ArticlePDF Available


from technical details, such as SNMP primitives or CMIS/CMIP terms, and translating them into a more understandable form for human operators.
Intelligent Agents
An Agent Future for Network Control?
Steven Willmott, Monique Calisti
In an age of rapidly increasing network complexity and diversity, the idea of “bringing intelligence to the
network” is becoming more of a necessity than “future work”. Since the early 1990’s, agent technology has
often been proposed as a way of achieving this more effective, robust and above all autonomous network con-
trol. This article provides a brief tour of current trends in network development and discusses the potential
for agent based solutions to some of the most pressing communications network problems.
From a network engineer’s point of view, a network is a
complex system requiring complicated management under
very trying domain constraints. To a Distributed Articial Intel-
ligence researcher, a network is a highly distributed, complex
and challenging environment for the application of intelligent
systems ([Lewis95], [Maes94], [Weihmayer/Velthuijsen 94]).
The idea of distributing communication network control and
management tasks by deploying “smart”, “cooperative” and
“autonomous” entities in network infrastructures has thus
received considerable attention from both the Distributed
Articial Intelligence (DAI) and the Communications Network
(CN) communities.
As networks become increasingly complex and difcult to
control, the ideal of a distributed, intelligent network manage-
ment and control system is becoming more and more of a
necessity. Furthermore, new software and network technolo-
gies are revolutionising what can be deployed in the network
and even what we think of as the network itself. Despite the
lack of deployed systems, these trends make an “agent future
for network management” seem closer than ever.
This article does not aim to replicate the useful surveys
already completed in this area. Instead, the aim is to give a
briefer overview of the research eld which balances the tradi-
tionally separated CN and DAI viewpoints. Rather than going
into detail on individual research efforts we review the area by:
Identifying the current trends which suggest that agent tech-
nology may play an increasingly important role in network
control (Section 2).
Highlighting three key areas which might benet most from
agent technology: multi provider environments (Section
3.1), resource management (Section 3.2) and communica-
tions integration (Section 3.3).
Discussing the necessary steps for the deployment of agent
systems in future communications networks (Section 4.).
Those readers interested in more detailed accounts of
previous work should nd the following surveys useful starting
[Kumar/Venkataram 97], [Weihmayer/Velthuijsen 98] and a
recent volume of collected works [Hayzelden/Bigham 99]
all give useful DAI perspectives.
[Martin-Flatin/Znaty 2000] gives an overview of existing
network management paradigms which places work on
agents in a Network Management context. More specic
works on software agents for management operations can be
found in previous proceedings of the IATA
and DSOM
1.1. “Agent” Terminology
One of the unfortunate side effects of the separation of work
between the DAI and CN communities is confusion over
terminology – particularly surrounding the term “agent”. Many
similar terms (for example SNMP agents, mobile agents,
“intelligent” agents, agents, BDI agents) are used for different
purposes by the two communities. In this article we follow the
agent denition given in [Jennings/Wooldrige 98]. This deni-
1. Intelligent Agents for Telecommunications Applications.
2. Distributed Systems: Operations & Management.
Steven Willmott is a researcher in the Articial Intelligence
Laboratory at EPFL Lausanne. His primary research interests
centre on the control of distributed systems using Distributed Ar-
ticial Intelligence techniques – with particular focus on agent co-
ordination, organisation and language. In 1999 he chaired the 3rd
workshop on Articial Intelligence in Distributed Information
Networking (AiDIN’99 held at AAAI in Orlando, USA). Aside
from his academic activities he is also heavily involved in the
FIPA agent standardisation effort as the editor of the 1999 FIPA
Agent Message Transport Specication.
Monique Calisti has received her master degree in Electrical
Engineering at the University of Bologna (Italy). After having re-
ceived the pre-doctoral School diploma in Communication Sys-
tems at EPFL, she joined the Articial Intelligence Laboratory of
EPFL at the end of 1997. She is currently PhD student and her pri-
mary area of interest is the interaction of distinct telecommunica-
tion network providers. The focus is on the allocation of multi-
provider service demands with Quality of Service requirements.
The use of Distributed Articial Intelligence techniques for ena-
bling a exible and automatic provider-to-provider paradigm rep-
resents one of the main topic under investigation. She is also an
active member of the FIPA agent standardisation group as the ed-
itor of the 1999 FIPA Content Language Library Specication.
Intelligent Agents
tion has strong DAI roots and proposes that an agent is an entity
which is:
in some environment.
, in the sense that the system can act without
direct intervention from others (humans or other software
, which is further broken down into three properties:
(perceives its environment and responds to
changes in a timely fashion),
(exhibits opportun-
istic, goal-directed behaviour) and
(able to interact
with humans or other articial agents).
This denition might be considered consistent with that of an
“intelligent agent” in the CN community. The reason for
dropping the “intelligent” is that the notion of intelligence is
difcult to pin down. Often, this becomes subdivided into
arbitrary “types of intelligence” or “levels of intelligence”. The
agent denition may also be loose enough to incorporate some
of the agents not considered “Intelligent” by the CN communi-
ty. It should also be noted that this denition considers mobility
as an optional property of an agent (or of a piece of code) rather
than something which denes a whole class of agent; helping
avoid problems such as classifying “intelligent mobile agents”.
This denition, we feel, is particularly useful since it reinforces
the view of an agent as a piece of software disposing a
fundamental set of properties. Entities displaying more or less
of these properties can then be considered more or less “agent-
like” (see Section 4).
