An Agent-based Framework for Multi-domain Service Networks
Eduroam Case Study
Ameneh Deljoo
1
, Leon Gommans
2
, Tom van Engers
3
and Cees de Laat
4
1
Institute of Informatics, University of Amsterdam, Amsterdam, Netherlands
2
Air France-KLM, Amsterdam-Schiphol, Netherlands
3
Leibniz Center for Law, University of Amsterdam, Amsterdam, Netherlands
4
Institute of Informatics, University of Amsterdam, Amsterdam, Netherlands
Keywords:
Normative Agents, Agent-roles, Service Provider Group, Story Animation, Petri Nets, Distributed Simulation
Models.
Abstract:
This paper introduces a methodology for the acquisition of the computational model of a service provider
group and its transformation into agent-based model. The methodology is as follows. First, we analyze the
case at the signal layer, i.e. the message exchange between actors, and model them with the components
of “belief, desire and intention (BDI)” agent architecture. In the next step, we identify the implicit actions,
intentions, and conditions which are necessary for the story to occur. These steps correspond to descriptions
of agent-roles observed in the case study. As a concrete result, a preliminary implementation of the framework
has been developed with Groovy.
1 INTRODUCTION
A priori identification of benefits and risks to stake-
holders that collaborate to provide a service across
multiple service domains is a problem that depends on
the goals, benefits and capabilities of multiple service
providers. In this paper, we will explore the feasibility
of Agent Based Modeling(ABM) use as a first step to
identify the benefits and risk in such an open domain
system. ABM has been introduced to model an open
domain where agents are self-governed autonomous
entities that pursue their own individual goals based
only on their own beliefs and capabilities (Abdelka-
der, 2003). Open domain systems have an intuitive
mapping onto an ABM. An open domain system con-
sists of smart, cooperative and autonomous agents
where each of them has its own goal to achieve.
Agents present specific roles in this system and inter-
act with others as a means to accomplish their goals.
Modeling such systems receives considerable atten-
tion from both the Artificial Intelligence (AI) and the
communications network communities (Abdelkader,
2003; Dignum et al., 2005). ABM provides a way to
investigate the benefits of collaborating autonomous
agents. A service network is an example of an open
system. Service networks are composed of competi-
tive service providers that see the benefit in collabo-
ration. It is important to note that in such networks,
each member cannot provide a service on its own and
collaboration provides benefits such as reduced cost
or increased revenue. For instance, providing authen-
ticated Eduroam WiFi access to visiting students is an
example of a campus IT service that a single univer-
sity is unable to provide on its own without having
a collaboration with other universities. The Service
Provide Group (SPG) (Gommans et al., 2015) is a
way to describe such collaboration. The SPG frame-
work provides a way to organize thinking about multi-
domain service network and can be used to describe
the structure of such a collaboration. Eduroam is a
good example of such collaborations. In this paper,
we take the Eduroam confederation as an example of
open domain system which consists of multiple au-
tonomous agents, where each of them have their own
goal and intent to collaborate.
ABM is an effective platform for the SPG because
they provide mechanisms to allow organizations to
advertise their goal, negotiate their terms, exchange
rich information, and synchronize processes at a high-
level of abstraction (Preece et al., 1999). A compre-
hensive model for ABM must be able to express the
global goal and the requirements of the domain in a
distributed way by considering the autonomy behav-
ior of the SPG.
Deljoo, A., Gommans, L., Engers, T. and Laat, C.
An Agent-based Framework for Multi-domain Service Networks - Eduroam Case Study.
DOI: 10.5220/0005821502750280
In Proceedings of the 8th International Conference on Agents and Artificial Intelligence (ICAART 2016) - Volume 1, pages 275-280
ISBN: 978-989-758-172-4
Copyright
c
2016 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
275
This paper aims at presenting an ABM for multi-
domain service network. We demonstrate the trans-
formation of a sequence of inter-agent interactions
into intra-agent characterizations. The paper is orga-
nized as follows. The methodology is introduced in
section 2. In section 3, we present our SPG case study.
We present it at the signal layer, defining the topol-
ogy and the story line. In the following we show how
to enrich the previous representations with an inten-
tional layer, integrating institutional concepts as well.
In section 4 we provide elements for the transforma-
tion of the previous models into scripts for cognitive
agents. Discussion and further steps will be presented
at the end.
