A Survey of Autonomic Communications
SIMON DOBSON FABIO MASSACCI
UCD Dublin, IE Universita` di Trento, IT
SPYROS DENAZIS PADDY NIXON
University of Patras, GR and UCD Dublin, IE
Hitachi Research Europe, FR
ANTONIO FERN ´ANDEZ
FABRICE SAFFRE
Universidad Rey Juan Carlos, ES
BT Group plc, UK
DOMINIQUE GA¨ITI
NIKITA SCHMIDT
Universite´ de technologie de Troyes, FR
UCD Dublin, IE
EROL GELENBE
and
Imperial College London, UK
FRANCO ZAMBONELLI
Universita` di Modena e Reggio
Emilia, IT
Autonomic communications seek to improve the ability of network and services to cope with un-
predicted change, including changes in topology, load, task, the physical and logical characteristics
of the networks that can be accessed, and so forth. Broad-ranging autonomic solutions require
designers to account for a range of end-to-end issues affecting programming models, network and
contextual modeling and reasoning, decentralised algorithms, trust acquisition and maintenance—
issues whose solutions may draw on approaches and results from a surprisingly broad range of disci-
plines. We survey the current state of autonomic communications research and identify significant
emerging trends and techniques.
A. Ferna´ndez is partially supported by the Spanish MEC under grant number TIN2005-09198-
C02-01, and by the Comunidad de Madrid under grant number S-0505/TIC/0285. E. Gelenbe,
F. Massacci, F. Saffre, and F. Zambonelli wish to acknowledge the CASCADAS (IST-027807) Project
funded by the Future and Emerging Technologies Programme of the European Commission.
E. Gelenbe’s work was supported by grants from EPSRC (UK) GR/S52360/01 and EU FP6 SAPAD
MIRG-CT-2004-506602 on “Self-Aware Networks and Quality of Service”. F. Saffre acknowledges
the EPSRC grant EP/D003105/1. N. Schmidt is funded by Science Foundation Ireland under
grant 04/RPI/1544 on “Secure and Predictable Pervasive Computing”. S. Dobson, S. Denazis and
P. Nixon are partially supported by the ACCA co-ordination action on autonomic communications
(IST-6475) funded by the Future and Emerging Technologies Programme of the European Com-
mission. S. Dobson and P. Nixon are also partially supported by Science Foundation Ireland under
grant number 03/CE2/1303-1, “LERO: the Irish Software Engineering Research Centre”.
Author’s address: S. Dobson, Systems Research Group, School of Computer Science and Informatics,
UCD Dublin, Belfield, Dublin 4, Ireland; email: simon.dobson@ucd.ie.
Permission to make digital or hard copies of part or all of this work for personal or classroom use is
granted without fee provided that copies are not made or distributed for profit or direct commercial
advantage and that copies show this notice on the first page or initial screen of a display along
with the full citation. Copyrights for components of this work owned by others than ACM must be
honored. Abstracting with credit is permitted. To copy otherwise, to republish, to post on servers,
to redistribute to lists, or to use any component of this work in other works requires prior specific
permission and/or a fee. Permissions may be requested from Publications Dept., ACM, Inc., 2 Penn
Plaza, Suite 701, New York, NY 10121-0701 USA, fax +1 (212) 869-0481, or permissions@acm.org.
C© 2006 ACM 1556-4665/06/1200-0223 $5.00
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006, Pages 223–259.

224 • S. Dobson et al.
Categories and Subject Descriptors: C.2.1 [Computer-Communications Networks]: Network
Architecture and Design
General Terms: Algorithms, Design, Management
Additional Key Words and Phrases: Autonomic communication
1. INTRODUCTION
Modern network infrastructures have achieved a small miracle in presenting
a simple and uniform fa cade to applications. In many ways, the programmer’s
view of the network has become simpler over time with TCP/IP acting as a de
facto gateway to a wide range of network technologies.
This external simplification has not unfortunately been matched by a cor-
responding simplification in the construction, management, and extension of
the network from a provider’s perspective. Adding a new network segment, a
new protocol, a new kind of element, or support for a new user- or system-level
application have become fraught exercises in managing the complexity of in-
teractions between elements. This in turn both reduces innovation in networks
and network-centric services and can directly affect the economic viability of
products and services that rely directly on IT and communications agility.
The development of self-managing self-configuring, and self-regulating net-
work and communications infrastructures—collectively referred to as auto-
nomic communications—is an area of considerable research and industrial
interest. By analogy to the human autonomic nervous system, which regu-
lates homeostatic functions without conscious intelligent control, autonomic
communications seeks to simplify the management of complex communica-
tions structures and reduce the need for manual intervention and manage-
ment. It draws on a number of existing disciplines including protocol de-
sign, network management, artificial intelligence, pervasive computing, control
theory, game theory, semantics, biology, context-aware systems, sensor net-
works, trust, and security. The distinguishing feature is the fusion of techniques
from these fields in pursuit of a goal of simplified systems deployment and
management.
It is clear that a topic drawing so diversely from existing disciplines presents
a serious learning curve for anyone wanting work with autonomic techniques.
Our goal in this article is to reduce this learning curve by presenting a com-
prehensive survey of the current state of research in autonomic communica-
tions. We motivate both autonomic communications and our approach to the
literature in Section 2, and then address the five interlinked perspectives of
the design and analysis of decentralized algorithms (Section 3); the model-
ing, handling and use of context (Section 4); novel and extended programming
approaches (Section 5); issues and approaches for addressing security and trust
(Section 6); and systems evaluation and testing (Section 7). In Section 8, we
highlight some of the emerging trends with a view to informing the evolving
research program and conclude with some observations on the potential impact
of autonomic communications and the fundamental research challenges that
remain.
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

A Survey of Autonomic Communications • 225
2. ISSUES IN AUTONOMIC COMMUNICATIONS
The increasing density of the global communications network offers indus-
try, network operators, developers, and users both dramatic advantages and
significant challenges.
For industry. The need to maintain diverse and complex networks is often a
significant (and increasing) cost of doing business. An infrastructure to reduce
these costs and facilitate new opportunities is urgently needed, and it must
at the same time be sufficiently flexible, robust, and secure for use across the
spectrum of corporate communications.
For operators. Increasing interconnectivity potentially allows improved ro-
bustness and bandwidth, but also increases the complexity of management and
the fragility of protocols in coping with a highly dynamic and largely scale-free
environment composed of diverse networks and technologies. Finer-grained mo-
bility and roaming require that the relationships between operators as well as
between operators and users be extensively rethought.
For developers. Mobile and pervasive networks allow applications and ser-
vices to extend into the environment, both providing and benefiting from sens-
ing capabilities and closer integration with the personal and social goals of
users, but at the cost of massively increased programming and configuration
complexity.
For users. Mobility and ubiquity tilt the balance of communications systems
in the users’ direction, placing individually- and socially-focused adaptations
at the core of the systems architecture, but with the danger that the increased
potential for surveillance and complexity will erode the privacy of individual
and further disenfranchise entire social groups.
Existing network paradigms deal poorly with this multilevel tension between
complexity and simplicity, diversity and ubiquity.
Traditional networks have been constructed and coordinated centrally ac-
cording to a single plan and can consequently be architected using a homoge-
neous population of components with common technical standards and manage-
ment goals. By contrast, next-generation networks are expected to grow more
chaotically with no centrally-mandated goals or levels of service, no universally-
agreed upon protocols or other technical standards, and no a priori knowledge
of the topology or component population. This freeing of central control over
networks has the potential to release an enormous burst of creativity and new
economic activity that is impossible to achieve in a more constrained environ-
ment and, consequently, has the potential to make networking a vehicle of
economic growth and social change.
It is clear, however, that the mathematical, economic, and technical bases of
networking must be changed radically to address the implied challenges. Specif-
ically, the next-generation network must be radically distributed and decentral-
ized, self-describing, self-organizing, self-managing, self-configuring, and self-
optimizing, providing a seamless communications infrastructure composed of
multiple technologies and able to leverage local information and decisions with-
out sacrificing global performance, robustness, and trustworthiness.
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

226 • S. Dobson et al.
2.1 The Emergence of Autonomic Systems
The notion of using autonomic techniques—of deploying technology specifically
to manage and optimize the functioning of other technology on an ongoing
basis—has its roots in work on control theory and managed elements. While
control theory can provide excellent descriptions of closed systems whose com-
ponents and desired properties are known and described by certain classes of
linear or nonlinear mathematical models, it deals poorly with general systems
(e.g., discrete and continuous, time-varying, having delayed or uncertain infor-
mation) even when they can be characterized mathematically. Control theory
encounters even greater difficulty when the system structure is unknown and
is being constantly discovered and modified. Managed elements are essential
for controlling a system but typically require extensive human guidance.
Autonomic design seeks to generalize the control-theoretic view of control
by enabling more flexible and adaptive functions in the underlying system. By
leveraging richer information sources than are typically considered in control
systems, autonomic systems should be able both to react to evolving situations
and, to some extent, preempt expected future demands.
An autonomic system offers an open environment for rapid and dynamic re-
source integration where federations of heterogeneous systems are formed with
no central authority or unified infrastructure, a similar situation to that which
pertains to pervasive and ubiquitous computing. The architecture of autonomic
system, in general considers them as consisting of autonomic elements, each
performing a fixed function and interacting with other elements, possibly in a
very dynamic environment. An autonomic element is commonly viewed as be-
ing comprised of one or more managed elements (also referred to as functional
units) that perform the element’s operational function and an autonomic man-
ager (management unit) that controls the managed elements’ configuration,
inputs, and outputs.
Autonomic systems form a feedback loop (Figure 1). The system collects in-
formation from a variety of sources including traditional network sensors and
reporting streams but also including higher-level device and user context. These
are analyzed to construct a model of the evolving situation faced by the network
and its services with this model used as a basis for adaptation decisions. These
decisions are actuated through the network and will potentially be reported to
users or administrators. The impact of the decisions can then be collected to
inform the next control cycle.
A high profile use of autonomic techniques is provided by IBM’s autonomic
computing initiative [Kephart and Chess 2003]. Autonomic computing is seen
as a way of reducing the total cost of ownership of complex IT systems by
allowing reconfiguration and optimization to proceed on an ongoing basis driven
by feedback on the system’s ongoing behavior. It combines a technological vision
with a business rationale for increasing the coupling between business goals
and IT services.
Autonomic communication, by contrast, generally refers to all these research
thrusts involved in a deep foundational rethinking of communication, network-
ing, and distributed computing paradigms to face the increasing complexities
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

A Survey of Autonomic Communications • 227
Fig. 1. Autonomic control loop.
and dynamics of modern network scenarios. The ultimate vision of autonomic
communication research is that of a networked world in which networks and
associated devices and services will be able to work in a totally unsupervised
manner, able to self-configure, self-monitor, self-adapt, and self-heal—the so-
called self-∗ properties. On the one hand, this will deliver networks capable of
adapting their behaviors dynamically to meet the changing specific needs of
individual users; on the other, it will dramatically decrease the complexity and
associated costs currently involved in the effective and reliable deployment of
networks and communication services.
Despite their evident similarities, there are significant differences between
autonomic computing and communication. While autonomic communication
is more oriented towards distributed systems and services and to the man-
agement of network resources at both the infrastructure and the user levels
[Quitadamo and Zambonelli 2007], autonomic computing is more directly ori-
ented towards application software and management of computing resources.
Nevertheless, both research areas recognize that traditional software systems
are facing a decreasing incremental benefit from technological advances (pow-
erful CPUs, large memories, and so forth) because the complexities of develop-
ment and management are overwhelming the technical gains. Accordingly, the
twin visions of autonomic communications and computing are aligned in iden-
tifying the need for decentralized algorithms and control, context-awareness,
novel programming paradigms, end-to-end privacy management, and compre-
hensive evaluation in order to deliver the desired self-∗ properties.
In the communications arena, the traditional architecture of control and data
planes has been expanded in a number of ways. Clark’s influential vision of a
knowledge plane [Clark et al. 2003] provides architectural support for integrat-
ing low-level (transport and network) knowledge with higher-level applications
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

228 • S. Dobson et al.
and user context. One may also view this from a context-aware systems per-
spective as making the meaning of the operations a network is carrying out
available to influence how it handles those operations [Dobson 2005]. The use
of context from beyond the network makes models situated in that they have a
model of their place in a wider scheme.
2.2 Challenges to Theory and Practice
Networks are traditionally described using a number of theories and technolo-
gies, the most influential of which are classical (Shannon) information theory,
communication theory, queuing theory and the IP suite. However, such foun-
dational insights must also be accessible to the programmers tasked with gen-
erating network services.
What are the core challenges for next-generation networking? Network re-
searchers have to a large extent developed approaches for tackling purely tech-
nical challenges such as bandwidth and authentication. We consider that many
of the core challenges lie at the boundary between networking and what might
traditionally be regarded as systems or applications, including, the following.
Interaction with strangers. Authentication allows one to identify users but
not their motives and can build secure paths between services without deciding
whether the services selected are the most reliable and least prone to informa-
tion leakage. Next-generation wireless, mobile, and ad hoc networks will need
to manage the trust and privacy of users and services end-to-end, without any
a priori knowledge of the parties involved.
Information reflection and collection. In order to decide between competing
demands, the network needs to embody information about the data it is trans-
porting and the uses being made of the data. It must also reflect on its own
behavior so as to, for example, recover from—or preferably predict and avoid—
routing failures and other events.
Lack of centralized goals and control. Decentralization increases the robust-
ness of individual services at a microscale and, as was seen with the emer-
gence of the World Wide Web, encourages new applications and services at a
macroscale.
Meaningful adaptation. Designing an adaptive system implies being able to
gain confidence that the system will adapt correctly to given stimuli, main-
tain key behaviors and avoid deleterious ones. Designers also need to un-
derstand adaptation at a system level so that adaptations that optimize in-
dividual services do not cause undesirable interactions with those of other
services.
Cooperative behavior in the face of competition. Open systems architectures
allow agents to join networks dynamically and both offer and consume services.
The growth of peer-to-peer services and the withering of centralized control and
billing both make it vital that cooperation structures are intrinsically proof
against free-riding and other selfish behaviors.
Heterogeneous services and semantics. Agents cannot guarantee the exis-
tence of particular services or their precise behaviors.
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

