An XML-based multi-agent system for the user-oriented management of QoS in telecommunications networks

Abstract

In this paper, we propose XICOMASQ, an XML-based multi-agent system for the user-oriented management of QoS in telecommunications networks. XICOMASQ is characterized by the following features: (i) it is strongly subjective, i.e. it handles network management activities according to user preferences represented by a user profile; (ii) it can operate on a large variety of telecommunications networks; (iii) it is semi-automatic; (iv) it is XML-based, i.e. it exploits XML for guaranteeing a light, versatile and standard mechanism for information management and exchange.

Full text

Page 1

An XML-based multi-agent system for the user-oriented management of QoS in
telecommunications networks
Pasquale De Meo
DIMET
Universit` a di Reggio Calabria
Via Graziella, Localit` a Feo di Vito
89060 Reggio Calabria, Italy
demeo@ing.unirc.it
Jameson Mbale
Department of Computer Science
Harbin Institute of Technology
92 Xidazhi Street, Nangang District, Box 773
Harbin 150001, China
mbalej@www.com
Giorgio Terracina
Dipartimento di Matematica
Universit` a della Calabria
Via Pietro Bucci
87036 Rende (CS), Italy
terracina@mat.unical.it
Domenico Ursino
DIMET
Universit` a di Reggio Calabria
Via Graziella, Localit` a Feo di Vito
89060 Reggio Calabria, Italy
ursino@unirc.it
Abstract
In this paper we propose XICOMASQ, an XML-based
multi-agent system for the user-oriented management of
QoS in telecommunications networks. XICOMASQ is
characterized by the following features: (i) it is strongly
subjective, i.e. it handles network management activities
according to user preferences represented by a user profile;
(ii) it can operate on a large variety of telecommunications
networks; (iii) it is semi-automatic; (iv) it is XML-based,
i.e. it exploits XML for guaranteeing a light, versatile and
standard mechanism for information management and ex-
change.
1Introduction
In the last years “global connectivity”, and the conse-
quent strict integration of computers and telecommunica-
tions networks, isbecoming a challenging issue in advanced
informationandcommunicationtechnologies. Adirectcon-
sequence of this trend is the growing role that information is
playing in these application contexts; indeed, it is possible
to predict that the present and the future technology scenar-
ios will be characterized by the presence of “information at
any time, at any place, at any form” [12].
“Global connectivity” is aided by new technologies and
systems (e.g., UMTS [9]) capable of effectively supporting
multimedia data transfer even under “hostile conditions”
(e.g., users traveling on board fast transportation means,
operating in locations where the development and the ex-
ploitation of wired network backbones is difficult and/or
expensive). The enormous development of services char-
acterized by a strong multimedia connotation undoubtedly
creates new scenarios and opportunities but also amplifies
some, already existing, problems.
In such a context, one of the most relevant challenges is
the management of the Quality of Service (hereafter QoS) in
a telecommunications network. This implies some specific
parameters, “measuring” the satisfaction degree of a user, to
be kept at values that fall within an interval she/he specifies.
More specifically, a dynamic, adaptive and subjective pro-
tocol for QoS management appears to be mandatory [16].
A QoS management policy is said: (i) dynamic, if there
exist some mechanisms capable of reacting to variations of
the environment parameters; (ii) adaptive, if it is capable of
re-distributing the available resources for adapting the of-
fered service to the network modifications; (iii) subjective,
ifthenetworkresourceassignmentpolicytakesintoaccount
the user preferences, generally represented by a user profile.
Dynamism and adaptivity can be effectively handled by
means of agent technologies. Indeed, in the literature, some
agent-based systems have been proposed for QoS manage-
mentinspecifictypologiesoftelecommunicationsnetworks
(see, for example, [7, 11, 13, 15, 19]).
All of these approaches are interesting attempts to use
agents for supporting a user in her/his exploitation of
telecommunications networks.
suited to handle dynamism and adaptivity; most of them
are also generic, i.e., capable of operating in a large vari-
They are generally well
Proceedings of the IEEE/WIC International Conference on Intelligent Agent Technology (IAT’03)
0-7695-1931-8/03 $ 17.00 © 2003 IEEE
Authorized licensed use limited to: UNIVERSITA DELLA CALABRIA. Downloaded on March 30,2010 at 05:49:45 EDT from IEEE Xplore. Restrictions apply.