Trends in Network Development
The main factors behind the increasing interest in agent
technology for network control can be divided into two
application pull
– the need for innovative solutions
to increasingly urgent network problems and
technology push
– the development of new techniques which make agent
deployment a real possibility.
There are three main factors which are generating potential
need for agent based solutions are:
Market liberalisation:
The deregulation of telecommunica-
tions markets has forced major changes to the roles, business
models and operational practices of network operators and
service providers alike (in fact until the early 1990s Network
Operators and Service Providers were often one and the
same). Competition is erce and has kick started an indus-
try-wide drive for efciency.
Rapidly changing technology:
The number and diversity of
deployed network technologies is continually growing. This
diversication is creating a complex heterogeneous network
infrastructure and serious technological challenges in
providing uniform and coherent services. There is often not
enough time for the industry to develop guidelines before
those guidelines are already obsolete. Standard bodies (such
as the ITU, ISO, ANSI, ATM Forum and IETF) are having
to catch up with common practice rather than setting the
Increasing exibility in usage requirements:
With market
liberalisation and increasing customer demand comes a need
for exible service deployment. Networks need to be
adapted to provide what customers are demanding, cope
with uctuations in usage and handle the introduction of
new multimedia services (such as video, audio, Internet
telephony and e-commerce related communications).
These three factors together are combining to produce very
complex network architectures and requirements. Issues of
scalability, reliability, security and interactions between
services are increasingly replacing any other concerns network
operators may have had. Section 3 picks out some key areas
where agent technology may be able to play a leading role in
solving some of the most pressing network control problems.
Until recently, many agent applications have remained
nothing more than small pilot projects in the research laborato-
ry. One key reason behind this is that the necessary network
architecture for agent deployment was just not available. This
is changing. There are three main areas of technology push:
Mobile Agents:
The utility of mobile agents and mobile code
for network control has been a recurring theme since the
early 1990s. This paradigm is now beginning to gain wider
acceptance in the CN community (see works such as [Baldi
et al. 97] and [Breugst/Magedanz 98] for example). Conse-
quently, the likelihood that agent capable platforms might be
supported by future networks is increasing.
Programmable Networks:
Researchers in the relatively new
eld of Active Networks [Tennehouse et al. 97] argue for
programmable networks which can receive and execute
code on time scales down to single packet arrival. Programs
can be downloaded to a router using a “backdoor” mecha-
nism or injected into the network in the headers of individual
data packets. Either way, this type of programmable network
would greatly increase the scope for the deployment of
agent based network control services into the network
Ongoing standardisation efforts within
bodies such as OMG and FIPA are providing standard inter-
action mechanisms for agent based software. These efforts
to provide interoperability for agent applications are a key
factor in enabling the use of agent technology for a large
range of tasks – including network related applications (also
see Section 4).
Above all, agent technology is maturing as a software
development paradigm. Developping environments and stand-
ards is becoming available. This trend is likely to build con-
dence in agent techniques and allow more wide-spread experi-
Key Application Areas
Agent technology has been proposed for a large number
of network related areas (publications easily run to the 100s).
This section picks out three areas of network control in which
agent technology may have real potential to make a difference.
3. Some CISCO Systems routers in fact already include Java Virtual
Machines, however their interfaces have not yet been made avail-
able to network engineers.
Intelligent Agents
3.1 Multi Provider Environments
Market liberalisation and increasing demands for the
allocation of services which span several networks are pushing
every network provider to evolve the way they interact with
peer operators. In order to understand what “to evolve” means,
several factors have to be taken into account.
What were once state monopolies control-
ling everything from end user access down to the copper
wires have become several layers of competing rms –
service providers, networks providers, brokers, etc. Distinct
networks can be based on different technologies and can
deploy different network management platforms. This
implies heterogenity also of the information models used in
different networks, i.e., different Management Information
Databases (MIBs).
Distribution of resources:
Network resources can be owned
by many different “authorities”, that need to be made to
work together to support advanced services spanning several
domains (Virtual Private Networks crossing different
networks for example). This task is even more delicate for
networks which aim to provide any kind of Quality of
Service guarantees, since individual providers are unwilling
to release detailed information about the state or topology of
their internal network.
Currently, many aspects of the interaction
between distinct networks, are statically xed by contracts
(number and available capacity of links connecting one
network domain to another, prices, etc.) and many steps of
the interaction are regulated by human operators via fax, e-
mail, etc. This makes the overall inter-interoperability
process very slow (several months can pass before effective
inter-domain network conguration changes take place) and
A further major problem is that there is little or
no infrastructure to support robust information exchange or
coordination between different service and network
providers. In the best case, TMN compliant networks use
standard TMN-X interfaces which provide a rudimentary
low level interface for synchronising the settings in routers
and other network elements. A common database, the
Shared Management Knowledge (SMK), allows the
visualisation of a minimal amount of information that needs
to be shared for the interaction. Even in this case however,
humans are responsible for supervising and controlling the
Considering these aspects, what seems more suitable for
future networks is a management solution based on static
and/or mobile software entities, collecting network state infor-
mation and which have the ability to directly invoke effective
changes to switch controllers, without the interaction of a
human operator (see [Posladt al. 99], [Corley et al. 98],
[Calisti/Faltings 99] and several works in [Hayzelden/Bigham
99]). Software agents have strong potential since they can be
distributed, intelligent, expert, heterogeneous, self-learning
and dynamic [Bigham et al. 99]. As concrete examples of
potential agent usage in the multi-provider framework, (see
Figure 1) software agents acting on behalf of every network
operator could:
Reduce the need of human interventions and communica-
Abstract from technical details, such as SNMP primitives or
CMIS/CMIP terms, and translating them into a more under-
standable form for human operators.