A Survey of Autonomic Communications • 229
Fig. 2. Cross-cutting issues
It would be attractive if each of these challenges could be dealt with indepen-
dently, but unfortunately this does not seem to be the case. Routing involves
measuring the network, interacting with untrusted nodes, and applying adap-
tive strategies; service provision must deal with noncooperative behavior, het-
erogeneity, and a lack of centralized direction. And the same seems to be true
for most other communications activities. To borrow a phrase from computer
science, the concerns of autonomic communications are cross-cutting and must
be woven into a coherent solution.
To focus on the cross-cutting concerns is to risk fragmenting the discussion
across accepted technical domains and so, for the remainder of this article, we
focus on five core domains within the research literature: algorithm design, con-
text collection and modeling, novel programming paradigms, trust, and testing
and evaluation. This allows the reader to focus on their area of technical in-
terest while still being exposed to the core cross-cutting concerns of autonomic
communications.
Figure 2 presents a road map to the rest of the article, matching the chal-
lenges against the cross-cutting issues and showing the technical ideas emerg-
ing from each issue when addressing each challenge.
Why these five domains? Although any choice is somewhat arbitrary, we
believe that these best capture the current research themes within the evolving
body of knowledge.
The characteristic features of autonomic communications are the use of
highly decentralized algorithms that have desirable emergent properties while
retaining both a high level of global predictability and a close integration with
cognitive and other contextual goals (issues addressed in Section 3). The emerg-
ing theory and protocols underlying autonomic communications must therefore
build on the classical understanding by synthesizing results from a range of ad-
ditional disciplines that are typically pursued independently.
Some of these topics touch on the most advanced aspects of communication
dealing with the problem of signal coexistence without a priori agreement.
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

230 • S. Dobson et al.
Other areas are more traditionally concerned with programming, reflecting
the fact that the availability of robust link-level primitives can be exploited by
autonomicity to bring more meaning into the network rather than treating it
purely as an uninterpreted channel. We address the issues of collecting and
modeling context in Section 4. The use of overlays and semantically-informed
protocols and algorithms allow us to treat the network in some sense as a
programming language to address specific problems using both programmatic
and communications-driven paradigms cooperatively. We discuss programming
paradigms in Section 5. However, such approaches also raise the importance
of being cognizance of security and privacy concerns in the core of applications
technology, a topic we address in Section 6.
The final issue concerns correctness and validation. Adaptive systems offer
additional failure modes over and above those of traditional systems. While a
traditional system may be tested or proven to be correct, such a point solution is
not acceptable in the face of adaptation. Each of a system’s adaptive behaviors
must be correct; in addition, the adaptations must occur correctly, a situation
referred to as process correctness. The testing and evaluation of autonomic
systems is still very much in its infancy, and we examine some approaches in
Section 7.
3. ANALYSIS AND DESIGN OF AUTONOMIC ALGORITHMS
It has been the rule for many years to assume that communication systems
could be monitored and controlled by a central entity. However, with the growth
of the size and complexity of communication systems, it has become clear that
this assumption is no longer valid. One of the paradigmatic cases of a communi-
cation system that has had to evolve to become autonomic and self-organized is
the Internet whose initial centralized structure has had to evolve for fully dis-
tributed control and management. This trend is slowly expanding to all kinds
of communications systems. It has given way to new models and paradigms
and has open the door to migrating new research techniques from other fields
for use in these systems.
In this context, new algorithms that take into account the complexity of the
communication systems have to be developed and analyzed. These algorithms
will have to guarantee the emergent properties of the system with limited con-
textual information and local control, making the system self-organizing in a
way that is atypical for most algorithm styles. Similarly, in most cases, they
will have to deal with very heterogenous and changing environments and will
be in charge of providing reliability and self-correction capabilities to the sys-
tems. All this has to be attained for systems that may be formed of millions of
different elements, hence requiring an extraordinary level of scalability.
As an example, classical communication paradigms are adapting to these
new models. Classical fault-tolerant communication systems and primitives for
group communication (see, e.g., Chockler et al. [2001], De´fago et al. [2004], and
Birman [2005]) which offered very interesting features such as deterministic
guarantees, self-repair, load distribution, and flow control are being rethought
due to their lack of scalability in a wider and (especially) ad hoc networking
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

A Survey of Autonomic Communications • 231
context. Hence the appearance of group and multicast communication sys-
tems and algorithms with probabilistic instead of deterministic guarantees,
like the Spinglass system [Spinglass Project 2005] or the algorithms discussed
in Eugster et al. [2003]. A similar evolution has been observed in publish-
and-subscribe systems and algorithms (see Eugster et al. [2003]) to yield, for
instance, epidemic approaches like Newscast [Jelasity et al. 2003].
One very desirable design objective would be that algorithms for autonomic
self-organized communication he as independent as possible of the specific com-
munication technology. The context in which these algorithms will have to op-
erate range from classical packet-switching networks, in which routers have to
act and react locally to different network traffic behaviors, to sensor networks
that have to organize themselves in order to operate. In between these extremes
we have, for instance, peer-to-peer and overlay networks with high scalability
issues and mobile and ad hoc networks that are very dynamic. Ideally, the same
algorithms that are used in these set-ups will be useful in other contexts and in
all of them will provide dependability and efficiency. Unfortunately, new algo-
rithms are currently almost always proposed with a very specific kind of com-
munication system in mind which prevents its use in other systems. It would
be of great interest to define general weak models of communications systems
so that any algorithm that works for these models can be used in several real
communication systems of different classes.
3.1 Extending Classical Design Techniques
Most algorithms for large communication systems proposed use classical ap-
proaches but with extensions to deal with (and hopefully profit from) the large
availability of collaborating agents, hence the proposals to use multihop rout-
ing in overlay networks to circumvent Internet connectivity failures such as in
Andersen et al. [2002] that profit from the availability of alternative routes.
Similarly, there is a body of algorithms to build and maintain peer-to-
peer systems based on distributed hash tables (see, e.g., those presented in
Balakrishnan et al. [2003]) and to manage ad hoc and sensor networks [Chrobak
et al. 2004; Chlebus et al. 2005; von Rickenbach et al. 2005; Tschudin et al. 2005;
A`lvarez et al. 2004; Tschudin et al. 2005; Chelius et al. 2005]. Other initiatives
include the use of spiked neural networks distributed throughout the network
routers which learn from online measurements using reinforcement learning so
as to achieve adaptive routing and address the users’ QoS needs [Gelenbe 2004a,
2004b, 2004c; Gelenbe et al. 2001, 2004; Gelenbe and Lent 2004; Gelenbe and
Nunez 2003]. Genetic algorithms have also been used and evaluated to discover
new network paths that offer potentially QoS from previously discovered paths.
However, other completely new paradigms of algorithm design are also being
considered, some of which are in the early stages of development. One such
approach is to base the algorithm’s behavior on nature, which has given way
to a collection of new models, systems, and algorithms (e.g., see Section 5 and
Babaoglu et al. [2005]).
Another line of research is networks with autonomic self-adapting topolo-
gies. A thorough study of the optimal topologies for each system configuration
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

232 • S. Dobson et al.
would lead to a dynamic adaptation of the topology to the system status [Cholvi
et al. 2005; Rodero et al. 2006]. This seems to be a promising line of research
for applying the increasing knowledge on small-world and scale-free networks.
Another area to continue exploring is the application of nonlinear (chaotic)
dynamics to communication systems, mainly to the lowest levels. Chaotic dy-
namics provide a way to optimize performance in DS-CDMA-based links used
in 3G mobile telephony systems [Setti et al. 2002; Kennedy et al. 2000], which
may be extended to other physical- and transport-layer schemes that exploit
simultaneously the code-frequency-time-space diversity that are likely to be
adopted for forth-generation telecommunications systems [Setti et al. 2004;
Rovatti et al. 1998, 2000, 2001, 2004a, 2004b; Mazzini et al. 1997, 1999, 2000,
2001].
Decentralization raises significant consistency challenges which are perhaps
encountered most strongly in the area of distributed databases [Milan-Franco
et al. 2004; Jime´nez-Peris et al. 2002; Baldoni et al. 2004]. Most algorithms
for transactions and other classical approaches to consistency rely on extensive
barrier synchronization which is difficult if not impossible in a decentralized
and low-reliability context.
3.2 Collaboration Issues in Algorithmic Design
As well as the more obvious security and trust issues (deferred until Section 6),
algorithm development in the face of autonomic adaptation poses some unique
problems. Most algorithm approaches assume a certain degree of collaboration
among the different communicating agents. However, in a large communication
system like the Internet, one may expect to have users with different interests
and with the capability of adapting the programs they run on their computers
to match their desired behavior. This will certainly lead to assumptions of ad-
versarial behavior when designing the algorithms. We have examples of this
need in the Berkeley Open Infrastructure for Network Computing (BOINC)
[BOINC Project 2006], used to perform scientific computations with high CPU
time requirements in the computers of volunteers, and which had to introduce
security mechanisms to prevent deception caused by (possibly malicious) wrong
solutions returned from the (untrusted) clients. In this case, the problem is not
hard to solve (probabilistically) by assigning the same computation to several
users and using a voting strategy to choose the good replies [Ferna´ndez et al.
2005, 2006].
The problem becomes harder when there is to be a continuous multiway col-
laboration and we want to have all users collaborating in the overall system
(e.g., a service like the BOINC system without central control). In these models,
it seems natural to introduce game theory into the design of any algorithm that
it is suitable to tamper with. A well thought our algorithm in this context will
guarantee that even a selfish user will obey the rules of the algorithm because
it is in its best interest to do so [Lu¨cking et al. 2004]. An example of game the-
ory in action is the incentives to collaborate included in the BitTorrent system
[Cohen 2003]. Interestingly, BitTorrent can still suffer from free riding [Qiu
and Srikant 2004], which attests to the difficulty of designing selfishness-proof
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

A Survey of Autonomic Communications • 233
algorithms. Game theory is also being considered in most aspects of commu-
nication systems (see, e.g., Lu¨cking et al. [2004], Laoutaris et al. [2004a], and
Siris and Courcoubetis [2004])
Game theory can also be applied to enhance existing protocols. An example
is extending the distributed replication group [Leff et al. 1993] to the case that
individual nodes act selfishly, catering to the optimization of their individual
local utilities. Game theory may be used [Laoutaris et al. 2004, 2004b, 2005]
to derive equilibrium object-placement strategies that can guarantee improved
local utilities for all nodes concurrently as compared to the corresponding local
utilities under greedy local object placement. Such approaches do not suffer
from potential mistreatment problems inherent in centralized strategies that
aim at optimizing the social utility and yet do not require the existence of
complete information at all nodes.
Economic models are another area of significant interest, essentially treating
an algorithm as an economic system. In the area of wireless networks, economic
and game-theoretic models can be combined to capture the interaction between
network control mechanisms operating at different levels, such as power con-
trol, channel selection, and rate control, as well as the interaction between
distributed autonomous devices, in order to jointly optimize their overall per-
formance and share resources in an efficient and equitable manner [Siris and
Courcoubetis 2004; Siris 2002; Siris et al. 2002].
3.3 Limits
Of special interest to the research conducted in this field is the exploration of the
limits to the amount of information that can be exchanged in a communication
system like those considered here and the techniques required to approach
these limits [Mestre et al. 2003; Maduen˜o and Vidal 2005]. This is closely related
to the concept of (network) information theory which studies these issues for
single (multipoint) channel communication. The extrapolation of the classical
(network) information theory to autonomic networks will lead to a new network
information theory discipline.
Finally, we believe that new coding advances at the application level will
strongly influence the design and evolution of future algorithms for communi-
cation systems. Specific new families of codes are Digital Fountain codes [Luby
2002, 2003] and Network Coding [Ahlswede et al. 2000]. These are already
influencing the next generation of algorithms for peer-to-peer content distri-
bution [Byers et al. 2004; Gkantsidis and Rodriguez 2005]. Furthermore, it is
possible that they will influence the future evolution of the Internet and one of
its main protocols, TCP [Byers et al. 2002; Lo´pez et al. 2005].
3.4 Stability and Reliability
Autonomic systems strive to provide self-managing and self-optimizing ser-
vices. The most basic service any network can provide is communication be-
tween nodes with traffic management and routing providing the most visible
target for decentralized algorithms. A basic question for such systems is the
extent to which it is possible to maintain a given quality of service in the face of
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

234 • S. Dobson et al.
changing conditions. This essentially characterizes the stability of the system
under perturbation, and often, specifically, its reliability.
An algorithm designer can adopt two basic positions in the face of a given
quality requirement. On the one hand, an algorithm may adapt its behavior
to achieve the best performance available under the prevailing circumstances.
This may involve trading off conflicting requirements at both network and ap-
plication/user levels, for instance, reducing sound fidelity (at application level)
to maintain video frame rate (a user preference) in the face of packet loss (a
network concern). On the other hand, an algorithm may fail in a controlled
fashion if the required quality of service cannot be maintained. Both of these
approaches might be termed reliable: the former focuses on best-effort behavior
no matter what the circumstances, while the latter will not display an undesir-
able behavior.
Somewhat orthogonally to these two approaches, a designer need to under-
stand how a given stimulus will affect an algorithm’s behavior. Adaptation is
not an excuse for incorrectness: one wants an algorithm’s behavioral “envelope”
to be well-defined, regardless of the possible conditions that occur in the net-
work [Dobson and Nixon 2004]. Allowing free adaptation within an envelope
provides a more precise notion of reliable behavior, while placing clear limits
on the degree to which self-management is allowed to weaken (or compromise)
service guarantees.
3.5 Summary
Traditional algorithms tend to require steering or placement in order to be-
have optimally in a given configuration. The challenge of autonomic self-
management and self-optimization is to provide these optimizations automati-
cally and with dynamic self-reconfiguration as the platform changes. It is also
not possible to assume that all agents will behave cooperatively. Recognizing
these constraints explicitly in algorithm design leads to a stress on algorithms,
which are inherently self-stabilizing, in drawing on insights from other self-
stabilizing systems in games, economics, and biology, and also drawing on
cross-layer requirements and context information necessary to address con-
cerns across the complete system.
4. CONTEXT AWARENESS AND SEMANTICS IN COMMUNICATION
Autonomic communications implies a stronger degree of self-management and
self-optimization than is found in conventional networks which are divorced
from human intervention (and, in many cases, from the possibility of such
intervention). To provide self-management and optimization capabilities, it
is necessary to investigate the context-aware approach to improve network-
ing properties. Software entity, network components, and software agents are
used to collect context information related to the presence, the location, the
identity, and the profile of users and services. A typical context use involves
locating services and users, calling-up services according to user behavior,
providing information for service composition, facilitating ad hoc communi-
cation mechanisms between users, and adaptation of the qualities of service to
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

