THE QUEST FOR SELF-MODEL IN SELF-MANAGING NETWORKS
Burak Simsek
FOKUS - Fraunhofer Institute for Open Communication Systems - SATCOM
Schloss Birlinghoven, Germany
Hakan Coskun
Department of Design and Test of Telecommunication Systems
Technical University of Berlin, Germany
Keywords:
Self-management, self-model, autonomicity, wireless networks, 802.11e.
Abstract:
Although autonomic networking has been discussed in related research activities in the last few years, there
is neither a commonly accepted model nor a common understanding of the necessary components of an au-
tonomic networking environment, which are also practically applicable. We present a simple model based on
the notion of self that is perceived by human beings and deduce its components from the originating point
which is the need for autonomic behavior. We also introduce three use cases that we implemented for solving
different problems of IEEE 802.11e networks.
1 INTRODUCTION
Autonomy and self-awareness are gaining momentum
in computer science, mainly because of the ever grow-
ing complexity of computer systems and networks.
New approaches tend to support or even eliminate the
human operator by moving the control intelligence to-
wards the inside of the system. Entities no longer are
dummy’ objects or components that are completely
managed from ’outside’, they become self-aware and
even self-managing meaning that the own behavior is
understood and changed or adapted to meet certain
criteria respectively (Sterritt and Hinchey, 2005).
Started some years ago, research activities try
to answer the question, how computer systems and
networks have to be designed and build in order
to provide capabilities that fulfil the requirements
of existing but also emerging requirements, such
as ad-hoc structures, increasing management com-
plexity, dynamic on-demand service composition and
ever growing quality of service (QoS) expectations.
Big vendors like SUN, HP and IBM on the one
side and the research community on the other side
have come up with different answers to those prob-
lems. Autonomic computing and utility computing
This work has been supported by the ”Krupp von
Bohlen und Halbach-Stiftung”
or autonomic communications respectively have been
proposed as solutions (Kephart and Chess, 2003),
(Smirnov, 2004), (Rappa, 2004). Although the is-
sues covered in these approaches are basically dif-
ferent (Abbas, 2003), their underlying concepts point
at the same direction; the increasing system com-
plexity can only be managed by making the systems
aware of their own operations. This is referred to self-
management. Therefore, it’s not surprising to see that
all novel principles are centered around self’, e.g.
self-configuration, self-healing, self-organization.
Despite the similarity of the mentioned proposed
solutions, there exists neither a common understand-
ing of the notion of self nor a framework for the de-
velopment of self managing systems. Approaches
so far also do not explain the intersecting issues of
those systems with the already existing artificial intel-
ligence methods. Therefore, it is difficult to figure out
the innovative sides of these approaches, except the
eight key elements of an autonomic-computing sys-
tem (Horn, 2001) which do not go beyond being re-
quirements for self managing environments. Hence,
although theoretically the concept of autonomic com-
puting has been stated expressly, practical issues are
still far from being mature.
Being aware of this issue, this paper is dedicated
to the presentation of a self model that we have used
for managing QoS over IEEE 802.11e networks. In
284
Simsek B. and Coskun H. (2007).
THE QUEST FOR SELF-MODEL IN SELF-MANAGING NETWORKS.
In Proceedings of the Second International Conference on Wireless Information Networks and Systems, pages 268-274
DOI: 10.5220/0002151302680274
Copyright
c
SciTePress
this way, we fill in the empirical gap of autonomic
communications and enable focusing more on prob-
lem specific issues.
In the rest of the paper, we summarize the notion
of self as perceived by human beings and induce our
self model based on these perceptions. In doing this,
we take into account the requirements and charac-
teristics of today’s communication environments and
correspondingly reduce the self model down to a set
of sine qua non enabling high scalability and com-
prehensiveness, which are essential for practice. We
clarify points intersecting with the classical artificial
intelligence methods and explain how this self model
can be extended with respect to problem specific sit-
uations. In the third section, we present three prob-
lems of the IEEE 802.11e standard in order to illus-
trate the introduced properties of the model, and build
corresponding self managing systems which we al-
ready implemented using network simulation tools.