Page 2

ety of telecommunications environments. However, in our
opinion, further efforts are necessary for constructing more
subjective approaches.
The following example should clarify this assertion. As-
sume that a user, say U, is accessing a telecommunica-
tions network. Assume, also, that, at a certain time, a net-
work bandwidth reduction occurs (e.g., because a network
fault has happened) and that the bandwidth available for U
is reduced. Most of the approaches exploited in telecom-
munications research try to uniformly reduce the quality
of transmitted information (e.g., they reduce image reso-
lution, chrominance, audio codex, frame rate, etc.). Some
approaches (e.g, the multi-bearer approaches for QoS man-
agement) have been also proposed which handle a (gener-
ally simple) profile of U in order to verify if she/he prefers
to sacrifice one transmission medium (e.g., video) and to
maintain the other ones at the same quality they had before
the network reduction.
In our opinion, this last is a very interesting idea lead-
ing to a more complete, partially subjective, definition of
QoS. We argue that it is possible to further improve it in or-
der to obtain a fully subjective QoS definition and manage-
ment. More specifically, in our opinion, a rich user profile,
storing the information contents she/he considered more in-
teresting in the past, could be constructed and maintained
by applying user modeling approaches [10]. When a net-
work reduction occurs, it could be possible (i) to reduce the
quality of transmitted information; (ii) to sacrifice a trans-
mission medium and to maintain the other ones at the same
transmission quality; (iii) to reduce the transmitted infor-
mation contents, by eliminating those ones appearing less
interesting for the user, and to maintain the same transmis-
sion media and quality.
This paper is an attempt to develop these ideas.
More specifically, it presents XICOMASQ(XML-based
Information Content Oriented Multi-Agent System for
QoS management).
XICOMASQis a multi-agent system characterized by
the following features: (i) It is highly subjective; in particu-
lar, it handles quite a rich and detailed user profile which
records her/his preferences and past behaviour and plays
a key role in the network resource management. (ii) It
is generic, i.e., it is capable of operating in a large va-
riety of telecommunications environments. (iii) It is dy-
namic, since it is provided with mechanisms for reacting
to variations of the environment parameters. (iv) It is semi-
automatic; in particular, as a consequence to its dynamism,
it autonomously makes most of its decisions. However, it
provides the user with the possibility of modifying and/or
negotiating them. (v) It is adaptive, i.e., it re-distributes the
available resources for adapting the offered service to the
network constraint modifications. (vi) It is XML-based, i.e.,
it exploits XML for storing and handling agent ontologies
as well as for managing agent communication.
2The XICOMASQarchitecture
XICOMASQconsists of the following agent typolo-
gies:
• A User-Device Agent (hereafter UDA), that handles a
connection session carried out by a user U by means
of a device D.
• A Network Agent (hereafter NA), that manages the
network resources.
• A Fault Agent (hereafter FA), that is a reactive agent
activated whenever a network fault occurs (resp., is re-
paired) for managing the corresponding bandwidth re-
duction (resp., augmentation).
• A Provider Agent (hereafter PA) which is associ-
ated with each supplier of information and/or services
available over the network and collaborates with User-
Device Agents for allowing a user to exploit the corre-
sponding information and/or services.
• AnInformationFilteringAgent(hereafterIFA)which
re-defines resource distribution whenever a fault has
occurred in the network.
As previously pointed out, the role of XML in
XICOMASQis crucial. Indeed:
• The agent ontologies are stored as XML documents;
as a consequence, they are light, versatile, easy to be
exchangedandcanresideondifferentdevicesandsoft-
ware platforms. In spite of this simplicity, the informa-
tion representation rules embodied in XML are power-
ful enough to allow a sophisticated information man-
agement.
• The agent communication language is ACML [6]; this
is the XML encoding of FIPA Agent Communication
Language [2]. The exploitation of ACML guarantees
various benefits to XICOMASQ; two of the most rel-
evant ones are the following:
– Developing and managing tools capable of carry-
ingoutACMLmessageparsingisextremelysim-
ple; indeed, these tools can be constructed by ex-
ploiting the numerous off-the-shelf XML parsers
available over the Internet. Vice versa, in order
to construct parsers for not XML-based ACL ver-
sions, it is generally necessary to exploit a Lisp-
like encoding (see [6] for all details) whose sup-
ports are more difficult to be found over the In-
ternet.
– Integrating Agents with a large variety of Web
technologies (suchasSecure Socket Layer-SSL,
for handling both the authentication of agents’
Proceedings of the IEEE/WIC International Conference on Intelligent Agent Technology (IAT’03)
0-7695-1931-8/03 $ 17.00 © 2003 IEEE
Authorized licensed use limited to: UNIVERSITA DELLA CALABRIA. Downloaded on March 30,2010 at 05:49:45 EDT from IEEE Xplore. Restrictions apply.