Automate the control of switches and routers, i.e., active
Provide automatic service negotiation with both peer opera-
tors and nal end-users.
In [Calisti et al. 99] a multi-agent paradigm for the automatic
allocation of inter-domain service demands is dened. Here
one of the main challenges is to nd a way of making use of
restricted information to make adequate routing decisions
when passing through domains controlled by several different
authorities (see [Calisti/Faltings 99] for more concrete results).
3.2 Resource Management
Despite predictions of bandwidth glut ([Smith 99] amongst
others), network resource management remains a very
challenging area. In the United States, backbone deployment is
a race against time and user demand, uctuations and routing
errors can have catastrophic results
. With the rapid rise of the
Internet as an essential business tool there are also concerns
about the potentially serious effects of prolonged periods of
poor or reduced service. This factor, above all, is driving
companies to demand 1) improvements in the overall service
quality the Internet provides and 2) the minimisation of
potentially damaging periods of poor service. These demands
need to be met with increasingly sophisticated techniques for
resource management (such as the efforts now going on under
the umbrella term of Trafc Engineering [Bhaniramka et al.
99]) both at the backbone level and at the IP network level.
These efforts correspond to
controlling resource allocations
the network to improve the use of the available infrastructure.
This is either done using reservation protocols (such as RSVP
[rfc2205]) or in an across the board fashion. Agent technology
has already been applied to several problems in this area:
IP routing:
there have been various approaches to routing
problems but perhaps amongst the most intuitively appeal-
ing are those based on the use of mobile agents to mimic
“ant like” behaviour. First proposed in [Appleby/Steward
94] and [Schoonderwoerd et al. 97] and continued by several
other research groups, this approach simulates the trail
laying behaviour of social insects such as ants in simple
mobile agents. Individual ants migrate around the network
laying and reinforcing trails on their chosen paths. Packets
4. Note also that many “network supported” agent applications are
also being developed (such as agent based information services,
web auction houses, information lters etc.) which have greater or
lesser contact with the network infrastructure. This work is not
treated here. See sources such as proceedings of the PAAM series
of conferences (
for this type of application.
5. Such as the almost global outage due to an single erroneous
router conguration on March 25, 1997 [Labovitz et al. 98].
Intelligent Agents
(or calls) can then be routed along the strongest re-
reinforced trails which are biased to be the shortest and
encounter the least congestion (ants leave stronger signals
when arriving at a destination more quickly than their
counterparts). This approach has recently been adapted to IP
networks [Subramanian et al. 97] and developed to produce
more general protocols [Chen et al. 99].
Bandwidth commerce:
is a newly emerging model for net-
work resource management which is based on owners of
network infrastructure selling spare capacity in open mar-
kets. This trade is already happening
and is currently car-
ried out by human operators bidding for bandwidth goods.
Agent based interactions could potentially provide much
more exible interaction using standard agent interfaces, au-
tomatic negotiation, bargaining over bundles of goods
(which involves complex reasoning) and, not least, saving
the patience of human operators. Projects that are addressing
these possibilities include MACH
. Market based resource
allocation has also previously been tried for off-line alloca-
tions in work such as [Gibney/Jennings 98] and [Wellman
Connection-oriented networks:
work including [Hay-
zelden/Bigham 98] and [Willmott et al. 99] has shown how
agent systems could be used to control resources in back-
bone networks (based on ATM or other con-
nection oriented technologies). These
methods are particularly applicable for net-
works where large amounts of state data is
generally needed to make routing deci-
sions. To fully automate the network it
would seem a logical progression to allow
agents managing IP network resources to
communicate with agent counterparts
charged with managing backbone resourc-
In general terms, network resource manage-
ment can be described at many levels of gran-
ularity (from the routing of a single packet and
the routing of a ow up to the implementation
of network operator allocation policies).
Agents with properties such as those described
in Section 1.1 are more appropriate at the
higher levels of this description.
Figure 2 shows a two tier model often
applied in control problems (also adopted in
[Hayzelden/Bigham 98] and our own work).
Control is divided into two systems: an on-line
system responsible for fast time scale alloca-
tions (packet route decisions for example) and
a background system which monitors, controls
and updates the faster on-line system. The on-
line system makes most of the day to day rout-
ing allocations, however the supervisory sys-
tem would intervene if (for example): failures occur, operator
policies change, trafc congestion appears to be building or
certain types of trafc need to be specially treated. For this ap-
plication agents appear to be particularly suitable for the super-
visory style of control system since they can:
monitor and react to the environment – hence pro-actively
deal with undesirable trafc patterns in the network,
provide control in localised areas of the network using only
local information,
communicate with each other to ensure that a more
coherent, global resource allocation policy is enforced.
3.3 Communications Integration
There is a clear trend towards providing the user with servic-
es rather than network access or bandwidth. Furthermore, users
are increasingly demanding that information services (such as
news, e-mail, fax, telephone etc.) are integrated seamlessly.