We conclude our discussion with the summary of the
handicaps we encounter during the modeling of auto-
nomic networks and our contribution for overcoming
those handicaps.
2 SELF MODEL
The dichotomy of self and non-self is the basis
of an entity which is said to be self-aware (Mul-
hauser, 1998). It is not easy to define what self is.
In philosophy and psychology self is the mental
and conceptual awareness an individual holds with
regard to his/her own existence. It refers to the
conscious, reflective personality determining his/her
identity (Tauber, 2002). Although we cannot de-
scribe the term self for non-human living things in
the same manner, we still impose on them the notion
of self based on our perception of their possessions.
Consequently, the notion of self has a relatively large
spectrum, from being self-aware or autonomous to
self-unaware or allonomous
2
. This is also the crucial
distinction between intelligence and unintelligence.
One can argue that there is no need for defining the
so called selfs (Metzinger, 2003). The fact that we
call the entity being developed self is not the cru-
cial point here at all. The goal is also not the cre-
ation of self as in the case of some artificial intelli-
gence techniques, but building methods which would
decrease the complexities of the management and the
utilization of the systems that we already built up. In
these terms, the issue of self is rather deduced from
the need for autonomy. Considering this fact, we will
2
governed by external stimuli
try to figure out the structure of the self based on our
needs for autonomic behavior in the following sub-
sections.
2.1 Notion of Emotion
It is a wonder of the nature that each existence has its
reasons or more strongly said defined targets’. The
target is intrinsically tied with the existence itself.
This makes it crucial to regard the target as the high-
est order feature of the self’. Based on this fact, most
of the artificial intelligence techniques are built on
the definition of targets (Ferber, 1999), (Keiblinger,
2000). Nevertheless, the introduction of targets poses
many obstacles.
The questions, how to tell the self about his tar-
gets and how these targets should be reached proved
to be extremely difficult to answer leading to difficul-
ties in the implementation of those techniques. For
this reason, ontology based languages are proposed
and corresponding evaluation algorithms are devel-
oped (Bratko, 2000), (Muggleton, 1999). Neverthe-
less, in practice the use of these languages also proved
to be extremely awkward (Scott, 2000).
If we have a look at the animal world, instead of
targets, we witness the existence of instincts that au-
tomatically tell animals what to do next as some trig-
gering factors emerge. In fact a leopard does not think
about his targets, which may be for him to survive.
Or how do his babies know where the milk is coming
from and why at all? Targets enriched with posses-
sions such as the ability to run or eat are already there
as they are born. This is a start-up process (bootstrap-
ping) in the lives of all living things.
In (Damasio, 2005) emotion is defined as
an intense mental state that arises auto-
matically in the nervous system rather than
through conscious effort, and evokes either a
positive or negative psychological response.
Taking this definition of emotion as the basis, we
can say that the impulses occurring in the form of
emotions are directly coupled with all the actions
of the living things following their births (Damasio,
2005), (Glasser, 1999).
The initial emotions give them a taste of what a
satisfaction would mean and autonomic reactions to
the sensed emotions show the initial solution propos-
als to these impulses (pleasure and pain). With time
and experience the causalities behind the possessed
functions, new ways of satisfaction and how to get rid
of dissatisfaction are learned.
The phenomenon which results emotions is the
state being sensed by the self’. Targets are also de-
fined as states. Therefore, targets of a self can be di-
THE QUEST FOR SELF-MODEL IN SELF-MANAGING NETWORKS
285
rectly coupled with a mechanism of emotion. If there
exists a measurable emotion for each state, on which
the behaviour of self depends and if it is also pos-
sible to manage the level of emotion at each state,
then it is also possible to lead the self to its targets
by only using emotion levels. This eliminates the
need for defining targets separately, which would oth-
erwise require additional definition, semantics, rep-
resentation and interpretation mechanisms (Levesque
et al., 1998).
Consequently, it is possible to model the self
without explicit targets, but with a well defined emo-
tion map over a state space by introducing the primary
target as the willingness to maximize satisfaction.