Page 3

identities and the encryption of ACL messages)
is very simple to be realized. Vice versa, address-
ing the same issues with a not XML-based Agent
Communication Language would imply heavy
constraints on the agent infrastructure (think, for
example, to the great overload to be put in the
ACL messages for handling these issues).
• The extraction of information from the various data
structures is carried out by means of XQuery [3]. This
is becoming the standard query language for the XML
environment. SinceitisbasedontheXMLframework,
XQuery can handle a large data variety. It has capa-
bilities typical of database query languages as well as
features typical of document management systems. Fi-
nally, it is provided with various high level constructs
for simplifying querying over the Web; among them,
we cite constructors, that allow the creation of XML
structures within a query, and FLWR expressions, that
support iteration and variable binding.
• The manipulation of agent ontologies is performed by
means of the Document Object Model (DOM) [1].
This is a platform- and language-neutral interface that
allows programs and scripts to dynamically access and
update the content, structure and style of XML doc-
uments. DOM makes it possible for programmers to
write applications working properly on all browsers
and servers as well as on a large variety of hardware
and software platforms.
In the following sections we provide a detailed description
of the various agent typologies which XICOMASQcon-
sists of.
3The Agents involved in XICOMASQ
3.1The User-Device Agent
A User-Device Agent UDAijis associated with a user
Ujexploiting a device Di; it supports Ujduring her/his ac-
tivities on the network carried out by means of Di. It is
worth pointing out that more than one User-Device Agent
can be associated with the same user, if she/he exploits
different devices; analogously, more than one User-Device
Agent can be associated with the same device, if it is ex-
ploited by different users.
3.1.1Ontology
The ontology of UDAij consists of a pair ?DPi,UPj?,
where DPi(resp., UPj) indicates the profile of Di(resp.,
Uj).
A Device Profile DPistores the characteristics of the
device Di. More specifically, a tuple
?DIdi,BMaxi,PPBi?
is associated with Di. Here:
• DIdiis the Identifier of Di;
• BMaxi is the Maximum Bandwidth which Di can
guarantee for accessing or downloading data from the
network;
• PPBiis the Price Per Byte of Di; it indicates the cost
for downloading or sending a byte of data by means of
Di.
A User Profile UPjcan be described, in its turn, by a tuple
?UIdj,ICSetj?. Here:
• UIdjis a code identifying Uj.
• ICSetj= {IC1,IC2,...,ICp} is the set of Informa-
tion Contents of interest for Uj.
A tuple:
?ICIdl,ICNamel,ICSizel,FV TSl,LV TSl,
V Cl,ATIl,SubjQoSl,ObjQoSl,QoSl?
is associated with each information content ICl∈ ICSetj.
Here:
• ICIdlis the Information Content Identifier of ICl.
• ICNamelis the Information Content Name of ICl.
• ICSizel is the Information Content Size associated
with ICl; it represents the size, in bytes, of ICl.
• FV TSlis the First Visit Time Stamp associated with
ICl; it represents the time instant when IClhas been
accessed for the first time by Uj.
• LV TSlis the Last Visit Time Stamp relative to ICl; it
indicates the time instant when IClhas been accessed
for the last time by Uj.
• V Clis the Visit Counter of ICl; it represents the total
number of accesses to IClcarried out by Uj.
• ATIlistheAccessTimeIndicatorassociatedwithICl.
Our model takes into account that the network usage is
highly influenced by both the characteristics of the ex-
ploited device and the network costs. As an example,
assume that a user takes n seconds for accessing an
information content. Assume also that this access is
repeated two times. During the first occurrence she/he
exploits a device having a low bandwidth and a high
price per byte. During the second occurrence, she/he
uses a different device supporting a connection charac-
terized by a high bandwidth and a low price per byte. It
is possible to argue that the interest the user exhibited
for the information content during the former access is
larger than that she/he exhibited during the latter one.
This brief example shows that it is necessary to “nor-
malize” the time exploited for accessing an informa-
tion content w.r.t. both the characteristics of the device
Proceedings of the IEEE/WIC International Conference on Intelligent Agent Technology (IAT’03)
0-7695-1931-8/03 $ 17.00 © 2003 IEEE
Authorized licensed use limited to: UNIVERSITA DELLA CALABRIA. Downloaded on March 30,2010 at 05:49:45 EDT from IEEE Xplore. Restrictions apply.

