lee, “Computing foundations and practice for cyber-
tectural approach towards the future Internet of Things,” in
physical systems: a preliminary report,” Tech. Rep. UCB/EECS-
Architecting the Internet of Things, pp. 1–24, Springer, Berlin,
2007-72, University of California, Berkeley, Berkeley, Calif,
Germany, 2011.
USA, 2007.
[8] S. Evdokimov, B. Fabian, S. Kunz, and N. Schoenemann,
[23] R. Al Ali, T. Bures, I. Gerostathopoulos et al., “DEECo: an
“Comparison of discovery service architectures for the internet
ecosystem for cyber-physical systems,” in Proceedings of the 36th
of things,” in Proceedings of the IEEE International Conference
International Conference on Software Engineering, pp. 610–611,
on Sensor Networks, Ubiquitous, and Trustworthy Computing
ACM, Hyderabad, India, June 2014.
(SUTC ’10), pp. 237–244, IEEE, Newport Beach, Calif, USA,
June 2010. [24] J. C. Eidson, E. A. Lee, S. Matic, S. A. Seshia, and J. Zou,
“Distributed real-time software for cyber-physical systems,”
[9] A. Katasonov, O. Kaykova, O. Khriyenko, S. Nikitin, and V. Proceedings of the IEEE, vol. 100, no. 1, pp. 45–59, 2012.
Y. Terziyan, “Smart semantic middleware for the internet of
things,” in Proceedings of the 5th International Conference on [25] P. Horn, “Autonomic computing: IBM’s perspective on the state
Informatics in Control, Automation and Robotics (ICINCO- of information technology,” Computing Systems, vol. 15, pp. 1–
ICSO ’08), vol. 8, pp. 169–178, Funchal, Portugal, May 2008. 40, 2001.
[10] L. Roalter, M. Kranz, and A. Möller, “A middleware for intelli- [26] A. G. Ganek and T. A. Corbi, “The dawning of the autonomic
gent environments and the Internet of Things,” in Ubiquitous computing era,” IBM Systems Journal, vol. 42, no. 1, pp. 5–18,
Intelligence and Computing, vol. 6406 of Lecture Notes in 2003.
Computer Science, pp. 267–281, Springer, Berlin, Germany, 2010. [27] J. O. Kephart and D. M. Chess, “The vision of autonomic
[11] L. Atzori, A. Iera, and G. Morabito, “The internet of things: a computing,” Computer, vol. 36, no. 1, pp. 41–50, 2003.
survey,” Computer Networks, vol. 54, no. 15, pp. 2787–2805, 2010. [28] B. Biskupski, J. Dowling, and J. Sacha, “Properties and mech-
[12] L. Tan and N. Wang, “Future Internet: the internet of things,” anisms of self-organizing MANET and P2P systems,” ACM
in Proceedings of the 3rd International Conference on Advanced Transactions on Autonomous and Adaptive Systems, vol. 2, no.
Computer Theory and Engineering (ICACTE ’10), pp. V5376– 1, article 1, 2007.
V5380, Chengdu, China, August 2010. [29] D. Kurzyniec, T. Wrzosek, D. Drzewiecki, and V. Sunderam,
[13] T. Erl, Service-Oriented Architecture: Concepts, Technology, and “Towards selforganizing distributed computing frameworks:
Design, Pearson Education India, 2005. the H2 O approach,” Parallel Processing Letters, vol. 13, no. 2, pp.
273–290, 2003.
[14] G. Eleftherakis, O. Paunovski, K. Rousis, and A. J. Cowl-
ing, “Emergent distributed bio-organization: a framework for [30] E. Di Nitto, C. Ghezzi, A. Metzger, M. Papazoglou, and K.
achieving emergent properties in unstructured distributed sys- Pohl, “A journey to highly dynamic, self-adaptive service-based
tems,” in Intelligent Distributed Computing VI, G. Fortino, C. applications,” Automated Software Engineering, vol. 15, no. 3-4,
Badica, M. Malgeri, and R. Unland, Eds., vol. 446 of Studies pp. 313–341, 2008.
in Computational Intelligence, pp. 23–28, Springer, Berlin, Ger- [31] A. J. Chakravarti, G. Baumgartner, and M. Lauria, “The organic
many, 2013. grid: self-organizing computation on a peer-to-peer network,”
International Journal of Distributed Sensor Networks 17