A Survey of Autonomic Communications • 235
changes in the environment as a result of user and service mobility [Coutaz
et al. 2005].
Two types of context-aware infrastructure can be proposed.
—Passive context-aware infrastructure. Context is raw information that, when
correctly interpreted, identifies the characteristics of an entity. An entity can
be a person, place, a device, or any object that is relevant to the interaction
between a user and the services. Context is a function of time and environ-
ment. The environment is in turn a function of the users, services, resources,
and other entities in the environment. In this phase, the focus will be on the
context gathering and representation. A data model and dissemination pro-
tocol represent, store, and manage context information. This includes classi-
fying context sources, providing a unified context structural representation,
and developing mobile storage strategies with data replication techniques to
insure the availability and the proximity of context information.
—Active context-aware infrastructure. An alternative technique is to extend
passive collection with smart context information delivery. A context-level
agreement protocol can provide automatic context matching with the user’s
profile, terminal capabilities, and service requirements and offering. The pri-
mary aim of such a protocol is the adaptive distribution of context information
among multiple mobile and fixed sources and destinations (e.g. devices, ser-
vice components) using (negotiated) specific dissemination attributes such as
power saving and cost. Context dissemination can be achieved in both “pull”
and “push” modes.
In order to provide self-adaptive behavior, an autonomic system must be
able to reason about both its context and its behavior relative to that context.
This does not necessarily imply a symbolic structure, but does suggest that the
system must be able to reflect on its environment and behavior in some sense
and generate feedback as a result [de Castro et al. 2004]. Thus the autonomic
network must accept goals and constraints from, and petition for the attention
of, its human governors using terms that are meaningful to their needs and
cognitive abilities [Pujolle et al. 2004]. However, this must be balanced against
the need for the autonomic network to map these semantic terms determinis-
tically to and from the self-managing capabilities of its elements. This requires
a high degree of semantic interoperability between the expression of adaptive
behavior and the changing context that drives such adaptation. As context-
awareness and semantic-based reasoning concerns adaptive, networked sys-
tems, it requires research about models and languages for representing their
behavior expressions and methods for adaptation that operate on such, possibly
semantically rich, representations.
The representation of the information in the autonomic network in itself rep-
resents a significant challenge. For instance, in an optimization problem, opti-
mization always occurs relative to some environment or property: one optimizes
for performance, or for robustness, and so forth, with the improvement in one
property often coming at the expense of another. Determining which properties
to optimize requires that the optimizing agent understands the relative priori-
ties that should be given to the various possible optimization targets which, in
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

236 • S. Dobson et al.
turn, implies that the optimizing agent is aware of the meaning of the informa-
tion within the network and its place in the ongoing user tasks. In an autonomic
system, the optimizing agent is software, implying that metadata about the
meaning of information, streams, and operations is injected into the network in-
frastructure and used to inform network-level decisions [Merghem et al. 2003].
Adaptive pervasive computing systems attempt to provide information and
services that are distraction-free for users. Approaches to adaptability may be
loosely classified as closed-adaptive when all adaptations are prespecified and
open-adaptive when new adaptations can be discovered [Oreizy et al. 1999].
Representations of context have varied between subsymbolic uses of neural
networks to more traditional symbolic AI [Henrickcon et al. 2002] with an ap-
parent consensus favoring concept graphs modeled (at least externally) using
the Resource Description Framework (RDF) [Lassila and Swick 1999]. This has
the advantage of providing a well-understood, triple-based representation to-
gether with an open exchange format to facilitate integration into the larger
system context—an important consideration given the small part that even
highly-contextualized services play in an enterprise-scale architecture.
Although pervasive computing is generally regarded as distinct from net-
working, there is significant convergence. A network is essentially a sensorized
system which can observe its own low-level activities and constraints. This may
be combined with higher-level contextual information about users, services, and
applications within a framework of uncertain reasoning.
It is widely recognized that managing the structures of context is a significant
challenge [Coutaz et al. 2005]. Many systems draw a distinction between con-
text (the low-level information observed or inferred about an environment) and
the situation (the high-level scenario in which the system is involved) [Gonzalez
and Ahlers 1999]. This frequently involves combining information at different
semantic levels. The trails model [Driver and Clarke 2004] provides one such
fusion, based on the insight of the nonhierarchical nature of contextual infor-
mation and the need to adopt cross-ontology, cross-layer, and cross-tool views
in order to obtain meaningful results. An alternative approach is to model com-
plete adaptive spaces whose properties may be analyzed a priori for conflicts
and other properties [Jensen and Milner 2003; Dobson and Nixon 2004].
As observed in Section 3, context-driven adaptation must be carefully con-
trolled in order to generate intelligible behavior: context has a direct influence
on users’ ability to form well-founded conceptual models of systems, so adap-
tive behavior can be shown to have a direct impact on usability. For a system,
network, or service to be predictable and usable, there must be a clear link
between adaptations and their environmental causes both in terms of cau-
sation and in the details of the way the adaptation supports working in the
new context [Dobson and Nixon 2004]. The communications industry is accus-
tomed to defining system semantics in a formal or semiformal manner, rang-
ing from SDL and Z for signalling systems, TMN’s GDMO and the DMTF’s
CIM schema for management models, and service-oriented models like ODL
(Z.100), and the TeleManagement Forum’s NGOSS technology-neutral model
for component-oriented communications software. However, the diversity of
these languages reflects the sectorial divisions within the communications
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

A Survey of Autonomic Communications • 237
industry that research into SAC must challenge. This challenge is being ex-
acerbated by the introduction of a variety of multi-agent technologies to the
communication domains, which introduce further language for capturing se-
mantics such as ACLs and KIF.
At the same time, however, the W3C’s semantic Web initiative is addressing
semantic interoperability issues through the standardization of a family of on-
tology languages for encoding knowledge and services on the Web: RDF, RDFS,
OWL, and OWL-S. There has been little attempt to define generic expression
of adaptive behavior, though development on a Semantic Web rule Language
may provide a suitable starting point [de Bruijn et al. 2005].
These technologies benefit from wide acceptance and improving toolsets and
have already been suggested as playing an important role in future communi-
cation architecture [Clark et al. 2003]. However, though some initial work has
been done is applying ontology-based semantics to communications problems
[de Vergara et al. 2004; Lewis et al. 2005], the suitability of these languages for
the demanding scalability and real-time requirements of this domain is yet to
be proven.
Recently there has been a growing number of propositions to model con-
text for context-aware systems in semantic Web languages like OWL [Strang
et al. 2003]. Although, since the introduction of the context-aware term [Weiser
1991], numerous approaches for context modeling have been proposed in com-
munications society (pervasive computing, peer-to-peer, and so forth), the most
popular of which is viewing context as some function or mode of the parameters
of the environment such as time, place, etc. Successful creation of autonomic
communications will require fuller interpretation of context as a dynamically-
changing concept rather than static one [Sterritt et al. 2005].
4.1 Summary
Context models provide explicit representations of concerns from a number of
different semantic levels. While not all algorithms require explicit context mod-
eling in order to exhibit self-managing behavior, a context model can uniformly
inform autonomic decision-making across the spectrum, allowing whole-system
self-optimization. It also provides a useful approach for open-adaptive behavior
and collaboration across tools through standard formats and protocols.
5. NOVEL COORDINATION AND COMMUNICATIONS
PROGRAMMING MODELS
The core challenges of autonomic communications calls for novel paradigms of
communication and coordination and, consequently, calls for novel modeling
approaches and novel supporting infrastructures and programming languages.
5.1 Inadequacy of Traditional Models
Traditional paradigms based on approaches such as message-passing, client-
server, or distributed shared memory—on which most practice of distributed
programming has relied so far—appear inadequate when dealing with the new
core challenges being faced.
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

238 • S. Dobson et al.
First, traditional programming models typically rely on static assumptions
and a priori knowledge about the system, that is, spatial-temporal coupling
[Cabri et al. 2000] and referential awareness [Tanenbaum 2004]): components
are assumed to live in the same network at the same period and are assumed
to know each other. Such assumptions can hardly apply to modern and future
network scenarios where interaction with strangers is the norm and where the
dynamics of the network (due to the presence of mobile and ephemeral nodes)
and its lack of centralized control make the adoption of any a priori strategy
useless. Rather, suitable programming and coordination models must assume
a dynamic treatment of components’ identification and location and must pay
attention to overall collective behaviors and coordination rather than individ-
ual behaviors and pairwise interactions. While Web services provide a partial
approach to this problem, they still rely to some extent on shared assumptions
about partners’ semantic capabilities in a manner that is in some senses very
similar to that of traditional distributed middleware [Baker and Dobson 2005].
Second, traditional models such as message-passing or distributed shared-
memories do not account in any way for context-awareness and meaningful in-
teractions [Mamei and Zambonelli 2004]. Put simply, components are assumed
to live in a void, where the only things that exist are the other components (or
some shared portions of memory). Such models do not enforce per se any form
of context-aware computing or context-aware interactions nor do they account
for the presence of infrastructures to support context-awareness (as from previ-
ous section). Clearly, this makes it very hard to enforce reflective behaviors via
analysis of current context. Also, it prevents (both conceptually and in practice)
the enforcement of distributed applications, and it services all those properties
of self-configuration, self-adaptation, self-healing that by definition (having to
rely on phenomena of adaptive self-organization) are need the modeling of com-
ponents situated in some environments and capable of reacting to its properties.
Third, and again strictly related to the adaptation issue, the previously
discussed traditional models rely on a traditional layered perspective of com-
munication systems. This typically prevents the communication medium from
adapting to network and application dynamics. On the one hand, the layered
architecture makes the higher layers blind with regard to underlying chang-
ing conditions (such as a bandwidth reduction caused by a network glitch); on
the other, lower layers are unaware of the kinds of services in which they are
involved, and so they cannot customize their activities accordingly (e.g., by the
transport layer switching autonomously between TCP and UDP depending on
the application it is supporting). The vision of autonomic communication affects
both the lower network layers—calling for programmable components capable
of adapting their behavior in a concerted way to provide autonomic features—
as well as the higher layers (from transport to application), in that software
components have to interact in a very dynamic world. This implies rethinking
the traditional layered network models: strong cross-layer interactions between
application components and network components are required to: (1) have ap-
plications access and underztand low-level information about the situation of
the network and vice versa; (2) achieve cross-layer tuning of their respective
behaviors that enables services to adapt to the current network characteristics;
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

A Survey of Autonomic Communications • 239
(3) conversely, to have the network understand the current needs of services and
adapt to them, and thus achieve overall orchestrated activity of the network as
a whole.
The previous considerations provide a compelling case for novel program-
ming and communications approaches in that they must enable a vision and a
programming style in which components must be able to interact in dynamic
scenarios, where the distinction between network and service components is
blurred, and where network components will become an integral part and inter-
act with those software components that execute on them in a semantic world.
5.2 Uncoupling Coordination
Although not explicitly conceived to face the challenges of autonomic commu-
nications, some well-assessed coordination paradigms exist that, by enforcing
uncoupled and dynamic coordination, can to some extent suit modern network
scenarios a bit more than message-passing or distributed shared memories do.
—Event-based models. In event-based publish/subscribe models, a distributed
application is modeled by a set of components interacting with each other by
generating events and by reacting to events of interest. This clearly supports
flexible and uncoupled interactions which are suitable for a dynamic network
scenario.
Modern distributed event-based approaches (such as Milan [G. Picco 2003;
Eugster et al. 2003] or Siena [Carzaniga et al. 2001]) also support an event-
based style of programming by providing distributed event-dispatching ser-
vices in the context of mobile ad hoc networks in which mobile nodes engage a
distributed algorithm to self-organize event-dispatching routes and to main-
tain such routes despite network dynamics.
Clearly, there is considerable scope for adaptation within such systems, as
the event service can abstract most of the network issues. However, top-down
influences are weak, and typically the event service will support only a small
number of interaction patterns efficiently, while several phenomena of self-
organization can hardly be mapped in terms of publish-subscribe patterns
because of the lack of any high-level concept of environment.
—Tuple-space models. Tuple-space-based coordination models exploit localized
data structures (tuple spaces) in order to let agents deposit information,
access information via a pattern-matching mechanism, and thus achieve
both some forms of context-awareness and the possibility of interacting with
unknown agents in a fully uncoupled way. These are indeed desirable char-
acteristics for an autonomic communications programming and coordination
model.
Although early proposals consider centralized tuple space or a limited set
of well-localized distributed tuples space, more recent proposals adapt the
model to better fit dynamic network scenarios. Systems such as Lime and
Egospace [Curino et al. pear; Picco et al. 2001; Roman et al. 2002]) exploit
transiently-shared tuple spaces as the basis for programming interactions
in dynamic network scenarios. Each mobile device, as well as each network
node, owns a private tuple space. Upon connection with other devices or with
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