Page 4

and the navigation costs; the Access Time Indicator
ATIlhas been conceived for this purpose. In particu-
lar, it is defined as ATIl=?V Cl
the kthaccess whereas PPBk
the Price Per Byte and the Maximum Bandwidth of the
Device Diexploited by Ujfor visiting IClduring the
kthaccess.
• SubjQoSldenotes the Subjective QoS of Ujfor ICl.
It specifies the subjective component of QoS, i.e. that
one taking into account the specific interests of Uj. Its
formulation has been inspired from the following con-
siderations, partially derived from [4]:
– the interest a user shows for an information con-
tent IClis related to the time she/he spent for
visiting it. As a consequence, SubjQoSlis di-
rectly proportional to ATIl;
– theinterestofauserforIClgrowswiththegrow-
ing of the number of accesses to it. Therefore,
SubjQoSlis directly proportional to V Cl.
– the more remote (resp., recent) the First Visit
Time Stamp (resp., the Last Visit Time Stamp)
is, the higher the interest is. A good indicator of
this intuition is given by the ratioLV TSl−FV TSl
here, TS denotes the current Time Stamp.
k=1
Tk
l×PPBk
BMaxk
i
i
; here, Tk
l
represents the time Ujspent for accessing IClduring
iand BMaxk
irepresent
TS−FV TSl
;
Fromtheconsiderationsoutlinedabove, SubjQoS can
be computed as1:
?p
the interval [0,1].
• ObjQoSldenotes the Objective QoS of Uj for ICl;
this parameter specifies the QoS guaranteed to Ujfor
IClby the network in the past. It depends only on net-
work characteristics and is the only QoS component
considered by the approaches previously proposed in
the literature. In our application context, the formu-
lation of ObjQoSlhas been derived from the follow-
ing considerations: (i) QoS is directly proportional to
the available network bandwidth; (ii) QoS is strictly
related to the characteristics of Di; (iii) QoS is nega-
tively influenced by the quality degradations imposed
by the network. From these considerations, ObjQoSl
can be formulated as ObjQoSl =
here, ObjQoSk
in the session kth. ObjQoSk


is composed by p elements.
SubjQoSl=
ATIl
k=1ATIk×
V Cl
?p
k=1V Ck×?LV TSl−FV TSl
TS−FV TSl
?
Observe that SubjQoSlis a real number belonging to
?V Cl
k=1ObjQoSk
V Cl
l
;
lis the Objective QoS relative to ICl
lis defined as:
if no degradation occurred
in the kthvisit of IClby Uj
Bk
BMaxk
i
ObjQoSk
l=

1
otherwise
1Recall that we have assumed the set of Information Contents ICSetj
Here, Bkdenotes the bandwidth available when the
last degradation occurred during the kthvisit of ICl;
Bkis measured by applying the framework presented
in [14]. BMaxk
Device Diexploited by Uj during the kthaccess to
ICl. Observe that ObjQoSlis a non-negative number
belonging to the real interval [0,1].
• QoSlindicates the overall Quality of Service Ujasso-
ciates with ICl. As pointed out in the Introduction, our
approach aims at combining a subjective and an objec-
tiveQoScomponent; indeed, SubjQoSlandObjQoSl
have been introduced for taking into account these two
aspects. As a consequence, QoSlcan be defined as:
iis the Maximum Bandwidth of the
QoSl= q × ObjQoSl+ (1 − q) × SubjQoSl
Here, q is a real number belonging to the interval [0,1].
If q is high, ObjQoSlis the dominant term in QoSl
and the network performance is crucial for defining
the QoS status. Vice versa, if q is low, SubjQoSlis
the dominant term and the interest of Ujtowards the
information contents she/he is accessing is more im-
portant than the network performance.
InordertosimplifythedescriptionoftheontologyofUDA,
we have represented it by means of abstract data structures,
such as sets. However, as previously pointed out, it is actu-
ally stored as an XML document; the corresponding XML
Schema is shown in Figure 1.
3.1.2Behaviour
UDAijis activated whenever Ujrequires to access a net-
workproviderbymeansofDi, aswellaswhenshe/heneeds
to modify the usage condition of a previously started access.
More specifically, UDAijis activated at the beginning
of a new working session; in this case, Ujspecifies some
high-level parameters (such as horizontal and vertical video
resolution, video chrominance, audio codex, data rate, etc.)
allowing to determine the QoS she/he desires; these are
sent by UDAij to NA. NA verifies if the requirements
of Uj can be satisfied; in the affirmative case it allocates
or reserves the necessary resources; otherwise, it informs
UDAijabout the impossibility to satisfy user requirements.
In this latter case Ujcan either renounce to exploit the net-
work resources or accept to reduce her/his requirements.
UDAijcould be activated also by IFA when this latter
needs information about the profile of Uj. If this happens,
UDAijsupplies all required information.
Finally, UDAij monitors the behaviour of Uj during
her/his access and exploitation of network resources and
updates her/his profile accordingly. More specifically, as-
sume that Ujvisits the Information Content ICtfrom the
time instant t1to the time instant t2. Assume, also, that
Proceedings of the IEEE/WIC International Conference on Intelligent Agent Technology (IAT’03)
0-7695-1931-8/03 $ 17.00 © 2003 IEEE
Authorized licensed use limited to: UNIVERSITA DELLA CALABRIA. Downloaded on March 30,2010 at 05:49:45 EDT from IEEE Xplore. Restrictions apply.