These demands require multiple services provided by various
network technologies to be coupled together effectively. The
types of integration required can be broadly classied into two
interface integration
network integration
Interface integration:
the integration of many network
services requires complex coordination between network
infrastructure, end devices and services. Agent based
approaches have already been tested for this type of problem
(see [Abu-hakima et al. 96] for example which uses a
purpose built LAN). The integrated network service should
6. See,, and for example.
7. calisti/MACH/mach.html
Fig. 1: “Agentication” of the future networks: traditional management tasks and
typical human interactions are carried out by software agents.
management system
management system
human-agent interface
Gradually replaces human interaction
Network A
Network B
e.g. SNMP
Agent Switch
Intelligent Agents
Allow the addition of new services, network technologies
and end devices (such a new pager) dynamically – deliv-
ering the communications service over the newly added
medium when appropriate.
Group together existing services to make them appear as
on “virtual service”, for example integrating voice mail,
fax and e-mail by delivering the messages arriving over all
three media in whichever of the three formats is currently
most appropriate.
The top level of this service integration is human-machine
interaction which is in turn supported by coordination in the
network infrastructure to carry out the required services.
The key advantages of agents here are in their pro-activity
and exible interaction with the environment. Agents enable
the integration of humans and diverse hardware or software
entities by adapting their behaviour to individual preferenc-
es, characteristics of users and characteristics of the network
hardware [Fipa 98].
A concrete example of this kind of integration is the effort to
provide a Virtual Home Environment (VHE) for 3rd
generation mobile phone systems (UMTS). The aim is to
have mobile phone users presented with the same options,
services and interfaces wherever he or she is in the world
and whichever mobile phone provider is currently providing
these services. Both static [Lloys/Pearmain 99] and mobile
agent (the EU ACTS “On the Move”
project for example)
approaches have been proposed for this problem.
Network integration:
as the number of deployed network
technologies grows, providing homogeneous services
requires abstraction from technological details and stand-
ardised models for communication. Additionally, different
functional parts of the infrastructure may be owned by
different companies with, for example, service providers
leasing bandwidth from network operators.
Applying agents to integrate heterogeneous networks and
network technologies has been proposed both within the CN
community (in the TINA framework
) and within the DAI
community (with the FIPA agent network management
model, Section 7 of the 1997 FIPA Specication [Fipa 97]).
In the TINA architecture, software entities interact with both
humans and physical network devices, communicating over
a distributed execution environment. The FIPA architecture
further encompasses the notion of different authorities
owning different levels and parts of the network and looks to
address the question of establishing end-to-end services
over several (separately owned) networks (hence similar to
the issues discussed in Section 3.1).
This trend towards integration in all directions looks set to
continue and is perhaps one of the most challenging problems
networks of the future will have to face. In this area the concept
of agent middleware which bridges the technological and
architectural gaps in current systems seems to have great
potential (see [Poslad et al. 99]). Agents provide a means of:
Abstracting from the technological idiosyncrasies of differ-
ent network technologies to improve their interoperation.
Enabling richer and more exible interaction between both
user and system (user network service access) and system
and system (automatically exchanging tasks between
different agents to customise service delivery).
The Agent Future?
The three key application areas discussed in Section 3
cover a large part of the communications network infra-
structure, however this is not intended to advocate the use of
“agents everywhere”. The type of software agent which ts the
denitions given in Section 1.1 would arguably be inappropri-
ate for tasks which:
Required vary fast repetitive processing:
the utility of using
agents is generally in providing exible execution behaviour
to function correctly in a dynamic environment. By its very
nature this type of processing is likely to be less efcient for
highly constrained, repetitive tasks (such as packet
Required rapid and precise information exchanges:
larly in cases where agents communicate using standard
agent communication languages such as KQML [Finin et al.
93] and FIPA ACL [Fipa 97], the exibility in agent
communication may be problematic. For many network
tasks, highly constrained, concise protocols are the best way
of exchanging information.
Need to execute on very low specication devices:
may well be pieces of software of substantial complexity
and not be able to run everywhere. This obstacle is gradually
being removed by smaller footprint agent platforms and
more performant network devices.
8. onthemove/
10. This is not to say agents cannot also employ these protocols, how-
ever DAI purists might argue these do not completely ll the role
of agent communication.
Fig. 2: The routers in the network each have an on-line
allocation mechanism. Agents communicate with each other
to resolve longer term allocation problems, occasionally
intervening in the on-line system’s operation.
Agent Architecture
Running Network
Network Node with
on-line routing
Agent Controller with
background influence
over a set of network
Intelligent Agents
Passing down the network stack and to operations which
need to be carried out at faster and faster time scales, one would
expect control software to have less and less of the features
listed in Section 1.1. However, this change is likely to be a
continuum rather than a sharp break and deciding where the
dividing line between “agent” and “non-agent” will perhaps
become somewhat academic (Figure 3).
In summary, entities near the bottom of the stack (such as an
SMNP agent or deployed service code) may be mobile and/or
have very limited tasks where as entities higher in the stack
(such as top level managers or user interface agents) may begin
to have properties which a DAI researcher might nd agent-
like. The two tier resource management model given in Section
3.2 illustrates this idea: agents are applied as layers of super-
visory systems controlling layers of increasingly constrained
and optimised on-line systems. This model is analogous to
what already goes on in networks today and is characterised by
[Musliner et. al. 95] as
intelligent reasoning about real-time
Aside from determining where in the network agents should
be deployed, there are also wider considerations which need to
be addressed before agent technology can realise its full
potential in communication networks. The future deployment
of agent technology rests critically on building increased
cooperation between the traditionally separate DAI and CN
communities. Apart from the terminology problems already
mentioned, the division has created other obstacles to
Continuing (to some extent justied) scepticism on the part
of communications network engineers as to the utility and
suitability (in terms of security, robustness, speed of
operation etc.) of agent technology. This has resulted in a
lack of tested practical solutions and many approaches
which have never made it beyond the test bed stage.