IEEE Transactions on Systems, Man, and Cybernetics Part A: [45] J. Kho, A. Rogers, and N. R. Jennings, “Decentralised adaptive
Systems and Humans., vol. 35, no. 3, pp. 373–384, 2005. sampling of wireless sensor networks,” in Proceedings of the
1st International Workshop on Agent Technology for Sensor
[32] H. Schmeck, “Organic computing—a new vision for distributed
Networks, pp. 55–62, Honolulu, Hawaii, USA, May 2007.
embedded systems,” in Proceedings of the 8th IEEE International
Symposium on Object-Oriented Real-Time Distributed Comput- [46] O. Paunovski, G. Eleftherakis, K. Dimopoulos, and T. Cowling,
ing (ISORC ’05), pp. 201–203, IEEE, May 2005. “Evaluation of a selective distributed discovery strategy in a
fully decentralized biologically inspired environment,” Informa-
[33] P. Boonma, P. Champrasert, and J. Suzuki, “BiSNET: a
tion Sciences, vol. 180, no. 10, pp. 1865–1875, 2010.
biologically-inspired architecture for wireless sensor networks,”
in Proceedings of the International Conference on Autonomic and [47] B. Beverly Yang and H. Garcia-Molina, “Designing a super-peer
Autonomous Systems (ICAS ’06), 54, IEEE, Silicon Valley, Calif, network,” in Proceedings of the 19th International Conference
USA, July 2006. on Data Engineering, U. Dayal, K. Ramamritham, and T.
M. Vijayaraman, Eds., pp. 49–60, IEEE Computer Society,
[34] K. Aberer, M. Hauswirth, and A. Salehi, “Middleware support Bangalore, India, March 2003.
for the ‘Internet of Things’,” in 5th GI/ITG KuVS Fachgespräch
“Drahtlose Sensornetze”, LSIR-CONF-2009-017, Stuttgart, Ger- [48] I. Crnkovic, J. Stafford, and C. Szyperski, “Software components
many, 2006. beyond programming: from routines to services,” IEEE Soft-
ware, vol. 28, no. 3, pp. 22–26, 2011.
[35] J. I. Vazquez, A. Almeida, I. Doamo, X. Laiseca, and P. Orduña,
“Flexeo: an architecture for integrating wireless sensor networks [49] L. Braubach and A. Pokahr, “Developing distributed systems
into the Internet of Things,” in 3rd Symposium of Ubiquitous with active components and Jadex,” Scalable Computing: Prac-
Computing and Ambient Intelligence 2008, vol. 51 of Advances in tice and Experience, vol. 13, no. 2, pp. 100–119, 2012.
Soft Computing, pp. 219–228, Springer, Berlin, Germany, 2009. [50] Z. Zhang, Y. Wang, F. Song, and W. Zhang, “An energy-
balanced mechanism for hierarchical routing in wireless sensor
[36] O. Dagdeviren, I. Korkmaz, F. Tekbacak, and K. Erciyes, “A
networks,” International Journal of Distributed Sensor Networks,
survey of agent technologies for wireless sensor networks,” IETE
vol. 2015, Article ID 123521, 10 pages, 2015.
Technical Review, vol. 28, no. 2, pp. 168–184, 2011.
[51] K. Kim, C. Ha, and C. Ok, “Network structure-aware ant-based
[37] C.-L. Fok, G.-C. Roman, and C. Lu, “Agilla: a mobile agent
routing in large-scale wireless sensor networks,” International
middleware for self-adaptive wireless sensor networks,” ACM
Journal of Distributed Sensor Networks, vol. 2015, Article ID
Transactions on Autonomous and Adaptive Systems, vol. 4, no.
521784, 16 pages, 2015.
3, article 16, 2009.
[52] A. Basholli, T. Lagkas, G. Eleftherakis, and P. A. Bath, “Wireless
[38] A. Boulis, C.-C. Han, R. Shea, and M. B. Srivastava, “Sensor- monitoring systems for enhancing national health services in