Page 5

<?xml version=‘‘1.0’’ ?>
<xs:schema xmlns:xs=‘‘http://www.w3.org/2001/XMLSchema’’
elementFormDefault=‘‘qualified’’>
<xs:complexType name=‘‘Device\_ProfileType’’>
<xs:attribute name=‘‘Device\_Identifier’’ type=‘‘xs:ID’’
use=‘‘required’’/>
<xs:attribute name=‘‘Maximum\_Bandwidth’’ type=‘‘xs:nonNegativeInteger’’
use=‘‘required’’/>
<xs:attribute name=‘‘Price\_Per\_Byte’’ type=‘‘xs:float’’
use=‘‘required’’/>
</xs:complexType>
<xs:complexType name=‘‘Information\_Content\_Type’’>
<xs:attribute name=‘‘Information\_Content\_Identifier’’ type=‘‘xs:ID’’
use=‘‘required’’/>
<xs:attribute name=‘‘Information\_Content\_Name’’ type=‘‘xs:string’’
use=‘‘required’’/>
<xs:attribute name=‘‘Information\_Content\_Size’’
type=‘‘xs:nonNegativeInteger’’ use=‘‘required’’/>
<xs:attribute name=‘‘First\_Visit\_TimeStamp’’ type=‘‘xs:dateTime’’
use=‘‘required’’/>
<xs:attribute name=‘‘Last\_Visit\_TimeStamp’’ type=‘‘xs:dateTime’’
use=‘‘required’’/>
<xs:attribute name=‘‘Visit\_Counter’’ type=‘‘xs:nonNegativeInteger’’
use=‘‘required’’/>
<xs:attribute name=‘‘Access\_Time\_Indicator’’ type=‘‘xs:float’’
use=‘‘required’’/>
<xs:attribute name=‘‘Subjective\_QoS’’ type=‘‘xs:float’’
use=‘‘required’’/>
<xs:attribute name=‘‘Objective\_QoS’’ type=‘‘xs:float’’
use=‘‘required’’/>
<xs:attribute name=‘‘QoS’’ type=‘‘xs:float’’ use=‘‘required’’ />
</xs:complexType>
<xs:complexType name=‘‘User\_ProfileType’’>
<xs:sequence>
<xs:element name=‘‘Information\_Content’’
type=‘‘Information\_ContentType’’ minOccurs=‘‘0’’
maxOccurs=‘‘unbounded’’/>
</xs:sequence>
<xs:attribute name=‘‘User\_Identifier’’ type=‘‘xs:ID’’
use=‘‘required’’/>
</xs:complexType>
<xs:element name=‘‘UDA\_Ontology’’>
<xs:complexType>
<xs:sequence>
<xs:element name=‘‘Device\_Profile’’ type=‘‘Device\_ProfileType’’
minOccurs=‘‘1’’ maxOccurs=‘‘1’’/>
<xs:element name=‘‘User\_Profile’’ type=‘‘User\_ProfileType’’
minOccurs=‘‘1’’ maxOccurs=‘‘1’’/>
</xs:sequence>
</xs:complexType>
</xs:element>
</xs:schema>
Figure 1. The XML Schema relative to the ontology of UDA
in the ontology of PA the Identifier, the Name and the In-
formation Content Size of ICtare ICIdt, ICNametand
ICSizet.
Ifno tuple relative
ICSetj,
UDAij
computes
ObjQoSt
and
QoSt
by
troduced previously;after this,
?ICIdt,ICNamet,ICSizet,t1,t2,1,ATIt,SubjQoSt,
ObjQoSt,QoSt? into ICSetj.
Vice versa, if the tuple ?ICIdt,ICNamet,ICSizet,
FV TSt,LV TSt,V Ct,ATIt,SubjQoSt,ObjQoSt,QoSt?
is already present in ICSetj, UDAijupdates it by regis-
tering information about the last access; more specifically
it sets LV TSt= t2, V Ct= V Ct+ 1 and updates ATIt,
SubjQoSt, ObjQoSt and QoSt by applying formulas
introduced previously.
to
ICt
ATIt,
applying
ispresent
SubjQoSt,
formulas
it adds the tuple
in
in-
3.2 Network Agent
The Network Agent NA acts as a “mediator” between
users and network. It is responsible of the reservation, al-
location and de-allocation of network resources according