2 METHODOLOGY
The method we used to provide an ABM model of
a SPG is as follows: the case was analyzed at the
signal layer, i.e.we identified all the events and their
messages; then we visualized them by using a mes-
sage sequences chart (MSC). Next, we integrated the
previous layer with an internal behavioral characteri-
zation, a specification and we constructed an agentic
layer where we addressed intentions of agents. For
this purpose, we defined the mental objects and events
from a BDI perspective, maintaining the relationship
between components. Finally, we embedded an in-
stitutional layer to consider the normative aspect. It
should be noted that our conceptual framework covers
several realities: physical (message exchange), men-
tal (social network), institutional and story line.
3 Eduroam AS A SERVICE
PROVIDER GROUP
This study has been motivated by Leon Gom-
manse (Gommans, 2014), who is a researcher on
multi domain authorization frameworks. His SPG re-
search considers the role of trust and power to autho-
rize a service delivery by multiple autonomous ser-
vice provider. In general terms, Eduroam [EDUR]
allows students, researchers and staff from participat-
ing institutions to obtain Internet connectivity (WiFi)
across campus and when visiting other participating
institutions by simply opening their laptop. Eduroam
allows participating research and education institu-
tions, known as an eduroam SPG, to provide internet
access for students from any other participating insti-
tute that acts as Identity Provider (IdP). To reduce the
complexity of case study, we only consider one SP
Figure 1: Example of occurrence: a WiFi connection in-
stance.
(university) and students who are willing to use this
free service. In the following section, we describe the
story of Eduroam
1
Eduroam network is an example of
a SPG and works as follows:
A university offers a WiFi access to students
and staff who are registered in the Eduroam
identity database. The student accepts the
term and condition of this free service. The
university requests an identity from the stu-
dents. The student provides a valid identity.
Finally, the university will deliver the service
(free WiFi).
A successful WiFi connection is a fundamental cross
domain transaction. Consequently, what the case de-
scribes is a collaboration among the SPG (university)
and their users (students) which is just one of many
other possible scenarios. Figure 1 shows a successful
Eduroam connection process.
3.1 Signal Layer
In order to initiate the modeling, first we look at the
speech acts of agents and all the events to illustrate the
first layer of our framework (called signal layer). As
a first definition, we may consider a story as a chain
of events which act as functions to bring the story
from a set of initial conditions to a certain conclusion
(from this perspective, we can consider all of them
as acts of communication, as messages going from
a sender to a receiver entity). In addition, each ac-
tion is coupled with the acknowledgment by the other
party. The Eduroam service delivery process is ba-
sically characterized by the actions offer, accept, re-
quest, provide and open access, performed by one of
the parties as represented in Figure 1. This process
is protected by an Eduroam confederation agreement,
when a mandatory action is not executed, the party
who was expected to benefit from the act can enforce
on a failed performance. In order to trigger the perfor-
mance of the reported acts, there might be other con-
1
Eduroam is a federated roaming service that provides
such secure network access by authenticating a user with
their own credentials issued by their IdP. A group of Na-
tional Research and Education Networks (NRENs) are in
essence providing this service for their participating educa-
tional institutes under the eduroam brand arranged by TER-
ENA (L
´
opez et al., 2008; Wierenga and Florio, 2005; Gom-
mans et al., 2015).
ICAART 2016 - 8th International Conference on Agents and Artificial Intelligence
276
University
Student
Out
In
Out
offer
Provide ID
Accept
In
deliver
In
Out
Provide ID
deliver(Access)
University
Out
In
out
offer
Deliver(Open Access)
Accept
Provide ID
In
Student
Figure 3: Topology of an Eduroam story: direct and indirect communications.
University
Student
offer
Accept
Deliver(Open Access)
Provide ID
request ID
Approved
Figure 2: Message sequence chart of a story about an
Eduroam WiFi service.
ditions or hidden acts to be taken into account. Thus,
the story can be illustrated in Figure 2. A student
usually accepts a term and condition only if he/she
holds a valid identity and physically presents at the
Eduroam WiFi range. To be Registered as a student
is a precondition for successfully perform the autho-
rization steps. Both physically presents and holds a
valid Id are examples of critical conditions for com-
pleting the process by this story. Such critical condi-
tions are in general associated to the ability or, more
in general, to the power of the agent, in a specific con-
text ”agent+environment”. They identify the proposi-
tions that should be true in the story-system, so that
the agent is successful in the performance of the asso-
ciated action.