240 • S. Dobson et al.
network nodes, the privately-owned tuple spaces can merge in a federated
tuple space to be used as a common data space to exchange information.
Tuple-space models can be encapsulated within higher-level environments,
for example, within Sun’s Jini framework. There remain issues in providing
efficient and fault-tolerant implementations [Rowstron 1999], especially in
highly-dynamic environments as the semantics of tuple-space interactions
require extensive use of synchronization. Also, tuple space models can hardly
be used to effectively program and enforce phenomena of self-organization
and self-adaptation. In fact, although the tuple space can act as a sort of
shared environment, its lack of structuring make mapping self-organization
phenomena on it very hard.
5.3 Emerging Models
Beside event-based and tuple-based models, a variety of research groups have
started proposing a number of innovative communication and coordination
models, and the associated programming languages and infrastructures which
appear much more suitable to the needs of future situated and autonomic com-
munication scenarios. In a word, all these innovative models build on the lessons
of event-based and tuple-space models and enrich them by giving meaning
and/or some structure to interactions.
—Field- and morphogen-based models. Field-based approaches (for example,
Co-Fields, TOTA and MMASS [Bandini et al. 2002; Mamei and Zambonelli
2004; Mamei et al. 2004]) can be regarded as a general framework to program
and engineer coordinated behaviors in dynamic and distributed computing
systems. The key idea in field-based coordination is to have components’
actions driven by computational force fields generated by the components
themselves and/or by some infrastructure and propagated across the envi-
ronment according to specific propagation structures. To some extent, fields
can be considered as sorts of distributed data structures that can play both
the role of events and of shared data with the added value of giving a meaning
to their distributed structuring over the network.
Field-based approaches enable the programming of adaptive and effec-
tive coordination schemes ranging from motion coordination to routing
in dynamic networks. Middleware infrastructures like TOTA [Mamei and
Zambonelli 2004] allow services to define and propagate field data struc-
tures across a dynamic network and to maintain such data structures
automatically in the presence of local failures. If one of the nodes support-
ing the field fails, for example, the field will reconfigure to reflect this local
change in the global configuration of the field, while local instabilities are
damped at the global level.
Morphogen-gradient approaches (e.g., Stoy and Nagpal [2004a, 2004b])
draw their inspiration from the original works on the amorphous computing
project [Abelson et al. 2000]. They propose driving the activities of auto-
nomic components in dynamic networks by means of data structures sim-
ilar in concept to fields and define specific gradient-oriented programming
languages accordingly. The primary application scenario addressed is that
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

A Survey of Autonomic Communications • 241
of pattern formation among mobile robots and computational particles, but
the model also finds useful applications in sensor networks and pervasive
computing.
—Biologically-inspired models. There is an increasingly widespread agreement
within the community that biologically-inspired (or biomimetic) solutions are
likely to play a key role in autonomic computing and communication (see, e.g.,
Babaoglu et al. [2005]. Indeed, biologically complex systems tend to exploit
fully decentralized and uncoupled coordination models, relying primarily on
environment-mediated local information transfer. This translates into desir-
able properties such as scalability, adaptability to changing conditions and
dynamic scenarios, and robustness to partial failure and/or hostile disruption
of normal activity.
A historical example of using biologically-inspired models to devise original
and efficient solutions to relevant routing and scheduling problems in net-
works is provided by the so-called swarm intelligence or ant colony paradigm
[Bonabeau et al. 1999; Merloti 2004]. However, many studies in the field
are explicitly targeting toy problems like, for instance, the Travelling Sales-
man Problem [Dorigo and Gambardella 1997] or the Graph Coloring Problem
[Costa and Hertz 1997]. Such approaches often incorporate some form of re-
inforcement learning as in the distributed neural network approach that has
been implemented in a large packet network testbed [Gelenbe et al. 2001;
Gelenbe 2004a]; in view of the encouraging experimental results that have
been already obtained with this bio-inspired approach, further investigation
of the interaction between neural network control and packet networks seems
well justified. Other very encouraging results on concrete applications (e.g.,
server farm management [Nakrani and Tovey 2004]) have been obtained
using Monte Carlo simulation techniques but have never been tested exper-
imentally.
So, beyond identifying the many similarities between the problems faced
by autonomic distributed systems and those already solved by their biolog-
ical counterparts (see, e.g., Shackleton et al. [2004]), there remains a criti-
cal need for in-depth, quantitative investigation of the performance of spe-
cific biomimetic algorithms in a practical deployment scenario. This is made
especially challenging by the fact that evaluating the complex system prop-
erties of bio-inspired solutions requires a paradigm shift from deterministic
to statistical predictions (see, e.g., Bullock and Cliff [2004]), which entails ac-
cepting some level of uncertainty as far as the behavior and fate of individual
system components is concerned. In practice, using biomimetic techniques in
artificial systems requires strict evaluation of the cost of trial-and-error ap-
proaches to problem-solving, as well as that of losing some units in the process
of system self-organization, both fundamental features of most biological mod-
els from morphogenesis to collective phenomena. Temporarily increased delays
and waste of resources (bandwidth, storage, CPU) must be adequately compen-
sated by improved long-term efficiency and/or responsiveness to unpredictable
fluctuations for biologically inspired solutions to outperform conventional, cen-
tralized alternatives.
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

242 • S. Dobson et al.
In the context of network-based services, Saffre and Blok [2005] describe a
biomimetic peer-to-peer provision framework (SelfService) based on on-demand
service instantiation and featuring emergent load-balancing. However, the high
rate of creation and termination of local access points, as well as the demands
on bandwidth incurred by broadcasting requests in the early stages of the sys-
tem’s evolution toward steady state, dictate that SelfService is only a viable
option if fluctuations in demand are genuinely unpredictable and access point
creation/termination is relatively lightweight. So, in this particular case, the
biomimetic approach could be comparatively powerful in a highly-dynamic, low
security environment, but probably not in a more static resource-sharing sce-
nario or if confidentiality considerations require a participating peer to be taken
offline and scrubbed of any sensitive data each time that it stops hosting a par-
ticular service.
Stigmergic approaches (e.g., Anthill and Swarmlinda [Babaoglu et al. 2002;
Menezes and Tolksdorf 2003; Omicini et al. 2004; Trianni et al. 2004]) rely on
stigmergic coordination derived from interactions in insect colonies to drive the
activities of autonomic application components in dynamic networks. As an ex-
ample of this class of approaches (based on artificial ants), Anthill supports the
design and development of adaptive peer-to-peer applications by relying on dis-
tributed mobile components (ants) that can travel and can indirectly interact
and cooperate with each other by leaving and retrieving bunches of informa-
tion (to act as synthetic pheromones) in the visited hosts. The key objective
of anthill is to build robust and adaptive semistructured networks of peer-to-
peer services by exploiting the capabilities of ants to reorganize their activity
patterns accordingly to the changes in the network structure. As another ex-
ample, SwarmLinda is an ant-inspired system to program distributed applica-
tion components that can adaptively coordinate with each other. Application
components on the Internet can access a global distributed tuple space that
is realized by a set of independent local tuple spaces to retrieve and deposit
information. Swarms of ant-agents that represent tuples or templates roam
across the network of spaces performing a kind of foraging activity that cre-
ate routes to guide application components in accessing the proper tuple-space
location.
—Probabilistic and metabolic approaches. Epidemic and probabilistic ap-
proaches [Castro et al. 2003; Costa and Picco pear; Eugster et al. 2004;
Eugster et al. 2004], while not directly related to programming, aim at over-
coming the burden related in maintaining global data structures such as rout-
ing tables and pheromone paths over dynamic networks, as may be required
in stigmergic and field-based approaches, and thus may impact on nearly all
models presented previously. They propose relying on epidemiology theories
to provide a probabilistic guarantees that data structures and routes will be
eventually maintained.
Metabolic approaches such as Fraglets [Tschudin and Yamamoto 2004]
apply a chemical execution model to the implementation of communica-
tion protocols. One guiding question is the creation of robust execution cir-
cuits which can distribute over a dynamic network and which continue their
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

A Survey of Autonomic Communications • 243
service despite parts of the implementation being knocked out. Like pack-
ets that can be lost (which can be recovered by the appropriate protocols),
it is possible to envisage an environment where parts of a protocol’s execu-
tion can be lost. The remaining implementation elements should continue
to operate and be able to recover by themselves for restoring full services
again.
—Structurally-based approaches. A very general issue in distributed program-
ming of large systems is the possibility for designers to predict and guarantee
that a service will exhibit and maintain certain desirable properties over its
lifetime regardless of any adaptations it might make. One approach is to
encode explicitly the context to which a system will adapt and to derive its
adaptive behavior in a way that respects both this structure and the over-
all goals of the service. Early research on using category theory to describe
adaptive behavior [Dobson and Nixon 2004] suggests that such approaches
may combine adaptivity with stronger guarantees on the adaptive envelope
of systems, although this remains to be demonstrated in larger cases and
may not provide sufficient dynamism for many applications.
Spatial and environment-based approaches, of which approaches based on
overlay networks are a specific case [Bandini et al. 2002; Castro et al. 2003;
Rao et al. 2003; Weyns et al. 2005; Zambonelli and Mamei 2004], propose to
exploit spatial and environmental abstractions as a primary means to drive
components interactions. In these approaches, spatial concepts are realized
by means of self-organizing and self-adapting overlay data structures that
provide components with context information suitable for driving and coor-
dinating their activities. Overlay data structures are distributed data struc-
tures that generalize the idea of overlay networks [Ratsanamy et al. 2001,
2002; Rowstron and Druschel 2001]. Overlay networks provide distributed
routing management, providing components with a suitable application-
specific or network-specific view of the network (e.g., providing the perception
of a specific, application-specific overlay topology of the network). To some
extent, spatial approaches can be considered as a specific instance of the
general concept of semantic-oriented autonomic communications in which
the adaptive space is limited to physical (metric) spaces.
5.4 Summary
Autonomic communications presents very different challenges to traditional
desktop, server, or embedded programming, and it is perhaps unsurprising
that novel programming models are evolving to meet them. Such models can
provide a useful substrate on which to construct autonomic control systems and
algorithms, and, in many cases, may embody the structure of a given class of al-
gorithms to simplify their construction. In and of itself, however, a programming
model will not give rise to self-organizing or self-optimizing behavior without
suitable algorithm design and context modeling: it is important that the model
complements the algorithms and vice versa, and it is unclear which models will
prove most suited to given applications.
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

244 • S. Dobson et al.
6. TRUST AND SECURITY
As mentioned earlier (Section 2), the components of an autonomic system are
typically regarded as consisting of a functional and a management component
with the latter taking responsibility for monitoring and influencing the be-
havior of the former. In this architecture, security tasks are performed by the
management unit governed by the element’s and the system’s policies, of which
security policies are a subset [Chess et al. 2003].
Setting aside the standard properties of integrity, confidentiality, and au-
thentication residing on message and network levels and assuming the avail-
ability of standards and protocols for that (e.g., WS-Security, SSL etc.), new
security challenges are placed ahead by autonomic communication.
6.1 Identity Management
Essentially all security properties and services—integrity, confidentiality, au-
thentication, trust, reputation—hinge on identity. There is hardly any point
in encrypting communication if we are not sure who we are talking to. While
in a static scenario digital identity management does not present much of a
problem, it emerges as an important issue when autonomic nodes dynamically
join different alliances.
A very widespread technique of identity management is the Single Sign-On
(SSO) mechanism. The main idea behind SSO is to eliminate the need of storing
and remembering multiple users’ IDs and passwords for each online service by
pushing the burden onto a trusted identity provider (IP) for a primary authen-
tication that is then forwarded to the partners offering the desired services.
OASIS Security Assertion Markup Language (SAML) [OASIS Security Ser-
vices TC 2004] is an open XML-based security standard that provides a
way of exchanging user authentication information. SAML on its own is the
most widely used standard for bilateral identity management. It offers one-
off SSO relationships in which two partners establish an SSO with each
other.
Microsoft .Net Passport is a proof-of-concept user identity management in-
frastructure. It takes a centralized approach and associates a unique ID with
every user (mapping this ID to the user’s personal profile) that is used for
signing in and accessing all online services that are also part of the .Net infras-
tructure. All users’ personal data is centrally stored on Microsoft servers which
play the role of IDs.
The Liberty Alliance project and WS-Federation [IBM, Microsoft, BEA, RSA
Security and VeriSign 2003] take a decentralized approach for cross-domain
identity management. It enables a multilateral federation of partners sharing
the same domain (circle) of trust. Each federation supports multiple identity
providers and within a federation (circle of trust) a user may traverse all in-
volved partners’ services with a single authentication. Liberty’s specifications
are based on SAML standard and extend it with a number of protocols and
features enabling multilateral identity management, while WS-Federation re-
lies on WS-Trust, WS-Policy, WS-SecureConversation etc. for describing trust
relationships and policies of entities in a federation.
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