to user requirements and profiles as well as network avail-
abilities. Such a task is carried out by applying classical
techniques for allocation, de-allocation and reservation of
resources in telecommunications networks (see, for exam-
ple, [5, 8, 18]).
3.2.1Ontology
The ontology of NA consists of two different components
managing user desires and network availabilities, resp.
More specifically,it
?User Management,Network Management?.
User Management component stores, for each user, the
maximum and the minimum required bandwidth as well
as the bandwidth presently assigned to her/him. The Net-
work Management component stores the total bandwidth
consists ofapair
The
available in the network, the total bandwidth presently
allocated to users and, finally, the average bandwidth
allocated to users in the past. The XML Schema associated
with the ontology of NA is analogous to that associated
with the ontology of UDA; due to space limitations, we do
not show it in the following.
3.2.2Behaviour
As previously pointed out, NA is in charge of the reserva-
tion, allocation and de-allocation of network resources.
It can be activated whenever a user Uj, exploiting a de-
vice Di, wants to allocate network resources; in this case
UDAij supplies information about the desires of Uj by
means of some high-level parameters, such as video hor-
izontal and vertical resolution, chrominance, audio codex
and data rate. High-level parameters are exploited by NA
for computing the maximum and the minimum bandwidth
required for satisfying Uj; such a computation is carried out
by applying formulas described in [17].
After this, NA examines the available resources and ver-
ifies if it can allocate or reserve a bandwidth greater than
the minimum one required by the user. In the affirmative
case NA allocates or reserves the maximum possible band-
width. In the negative case, NA informs UDAij which
sends a suitable message to Ujwho can decide to reduce
her/his requirements or, alternatively, to renounce to access
the network services.
NA can be also activated whenever a user needs to mod-
ify the quality or the quantity of network resources she/he
is already exploiting or reserving, as well as for requiring a
resource de-allocation, if she/he finished her/his activities.
In both cases, the behaviour of NA is analogous to that we
have seen above.
NA can be activated by IFA when this latter needs in-
formation about the maximum and the minimum require-
ments of each user as well as her/his current resource ex-
ploitation. In this case NA provides IFA with all required
Proceedings of the IEEE/WIC International Conference on Intelligent Agent Technology (IAT’03)
0-7695-1931-8/03 $ 17.00 © 2003 IEEE
Authorized licensed use limited to: UNIVERSITA DELLA CALABRIA. Downloaded on March 30,2010 at 05:49:45 EDT from IEEE Xplore. Restrictions apply.