2.2 Embedded Notion of Emotion
The sense of emotions is solely a triggering factor
for the functioning of the self. A deeper mechanism,
which exists in the core of the self is responsible for
the birth of the self. It has exclusively only the ca-
pability to ’trigger’ some functionality of itself with
respect to its feelings. In fact this entity or mecha-
nism is what the philosophers have been trying to find
out (Glasser, 1999). Instead of speculating about what
this entity might be for different living things, we just
try to model its functional properties that are crucial
to the formation of the self. We call this mechanism
’Embedded Notion of Emotion’ (ENE).
Although ENE is the core element of the self
model, it is an complete abstract entity. Its program
is simply based on the desire to maximize its reward
function and the ability to trigger the possessed func-
tionalities. Its existence gives birth to a living self
but has no direct influence on the possessions of the
self. These possessions are rather given. It might be
the case that, ENE is enriched with possessions which
enable the self to reach higher rewards by gaining new
possessions. This however defines the intelligence
level of the self which is not a necessary character-
istic for its existence.
Creating intelligent systems is in itself very inter-
esting, but this sound should not deflect one from the
actuator point that we should try to minimize the com-
plexity within highly entangled systems. Although
intelligence is a desired characteristic, it is most of
the times not true that the intelligent systems decrease
complexity. For this reason what we are search-
ing for is autonomy more than intelligence. Auto-
nomic systems can be composed of system manage-
ment methodologies, which are defined to be static.
However, the self model can also cover artificial in-
telligence methodologies in order to make ’learning’
possible so that knowledge and functions of the sys-
tem are adapted to new environments gradually and
become self-aware resulting in autonomous environ-
ments.
2.3 Sine Qua Non of Self
Based on the discussion of Sections 2.1 and 2.2, we
define three necessary components for the existence
of a self’. These are the reward mechanism defined
over a state space, a primary triggering function re-
flecting a targeted incentive running over the reward
mechanism (ENE) and a possession which can be an
abstract or a concrete one. Although these three com-
ponents are sufficient to build the very basics of self
regardless of what kind of possessions the self has or
how its reward map is designed, the notion of auton-
omy needs the definition of properties that enable the
perception and evaluation of the possessions by the
self. An illustration of this model can be seen in Fig-
ure 1 We define those properties in the following sec-
tion.
ENE
Observer
Function
Triggers
Emotion
Receives
“The self”
Figure 1: A simple illustration of ENE.
2.4 Observer
Self-awareness, the most fundamental requirement
for a self-managing system, is the continuous recog-
nition of the own possessions, their states and func-
tionalities abstracted from the environment in which
self exists. Correspondingly, the process of observing
the changes in own possessions, their states and func-
tionalities relative to the environment of existence is
called self-monitoring. During the process of self-
monitoring, the observer and the observed are one in
a process that recursively gives rise to each.
In the area of social sciences self-monitoring is
not a new issue. Human beings are self-aware by na-
ture. Psychology, sociology, medicine and even sport
deal with the monitoring of self But psychology and
sociology use the term ’moitoring’ with a different
meaning. Monitoring is not only dealing with measur-
ing and observing properties, according to the litera-
ture, self-monitoring involves three major and some-
what distinct grounds (Baron and Greenberg, 1990):
the willingness to be the centre of attention
WINSYS 2007 - International Conference on Wireless Information Networks and Systems
286
sensitivity to the reactions of others;
ability and willingness to adjust behaviour to in-
duce positive reactions in others.
Along these points, monitoring does not only deal
with observation, but also includes behavioural con-
trol tendencies. However, in the technical literature
these are two separate issues. In a technical sense self-
management is rather perceived to be the subsequent
use of the recognized possessions for controlling, pre-
serving and changing the recognized states and func-
tionalities.
Entities performing self-monitoring are separated
into two distinct types. High-monitors are aware
of their own presentation within their environment,
keeping track of external cues to regulate their be-
haviour. In contrast to that, low monitors are sensi-
tive and aware of their ’inner’ states resulting in a low
degree of public awareness. They pay ’less attention
to appropriate behaviour in social situations and try
to maintain consistent behaviour across all situations’
(Snyder, 1986). High-monitors deal mostly with pub-
lic awareness. The environment is monitored more
than themselves. These entities are concerned about
each situation and how to fit to it and choose the right
’face’. Low self-monitors focus on internal states and
cues as an indication of behaviour modification. Pri-
marily, these care for a consistent behaviour across
different situations.