A Survey of Autonomic Communications • 245
In a single federation, each service provider is responsible for the manage-
ment and enforcement of its own security policies with a high degree of auton-
omy. Hence, for many services, no partner can guess a priori what will be sent
by clients and clients may not know a priori what credentials are demanded
for completing a service, which may require the orchestration of many different
autonomic nodes.
The work on interactive access control [Koshutanski and Massacci 2004a,
2004b] proposes that servers should be able to get back to clients asking for
missing credentials, whereas the latter may decide to supply or decline the
requested credentials and so on until a final decision is taken. One may also
analyze the causal dependencies between software agents using techniques
such as versioning vectors [Almeida et al. 2002].
Though there are a number of industrial proposals for identity solutions,
such as those described, they do not cater well to dynamic autonomic scenarios.
Therefore a research challenge is to adapt and apply the existing technology to
the case of autonomic communication.
6.2 Trust Management and Negotiation
Trust is an important aspect for making decision on security in information
systems, particularly influencing the specification of security policies. Trust
management is an approach to managing distributed access control by combin-
ing policies, digital credentials, and logical deduction.
Since in an autonomic network there are multilateral communications
among self-managing and self-preserving partners, there is a pressing need for
suitable models/schemes for establishing and maintaining trust relationships
between those partners. The highly dynamic nature of autonomic systems re-
quires novel dynamic trust management models [English et al. 2003a, 2003b;
Cahill et al. 2003; Terzis et al. 2004] for establishing trust relationships and
managing access rights.
Capability-based systems approach distributed authorization by basing their
access decisions on the user’s capabilities (access rights) expressed as digital
credentials. So management of credentials emerges as the key issue for a dis-
tributed authorization framework, and credential-based access control [ITU-T
2001; Ellison et al. 1999; Thompson et al. 1999; Chadwick et al. 2003; Park and
Sandhu 1999; Karabulut 2003] becomes a suitable model for a trust manage-
ment system.
A number of frameworks have been developed for designing trust manage-
ment systems. They mainly focus on different aspects of trust by adopting dif-
ferent notions of trust relationships and so implementing different mechanisms
for propagating trust and deducing new security statements. The key focus of
these proposals is usually the policy and credential language as in KeyNote, Pol-
icyMaker and REFEREE [Blaze et al. 1999]. A number of later proposals have
refined the languages used for policies, single credentials, or hierarchies thereof
and for their evaluation [Li et al. 2002; Yao 2003; Becker and Sewell 2004].
Extending the concept of trust management from the level of transactions
and message exchange to the context of the semantics of communications opens
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

246 • S. Dobson et al.
up approaches to establish trust by agreeing on an upper-layer ontology. The
challenge is to find a level of abstraction that encompasses the range of concepts
used by different mechanisms to establish trust but which captures enough se-
mantics to usefully support interoperation between those different approaches
at a later date.
The last few years have seen the emergence of a new concept in trust manage-
ment methodology called trust negotiation [Seamons and Winsborough 2002;
Yu et al. 2003; Bonatti and Samarati 2002; Winsborough and Jacobs 2003;
Koshutanski and Massacci 2004c]. It enables iterative disclosures of creden-
tials between a requester and a provider in order to establish the necessary
level of trust to allow the exchange of data. This makes it particularly suitable
for autonomic communication systems.
Further, digital reputation methods (for preliminary results and related work
see Garg et al. [2004] and Michiardi and Molva [2002]) will be used for the
continual self-monitoring of autonomic entities (AEs) and the services they
provide. This system will rely on the dissemination, throughout the network,
of trust information gathered through transactions between AEs. In this way
AEs can build knowledge about the behavior of other AEs (with whom they
may never have interacted before) and the available services. This information
can then be used to decide whether to interact with another AE and whether
to continue supporting an existing service or to start supporting a new service.
This self-monitoring is done at two levels. The first level monitors long-term
actions such as the offering of new services and the discontinuing of unwanted
services. The services will need to be continually evaluated for the utility they
provide and the resources they consume. The second, short-term level is more
reactive with a shorter response time that will incorporate mechanisms to re-
alize services, improve them (e.g., by appropriating more resources for services
that are more popular), identify misbehaving AEs, modify resource allocation,
etc.
6.3 Self-Protection and Self-Healing
Recent work in security management has revealed the necessity of designing a
new generation of self-adaptive security solutions. In this context, these solu-
tions can be based on multi-agent systems and intelligent agent technology. An
example of this approach is the work of Gelenbe et al. [2001] which proposes and
evaluates a scheme for denial of service detection and defence based on a self-
healing autonomic approach, using the Cognitive Packet Network paradigm.
Biological models of resilience provide an analogy with nervous and immune
systems in biological organisms. A nervous system provides sensing (problem
detection) and self-protection through reflexes (autonomic responses). An im-
mune system is responsible for anomaly detection (self versus non-self) and self-
healing. Effective artificial immune systems have been developed which very
closely model elements and processes of natural immune systems, such as lym-
phocytes, their generation, maturation, circulation, binding to pathogens, and
activation, as well as both primary and secondary immune responses [Hofmeyr
and Forrest 2000; Esponda et al. 2004].
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

A Survey of Autonomic Communications • 247
Research on traditional intrusion detection systems in the context of au-
tonomic communications is looking at knowledge representation of service
providers, consumers, services, and threats. Object-oriented and ontology-based
models have been proposed [Undercoffer et al. 2003; McGibney et al. 2005].
6.4 Self-Organized Public Key Management
In general, the use of public key cryptography requires the presence of a central-
ized certification authority. However, such an authority (and the infrastructure
it requires) is incompatible with the decentralized nature of autonomic com-
munications. To solve this problem, some authors propose an approach similar
to the web of trust of Pretty Good Privacy (PGP) [Abdul-Rahman 1996] in the
sense that users issue certificates for each other based on their personal ac-
quaintances. However, unlike the PGP, certificates are stored and distributed
by the users themselves in a completely self-organized fashion. When two users
want to verify the public keys of each other, they merge their local certificate
repositories and try to find appropriate certificate chains within the merged
repository that make the verification possible. The success of this approach de-
pends very much on the algorithm for the construction of the local certificate
sets and on the characteristics of the certificate graph, that is, a graph whose
vertices represent public keys of the users and the edges represent public-key
certificates issued by the users. The analysis of the two typical algorithms shows
that even a simple construction algorithm can achieve high performance. More-
over, the certificate graph exhibits small-world features so that good scalability
of the approach is obviously expected.
6.5 Summary
Effective trust management is vital to the acceptability of highly pervasive ap-
plications and networking. Traditional trust and security models are highly
centralized, while autonomic systems require significantly more decentralized
approaches if they are to offer sufficient self-management. In particular, mech-
anisms are needed to establish rich collaborations between agents—not gener-
ally people—in a way that does not assume bona fides and allows autonomic
identification of unacceptable patterns of behavior.
7. EVALUATION AND TESTING
Autonomic networks open a new area of investigation for experimental net-
work evaluation because, contrary to research on conventional IP networks, it
is no longer sufficient to interconnect systems via existing wired, optical and
wireless modalities, and to measure the effect of carrying perhaps novel traffic
flows in the presence of incremental changes in the protocols. In autonomic
networks, both the user’s perception of context awareness, and the lower-level
network perception of autonomic access and management of resources need to
be addressed using novel approaches, and the methods used to evaluate them
need to be sufficiently pragmatic and empirical to be convincing.
Much of European academic research in computer and communication net-
works has been based in the past on the use of theoretical tools such as queueing
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

248 • S. Dobson et al.
models, software models and studies of protocols, and simulation studies. Net-
working research in industry and at telecommunications operators has used
similar approaches but, in addition, it has widely benefited from experimental
facilities and the possibility of testing experimental networks and measuring
networks which are already deployed. These differences have been justified in
the past by the high costs of hardware related to experimental networks and
to the need for software and system development when one builds or modi-
fies experimental systems. However, this balance has now been modified by
the advent of open-system platforms and the possibility we have of using low-
cost off-the-shelf hardware to design and interconnect network routers. Thus,
the threshold of resources needed to conduct meaningful research on network
evaluation using testbeds is now definitely lower (probably by an order of mag-
nitude) than a decade ago. Several recent projects in Europe consider low-cost
wireless testbeds [Vidales et al. 2005; Tsarampopoulos et al. 2005] or specific
fiber optics networks with local connectivity that could support experimenta-
tion on adaptive autonomic networks. Other work examines multimedia traffic
[Magedanz et al. 2005], while wireless power management and QoS are in-
vestigated in some existing projects from the autonomic perspective [Gelenbe
and Lent 2004; Gelenbe et al. 2004]. Precise traffic measurement on long haul
networks is considered in Morato´ et al. [2005]. An ambitious project is cur-
rently planning a large regional wired/optical and wireless testbed that would
support a variety of social activities [Carreras et al. 2005]. Finally, usability
issues for autonomic network in Europe could be developed as an extension
of work concerning virtual environments [O’Neill et al. 2005, 2006]. On the
other hand, the sheer size of US research in this area, with the possibility
of interconnecting hundreds of nodes, and the sophistication of some of the
Japanese testbeds with respect to context awareness and usability of autonomic
networking environments of some experimental US and Canadian projects
[Takai et al. 2005; Ionescu et al. 2005] should guide us to a higher level of
ambition.
The Ubiquitous Home [Yamazaki 2005] project in Japan has created a wire-
less home where context-aware services are offered dynamically and ubiq-
uitously. The self-aware network project at Imperial [Gelenbe et al. 2004]
investigates how lower-level networking functions can be implemented us-
ing context-aware measurement-based techniques combined with an adaptive
neural network-based reinforcement learning. Between these two extremes
at the high (application) and low (network) levels, there remains very sub-
stantial work to be done in all areas of the performance evaluation of au-
tonomic networks from services to protocols and from robust networking to
QoS.
7.1 Summary
In many ways, existing network testbeds provide a good basis for comparing
autonomic systems with other, more traditional approaches. The use of open
standards and formats simplifies simulation and allows realistic scale simula-
tion to answer quite complex questions about an autonomic system’s reactions
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

A Survey of Autonomic Communications • 249
to stimuli, although this realism may be limited by the fidelity of modeling the
user interactions.
8. CONCLUSION
The notion of autonomic computing arose in specific reaction to the increasing
cost of ownership of enterprise-grade systems and is leading to significant in-
novations in systems administration and application configuration. However,
emerging applications in domains such as pervasive computing, ad hoc network-
ing, and wireless sensors must place equal emphasis on their communications
elements whose exposure to change is at least as great as that of individual
computing elements. Autonomic communications encompasses a range of tech-
niques whose application impacts computing and communications equally.
In this article, we have surveyed the state-of-the-art in autonomic communi-
cations from five complementary perspectives. Following, we draw some specific
trends. As a general conclusion, however, these (and other) techniques funda-
mentally change the ways in which those designing, implementing, deploying,
administering, and using highly-distributed adaptive systems will interact with
those systems in the future. The emphasis is clearly shifting away from sys-
tems which are developed against a set of requirements agreed a priori, and
towards platforms that can adapt to the changing demands placed upon them
with greatly reduced human interaction and steering. Although more complex
in the development phases, such systems offer enormous labor, complexity, and
cost savings over the medium-and long-terms.
Does autonomic systems engineering constitute a discipline? Probably not:
there are as yet few techniques deriving directly from the study of autonomic
regulation of computing or communications. There is, however, a considerable
body of knowledge arising from the ways in which techniques from different
disciplines can interact to provide emergent properties and other autonomic
features. This in turn may generate insights that would not arise from the
individual disciplines in isolation.
8.1 Emerging Trends and Research Priorities
We conclude by extracting the trends and challenges emerging from our fore-
going survey of the autonomic communications landscape.
Decentralization, complexity and analysis. Traditional algorithm design
has focused on issues such as space- and time-complexities, but such analyzes
often require assumptions that are hard to guarantee in open environments.
Of particular significance is the breakdown of the assumptions of cooperative
behavior, which is being replaced by less trusting models. It seems likely that
game theory and economics-derived models of interaction will increasingly be-
come core components of algorithm analysis.
Classical (Shannon) information theory has provided an excellent basis for
modeling and reasoning about slowly-changing networks. Autonomic systems
do not, however, respect the Shannon view of communications on uninterpreted
channels: in an autonomic system, the message affects the medium so as to
improve the latter’s ability to function. There is as yet little understanding of
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

250 • S. Dobson et al.
the impact this will have on information-theoretic properties and, in particu-
lar, on the limits of the channel in terms of information flow and robustness.
Understanding these issues will require significant foundational research and
will inevitably involved a richer model of how information is described and
manipulated end-to-end throughout a network.
Decentralization is the sine qua non of autonomic systems: any centralized
resource or control point will act as a brake on a system’s ability to adapt, es-
pecially in terms of robustness of performance. While this statement is hardly
contentious, it is important to understand that decentralization is poorly under-
stood at the algorithmic, analytic, and programming levels. New techniques are
urgently needed to understand the exact robustness, performance, and complex-
ity characteristics of decentralized algorithms. Moreover, the existing repertoire
of programming techniques must be significantly enhanced to escape from the
tendency—sometimes very well concealed—to inadvertently introduce a cen-
tralized element into even the most distributed computation. Even superficially
innocuous techniques such as iteration can conceal problems which almost de-
mand a centralized resource of some kind.
Indeed, one might reasonably argue that decentralization is a goal worth
pursuing even at the cost of performance. Although many people advocate (for
example) peer-to-peer solutions on scalability grounds, such solutions are often
intrinsically more able to withstand adaptation, treating service relocation and
other changes as failures to be recovered from.
Cross-layer impacts. The traditional layered models of networks, while
valuable conceptually and pedagogically, were never realistic implementation
strategies, and this is even more true for autonomic systems. Layering breaks
up the holistic nature of context and reasoning, whereas valuable strategies can
derive from the ability to correlate (for example) traffic-level properties against
the applications that generate that traffic. A more integrated model of network
monitoring and control need not generate awkward dependencies and loops,
although that is, of course, a significant danger that must be avoided.
The use of context information is no longer confined to pervasive computing,
and is instead becoming a key part of systems design. Techniques developed
for pervasive applications can be applied to other, less interactive systems: the
core innovations are in the use of sensor fusion and uncertain reasoning, where
sensor is taken generally to mean any data-collecting element that can provide
information about the context of a system. (Contextor is another, perhaps less
loaded word.)
Many autonomic systems have only weak guarantees on the properties they
present. They may not, for example, be able to guarantee delivery of messages
under all circumstances or be able to bound end-to-end properties such as
latency or security. In part this is derived from the inadequate modeling mech-
anisms mentioned previously; in part, it derives from the increased dynamism
in the protocols and management approaches being utilized. However, users’
satisfaction with a system derives largely from such end-to-end properties.
Delayering and other techniques must be improved to allow better end-to-end
guarantees.
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