Page 6

information. The activation of NA could be also requested
by IFA at the end of a negotiation phase for modifying the
resource distribution. In such a case, NA receives, for each
user, the bandwidth modifications, varies the network re-
source allocation and de-allocation accordingly and updates
its ontology suitably.
Finally, NA could be activated by FA when a network
fault occurs (resp., is repaired). In this case FA provides
NA with the bandwidth modifications caused by the net-
work fault presence (resp., repair). In both cases NA must
modify its ontology for registering bandwidth reduction
(resp., augmentation) caused by the fault presence (resp.,
repair).
3.3 The Provider Agent
3.3.1
A Provider Agent PA is associated with a service provider.
Its ontology stores the set of Information Contents relative
to the corresponding provider. A tuple
?ICIdl,ICNamel,ICSizel? is associated with each In-
formation Content IClstored in the ontology of PA. The
XML Schema associated with this ontology is analogous
to that associated with the ontology of UDA; due to space
limitations we do not show it. The ontology of a PA is
created when the corresponding service provider is made
available over the network. It is modified whenever an In-
formation Content is created or removed or updated in the
corresponding service provider.
Ontology
3.3.2
A Provider Agent PA is activated by a User-Device Agent
UDAijwhenever a user Ujwants to know the information
contents associated with the corresponding provider; in this
case, PA supplies the required information.
PA is activated also by UDAijwhenever Ujaccesses
information and/or services associated with it; in this case
PA provides UDAijwith information about the behaviour
of Ujduring her/his access to the provider; this information
is exploited by UDAijfor updating the profile of Uj.
PA can be activated by IFA whenever this last must de-
fine a new resource distribution; in this case PA supplies to
IFA information about the contents of the corresponding
provider. PA is also activated by IFA at the end of a ne-
gotiation phase for suitably changing the set of Information
Contents supplied to UDAij.
Behaviour
3.4 The Information Filtering Agent
The capability of filtering the most relevant information
for a user when a bandwidth reduction occurs is one of the
most important features of XICOMASQ. Such a capa-
bility is handled by the Information Filtering Agent IFA
which is responsible of making the network resource distri-
bution compatible with the new network status and to max-
imize user satisfaction according to the new network con-
straints.
3.4.1
The Ontology of IFA stores the current and the maximum
bandwidth that the network can supply. The corresponding
XML Schema is trivial and, due to space constraints, is not
reported.
Ontology
3.4.2
IFA is activated by FA when a fault occurs or is repaired
in the network; FA supplies to IFA information about the
current network status. If a fault occurs, first IFA activates
NA and requires information about the users’ exploitation
of network resources before the fault occurrence. When
IFA receives this information, it verifies if the present net-
workavailabilitiesaresufficientforguaranteeingtoallusers
the QoS they had before the fault. In the affirmative case,
no further action is necessary. In the negative case, first
IFA “distributes” the bandwidth reduction caused by the
fault occurrence among all User-Device Agents and, then,
activates a negotiation with each of them.
Let UDAijbe one of these User-Device Agents. First
IFA activates UDAijand requires information about the
profile of Uj. Then, it activates all PA’s for determining the
set CurrICSetijof Information Contents which Ujis cur-
rently accessing by means of Di. IFA exploits this infor-
mation, along with the current bandwidth available for the
pair ?Di,Uj?, for defining the set of Information Contents
and the value of high-level parameters which maximize the
QoS for Uj, according to the bandwidth limitation.
In particular, first IFArequires to Ujto specify (through
UDAij) new values for high-level parameters. These must
be lesser than or equal to those specified before the fault
has occurred. After this, the objective QoS component
is fixed and, therefore, it is necessary to maximize only
the subjective one, according to the new bandwidth limi-
tations. This implies to select those information contents
of CurrICSetijwhose overall requirements can be satis-
fiedbythecurrentbandwidth availabilitiesandwhich, atthe
same time, maximize the satisfaction of Uj. Such a filtering
activity can be appropriately formalized as a 0/1 Knapsack
Problem (0/1 KP in the following) as:
maximize?|CurrICSetij|
k=1
ICSizek×δk≤ B×Tw
δk∈ {0,1} ∀k : 1 ≤ k ≤ |CurrICSetij|
Here, B is the bandwidth currently available for Uj
whereas Twis a time constant representing the maximum
delay for information loading considered acceptable by the
user.
Behaviour
k=1
SubjQoSk× δk
s.t.?|CurrICSetij|
Proceedings of the IEEE/WIC International Conference on Intelligent Agent Technology (IAT’03)
0-7695-1931-8/03 $ 17.00 © 2003 IEEE
Authorized licensed use limited to: UNIVERSITA DELLA CALABRIA. Downloaded on March 30,2010 at 05:49:45 EDT from IEEE Xplore. Restrictions apply.