The biggest stumbling block for DAI researchers has
perhaps been the technological complexity of the networks
being studied. It would be fair to say that several of the
promising methods developed by DAI researchers in the
past have met with little success due to failings in the
starting assumptions about the network domain.
Agent solutions which have been proposed by the CN
community have remained very simple and not leveraged
some of the more powerful techniques developed by the DAI
There are indications that this collaboration is increasing and
that the interests of the two communities are growing together.
The papers presented at the Smartnet
and DSOM
shops this year, for example, include a signicant number of
agent related papers. Furthermore, the continuing interest in
agent technology within the OMG and FIPA standards bodies
for example (both of whom list many member companies heav-
ily involved in communication networks including: British
Telecom, France Telecom, Nortel, Motorola and many others)
is encouraging. The European Union AgentLink project
also contributing to this collaboration with a special interest
group dedicated to the application of agent technology to tele-
communications networks.
Having covered three areas which might greatly benet
from the application of agent technology and discussed some
of the provisos in its application, we can conclude by advanc-
ing three main reasons for believing that future network
developments may include the deployment of agents:
Need for innovation:
Increasing competition, technological
complexity and usage requirements are all contributing to
increased strain on network infrastructure. This push is
making innovative solutions (and potentially agent
solutions) to network problems vital for ensuring continued
good service.
Technological feasibility:
Agent technology is maturing as a
software paradigm. Alongside the increasing availability of
development environments it is increasingly likely that
deployed network equipment will in the future be able to
support the computational needs of agents.
Increase industry openness:
As the continuing collabora-
tions within FIPA, OMG and many European projects show,
Fig. 3: In an “agentied” model of the network software entities
operate at different levels. Physical Agents (PA) might control
specic network elements (as simple input-output sensors for
example). Resource Agents (RA) might invoke changes in the
switches and routers using information coming from both the
higher and lower levels in the network. Mediator Agents (MA)
might be more sophisticated entities needing to be able and
inter-operate with other entities by using a common agent
language. Finally, Interface Agents (IA) could translate from
agent languages to more human understandable information.
The lower down through the layers an entity resides, the less
sophisticated it is and the less developed its “agent properties”
might be considered to be - the decreasing sophistication is
illustrated by an increasingly dashed line.
Osi reference model
Intelligent Agents
the communications network industry is increasingly open
to experimentation with agent based solutions.
Together, these reasons suggest that there could be a slow
agentication at least of the upper layers of the network infra-
structure – little by little – agents may begin to appear in our
networks. The technologically dynamic communications
industry is however known for its frequent changes of tack so
only time will tell.
Due to the limited space available this article can only give a
brief overview of the subject area. We hope to have included
enough pointers to literature (in particular see the survey
articles referenced in Section 1) to serve as a useful starting
point for further reading.
The authors would like to extend their thanks to the other partners
in the SPP-ICC IMMuNe project (which partly funds this work).
Funding for IMMuNe from the Swiss National Science Foundation
is also gratefully acknowledged.
[Abu-hakima et al. 96]
S. Abu-hakima, S. Liscano, and R. Impey. Cooperative Agents
that Adapt for Seamless Messaging in Heterogeneous Communi-
cation Environments. In Proceedings of the AAAI- 96 Workshop
on Cooperative Information Agents - Portlan Oregon. AAAI
Press, 1996.
[Appleby/Steward 94]
S. Appleby and S. Steward. Mobile Software Agents for Control
in Telecommunications Networks. British Telecom Technology
Journal, 12(2), 1994.
[Baldi et al. 97]
Mario Baldi, Silvano Gai, and Gian Pietro Picco. Exploiting code
mobility in decentralized and exible network management. In
Proceedings of the First International Workshop on Mobile
Agents, Berlin, Germany, April 1997.
[Bhaniramka et al. 99]
P. Bhaniramka, W. Sun, and R. Jain. Quality of service using
trafc engineering over mpls: An analysis, March 1999. draft-
[Bigham et al. 99]
J. Bigham, L. Cuthbert, A. Hayzelden, and Z. Luo. Multi-agent
system for network resource management. Lecture Notes in
Computer Science, 1597:514-526, 1999.
[Braden et al. 97]
R. Braden, L. Zhang, S. Berson, and S. Herzog, S. and Jamin.
Resource ReSerVation Protocol (RSVP) - Version 1 Functional
Specication. RFC 2205, IETF Network Working Group,
Standards track, September 1997.
[Breugst/Magedanz 98]
M. Breugst and T. Magedanz. On the Usage of Standard Mobile
Agent Platforms in Telecommunication Environments. In S.
Trigila et al., editor, Proceedings of 5th Int. Conference on
Intelligence in Services and Networks (IS&N), Lecture Notes of
Computer Sciences 1430, Intelligence in Services and Networks:
Technologies for Ubiquiteous Telecom Services, pages 275-286,
Antwerp, Belgium, May 1998. Springer Verlag.
[Calisti/Faltings 99]
M. Calisti and B. Faltings. A multi-agent paradigm for the Inter-
domain Demand Allocation process. DSOM’99, Tenth
IFIP/IEEE International Workshop on Distributed Systems:
Operations and Management, 1999.