A Survey of Autonomic Communications • 251
A consequence of this is that the traditional reductionist strategies for de-
signing, constructing, and managing systems break down as systems strive to
become more autonomic. The self-∗ properties are inherently holistic and re-
quire whole-system treatments to be developed. This poses a significant chal-
lenge for both theory and practice: autonomic systems need theories that can
handle information at multiple levels of abstraction that are traditionally de-
scribed using completely different formalisms, while any resulting theoretical
analysis must be suitable for being maintained dynamically and robustly within
an operating network.
Paradigms. Traditional programming languages evolved in a milieu radi-
cally different from that targeted by most autonomic systems, and it seems
unlikely that their abstractions and constructs will be adequate for dealing
with the emerging challenges. The range of new programming styles will re-
quire innovation in the core of programming language design, not simply new
libraries.
Biologically-inspired models are likely to play an increasingly important role
as the very concept of autonomic computing is itself inspired by biological mod-
els such as the autonomic nervous system. Swarm intelligence has been applied
to routing and scheduling in autonomic communications, and artificial immune
systems provide self-healing and self-protection. Biological models also tend to
exhibit scalable convergence properties which can simplify analysis, but the in-
teractions between subsystems can remain surprisingly subtle and need more
precise characterization.
Trust and security have never evoked the public response that the research
community feels they merit. This is certain to change as the permeation of
daily life by advanced IT support accelerates. The emerging standards provide
a platform for expressing the requirements and policies of applications—but
only if these policies can be captured and balanced on a systemwide basis. We
contend that the handling of trust issues must necessarily lead public con-
cerns since a high-profile failure could have significant consequences for future
adoption.
Despite significant existing research, we still lack clear and convincing
trust and privacy models for highly-interactive open systems. While the core
technologies may provide a proper basis, there is a clear challenge remain-
ing in engaging in a dialogue with stakeholders to acquire, maintain, and
evolve trust and privacy constraints in a simple, unobtrusive, and modular
fashion.
In principle, an adaptive system may be significantly less variable to a user’s
eyes than a traditional, nonadaptive system, as the adaptations will be used to
mask otherwise significant observable differences. It is vital that this paradox
be carried over into the development and management domains so that those
who develop deploy, and maintain complex networks can focus on the value
added that their activities bring without being consumed by the complexity
of the mechanisms that underlie them. It is only in this way that autonomic
communications systems can become leveraged partners in delivering the next
generation of pervasive and reliable services.
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

252 • S. Dobson et al.
ACKNOWLEDGMENTS
We would like to acknowledge the support of the EU Future and Emerging Tech-
nologies initiative and the assistance of the members of the ACEnet and ACCA
European project consortia, many of whose ideas have significantly informed
the understanding of the issues presented here.
REFERENCES
ABDUL-RAHMAN, A. 1996. The PGP trust model. http://www.cs.ucl.ac.uk/staff/F.
AbdulRahman/docs/pgptrust.html.
ABELSON, H., ALLEN, D., COORE, D., HANSON, C., HOMSY, G., KNIGHT, T., NAGPAL, R., RAUCH, E., SUSSMAN,
G., AND WEISS, R. 2000. Amorphous computing. Comm. ACM 43, 5 (May), 74–82.
AHLSWEDE, R., CAI, N., LI, S.-Y. R., AND YEUNG, R. W. 2000. Network information flow. IEEE Trans.
Inform. Theory 46, 4, 1204–1216.
ALMEIDA, P. S., BAQUERO, C., AND FONTE, V. 2002. Version stamps—decentralized version vectors. In
Proceedings of the 22nd International Conference on Distributed Computing Systems (ICDCS’02).
IEEE Computer Society, 544.
A`LVAREZ, C., Dı´AZ, J., PETIT, J., ROLIM, J., AND SERNA, M. 2004. Efficient and reliable high level
communicatioon in deployed sensor networks. In ACM International Workshop on Mobility Man-
agement and Wireless Systems (Mobiwac’04). ACM Press, 106–110.
ANDERSEN, D. G., BALAKRISHNAN, H., KAASHOEK, M. F., AND MORRIS, R. 2002. Resilient overlay net-
works. Comput. Comm. Rev. 32, 1, 66.
BABAOGLU, O¨., JELASITY, M., AND MONTRESOR, A. 2005a. Grassroots approach to self-management
in large-scale distributed systems. In Unconventional Programming Paradigms. Lecture Notes
in Computer Science, vol. 3566. Springer-Verlag, 286–296.
BABAOGLU, O¨., JELASITY, M., MONTRESOR, A., FETZER, C., LEONARDI, S., VAN MOORSEL, A. P. A., AND VAN
STEEN, M., EDS. 2005b. Self-star properties in complex information systems, conceptual and
practical foundations. Lecture Notes in Computer Science, vol. 3460. Springer.
BABAOGLU, O¨., MELING, H., AND MONTRESOR, A. 2002. A framework for the development of agent-
based peer-to-peer systems. In 22nd International Conference on Distributed Computing Systems.
Vienna, Austria. IEEE Computer Socity Press, 15–22.
BAKER, S. AND DOBSON, S. 2005. Comparing service-oriented and distributed object architec-
tures. In Proceedings of the International Symposium on Distributed Objects and Applications,
R. Meersman and Z. T. et al, Eds. Lecture Notes in Computer Science, vol. 3760. Springer Verlag,
631–645.
BALAKRISHNAN, H., KAASHOEK, M. F., KARGER, D. R., MORRIS, R., AND STOICA, I. 2003. Looking up
data in P2P systems. Comm. ACM 46, 2, 43–48.
BALDONI, R., QUERZONI, L., PATIN˜O-MART‘´NEZ, M., AND JI´MENEZ-PERIS, R. 2004. Low-load dynamic
hierarchical quorums for P2P networks. Tech. rep. (Dec).
BANDINI, S., MANZONI, S., AND SIMONE, C. 2002. Heterogeneous agents situated in heterogeneous
spaces. Appl. Artificial Intellig. 16, 9–10 (Oct.–Dec.), 831–852.
BECKER, M. Y. AND SEWELL, P. 2004. Cassandra: Flexible trust management, applied to elec-
tronic health records. In Proceedings of the 17th IEEE Computer Security Foundations Workshop
(CSFW). 139–154.
BIRMAN, K. 2005. Reliable Distributed Systems Technologies, Web Services, and Applications.
Springer.
BLAZE, M., FEIGENBAUM, J., IOANNIDIS, J., AND KEROMYTIS, A. D. 1999. The role of trust management
in distributed systems security. In Secure Internet Programming: Security Issues for Mobile and
Distributed Objects. Springer-Verlag, 185–210.
BOINC PROJECT. 2006. Berkeley open infrastructure for network computing. http://boinc.
berkeley.edu/.
BONABEAU, E., DORIGO, M., AND THERAULAZ, G. 1999. Swarm Intelligence: From Natural to Artificial
Systems. Oxford University Press.
BONATTI, P. AND SAMARATI, P. 2002. A unified framework for regulating access and information
release on the web. J. Comput. Security 10, 3, 241–272.
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

A Survey of Autonomic Communications • 253
BULLOCK, S. AND CLIFF, D. 2004. Complexity and emergent behaviour in ICT systems. Tech. rep.
HPL-2004-187.
BYERS, J., HORN, G., LUBY, M., MITZENMACHER, M., AND SHAVER, W. 2002. FLID/DL: Congestion
control for layered multicast. IEEE J. Select. Areas Comm. 20, 8 (Oct.), 1558–1570.
BYERS, J. W., CONSIDINE, J., MITZENMACHER, M., AND ROST, S. 2004. Informed content delivery across
adaptive overlay networks. IEEE/ACM Trans. Netw. 12, 5, 767–780.
CABRI, G., LEONARDI, L., AND ZAMBONELLI, F. 2000. Mobile-agent coordination models for internet
applications. IEEE Comput. 33, 2.
CAHILL, V., SHAND, B., GRAY, E., DIMMOCK, N., TWIGG, A., BACON, J., ENGLISH, C., WAGEALLA, W., TERZIS, S.,
NIXON, P., BRYCE, C., DI MARZO SERUGENDO, G., SEIGNEUR, J.-M., CARBONE, M., KRUKOW, K., JENSEN, C.,
CHEN, Y., AND NIELSEN, M. 2003. Using trust for secure collaboration in uncertain environments.
IEEE Pervasive Comput. 2, 3 (July), 52–61.
CARRERAS, I., GRASSO, R., AND SZABO, C. A. 2005. Design considerations on the CREATE-NET
testbed. In Proceedings of TRIDENTCOM 2005. 44–53.
CARZANIGA, A., ROSENBLUM, D., AND WOLF, A. 2001. Design and evaluation of a wide-area event
notification service. ACM Trans. Comput. Syst. 19, 3 (Aug.), 332–383.
CASTRO, M., JONES, M., KERMARREC, A., ROWSTRON, A., THEIMER, M., WANG, H., AND WOLMAN, A. 2003.
An evaluation of scalable application-level multicast built using peer-to-peer overlays. In IEEE
Infocom. San Francisco, CA. IEEE Computer Society Press.
CHADWICK, D., OTENKO, A., AND BALL, E. 2003. Role-based access control with X.509 attribute
certificates. IEEE Internet Comput. 7, 2 (March/April), 62–69.
CHELIUS, G., JELGER, C., FLEURY, E., AND NOEL, T. 2005. IPv6 addressing scheme and self-
configuration for multi-hops wireless ad hoc network. In Proceedings of the International Con-
ference on Information Networking (ICOIN’05). Jeju, Korea.
CHESS, D. M., PALMER, C. C., AND WHITE, S. R. 2003. Security in an autonomic computing environ-
ment. IBM Syst. J. 42, 1, 107–118.
CHLEBUS, B. S., GASIENIEC, L., KOWALSKI, D. R., AND RADZIK, T. 2005. On the wake-up problem in
radio networks. In ICALP, L. Caires, G. F. Italiano, L. Monteiro, C. Palamidessi, and M. Yung,
Eds. Lecture Notes in Computer Science, vol. 3580. Springer, 347–359.
CHOCKLER, G., KEIDAR, I., AND VITENBERG, R. 2001. Group communication specifications: A com-
prehensive study. ACM Comput. Surv. 33, 4, 427–469.
CHOLVI, V., LADERAS, V., LO´PEZ, L., AND FERNA´NDEZ, A. 2005. Self-adapting network topologies in
congested scenarios. Physical Rev. 71, 3.
CHROBAK, M., GASIENIEC, L., AND RYTTER, W. 2004. A randomized algorithm for gossiping in radio
networks. Networks 43, 2, 119–124.
CLARK, D., PARTRIDGE, C., RAMMING, J., AND WROCLAWSKI, J. 2003. A knowledge plane for the inter-
net. In Proceedings of ACM SIGCOMM.
COHEN, B. 2003. Incentives build robustness in bittorrent. http://www.bittorrent.com/
bittorrentecon.pdf.
COSTA, D. AND HERTZ, A. 1997. Ants can colour graphs. J. Operation. Resear. Soc. 48, 295–305.
COSTA, P. AND PICCO, G. 2005 (to appear). Semi-probabilistic content-based publish-subscribe.
In IEEE International Conference on Distributed Computing Systems (ICDCS), Columbus, OH.
IEEE Computer Society Press.
COUTAZ, J., CROWLEY, J., DOBSON, S., AND GARLAN, D. 2005. Context is key. Comm. ACM 48, 3
(March), 49–53.
CURINO, C., GIANI, M., GIORGETTA, M., GIUSTI, A., MURPHY, A., AND PICCO, G. 2005. Tinylime: Bridging
mobile and sensor networks through middleware. In IEEE International Conference on Perva-
sive Computing and Communications (PerCom), Kauai Island, Hawaii. IEEE Computer Society
Press.
DE BRUIJN, J., LAUSEN, H., POLLERES, A., AND FENSEL, D. 2005. The WSML rule languages for
the semantic web. In Proceedings of the W3C Workshop on Rule Languages for Interoperability.
Washington, DC.
DE CASTRO, M. F., MERGHEM, L., GAI¨TI, D., AND MHAMED, A. 2004. The basis for an adaptive IP QoS
management. IEICE Trans. Comm. E87-B, 3 (March), 564–572.
DE VERGARA, J., VILLAGRE´, V., AND BERROCAL, J. 2004. Applying the Web Ontology Language to
management information definitions. IEEE Comm. Mag. 42, 7 (July), 68–74.
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