Although monitoring within ENE is not a must,
its existence is necessary during implementations in
order to generate corresponding emotion stimuli. Ad-
ditionally, in case of the need for behavioural intelli-
gence it is also necessary that the self is aware of
its state. Depending on the environment in which
the self exists, both high and low monitoring must
be made available. The generation of corresponding
emotion stimuli is a problem specific issue. Although
low monitoring takes a role in generating emotion
stimuli always, high monitors are generally used dur-
ing decision making as additional information. Only
in case of social selfs it might be required to include
high monitoring for emotion stimuli generation.
3 A SIMPLE SELF MODEL
USING IEEE 802.11E
The amendment IEEE 802.11e extends the existing
IEEE 802.11 standard by adding new functions tar-
geting both differentiated and integrated services. In
this way, IEEE 802.11e enables QoS enhanced ac-
cess points (QAP) to cope with real-time traffic that is
delay-sensitive, jitter-sensitive and error-prone, such
as voice and video streams (see (Chalmers and Slo-
man, 1999) for a detailed overview).
New primitives of IEEE 802.11e, such as MAC
Layer Management Entity ADD Traffic Stream re-
quest (MLME-ADDTS.request) allow negotiations
between a QAP and a QoS enhanced station (QSTA)
such as required maximum MAC Service Data Unit
(MSDU) size, data rate, burst size and average de-
lay. Additionally the QoS enhanced basic service set
(QBSS) load element being advertised within beacon
frames periodically gives information about the situa-
tion of an access point. This enables decision making
prior to attachment using parameters other than signal
strength. Furthermore, there is a number of new QoS
related parameters used by a central control mecha-
nism, the hybrid coordinator (See 1 for more detail
on those parameters). Hybrid coordinator (HC) is re-
sponsible for managing those parameters in order to
assure the best possible QoS over the QAP.
The number of available parameters that can be
reconfigured during run time, the definition of a clear
target, which is the maximization of QoS offered to
the attached mobile stations and the existence of cor-
responding negotiation and control mechanisms make
IEEE 802.11e a very good candidate for the deploy-
ment of a self managing wireless network. There-
fore we adapted our model ENE to the IEEE 802.11e
implementation of the ns2 network simulator. In
the following sections we introduce three problems
of IEEE 802.11e and corresponding ENE models to
solve those problems.
Within our simple scenario there exist a QAP, a
number of stations already connected with the QAP
and a QSTA which has a new traffic stream to transfer.
There are three problems for the establishment and
preservation of the connection with the QSTA.
3.1 The Scenario
Within our simple scenario there exist a QAP, a num-
ber of stations already connected with the QAP and
a QSTA which has a new traffic stream to transfer.
There are three problems for the establishment and
preservation of the connection with the QSTA.
1. The QSTA should decide if the QAP is the right
choice for making the connection in terms of QoS.
2. The QAP must decide on accepting the new re-
quest coming from QSTA or not.
3. During the data transfer from QSTA, QAP has to
make sure that its parameters are configured in a
way that the negotiated QoS level is preserved.
The first two problems deal with the interac-
tion between two different selfs (high monitoring),
THE QUEST FOR SELF-MODEL IN SELF-MANAGING NETWORKS
287
whilst the last one only deals with a problem within
a self (low monitoring). In order to implement these
scenarios, we used network simulator ns2. We present
here only the models, since this paper aims at present-
ing the model ENE. Real implementations and their
results are referred after each problem.
3.2 Problem 1
In the first problem, the QSTA has to decide whether
it would want to connect to the candidate QAP or not.
The choice of the QAP over which the QSTA sends
its frames is left to vendors within IEEE 802.11e.