Ware: programming sensor networks beyond code update and developing regions,” in Proceedings of the International Confer-
querying,” Pervasive and Mobile Computing, vol. 3, no. 4, pp. ence on Health Informatics (HEALTHINF ’14), M. Bienkiewicz,
386–412, 2007. C. Verdier, G. Plantier, T. Schultz, A. L. N. Fred, and H. Gamboa,
[39] T. Harman, J. Padget, and M. Warnier, “A multi-layered Eds., pp. 511–516, ESEO, SciTePress, March 2014.
semantics-ready sensor architecture,” in Proceedings of the
3rd International Workshop on Agent Technology for Sensor
Networks (ATSN ’09), Budapest, Hungary, May 2009.
[40] J. W. Hui and D. Culler, “The dynamic behavior of a data
dissemination protocol for network programming at scale,” in
Proceedings of the 2nd International Conference on Embedded
Networked Sensor Systems, pp. 81–94, ACM, November 2004.
[41] T. Liu and M. Martonosi, “Impala, a middleware system for
managing autonomic, parallel sensor systems,” in Proceedings
of the 9th ACM SIGPLAN Symposium on Principles and Practice
of Parallel Programming (PPoPP ’03), vol. 38, pp. 107–118, San
Diego, Calif, USA, June 2003.
[42] R. Willett, A. Martin, and R. Nowak, “Backcasting: adaptive
sampling for sensor networks,” in Proceedings of the 3rd Interna-
tional Symposium on Information Processing in Sensor Networks
(IPSN ’04), pp. 124–133, ACM, Berkeley, Calif, USA, April 2004.
[43] G. Mainland, D. C. Parkes, and M. Welsh, “Decentralized, adap-
tive resource allocation for sensor networks,” in Proceedings of
the 2nd conference on Symposium on Networked Systems Design
& Implementation-Volume 2, pp. 315–328, USENIX Association,
2005.
[44] P. Padhy, R. K. Dash, K. Martinez, and N. R. Jennings, “A utility-
based sensing and communication model for a glacial sensor
network,” in Proceedings of the 5th International Joint Conference
on Autonomous Agents and Multiagent Systems (AAMAS ’06),
pp. 1353–1360, ACM, May 2006.
 International Journal of

Rotating
Machinery

International Journal of
The Scientific
Engineering Distributed
Journal of
Journal of

Hindawi Publishing Corporation

World Journal
Hindawi Publishing Corporation Hindawi Publishing Corporation
Sensors
Hindawi Publishing Corporation
Sensor Networks
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

Journal of

Control Science
and Engineering

Advances in
Civil Engineering
Hindawi Publishing Corporation Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

Submit your manuscripts at

http://www.hindawi.com

Journal of
Journal of Electrical and Computer
Robotics
Hindawi Publishing Corporation
Engineering
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

VLSI Design
Advances in
OptoElectronics
International Journal of

International Journal of
Modelling &
Simulation
Aerospace
Hindawi Publishing Corporation Volume 2014
Navigation and
Observation
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
in Engineering
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Engineering
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Hindawi Publishing Corporation
http://www.hindawi.com
http://www.hindawi.com Volume 2014

International Journal of
International Journal of Antennas and Active and Passive Advances in
Chemical Engineering Propagation Electronic Components Shock and Vibration Acoustics and Vibration
Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014