3.2 Topology
The topology is drawn based on the collection of mes-
sages. The topology serves as a still picture of the
whole story-system, and shows how signals are dis-
tributed over the agent-roles (Boer and Van Engers,
2011). In our approach, topology helps to identify an
agent in a certain agent-role which is shown by the
MSC (e.g. Figure 2). This part of the research has
been inspired by an ”actor model” of (Hewitt et al.,
1973). For simplicity, we only consider two possi-
ble representations of the topology (direct and indi-
rect communication), which has been illustrated in
Figure 3. The figure’s right side, the little boxes are
messages queues and lines are communication chan-
nels. Also, all messages have a specific propositional
content. On the left figure, the dashed lines refer to
actions that have relevant outcomes besides the di-
rect communication. In order to take eventual side-
Figure 4: MSC with intentions and critical conditions.
effects into account, we introduced an explicit world
actor, dis-joining the sender from the receiver (which
is presented in the right figure). The world would play
as intermediary entity also in case of broadcast mes-
sages. In the Eduroam case, world plays as a role of
an IdP.
4 AGENT PERSPECTIVE
In the previous section we referred to the message ex-
change in this story (Eduroam WiFi access). When
we discover such messages, however, we apply an
intentional stance introduced by Dennett (Dennett,
1989). We started from a representation of the
Eduroam WiFi access story on the MSC chart and
we refined it with Petri nets patterns. Beside that we
want to extract agent-role descriptions from this rep-
resentation. Therefore, we write down typical scenar-
ios (e.g. Figure 1) with the typical roles that agents
play. Such roles are associated with certain beliefs,
plans (resulting in actual actions) and goals. An agent
is able to play simultaneously several roles and vice-
versa. From the agent point of view, the precondition
and ex-post intentional interpretation of the story re-
sults in a decomposition of the plans followed by the
agents. A possible result of the interpretation is pre-
sented in Figure 4. In order to do so, first externalized
intents have been considered as the events triggering
the processes of offer/open access between a univer-
An Agent-based Framework for Multi-domain Service Networks - Eduroam Case Study
277
sity and a student. The final results of those actions
are then reported with output messages at the end of
the chart. In the next step, we consider the possible
hidden acts. In this case, we know that the univer-
sity usually accepts student’s requests for using the
WiFi, once their identities have been approved by the
IdP. And finally, we use the critical grouping to high-
light conditions (in addition to sequential constraints),
which are necessary for the production of that mes-
sage. To sum up, we assume that: (a) the student per-
forms an evaluation of the offer (evaluation action),
(b) the student accepts the offer if it is acceptable for
him/her (term and acceptability condition), and (c)
the student provides the ID (the university provides
access) if he owns the requested information (valid
identity or ownership condition). The MSC diagram
in Figure 4 depicts a good conclusion for the story;
the inputs/outputs provide an intentional characteriza-
tion. The vertical bars indicate the ongoing activities,
while the messages refer to successful acts of emis-
sion and reception, whose occurrence is constrained
by the critical conditions. In the following sections
we will introduce some patterns to be attached to the
flow of the story. Instead of using just one visualiza-
tion, we provide alternative representational models.
In our model, we refer to four layers, each of which
addresses specific components:
the signal layer—acts, side-effects and failures
(e.g. technical failure,user abuse): outcomes of
actions,
the action layer—actions (or activities): perfor-
mances intended to bring about a certain result,
the intentional layer—intentions: commitments to
actions, or to build up intentions,
the motivational—motives: events triggering the
creation of intentions.
The last three layers compose the agentic layer. The
closure of the sensing-acting cycle of the agent is
guaranteed by the fact that certain signals, when per-
ceived by agents, becomes motives for action. In our
framework, motivation refers to conditions that makes
the agent sensitive to a certain fact, which becomes
the motive for starting an action. As we observed be-
fore, obligations are prototypical reason for actions.
Despite of that, not all obligations are followed by a
performance. People comply with obligations when
they have some motivation due to habit, convenience,
respect for authority, or in our case to use WiFi. Mo-
tivations however often remain implicit in the story
(Sileno et al., 2013).
4.1 Institutions
In general, we can say that an institution is an inten-
tional social collective entity (Boer, 2009), defined by
certain rules and some institutional facts. It is col-
lective and intentional, simply because a group of
people recognizes its existence. A complementary
perspective on institutions has been introduced by
Searle (Searle, 1969), and more recently by (Searle,
2009), as an outgrowth of his study. This concept of
institution unifies games, (social) informal norms and
legal norms. Terms like ”university” and ”student”
denote agents acting within the free WiFi institution.