[Calisti et al. 99]
M. Calisti, C. Frei, and B. Faltings. A distributed approach for
QoS-based multi-domain routing. AiDIN’99, AAAI-Workshop
on Articial Intelligence for Distributed Information Network-
ing, 1999.
[Chen et al. 99]
J. Chen, P. Druschel, and D. Subramanian. A new approach to
routing using dynamic metrics. In Proceedings of INFOCOM99.
[Corley et al. 98]
S. Corley, M.Tesselaar, J. Cooley, and J. Meinkoehn. The
application of intelligent and mobile agents to network and
service management. Lecture Notes in Computer Science,
1430:127-??, 1998.
[Finin et al. 93]
Tim Finin et al. Specication of the KQML Agent-Communica-
tion Language - plus example agent policies and architectures,
[FIPA 97]
Foundation for Intelligent Physical Agents (FIPA). FIPA Agent
specication 1997. Technical report, FIPA, homepage - http://
www., October 1997.
[FIPA 98]
Foundation for Intelligent Physical Agents (FIPA). FIPA Agent
specication 1998. Technical report, FIPA, homepage - http://
www., October 1998.
[Gibney/Jennings 98]
M. A. Gibney and N. R. Jennings. Dynamic Resource Allocation
by MArket-Base Routing in Telecommunications Networks. In S.
Albayrak and F. J. Garijo, editors, Proceedings Second Interna-
tional Workshop on Intelligent Agents for Telecommunications
Applications IATA’98, pages 102-117. Springer (as LNAI-1437),
[Hayzelden/Bigham 98]
A. L. G. Hayzelden and J. Bigham. Heterogeneous Multi-Agent
Architecture for ATM Virtual Path Network Resource Congura-
tion. In S. Albayrak and F. J. Garijo, editors, Proceedings Second
International Workshop on Intelligent Agents for Telecommuni-
cations Applications IATA’98, pages 45-59. Springer (as LNAI-
1437), 1998.
[Hayzelden/Bigham 99]
A. L. G. Hayzelden and J. Bigham. Software Agents for Future
Communication Systems. Springer Verlag, April 1999.
[Jennings/Wooldrige 98]
N. Jennings and M. (eds) Wooldrige. Agent Technology Founda-
tions, Applications, and Markets. Springer/UNICOM, February
[Kumar/Venkataram 97]
G. P. Kumar and P. Venkataram. Articial Intelligence Approach-
es to Network Management: Recent Advances and a Survey.
Computer Communications, 20:1313-1322, 1997.
[Labovitz et al. 98]
C. Labovitz, A. Ahuja, and F. Jahanian. Experimental study of
internet stability and wide-area backbone failures. Technical
Report CSE-TR-382-98, University of Michigan Department of
Electrical Engineering and Computer Science, December 16,
[Lewis 95]
L. Lewis. AI and Intelligent Networks in the 1990s and into the
21st Century. In J. Liebowitz and D. Prerau, editors, Worldwide
Intelligent Systems. IOS Press, 1995.
[Lloys/Pearmain 99]
S. Lloys and A. Pearmain. Multi Agent System for Establishing
‘Virtual Home Environments in the Convergence of Fixed and
Mobile Telecommunications Networks. In S. Willmott and S.
Abu-Hakima, editors, Proceedings of the 3rd Workshop on
14. Project Number SPP-ICC 5003-45311. See I
3/1999, p. 29.
Intelligent Agents
Articial Intelligence for Distributed Information Networks, held
at AAAI’99, pages 78-83. AAAI Press, 1999.
[Maes 94]
Pattie Maes. Agents that Reduce Work and Information Overload.
Communications of the ACM, 37(7):31-40, July 1994.
[Martin-Flatin/Znaty 2000]
S. Martin-Flatin and J. P. Znaty. Two Taxonomies of Distributed
Network Management Paradigms. In S. Erfani and P. Ray,
editors, Emerging Trends and Challenges in Network
Management. Plenum Press, New York, NY, USA, March 2000.
[Musliner et al. 95]
D. J. Musliner, J. A. Hendler, A. K. Agrawala, E. H. Durfee, J. K.
Strosnider, and C. J. Paul. The Challenges of Real-Time AI. IEEE
Computer, 28(1), January 1995.
[Poslad et al. 99]
S. Poslad, J. Pitt, R. Mamdani, A. Hadingham, and P. Buckle.
Agent-Oriented Middleware for Integrating Customer Network
Services. In A. Hayzelden and J. Bughma, editors, Software
Agents for Future Communciations Systems. Springer Verlag,
[Schoonderwoerd et al. 97]
Ruud Schoonderwoerd, Owen Holland, and Janet Bruten. Ant-
like agents for load balancing in telecommunications networks.
In W. Lewis Johnson and Barbara Hayes-Roth, editors, Proceed-
ings of the 1st International Conference on Autonomous Agents,
pages 209-216, New York, February5-8 1997. ACM Press.
[Smith 99]
J. M. Smith. Programmable Networks: Selected Challenges in
Computer Networking. IEEE Computer Magazine, 32(1):40-42,
January 1999.
[Subramanian et al. 97]
D. Subramanian, P. Druschel, and J. Chen. Ants and Reinforce-
ment Learning: A Case Study in Routing in Dynamic Networks.
In Proceedings of IJCAI’97, pages 832-838. 1997.