254 • S. Dobson et al.
DE´FAGO, X., SCHIPER, A., AND URBA´N, P. 2004. Total order broadcast and multicast algorithms:
Taxonomy and survey. ACM Comput. Sur. 36, 4, 372–421.
DOBSON, S. 2005. Putting meaning into the network: Some semantic issues for the design of au-
tonomic communications systems. In Proceedings of the 1st IFIP Workshop on Autonomic Com-
munications (Berlin, Germany), M. Smirnov, Ed. Lecture Notes in Computer Science, vol. 3457.
Springer Verlag.
DOBSON, S. AND NIXON, P. 2004. More principled design of pervasive computing systems. In Human
Computer Interaction and Interactive Systems, R. Bastide and J. Roth, Eds. Lecture Notes in
Computer Science, vol. 3425. Springer Verlag.
DORIGO, M. AND GAMBARDELLA, L. M. 1997. Ant colonies for the travelling salesman problem.
Biosystems 43, 73–81.
DRIVER, C. AND CLARKE, S. 2004. Context-aware trails. IEEE Comput. 37, 8 (Aug.), 97–99.
ELLISON, C., FRANTZ, B., LAMPSON, B., RIVEST, R., THOMAS, B., AND YLONEN, T. 1999. SPKI certificate
theory. IETF RFC 2693.
ENGLISH, C., TERZIS, S., WAGEALLA, W., LOWE, H., NIXON, P., AND MCGETTRICK, A. 2003a. Trust dy-
namics for collaborative global computing. In Proceedings of 12th IEEE International Workshop
on Enabling Technologies (WETICE’03), Linz, Austria. IEEE Computer Society. 283–288.
ENGLISH, C., WAGEALLA, W., NIXON, P., TERZIS, S., LOWE, H., AND MCGETTRICK, A. 2003b. Trusting
collaboration in global computing systems. In Proceedings of 1st International Conference on
Trust Management (iTrust’03), Heraklion, Crete, Greece. Lecture Notes in Computer Science,
vol. 2692. Springer, 136–149.
ESPONDA, F., FORREST, S., AND HELMAN, P. 2004. A formal framework for positive and negative
detection. IEEE Trans. Syst., Man, Cybernet. 34, 1 (Feb.).
EUGSTER, P., GUERRAOUI, R., KERMARREC, A., AND MASSOULIE, L. 2004. Epidemic information dis-
semination in distributed systems. IEEE Comput. 37, 5, 60–67.
EUGSTER, P., GUERRAOUI, R., AND KOUZNETSOV, P. 2004. D-reliable broadcast: A probabilistic mea-
sure of broadcast reliability. In IEEE International Conference on Distributed Computing Systems
(ICDCS). Tokio, Japan.
EUGSTER, P. T., FELBER, P., GUERRAOUI, R., AND KERMARREC, A.-M. 2003. The many faces of publish/
subscribe. ACM Comput. Surv. 35, 2 (June), 114–131.
EUGSTER, P. T., GUERRAOUI, R., HANDURUKANDE, S. B., KOUZNETSOV, P., AND KERMARREC, A.-M. 2003.
Lightweight probabilistic broadcast. ACM Trans. Comput. Syst. 21, 4, 341–374.
FERNA´NDEZ, A., GEORGIOU, C., LO´PEZ, L., AND SANTOS, A. 2005. Reliably executing tasks in the
presence of malicious processors. In DISC, P. Fraigniaud, Ed. Lecture Notes in Computer Science,
vol. 3724. Springer, 490–492.
FERNA´NDEZ, A., GEORGIOU, C., LO´PEZ, L., AND SANTOS, A. 2006. Reliably executing tasks in the pres-
ence of untrusted entities. In Proceedings of the 25th IEEE Symposium on Reliable Distributed
Systems (SRDS’06). Leeds, UK.
G. PICCO, G. CUGOLA, A. M. 2003. Efficient content-based event dispatching in presence of topo-
logical reconfigurations. In Proceedings of the IEEE International Conference on Distributed
Computing Systems, Providence, RI. IEEE Computer Society Press.
GARG, A., BATTITI, R., AND COSTANZI, G. 2004. Dynamic self-management of autonomic systems:
The reputation, quality and credibility (RQC) scheme. In Proceedings of the 1st IFIP TC6 WG6.6
International Workshop on Autonomic Communication (WAC). Springer-Verlag, Berlin, Germany.
GELENBE, E. 2004a. Cognitive packet network, U.S. Patent 6,804,201.
GELENBE, E. 2004b. Cognitive routing in packet networks. In ICONIP, N. R. Pal, N. Kasabov,
R. K. Mudi, S. Pal, and S. K. Parui, Eds. Lecture Notes in Computer Science, vol. 3316. Springer,
625–632.
GELENBE, E. 2004c. Sensible decisions based on QoS. Comput. Manage. Science 1, 1, 1–14.
GELENBE, E., GELLMAN, M., LENT, R., LIU, P., AND SU, P. 2004. Autonomous smart routing for net-
work QoS. In Proceedings of the First International Conference on Autonomic Computing. IEEE
Computer Society, 232–239.
GELENBE, E. AND LENT, R. 2004. Power aware ad hoc cognitive packet networks. Ad Hoc Netw.
J. 2, 3, 205–216.
GELENBE, E., LENT, R., AND NUNEZ, A. 2004. Self-aware networks and quality of service. Proceed-
ings of the IEEE 92, 9, 1479–1490.
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

A Survey of Autonomic Communications • 255
GELENBE, E., LENT, R., AND XU, Z. 2001. Design and performance of cognitive packet networks.
Perform. Evaluat. 46, 155–176.
GELENBE, E. AND NUNEZ, A. 2003. Adaptive web service for QoS improvement. In Proceedings of
IADIS International WWW/Internet Conference. Algarve, Portugal.
GKANTSIDIS, C. AND RODRIGUEZ, P. 2005. Network coding for large scale content distribution. In
IEEE/INFOCOM 2005. Miami, FL.
GONZALEZ, A. AND AHLERS, R. 1999. Context-based representation of intelligent behaviour in train-
ing simulations. Trans. Society Comput. Simulat. Int. 15, 4 (March).
HENRICKCON, K., INDULSKA, J., AND RAKOTONIRAINY, A. 2002. Modeling context information in per-
vasive computing systems. In Proceedings of the 1st International Conference on Pervasive Com-
puting Systems, F. Mattern and M. Naghshineh, Eds. Lecture Notes in Computer Science, vol.
2414. Springer Verlag.
HOFMEYR, S. A. AND FORREST, S. 2000. Architecture for an artificial immune system. Evolution.
Computat. 8, 4, 443–473.
IBM, MICROSOFT, BEA, RSA SECURITY AND VERISIGN. 2003. Web Services Federation Language
(WS-Federation). IBM, Microsoft, BEA, RSA Security and VeriSign. http://www-106.ibm.com/
developerworks/webservices/library/ws-fed.
IONESCU, B., IONESCU, M., VERES, S., IONESCU, D., CUERVO, F., AND LUIKEN-MILLER, M. 2005. A testbed
and research network for next generation services over next generation networks. In Proceedings
of TRIDENTCOM 2005. 21–31.
ITU-T. 2001. The directory: Public-key and attribute certificate frameworks. ITU-T Recommen-
dation X.509:2000(E) | ISO/IEC 9594-8:2001(E).
JELASITY, M., KOWALCZYK, W., AND VAN STEEN, M. November 2003. Newscast computing. Tech. rep.,
Vrije Universiteit IR-CS-006.
JENSEN, O. H. AND MILNER, R. 2003. Bigraphs and mobile processes. Tech. rep. UCAM-CL-TR-570,
University of Cambridge Computer Laboratory.
JIME´NEZ-PERIS, R., PATIN˜O-MARTI´NEZ, M., AND ALONSO, G. 2002. An algorithm for non-intrusive, par-
allel recovery of replicated data and its correctness. In Proceedings of 21st IEEE Int. Conference
on Reliable Distributed Systems (SRDS’02). Osaka, Japan, 150–159.
KARABULUT, Y. 2003. Implementation of an agent-oriented trust management infrastructure
based on a hybrid PKI model. In iTrust, P. Nixon and S. Terzis, eds. Lecture Notes in Computer
Science, vol. 2692. Springer, 318–331.
KENNEDY, M. P., ROVATTI, R., AND SETTI, G. 2000. Chaotic Electronics in Telecommunications. CRC
Press.
KEPHART, J. AND CHESS, D. 2003. The vision of autonomic computing. IEEE Comput. 36, 1, 41–50.
KOSHUTANSKI, H. AND MASSACCI, F. 2004a. E pluribus unum: Deduction, abduction and induction,
the reasoning services for access control in autonomic communication. In Proceedings of the 1st
IFIP TC6 WG6.6 International Workshop on Autonomic Communication (WAC). Springer-Verlag,
Berlin, Germany.
KOSHUTANSKI, H. AND MASSACCI, F. 2004b. Interactive access control for Web Services. In Proceed-
ings of the 19th IFIP Information Security Conference (SEC’04), Toulouse, France. Kluwer Press,
151–166.
KOSHUTANSKI, H. AND MASSACCI, F. 2004c. An interactive trust management and negotiation
scheme. In Proceedings of the 2nd International Workshop on Formal Aspects in Security and
Trust (FAST), Toulouse, France. Kluwer Press, 139–152.
LAOUTARIS, N., PANAGAKIS, A., AND STAVRAKAKIS, I. 2004a. Content distribution through autonomic
content and storage management. In WAC 2004, M. Smirnov, Ed. Lecture Notes in Computer
Science, vol. 3457. Springer, Berlin, Germany, 69–78.
LAOUTARIS, N., PANAGAKIS, A., AND STAVRAKAKIS, I. 2004b. Content distribution through autonomic
content and storage management. In WAC 2004.
LAOUTARIS, N., TELELIS, O., ZISSIMOPOULOS, V., AND STAVRAKAKIS, I. 2004. Local utility aware content
replication. In IFIP Networking 2005.
LAOUTARIS, N., TELELIS, O., ZISSIMOPOULOS, V., AND STAVRAKAKIS, I. 2005. Distributed selfish repli-
cation. IEEE Trans. Paral. Distrib. Syst. [under submission].
LASSILA, O. AND SWICK, R. 1999. Resource Description Framework model and syntax specification.
Tech. rep., World Wide Web Consortium.
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

256 • S. Dobson et al.
LEFF, A., WOLF, J., AND YU, P. 1993. Replication algorithms in a remote caching architecture. IEEE
Trans. Paral. Distrib. Syst. 4, 11 (Nov.), 1185–1204.
LEWIS, D., FEENEY, K., CAREY, K., TIROPANIS, T., AND COURTENAGE, S. 2005. Semantic-based policy
engineering for autonomic systems. In Proceedings of 1st IFIP WG6.6 International Workshop on
Autonomic Communication, M. Smirnow, Ed. Springer Verlag.
LI, N., MITCHELL, J. C., AND WINSBOROUGH, W. H. 2002. Design of a role-based trust-management
framework. In Proceedings of IEEE Symposium on Security and Privacy, 2002. S&P. IEEE
Press.
LO´PEZ, L., FERNA´NDEZ, A., AND CHOLVI, V. 2005. A game theoretic analysis of protocols based on
fountain codes. In The 10th IEEE Symposium on Computers and Communications (ISCC’05). La
Manga del Mar Menor, Spain.
LUBY, M. 2002. LT codes. In FOCS. IEEE Computer Society, 271.
LUBY, M. 2003. Fast, reliable data transport. In USENIX Symposium on Internet Technologies
and Systems.
LU¨CKING, T., MAVRONICOLAS, M., MONIEN, B., AND RODE, M. 2004. A new model for selfish routing.
In Proceedings of STACS 2004. 547–558.
MADUEN˜O, M. AND VIDAL, J. 2005. Joint physical-MAC layer design of the broadcast channel pro-
tocol in adhoc networks. IEEE J. Select. Areas in Comm. (Special Issue on Ad-Hoc Networking).
MAGEDANZ, T., WITASZEK, D., AND KNUETTEL, K. 2005. The IMS playground at FOKUS—an open
test-bed for next generation network multimedia services. In Proceedings of TRIDENTCOM
2005.
MAMEI, M. AND ZAMBONELLI, F. 2004. Programming pervasive and mobile computing applications
with the TOTA middleware. In IEEE International Conference On Pervasive Computing (Percom)
Orlando, FL. IEEE Computer Socity Press.
MAMEI, M., ZAMBONELLI, F., AND LEONARDI, L. 2004. Co-fields: A physically inspired approach to
distributed motion coordination. IEEE Pervasive Comput. 3, 2, 52–61.
MAZZINI, G., ROVATTI, R., AND SETTI, G. 2000. A tensor approach to higher order expectations of
chaotic trajectories—part II: Application to chaos-based DS-CDMA in multipath environements.
IEEE Trans. Circuits and Syst.—Part I 47, 1584–1596.
MAZZINI, G., ROVATTI, R., AND SETTI, G. 2001. Chaos-based asynchronous DS-CDMA systems and
enhanced rake receivers: Measuring the improvements. IEEE Trans. Circuits Syst.—Part I 48, 12,
1445–1453.
MAZZINI, G., ROVATTI, R., AND SETTI, G. 1999. Interference minimization by auto-correlation shap-
ing in asynchronous DS-CDMA systems: Chaos-based spreading is nearly optimal. IEEE Elec-
tronics Lett. 35, 13 (June), 1054–1055.
MAZZINI, G., SETTI, G., AND ROVATTI, R. 1997. Chaotic complex spreading sequences for asyn-
chronous DS-CDMA—Part I: System modeling and results. IEEE Trans. Circuits Syst.—
Part I 44, 10, 937–947.
MCGIBNEY, J., SCHMIDT, N., AND PATEL, A. 2005. A service-centric model for intrusion detection in
next-generation networks. Comput. Stand. Interf. 27, 5 (June), 513–520.
MENEZES, R. AND TOLKSDORF, R. 2003. A new approach to scalable linda-systems based on swarms.
ACM Symposium on Applied Computer, Orlando, FL. ACM Press, 375–379.
MERGHEM, L., GAI¨TI, D., AND PUJOLLE, G. 2003. On using multi-agent systems in end to end adap-
tive monitoring. In Proceedings of E2EMON (End to End MONitoring) Workshop, in Conjunction
with MMNS2003 (Belfast, UK). Lecture Notes in Computer Science, vol. 2839. Springer Verlag,
422–435.
MERLOTI, P. E. 2004. Optimization algorithms inspired by biological ants and swarm behavior.
MESTRE, X., FONOLLOSA, J. R., AND PAGE`S, A. 2003. Capacity of MIMO channels: Asymptotic eval-
uation under correlated fading. IEEE J. Select. Areas Comm. (Special Issue on MIMO Systems
and Applications) 21, 5 (June).
MICHIARDI, P. AND MOLVA, R. 2002. Core: A COllaborative REputation mechanism to enforce node
cooperation in mobile ad hoc networks. In Proceedings of Communication and Multimedia Secu-
rity Conference.
MILAN-FRANCO, J. M., JIME´NEZ-PERIS, R., PATIN˜O-MARTI´NEZ, M., AND KEMME, B. 2004. Adaptive dis-
tributed middleware for database replication. In Proceedings of 5th ACM/IFIP/USENIX Mid-
dleware Conference. Toronto, Canada, 175–194.
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