However the unit responsible for making this decision
is the station management entity (SME). For making
this decision, the SME is supposed to use the QBSS
load element. It includes three parameters: station
count, channel utilization and available admission ca-
pacity. The station count is the total number of sta-
tions currently associated with the access point. The
channel utilization gives the percentage of the time
the channel is sensed to be busy using either the phys-
ical or virtual carrier sense mechanism of the access
point. The available admission capacity gives the
amount of time that can be used by explicit admis-
sion control. We refer to (Simsek et al., 2006a) for a
detailed study of the QBSS load element and its usage
during decision making.
As we described in Section 2.2, the ENE is an
abstract entity which has no direct influence on the
possessions of the self. These possessions are rather
given to it. Within the first problem, the ENE is em-
bedded into the QSTA with four interfaces. One in-
terface is at the SME, where the decision about the
access point selection is given. The second interface
is for monitoring purposes. Parameters regarding the
characteristics (priority, burst rate, mean data rate and
delay bound) of traffic stream are visible to the ENE.
ENE regards both the function responsible for can-
didate access point selection and the parameters of
the traffic stream, hence the traffic stream itself as its
own possessions. The third interface that the ENE has
is the point where the QBSS load element advertise-
ments are received by the QSTA. ENE regards QBSS
load element as the ’other’ which is the QAP.
ENE is responsible for making a decision with re-
spect to the state which it perceives as a result of the
monitoring at the second and third interfaces and ap-
plies its decision on the first interface. The last in-
terface is indirectly bound with ENE. The QoS of the
traffic being transmitted is observed and satisfaction
level is calculated using mean opinion score (MOS).
This MOS level is then given to the ENE as the satis-
faction mechanism.
As can be seen from figure 2, the original ns2 im-
plementation is not changed except the point where
the selection decision is given. This is true regard-
less of the problem. This property is the reason for
the name ’embedded notion of emotion’, since ENE
is embedded into an existing system partially or en-
tirely, without modifying the system. As described in
Sections 2.1 and 2.2, the functionality of ENE is sim-
ply based on the desire for having more satisfaction
and on the ability to trigger the possessed functional-
ities of the system in which it is embedded. We refer
to our previous study (Simsek et al., 2006b) for the
implementation of this model and its results.
ENE
SME
Monitoring
Monitoring
(others)
Traffic Stream
Characteristics
QBSS Load
Element
Decision
enforcement
Emotion
ENE
SME
Monitoring
Monitoring
(others)
Traffic Stream
Characteristics
QBSS Load
Element
Decision
enforcement
Emotion
Figure 2: Representation of ENE within Problem 1.
3.3 Problem 2
Following the decision of QSTA for initiating the
transmission with a QAP, the QAP has to decide
whether it should admit the request or not, depend-
ing on the characteristics and if existing on the cor-
responding service level agreements with the mobile
user. During the first problem, the QSTA had only
limited information for making its decision, which is
the QBSS load element of the QAP. However, the
QAP has all the capacity related information (chan-
nel utilization of each priority and each traffic stream,
corresponding QoS metrics such as delay and loss
rates of each priority) and also its own admission con-
trol policy (service level agreements done with other
mobile users) (See Figure 3). Although, both prob-
lems are very similar in nature, namely, if the QAP
will be able to satisfy the QoS requirements of the
traffic stream or not, the amount of available informa-
tion in the second problem is significantly higher and
the point of view is different. This allows us to com-
pare behavioural differences of both parties (QSTA
and QAP) for the same problem. This is especially
interesting when learning is enabled, since the expe-
riences of QSTA and QAP and also the monitoring of
the behaviours from different perspectives are com-
pletely different. Furthermore, we can also analyze
the consequences of using different amount of infor-
mation.
WINSYS 2007 - International Conference on Wireless Information Networks and Systems
288
ENE
HC
Monitoring
Traffic Request
Characteristics
Decision
enforcement
QoS for each class, SLAs, QBSS
Load Element, Parameter
Configuration
Monitoring
(others)
Emotion
ENE
HC
Monitoring
Traffic Request
Characteristics
Decision
enforcement
QoS for each class, SLAs, QBSS
Load Element, Parameter
Configuration
Monitoring
(others)
Emotion
Figure 3: Representation of ENE within Problem 2.