[Tennenhouse et al. 97]
David L. Tennenhouse, Jonathan M. Smith, W. David Sincoskie,
David J. Wetherall, and Gary J. Minden. A survey of active
network research. IEEE Communications, 35(1):80-86, January
[Weihmayer/Velthuijsen 94]
Robert Weihmayer and Hugo Velthuijsen. Application of Distrib-
uted AI and cooperative problem solving to telecommunications.
In Proceedings of the 13th International Workshop on Distributed
Articial Intelligence, pages 378-402, Seatle, WA, July 1994.
[Weihmayer/Velthuijsen 98]
R. Weihmayer and H. Velthuijsen. Intelligent Agents in Tele-
communications. In N. R. Jennings and M. Wooldridge, editors,
Agent Technology Foundations, Applications and Markets, pages
201-217. Springer Verlag and UNICOM UK, 1998.
[Wellman 94]
M. P. Wellman. A Market-Oriented Programming Environment
and its Application to Distributed Multicommodity Flow
Problems. Journal of Articial Intelligence Research, 1(1):1-23,
[Willmott et al. 99]
S. N. Willmott, C. Frei, B. Faltings, and M. Calisti. Organisation
and Co-ordination for Online Routing in Communications
Networks. In A. L. G. Hayzelden and J. Bigham, editors,
Software Agents for Future Communication Systems, pages 130-
159. Springer Verlag, 1999.
... Components characteristics evaluate alternative situations within the fuzzy agent interface based on the events and defined decision strategies. The agent architecture and communication structure are evaluated by Wilmott and M. Calisti in general (2000)[1]. James S. Albus (2002)[2] presents the reference architecture model. This model enlightens the robot architecture mechanism. ...
... This function is represented as the intersection of r tuples, denoted as a decision measure, M(C i , b i ), involving objectives and preferences. Elements of preference set can be linguistic values such as none, low, medium, high, absolute, or perfect; or it could be values on the interval [0,1]; or it could be values on any other linearly ordered scale e.g., [-1, 1], [1,10] etc. These preferences will be attached to each of the objectives to quantify the decision maker's feelings about the influence that each objective should have on the chosen alternative. ...
... This function is represented as the intersection of r tuples, denoted as a decision measure, M(C i , b i ), involving objectives and preferences. Elements of preference set can be linguistic values such as none, low, medium, high, absolute, or perfect; or it could be values on the interval [0,1]; or it could be values on any other linearly ordered scale e.g., [-1, 1], [1,10] etc. These preferences will be attached to each of the objectives to quantify the decision maker's feelings about the influence that each objective should have on the chosen alternative. ...
Conference Paper
This study deals with the interface environment of an intelligent fuzzy agent model, including the evaluation of the fuzzy approach in decision making mechanism in the flexible manufacturing environment. The behavior style of this decision making mechanism is expressed in multi agent system. The fuzzy logic approach is preferred to expedite the process of decision making in the system. This will also reduce uncertainty in the process in the interface environment. The suggested system provides decision-making and evaluation of results for manufacturing activities in the fuzzy situation. The evaluation of the decision-making process includes the agent list, task list, rule list and knowledge list structures that are combined in the fuzzy inference mechanism.
... The data networks that provide us with ubiquitous access are an example which is already partly a multi-agent system because of the autonomous functioning of its routers. Current research tries to lever-age its services by means of agents which communicate with each other and intervene with the routers [3]. An other application area includes the electrical power networks. ...
Full-text available
The analysis and design of solutions in terms of multi-agent systems is an emerging technology. We have con-ducted a pilot in teaching a short agent-based simulation course at an applied sciences university. The involved teachers and students were neither professional pro-grammers nor computer specialists. By the content of its units and through the functionality of our custom simula-tion environment the course emphasised parallelism and uncertainty. Our research allowed us to collect some material on how teachers and students cope with the way we presented these fundamental principles. The discus-sion is aimed at improving the course but it could be also of interest to professionals that build multi-agent systems.
... It includes self-configuration, self-optimization, self-healing and self-protection. In terms of resource management, increasingly sophisticated techniques are needed to improve the overall service quality and minimize the effects of potentially damaging periods of poor service [3]. Vilà and al [4] have proposed a dynamic bandwidth management scheme in logical network such as ATM or MPLS based on distributed agents. ...
As Internet is becoming the global infrastructure for all media communications, ensuring the reliability and the quality of network services is far from being a secondary task. Indeed, many business activities rely on the network and any performance degradation can result in serious financial losses. The main objective of Internet Service Providers is to maintain permanently an acceptable service delivery for their subscribers and respect scrupulously the Service Level Agreements. To achieve this goal, just committing bandwidth resources is not sufficient since network services are not safe from breakdowns. The dysfunction of some of its elements or persistent overloads generated by bursts traffics can induce serious deterioration of the network performance and impact the end users’ applications. Continuous service monitoring is then required and an autonomous network resource regulation mechanism is necessary in such situation to avoid the degradation and the collapse of all the applications sharing the impacted resource. In this paper, we propose a decentralized approach for the monitoring and management of the network backbone shared resource by the mean of distributed autonomous agents. The agents are deployed at the edge routers and cooperate together in order to maintain a global acceptable level of service in critical situations where the current quality of service is less than the expected one. This allows self adaptive service and resource management with an interesting abstraction from network backbone heterogeneity and complexity. KeywordsAgents-Resource management-Network performance-Active probing
IntroductionIP networks and their managementThe multi-agent paradigmMAS for IP network managementPerspectives and conclusionBibliography
Conference Paper
This study deals with the interface environment of an intelligent fuzzy agent model, including the assessment of the fuzzy approach in decision making mechanism. The behavior style of this decision making mechanism is expressed in an analytic model. In general, the interface environment consists of the domain model, decision making mechanism, dynamic knowledge acquisition and case-based reasoning. The fuzzy logic approach is preferred to expedite the decision making process in the system. This will also reduce uncertainty in the process in the interface environment. An analytic model defines the response to the tasks, the evaluation of alternative situations and optimization of the decision strategies in a fuzzy situation. The sections in this study define the agent interface components as well as the agent activities such as perception, learning, making a decision, and assisting support units. The last section covers the evaluation of the whole system with the analytic model.