However, there is an intrinsic difference between the
actual participants and the role that they play. An
institution concerns persons, but not whole persons:
each one enters via an adequately trained and special-
ized part of himself. These parts are embodying spe-
cific institutional roles.
Agent-role model were first introduced by Boer
and Van Engers (Boer and Van Engers, 2011) with
the purpose of representing scenarios of compliance
and non-compliance elicited from legal experts. In
this work, an agent-role links the concepts of institu-
tional role, and intentional agent. In practice, we add
characteristics to the role that are important factors
according to the constructed normative theory, and
we describe its behavior by using an intentional ap-
proach. We start considering only the core functions
(events, acts) related to our character. This process
proceeds by using a common knowledge interpreta-
tion to define each agent’s intention. Then, the anal-
ysis of intentions allows us to reconstruct the goal
reduction process. Rationality is usually defined as
the ability of an agent to construct plans of actions
to reach the goal. Differently from objects (and ac-
tors), however, agent-roles are entities associated also
to motivational and cognitive elements like desires,
intents, plans, and beliefs.
Institutional roles are defined with certain abilities
(equivalently, powers), obligations and expectations
about the other roles. Following the description given
by the Eduroam WiFi agreement, if a student accepts
an offer, and if he/she does that, he/she has to provide
a valid ID, to expect an access to the WiFi from any
other participating institute (universities) that acts as
an IdP provider. Therefore, institutional roles are de-
fined in the first place by actions that may be taken, in
an adequate institutional setting, in order to achieve
certain goals. Furthermore, we observe that the pos-
sibility of WiFi connection exists because there is a
university who has offered and received acceptance
and finally delivers the service. Both roles are strictly
necessary: there cannot be a student without a univer-
ICAART 2016 - 8th International Conference on Agents and Artificial Intelligence
278
Figure 5: A WiFi connection transaction illustrated in terms of actions, with pre- and post conditions.
Figure 6: Full action pattern associated to an Eduroam WiFi connection.
sity in case of free WiFi. In this line of thought, WiFi
connection does not concern only one university with
its own students. Eduroam connection is a free WiFi
for all students all over the world and composed by
several competing actors (e.g. campuses). Although,
it is not explicitly present in the formal description of
the internet connection institution, the presence of IdP
and technical partners is obviously not negligible for
the institutional role. These relations are involved in
the evaluation of the offer (having a power to offer as
a university) and action which meant to judge the ac-
ceptability of the offer (being in the WiFi range and
holds a valid ID). Evaluation however, is not made
explicit in the definition of the free WiFi process. A
complete scheme about the process can be drawn uni-
fying procedural and institutional descriptions which
have been shown in Figures 5, 6. The university acts
as offeror and student acts as offeree. The gray circle
in Figure 5 shows that the performance of the action is
not sufficient to proceed, but it has to return a positive
result.
4.2 Visualization
We use Petri nets to visualize the flow of the
story (Fehling, 1993). The choice of Petri nets as rep-
resentational model has positive outcomes in itself.
First, allowing for descriptions in terms of localized
states rather than a global state comparable to Kripke
models(Esakia, 1974). Next, exploiting their topo-
logical characterization, can easily model the institu-
tional dynamic, i.e. timing/synchronization aspects.
Finally, Petri nets offer a direct visualization both of
the local structure and of the behavior of the system,
which can be useful for validation purposes.
4.3 Preliminary Implementation
For the model discussed in Figure 6 we have imple-
mented a generic framework in Groovy. Groovy is a
recent extension of Java which compiles on the Java
virtual machine (JVM). The language is suitable for
fast prototyping, and differently from Java, is a dy-
namic language, and can be used for scripting. It
provides the libraries implementing the actor compu-
tational model as described by Hewitt et al. (Hewitt
et al., 1973). This framework is able to take care
of all the parallel and concurrent actions that need to
be done in the case study. The Default-Actor class
was provided to be the base for our agent class. This
Agent class can be used to further implement differ-
ent agent roles and the interaction between them. For
space reasons, we are not able to present further ex-
amples of their use with Petri net models and source
code, but the interested readers can find some other
models on our website
2
and more information about
the project on our website
3
.