A Survey of Autonomic Communications • 257
MORATO´, D., MAGANA, E., IZAL, M., ARACIL, J., NARANJO, F. J., ASTIZ, P., ALONSO, U., CSABAI, I., HA´GA,
P., SIMON, G., STEGER, J., AND VATTAY, G. 2005. The European traffic observatory measurement
infrastructure (ETOMIC): A testbed for universal active and passive measurements. In Proceed-
ings of TRIDENTCOM 2005.
NAKRANI, S. AND TOVEY, C. 2004. On honey bees and dynamic server allocation in internet hosting
centers. Adaptive Behavior 12, 3–4, 223–240.
OASIS SECURITY SERVICES TC. 2004. Security Assertion Markup Language (SAML). http://www.
oasis-open.org/committees/security.
OMICINI, A., RICCI, A., VIROLI, M., CASTELFRANCHI, C., AND TUMMOLINI, L. 2004. Coordination arti-
facts: Environment-based coordination for intelligent agents. International Joint Conference on
Autonomous Agents and Multiagent Systems (AAMAS). New York (NY).
O’NEILL, E., KLEPAL, M., LEWIS, D., O’DONNELL, T., O’SULLIVAN, D., AND PESCH, D. 2005. A testbed
for evaluating human interaction with ubiquitous computing environments. In Proceedings of
TRIDENTCOM 2005.
O’NEILL, E., LEWIS, D., MCGLINN, K., AND DOBSON, S. 2006. Rapid user-centred evaluation for
context-aware systems. In Proceedings of the 13th International Workshop on Design, Specifi-
cation and Verification of Interactive Systems (DSVIS’06), G. Doherty and A. Blandford, Eds.
Lecture Notes in Computer Science, Springer Verlag.
OREIZY, P., GORLICK, M., TAYLOR, R., HEIMBIGNER, D., JOHNSON, G., MEDVIDOVIC, N., QUILLIEI, A., ROSEN-
BLUM, D., AND WOLF, A. 1999. An architecture-based approach to self-adaptive software. IEEE
Intel. Syst. 14, 3 (May/June), 54–62.
PARK, J. S. AND SANDHU, R. 1999. RBAC on the Web by smart certificates. In Proceedings of the
4th ACM Workshop on Role-Based Access Control. ACM Press, 1–9.
PICCO, G., MURPHY, A., AND ROMAN, G. 2001. Lime: A middleware for logical and physical mo-
bility. In Proceedings of the IEEE International Conference on Distributed Computing Systems,
Providence, RI. IEEE Computer Society Press, 524–536.
PUJOLLE, G., CHAOUCHI, H., AND GAI¨TI, D. 2004. Beyond TCP/IP: A context-aware architecture. In
Proceedings of NetCon 2004, Palma de Mallorca. Kluwer Academic Publishers, 337–347.
QIU, D. AND SRIKANT, R. 2004. Modeling and performance analysis of bittorrent-like peer-to-
peer networks. In SIGCOMM, R. Yavatkar, E. W. Zegura, and J. Rexford, Eds. ACM, 367–
378.
QUITADAMO, R. AND ZAMBONELLI, F. 2007. Autonomic communication services: a new challenge for
software agents. J. Autonom. Agents Multiagent Syst.
RAO, A., PAPADIMITRIOU, C., RATNASAMY, S., SHENKER, S., AND STOICA, I. 2003. Geographic routing
without location information. In ACM Mobicom. San Diego, CA. ACM Press.
RATSANAMY, S. AND AL. 2002. Ght: A geographic hash table for data-centric storage. In 1st ACM
International Workshop on Wireless Sensor Networks and Applications. ACM Press, Atlanta,
Georgia, USA.
RATSANAMY, S., FRANCIS, P., HANDLEY, M., AND KARP, R. 2001. A scalable content-addressable net-
work. In ACM SIGCOMM Conference, San Diego, CA. ACM Press.
RODERO, L., LO´PEZ, L., FERNA´NDEZ, A., AND CHOLVI, V. 2006. Dante: A self-adapting peer-to-peer
system. In Proceedings of the 5th International Workshop on Agents and Peer-to-Peer Computing
(AP2PC06), Hakodate, Japan. Lecture Notes in Computer Science. Springer.
ROMAN, G., JULIEN, C., AND HUANG, Q. 2002. Network abstractions for context-aware mobile com-
puting. 24th International Conference on Software Engineering, Orlando, FL. ACM Press.
ROVATTI, R., MAZZINI, G., AND SETTI, G. 2000. A tensor approach to higher order expectations
of chaotic trajectories—Part I: General theory and specialization to piecewise affine Markov
systems. IEEE Trans. Circuits Syst.—Part I 47, 1571–1583.
ROVATTI, R., MAZZINI, G., AND SETTI, G. 2001. Enhanced rake receivers for chaos-based DS-CDMA.
IEEE Trans. Circuits Syst.—Part I 48, 818–829.
ROVATTI, R., MAZZINI, G., AND SETTI, G. 2004a. On the ultimate limits of chaos-based asynchronous
DS-CDMA—Part I: Basic definitions and results. IEEE Trans. Circuits Syst.—Part I 52, 7, 1336–
1347.
ROVATTI, R., MAZZINI, G., AND SETTI, G. 2004b. On the ultimate limits of chaos-based asynchronous
DS-CDMA—Part II: Analytical results and asymptotics. IEEE Trans. Circuits Syst.—Part I 52, 7,
1348–1364.
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

258 • S. Dobson et al.
ROVATTI, R., SETTI, G., AND MAZZINI, G. 1998. Chaotic complex spreading sequences for asyn-
chronous DS-CDMA—Part II: Some theoretical performance bounds. IEEE Trans. Circuits
Syst.—Part I 45, 4, 496–506.
ROWSTRON, A. 1999. Mobile coordination: Providing fault tolerance in tuple space based coordina-
tion language. Coordination Languages and Models (Coordination’99), P. Ciancarini and P. Wolf,
Eds. Lecture Notes in Computer Science, vol. 1594. Springer-Verlag, 196–210.
ROWSTRON, A. AND DRUSCHEL, P. 2001. Pastry: Scalable, decentralized object location and routing
for large-scale peer-to-peer systems. 18th IFIP/ACM Conference on Distributed Systems Plat-
forms. Heidelberg, Germany. ACM Press.
SAFFRE, F. AND BLOK, H. R. 2005. SelfService—A theoretical protocol for autonomic distribu-
tion of services in P2P communities. In Proceedings of the 12th IEEE International Con-
ference and Workshops on the Engineering of Computer-Based Systems (ECBS’05). 528–
534.
SEAMONS, K. AND WINSBOROUGH, W. 2002. Automated trust negotiation. Tech. rep., US Patent and
Trademark Office. IBM Corporation, patent application filed March 7, 2000.
SETTI, G., MAZZINI, G., ROVATTI, R., AND CALLEGARI, S. 2002. Statistical modeling and design of
discrete time chaotic processes: Basic finite-dimensional tools and applications. In Proceedings
of the IEEE. Vol. 90. 662–690.
SETTI, G., ROVATTI, R., AND MAZZINI, G. 2004. Performance of chaos-based asynchronous DS-CDMA
with different pulse shapes. IEEE Comm. Lett. 8, 7 (July), 416–418.
SHACKLETON, M., SAFFRE, F., TATESON, R., BONSMA, E., AND ROADKNIGHT, C. 2004. Autonomic com-
puting for pervasive ICT—A whole system perspective. BT Tech. J. 22, 3, 191–199.
SIRIS, V. A. 2002. Resource control for elastic traffic in CDMA networks. In Proceedings of 8th
ACM Conference on Mobile Computing and Networking (MOBICOM’02). Atlanta, GA.
SIRIS, V. A., BRISCOE, B., AND SONGHURST, D. 2002. Economic models for resource control in wire-
less networks. In Proceedings of IEEE Int. Symposium on Personal, Indoor, and Mobile Radio
Communications (PIMRC’02).
SIRIS, V. A. AND COURCOUBETIS, C. 2004. Resource control for loss-sensitive traffic in CDMA net-
works. In Proceedings of 23rd Conference of the IEEE Communications Society (INFOCOM).
Hong Kong.
SPINGLASS PROJECT. 2005. Spinglass: Adaptive probabilistic tools for advanced networks. http:
//www.cs.cornell.edu/Info/Projects/Spinglass/.
STERRITT, R., MULVENNA, M., AND LAWRYNOWICZ, A. 2005. A role for contextualised knowledge in
autonomic communications. In Proceedings of 1st IFIP TC6 WG6.6 International Workshop on
Autonomic Communications, M. Smirnow, Ed. Springer Verlag.
STOY, K. AND NAGPAL, R. 2004a. Self-reconfiguration using directed growth. In 7th International
Symposium on Distributed Autonomous Robotic Systems (DARS). Toulouse, France. ACM Press,
149–160.
STOY, K. AND NAGPAL, R. 2004b. Self-repair through scale independent self-reconfiguration. In Pro-
ceedings of IEEE/RSJ International Conference on Robots and Systems (IROS), Sendai, Japan.
IEEE Press.
STRANG, T., LINNHOFF-POPIEN, C., AND FRANK, K. 2003. CoOL: A context ontology language to enable
contextual interoperability. In Proceedings of 4th IFIP International Conferece on Distributed
Applications and Interoperable Systems (DAIS’03).
TAKAI, M., BAGRODIA, R., GERLA, M., DANESHRAD, B., FITZ, M. P., SRIVASTAVA, M. B., BELDING-ROYER,
E. M., KRISHNAMURTHY, S. V., MOLLE, M., MOHAPATRA, P., RAO, R. R., MITRA, U., SHEN, C.-C., AND
EVANS, J. B. 2005. Scalable testbed for next generation wireless networking technologies. In
Proceedings of TRIDENTCOM 2005. 162–171.
TANENBAUM, A. 2004. Distributed Systems, 2nd ed. Addison Wesley.
TERZIS, S., WAGEALLA, W., ENGLISH, C., AND NIXON, P. 2004. Trust lifecycle management in a global
computing environment. In Global Computing, IST/FET International Workshop GC 2004,
Rovereto, Italy. 9–12 (March), C. Priami and P. Quaglia, Eds. Lecture Notes in Computer Science,
vol. 3267. 291–313.
THOMPSON, M., JOHNSTON, W., MUDUMBAI, S., HOO, G., JACKSON, K., AND ESSIARI, A. 1999. Certificate-
based access control for widely distributed resources. In Proceedings of 8th USENIX Security
Symposium (Security’99) (23–26). 215–228.
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.

A Survey of Autonomic Communications • 259
TRIANNI, V., LABELLA, T., AND DORIGO, M. 2004. Evolution of direct communication for a swarm-bot
performing hole avoidance. In ANTS Workshop. Brussels, Belgium.
TSARAMPOPOULOS, N., KALAVROS, I., AND LALIS, S. 2005. A low-cost and simple to deploy peer-to-peer
wireless network based on open source linux routers. In Proceedings of TRIDENTCOM 2005.
TSCHUDIN, C., GUNNINGBERG, P., LUNDGREN, H., AND NORDSTROM, E. 2005. Lessons from experimental
MANET research. Ad Hoc Netw. J. 3, 2 (March).
TSCHUDIN, C. AND YAMAMOTO, L. 2004. A metabolic approach to protocol resilience. In Proceedings
of the 1st IFIP Workshop on Autonomic Communication (WAC’04). Berlin, Germany.
TSCHUDIN, C. F., GUNNINGBERG, P., LUNDGREN, H., AND NORDSTRO¨M, E. 2005. Lessons from experi-
mental MANET research. Ad Hoc Netw. 3, 2 (March), 221–233.
UNDERCOFFER, J., JOSHI, A., AND PINKSTON, J. 2003. Modeling computer attacks: An ontology for
intrusion detection. In Proceedings of the 6th International Symposium on Recent Advances in
Intrusion Detection (RAID’03). Pittsburgh, PA. Lecture Notes in Computer Science, vol. 2820,
113–135.
VIDALES, P., MAPP, G., STAJANO, F., CROWCROFT, J., AND BERNARDOS, C. 2005. A practical approach for
overlay systems: Deployment of overlay networks. In Proceedings of TRIDENTCOM 2005.
VON RICKENBACH, P., SCHMID, S., WATTENHOFER, R., AND ZOLLINGER, A. 2005. A robust interference
model for wireless ad-hoc networks. In IPDPS. IEEE Computer Society.
WEISER, M. 1991. The computer for the 21st century. Scientific American 265, 3, 94–104.
WEYNS, D., PARUNAK, H., MICHEL, F., HOLVOET, T., AND FERBER, J. 2005. Environments for Multiagent
Systems, State-of-the-art and Research Challenges. Lecture Notes in Artificial Intelligence, vol.
3374. Springer Verlag, Berlin, Germany.
WINSBOROUGH, W. AND JACOBS, J. 2003. Automated trust negotiation in attribute-based access
control. In Proceedings of DARPA Information Survivability Conference and Exposition. Vol. 2.
IEEE Press, 252–257.
YAMAZAKI, T. 2005. Ubiquitous home: Real-life testbed for home context-aware services. In Pro-
ceedings of TRIDENTCOM 2005. 54–69.
YAO, W. 2003. Fidelis: A policy-driven trust management framework. In 1st International Con-
ference on Trust Management (iTrust). Lecture Notes in Computer Science, vol. 2692. Springer,
301–317.
YU, T., WINSLETT, M., AND SEAMONS, K. E. 2003. Supporting structured credentials and sensitive
policies through interoperable strategies for automated trust negotiation. ACM Trans. Inform.
Syst. Secur. (TISSEC) 6, 1, 1–42.
ZAMBONELLI, F. AND MAMEI, M. 2004. Spatial computing: An emerging paradigm for autonomic
computing and communication. In International Workshop on Autonomic Communication. Berlin,
Germany.
Received January 2006; revised July 2006; accepted August 2006
ACM Transactions on Autonomous and Adaptive Systems, Vol. 1, No. 2, December 2006.