Full-text available
From the perspectives of social science and computer science, the time has come to truly examine the social dimensions of machines. This paper argues that it is sensible to narrow a social science of men and machines down to a social science of men and agents. Agents can function as a single actor, or as parts of multi-agent environments, or as actors in society. A variety of theories serve to design these complex machines with communication, coordination, and cooperation as the fundamental dimensions of multi-agent environments. There is a wide and increasing range of mechanisms dealing with these problems. However, a general theory of multi-agent environments is absent. A general theory would be useful as the issues of communication, coordination, and cooperation involve a variety of problems that occur in most multi-agent environments but of which every specialised approach only tackles a subset. Eventually, as social actors in society, agents demonstrate features that distinguish them from traditional software on the one hand and from humans on the other hand. (PDF) Software Agents take the Internet as a Shortcut to Enter Society: A Survey of New Actors to Study for Social Theory.. Available from: [accessed Mar 11 2019].
. A particular challenging area where agent technology is increasingly applied is the Communication Networks field. As networks become increasingly complex, hard to manage and control the ideal of a distributed, intelligent network management and control system is becoming more and more of a necessity. This paper concentrates on the allocation of service demands spanning network domains owned and controlled by distinct operators. In particular, the focus is on an agent-based solution that has been defined in order to allow automatic intra-domain resource allocation and inter-domain negotiations between peer operators. Self-interested agents interact in order to define consistent end-toend routes that satisfy Quality of Service requirements, network resources availability and utility's maximisation. 1 Introduction Recently, many researchers have been investigating the potential of introducing autonomous software agents for distributing communication network control and mana...
Full-text available
There are a variety of forces causing previously disparate networks and services to be unified or integrated, including cost-reduction, ease of configuration and ease of maintenance. Communication networks of the future will consist of a wide variety of inter-linked computer networks, giving customers access to a huge range of potentially competing network services. From a customer perspective, the unification and integration of heterogeneous networks means that a service user (or value added service provider) can access any service from anywhere at any time. Examples of how services are coalescing include a diverse range of services from unified messaging services (incorporating FAX, voice-mail, voice and email) to the diversification of services offered by supermarkets.
Full-text available
A cursory survey of applications in the DAI literature suggests that a primary area of using DAI technology for real-world systems can be found in telecommunications. This paper explores the current state-of-the-art of DAI and cooperative problem solving in this domain. Telecommunication networks have proven in the last five years to be a fertile ground for problems involving the coordination of distributed intelligence. A wide range of problems from distributed traffic management to resolution of service interactions in intelligent networks have led to conceptual studies and prototype developments involving systems of cooperative agents. Although there are currently no fielded systems that can be said to be DAI-based, we believe these will begin to appear within the next five years. We give an overview of the full range of potential applications found in the literature and additionally consider four applications in more detail, including the authors contributions to the field.
From the Publisher: This monograph-like anthology is the first systematic introduction to software agents and their application to future communications systems. Fifteen coherently written chapters by leading software agent researchers provide complementary coverage of the relevant issues. Multi-agent systems and mobile agent approaches are presented, in a well-balanced way and applied to the most important topics in future communications systems. In addition, the volume editors have provided detailed introductory survey chapters.
The underlying fabric for communication among intelligent agents will in many cases be provided by telecommunication networks. But telecommunication networks have been seen as a natural domain for the investigation and application of intelligent agents technology as it emerged from the area of Distributed Artificial Intelligence (DAI). Telecommunication network administrations are vast organizations dedicated to operating and managing networks with broad functional segmentations: telephone network outside plant, switching and transmission plants, public network, all supporting different layers of specialized customer or service networks. These networks are organized into multiple physical and logical layers built with large quantities of repeated network elements and subnetwork structures. All these elements need to be configured, monitored, and controlled. In the future, this will preferably be done by automated operation support systems and without substantial human intervention.
With increased number of new services and users being added to the communication network, management of such networks becomes crucial to provide assured quality of service. Finding skilled managers is often a problem. To alleviate this problem and also to provide assistance to the available network managers, network management has to be automated. Many attempts have been made in this direction and it is a promising area of interest to researchers in both academia and industry. In this paper, a review of the management complexities in present day networks and artificial intelligence approaches to network management are presented.
Conference Paper
We present an approach to resource allocation in telecommunications networks based on the interaction of self-interested agents which have limited information about their environment. A system architecture is described which allows agents representing various network resources, potentially owned by different real-world enterprises, to coordinate their resource allocation decisions without assuming a priori cooperation. It is argued that such an architecture has the potential to provide a distributed, robust and efficient means of traffic management for telecommunications networks. Some preliminary work on the design of the trading behaviour of the agents in the economy is presented, including the results of experiments which investigate the relative performance of market-based agents compared with traffic management based on static routing.