2
https://sarnet.uvalight.net/
3
http://delaat.net/sarnet/index.html
An Agent-based Framework for Multi-domain Service Networks - Eduroam Case Study
279
5 DISCUSSION AND FURTHER
DEVELOPMENTS
Our research is intended to model normative reason-
ing in a complete distributed environment. In par-
ticular, we are interested in how to model the SPG
from the normative perspective to observe the agent
behavior and identify the benefits and risks. In the
current approach, typical strategy decision problems
for a given game do not take explicitly into account
the possibility of the player to behave avoiding a rule,
or forcing the interpretation of the rule toward its in-
terest, if the regulator (consciously or not) left some
ambiguity. Using our framework, agents models or
roles involved in a social scenario, outlined from a
story can be described. As an operative result, such a
simulation can help to understand the social (institu-
tional) dynamics: validating the domain of conceptu-
alization of the experts, making predictions, suggest-
ing improvements to regulations for the SPG frame-
work and spotting normative weaknesses and vulner-
abilities. This model is an essential step to provide
an ABM of cross domain framework which is one of
the directions of our future research. Along with this
work, a preliminary implementation has been devel-
oped, using an existing ABM platform. The current
ABM implementation was expressive enough to build
a first version of the generic ABM framework of SPG.
However, we have experienced some shortcomings in
expressivity, which are left outside the scope of this
paper, and we will address these issues as next steps
and present in a separate publication.
ACKNOWLEDGEMENTS
We would like to thank the Netherlands COMMIT/
program and NWO organization for making this re-
search possible. We also like to thank KLM for pro-
viding guidance and the context for this research.
REFERENCES
Abdelkader, G. (2003). Requirements for achieving soft-
ware agents autonomy and defining their responsibil-
ity. In Proc. Autonomy Workshop at AAMAS 2003,
volume 236.
Boer, A. (2009). Legal theory, sources of law and the se-
mantic web. In Proceedings of the 2009 conference on
Legal Theory, Sources of Law and the Semantic Web,
pages 1–316. IOS Press.
Boer, A. and Van Engers, T. (2011). An agent-based legal
knowledge acquisition methodology for agile public
administration. In Proceedings of the 13th Interna-
tional Conference on Artificial Intelligence and Law,
pages 171–180. ACM.
Dennett, D. C. (1989). The intentional stance. MIT press.
Dignum, V., V
´
azquez-Salceda, J., and Dignum, F. (2005).
Omni: Introducing social structure, norms and ontolo-
gies into agent organizations. In Programming multi-
agent systems, pages 181–198. Springer.
Esakia, L. (1974). Topological kripke models. In Soviet
Mathematics Doklady, volume 15, pages 147–151.
Fehling, R. (1993). A concept of hierarchical petri nets with
building blocks. In Advances in Petri Nets 1993, pages
148–168. Springer.
Gommans, L. (2014). Multi-domain authorization for e-
infrastructures,university of amsterdam: Phd thesis.
Gommans, L., Vollbrecht, J., Gommans-de Bruijn, B., and
de Laat, C. (2015). The service provider group frame-
work: A framework for arranging trust and power to
facilitate authorization of network services. Future
Generation Computer Systems, 45:176–192.
Hewitt, C., Bishop, P., and Steiger, R. (1973). A univer-
sal modular actor formalism for artificial intelligence.
In Proceedings of the 3rd international joint confer-
ence on Artificial intelligence, pages 235–245. Mor-
gan Kaufmann Publishers Inc.
L
´
opez, G., C
´
anovas,
´
O., G
´
omez-Skarmeta, A. F., and
S
´
anchez, M. (2008). A proposal for extending
the eduroam infrastructure with authorization mecha-
nisms. Computer Standards & Interfaces, 30(6):418–
423.
Preece, A., Hui, K., and Gray, P. (1999). Kraft: Supporting
virtual organisations through knowledge fusion. In Ar-
tificial Intelligence for Electronic Commerce: Papers
from the AAAI-99 Workshop, pages 33–38.
Searle, J. (2009). Making the social world: The structure of
human civilization. Oxford University Press.
Searle, J. R. (1969). Speech acts: An essay in the philosophy
of language, volume 626. Cambridge university press.
Sileno, G., Boer, A., and Van Engers, T. (2013). To-
wards a computational model for institutional sce-
narios. In BNAIC 2013: Proceedings of the 25th
Benelux Conference on Artificial Intelligence, Delft,
The Netherlands, November 7-8, 2013. Delft Univer-
sity of Technology (TU Delft); under the auspices
of the Benelux Association for Artificial Intelligence
(BNVKI) and the Dutch Research School for Infor-
mation and Knowledge Systems (SIKS).
Wierenga, K. and Florio, L. (2005). Eduroam: past, present
and future. Computational methods in science and
technology, 11(2):169–173.
ICAART 2016 - 8th International Conference on Agents and Artificial Intelligence
280