3.4 Problem 3
Different than the first and the second problems,
where the QSTA and the QAP had to make decisions
following the request for a traffic stream, the QAP is
responsible for preserving the promised QoS levels
for each traffic stream in a continuous manner within
the last problem. This requires permanent monitoring
of QoS levels of each active traffic and modification
of own parameters respectively.
ENE
Parameter
Configuration
Monitoring
QoS / SLA of each
stream
Decision
enforcement
HCCA EDCA
HC
Emotion
ENE
Parameter
Configuration
Monitoring
QoS / SLA of each
stream
Decision
enforcement
HCCA EDCA
HC
Emotion
Figure 4: Representation of ENE within Problem 3.
As seen from figure 4, the IEEE 802.11e in its
own has two functions, EDCA (Enhanced Distributed
Channel Access) and HCCA (Hybrid Coordinator
Controlled Channel Access). These functions are
used by the hybridc coordinator (HC) for differenti-
ated and integrated services respectively. The HC has
full control over HCCA and its schedule, whereas it
has only limited influence on the functioning of the
EDCA. We are going to describe how different lev-
els of control over the possessions of the self can
be managed in our next study. Nevertheless, although
HC cannot control the transmissions over EDCA, it
can restrict the use of some of the four priorities by
requiring for explicit negotiation for those priorities.
Additionally, it can change the contention window
sizes of EDCA for each priority and also alter the
TXOP limit. Hence, the HC can indirectly affect the
QoS of traffic using EDCA. Table 1 shows a list of the
most influential parameters of HCCA and EDCA that
we found to be most significant for the functioning of
IEEE 802.11e.
Within the third problem, ENE has two interfaces
with the QAP. The first interface is used for monitor-
ing the QoS levels of each traffic by the use of which
satisfaction level is given to ENE. The second inter-
face is with the parameters of table 1. ENE alters
the parameters continuously with respect to the sat-
isfaction level it expects. This problem is also known
as algorithm configuration problem. We are going to
present our implementation of the third problem and
its results within our next studies.
Table 1: List of HC configuration parameters.
HCCA
% of time reserved for HCCA
service interval
EDCA
CWmin
CWmax
TXOP Limit
priority restriction
4 CONCLUSION
Although there are plenty of models proposed so far
for building the intelligent environments of the fu-
ture which should reduce system complexities, these
models either could not bear down the conceptual
planning phase like many bio-inspired models (Sim-
sek and Albayrak, ) or they proved to be ineffi-
cient in terms of performance like the agent tech-
nology (Keiblinger, 2000). In building such models,
the research community ignored the following crucial
points which constitute the igniting factors of our pa-
per.
The effort to build and preserve the self manag-
ing components should be significantly lower than
traditional solutions. As mentioned above, the
agent technology is a very good example for an
unsuccessful effort in bringing this facility. Al-
though in terms of conceptual planning the agent
technology proved to reflect many requirements
that one may expect from autonomic environ-
ments, its development, application and mainte-
nance became to be a greater obstacle in reducing
system complexity.
The model for self managing components should
be system independent and work generically in or-
der to prove its ability to reduce system complex-
ity. Otherwise it is an additional load to software
developers and it makes more sense to use appli-
cation specific solutions. Therefore, the self man-
aging model should have the plug & play capa-
THE QUEST FOR SELF-MODEL IN SELF-MANAGING NETWORKS
289
bility which starts functioning after it is embed-
ded into the system being considered regardless
of the application type. However previous models
required the implementation of the systems from
scratch.
There must be clear metrics for defining self man-
aging components, their states, targets and be-
havioural directions. However there is so far no
such well defined metric for self managing com-
ponents.
It must be possible to define the behaviour of self
managing components in a formal manner so that
it becomes manageable. This behavioural man-
agement should be independent of the application
and easy to define so that the course of the actions
of the self managing components can be followed
in a causal manner during runtime.
The points mentioned above show how our de-
sire for a more intelligent computing environment can
easily become a fallacy in terms of application effi-
ciency. Hence, during the development of models for
autonomic computing it is essential to consider the
practical aspect primarily. This is especially impor-
tant for the new trend ’autonomic networking’.