Page 7

At the end of this task, IFA activates NA, for perform-
ing the corresponding modifications on network resources
distribution, and the various PA’s for carrying out the suit-
able modifications on the set of Information Contents sup-
plied to UDAij. If IFA is activated after a fault repair, it
behaves in a way analogous to that we have described pre-
viously.
4 Conclusions
In this paper we have presented XICOMASQ, an
XML-based multi-agent system for the user-oriented
management of QoS in telecommunications networks.
XICOMASQis characterized by various interesting fea-
tures such as: (i) management and exploitation of a user
profile; (ii) management of both a subjective and an ob-
jective QoS component; (iii) capability of handling various
network typologies; (iv) semi-automaticity; (v) exploitation
of XML for information storing and exchange.
In our opinion, XICOMASQis an interesting attempt
of creating a strict interaction among research efforts in
Telecommunications, Artificial Intelligence and Informa-
tion Systems for solving complex problems involving as-
pects relative to each of these research fields.
In the future we plan to make such an interaction even
stricter. More specifically, on the one side, we plan to define
user profile based techniques for more efficiently handling
various telecommunications aspects presently managed by
“objective” approaches. On the other side, we plan to de-
sign novel approaches capable of managing a large variety
of e-services by taking into account the characteristics of
the device the user is exploiting for accessing them.
Acknowledgment. This work was partially funded by the Infor-
mation Society Technologies programme of the European Com-
mission, Future and Emerging Technologies under the IST-2001-
33570 INFOMIX project and the IST-2001-37004 WASP project.
References
[1] Document Object Model. [http://www.w3.org/DOM/].
[2] Foundation for Intelligent Physical Agents (FIPA) Specifica-
tions. [http://www.fipa.org].
[3] WorldWideWeb Consortium
[http://www.w3.org/XML/Query].
[4] P. Chan. A non-invasive learning approach to building web
user profiles. In Proc. of KDD-99 Workshop on Web Usage
Analysis and User Profiling (WebKDD’99), pages 7–12, San
Diego, California, USA, 1999. Lecture Notes in Computer
Science, Springer.
[5] S. Choi and K. G. Shin. A comparative study of bandwidth
reservation and admission control schemes in QoS-sensitive
cellular networks. Wireless Networks, 6(4):289–305, 2000.
[6] B. Grosof and Y. Labrou. An approach to using XML and
a rule-based content language with an agent communication
language. In Proc. of IJCAI-99 Workshop on Agent Commu-
nication Language, pages 96–117, Stockolm, Sweden, 1999.
- XMLquery.
[7] D. Gurer, V. Lakshminarayan, and A. Sastry.
gent agent-based architecture for the management of hetero-
geneous networks. In Proc. of Ninth IFIP/IEEE International
Workshop on Distributed Systems: Operations & Manage-
ment(DSOM’98), Newark, Delaware, USA, 1998.
[8] A. Hac and A. Armstrong. Resource allocation scheme for
QoS provisioning in microcellular networks carrying multi-
media traffic. International Journal of Network Management,
11(5):277–307, 2001.
[9] H. Holma and A. Toskala. WCDMA for UMTS - Radio access
for third generation mobile communications. John Wiley and
Sons, Ltd., Baffins Lane (West Sussex), England, 2000.
[10] A. Kobsa. Generic user modeling systems. User Modeling
and User-Adapted Interaction, 11:49–63, 2001.
[11] M. Laukkanen, H. Helin, and H. Laamanen. Supporting no-
madic agent-based applications in FIPA agent architecture.
In Proc. of The First International Joint Conference on Au-
tonomous Agents & Multiagent Systems (AAMAS’02), pages
1348–1355, Bologna, Italy, 2002. ACM Press.
[12] T. Magedanz, K. Rothermel, and S. Krause.
agents: Anemergingtechnologyfornextgenerationtelecom-
munications? In Proc. of the Conference on Computer Com-
munications (INFOCOM’ 96), pages 464–472, San Fran-
cisco, California, USA, 1996. IEEE Press.
[13] P. Misikangas, M. Makela, and K. Raatikainen. Predicting
QoS for nomadic applications using intelligent agents. In
A. Hayzelden and R. Bourne, editors, Agent Technology for
Communication Infrastructures, chapter 15, pages 197–208.
John Wiley & Sons, 2001.
[14] T. Nakano, K. Harumoto, S. Shimojo, and S. Nishio. User
adaptive content delivery mechanism on the World Wide
Web. In Proc. of the 17th ACM Symposium on applied com-
puting (SAC 2002), pages 1140–1146. ACM Press, 2002.
[15] K. Nygren, I. Jonsson, and O. Carlvik.
tem for media on demand services.
national Conference on the Practical Application of Pro-
log/Constraints/Intelligent Agents and Multi-Agent Technol-
ogy (PAAM’96), London, England (UK), 1996.
[16] D. Reininger, R. Izmailov, B. Rajagopalan, M. Ott, and
D.Raychaudhuri. Soft QoS control in the WATMnet broad-
band wireless system. IEEE Personal Communications Mag-
azine, 1:34–43, 1999.
[17] R. Steinmetz and K. Nahrstedt. Multimedia: Computing,
Communication and Applications. Prentice Hall, Inc., Upper
Saddle River,NJ 07458,USA, 1995.
[18] J. Tajima and K. Imamura. A strategy for flexible channel
assignment in mobile communication systems. IEEE Trans-
actions on Vehicular Technology, 37(5):92–103, 1988.
[19] K. Trzec and D. Huljenic. Intelligent agents for QoS man-
agement. In Proc. of The First International Joint Conference
on Autonomous Agents & Multiagent Systems (AAMAS’02),
pages 1405–1412, Bologna, Italy, 2002. ACM Press.
An intelli-
Intelligent
An agent sys-
In Proc. of Inter-
Proceedings of the IEEE/WIC International Conference on Intelligent Agent Technology (IAT’03)
0-7695-1931-8/03 $ 17.00 © 2003 IEEE
Authorized licensed use limited to: UNIVERSITA DELLA CALABRIA. Downloaded on March 30,2010 at 05:49:45 EDT from IEEE Xplore. Restrictions apply.