In this paper we presented a model for autonomic
networking which is simplified down to the very ba-
sic elements of the self as perceived by human be-
ings. In doing this, we especially paid attention to
our originating point, which is the need for autonomic
behaviour. We kept the model as simple as possible
by considering the above mentioned problems so that
the model is scalable and applicable within the seven
OSI layers even with scarce resources. For illustration
purposes, we also summarized three implementations
of ours using the introduced ENE model for solving
three different problems of IEEE 802.11e. In this way,
we opened the way for more problem specific studies
in autonomic networking.
REFERENCES
Abbas, A. (2003). Autonomic computing report: Charac-
teristics of self managing it systems. Technical report,
Grid Technology Partners.
Baron, R. A. and Greenberg, J. (1990). Personality and or-
ganizational conflict: Effects of type a behavior pat-
tern and self-monitoring. Organizational Behavior
and Human Decision Processes, 44(2):204–206.
Bratko, I. (2000). Prolog Programming for Artificial Intel-
ligence. Addison Wesley.
Chalmers, D. and Sloman, M. (1999). A survey of quality
of service in mobile computing environments. IEEE
Communications Surveys and Tutorials, 2(2).
Damasio, A. (2005). Descartes’ Error : Emotion, Reason,
and the Human Brain. Penguin (Non-Classics).
Ferber, J. (1999). Multi-Agent Systems: An Introduction
to Distributed Artificial Intelligence. Addison-Wesley
Professional.
Glasser, W. (1999). Choice Theory: A New Psychology of
Personal Freedom. HarperCollins Publishers.
Horn, P. (2001). Autonomic computing: IBM perspec-
tive on the state of information technology. In
AGENDA’01, Scottsdale, AR.
Keiblinger, A. (2000). Agentenorientierter softwareentwurf
zur dynamischen konfiguration. Diploma Thesis.
Kephart, J. O. and Chess, D. M. (2003). The vision of auto-
nomic computing. Computer, 36(1):41–50.
Levesque, H., Pirri, F., and Reiter, R. (1998). Foundations
for the situation calculus.
Metzinger, T. (2003). Being No One: The Self-Model The-
ory of Subjectivity. The MIT Press.
Muggleton, S. (1999). Inductive logic Programming. In
The MIT Encyclopedia of the Cognitive Sciences
(MITECS). MIT Press.
Mulhauser, G. R. (1998). What is self awareness?
Rappa, M. A. (2004). The utility business model and the
future of computing services. IBM Syst. J., 43(1):32–
42.
Scott, M. L. (2000). Programming language pragmat-
ics. Morgan Kaufmann Pu blishers Inc., San Fran-
cisco, CA, USA.
Simsek, B. and Albayrak, S. Living factory: Back to
koestler in holonic manufacturing.
Simsek, B., Wolter, K., and Coskun, H. (2006a). Analy-
sis of the qbss load element parameters of 802.11e for
a priori estimation of service quality. International
Journal of Simulation: Systems, Science and Technol-
ogy, Special Issue: Performance Engineering of Com-
puter and Communication Systems, 7(2).
Simsek, B., Wolter, K., and Coskun, H. (2006b). Dy-
namic decision making for candidate access point se-
lection. In Ga
¨
ıti, D., Pujolle, G., Al-Shaer, E. S.,
Calvert, K. L., Dobson, S. A., Leduc, G., and Mar-
tikainen, O., editors, Autonomic Networking, volume
4195 of Lecture Notes in Computer Science, pages
50–63. Springer.
Smirnov, M. (2004). Autonomic communication: Research
agenda for a new communication paradigm.
Snyder, M. (1986). The psychology of self-monitoring.
W.H. Freeman & Company.
Sterritt, R. and Hinchey, M. (2005). Why computer-based
systems should be autonomic. In ECBS ’05: Proceed-
ings of the 12th IEEE International Conference and
Workshops on the Engineering of Computer-Based
Systems (ECBS’05), pages 406–412, Washington, DC,
USA. IEEE Computer Society.
Tauber, A. (2002). The biological notion of self and non-
self. The Stanford Encyclopedia of Philosophy (Sum-
mer 2002 Edition).
WINSYS 2007 - International Conference on Wireless Information Networks and Systems
290