A Systematic Review of Self-adaptation in Service-oriented Architectures
M. Pilar Romay
Dept. Inf. & Com. Syst. Engineering (DISIT)
St. Paul-CEU University
Boadilla del Monte, Madrid, Spain
pilar.romayrodriguez@ceu.es
Luis Fernández-Sanz, Daniel Rodríguez
Dept. of Computer Science
University of Alcalá (UAH)
Alcalá de Henares, Madrid, Spain
luis.fernandezs@uah.es, daniel.rodriguezg@uah.es
Abstract—The study of adaptivity, i.e., the capability to react to
changes in the environment, is becoming ever more important
in many fields of study, and in the development of software in
particular. This paper presents a systematic review in which
both the extension and complexity of this notion are examined.
After studying the influence from external fields, this review
checks the hypothesis of using the scope of service-oriented
architecture as a comparable model for the whole field. As part
of the systematic review, the influence of the most relevant
bibliography is considered, and the terminology is clarified.
Keywords – adaptivity; self-adaptation; service architecture;
autonomic systems; SOA; systematic review.
I. INTRODUCTION
The growing complexity, along with continuous
operation, of software systems –not only conventional ones,
but also the next complexity level, so-called large-scale
software systems [1] – has greatly increased the interest of a
series of techniques for self-managing system features. These
techniques make possible for them to guarantee a wide range
of properties, all by themselves. Traditionally, these
properties had been dealt with manually, or had to be
developed from specific requirements. Instead of that, the
new approach considers them as intrinsic system properties,
and thus they should be dealt with automatically, and
considered as just another issue in conventional software
systems development.
Systems conceived in such a way are generically known
as adaptive systems or, more specifically, as self-adaptive
systems [2].
Therefore, we have systems able to deal with faults and
critical situations (self-healing), able to control their own
behavior (self-managing) [3], able to observe and evaluate
their own performance (self-monitoring), able to modify their
own configuration to react to changes in their environment
(self-configuring), or even to automatically guarantee certain
system-level properties, such as protection, fault tolerance,
etc. (autonomic systems) [4] [5], among many others.
The wide range of systems which could make use of
these adaptive properties causes a great variability; therefore
many different approaches could be conceived. For this
reason, it is reasonable to focus our efforts on a specific area
of study: in our case, software services. This area has been
chosen because it still covers a wide range of systems and
shows a great variability itself, and therefore it can be
considered as a representative, even a lower-scale analog, for
the whole of the field of adaptive systems. The goal of this
work is, therefore, to study self-adaptive software services.
Software services define, due to their own properties, an
area of a great potential to describe and use adaptive (or
adaptation-related) features. Moreover, service and service-
oriented architectures are among the systems where the need
for these features is clearer, and more compelling: the nature
of services is inherently dynamic, and this implies the need
for adaptation; and also the structure of service architectures
requires the flexibility that self-adaptation provides. In short,
this make our specific goal (consider adaptation in services,
rather than in general systems) even more pragmatic. Finally,
considering the growing, relevance and broad dissemination
of service ecosystems, this is also the environment in which
this approach is currently pertinent and more interesting.
Adaptivity is often described at different levels, namely
at service level or the wider system level [6], but this will not
be the main interest of our study. Instead of that, we will
focus on exploring and analyzing adaptivity and all its
related properties, a set which is often generically known as
self-*.
Moreover, this paper will also consider the impact of
self-organization (considered as a related notion, rather than
as an adaptive feature) within the specific area of service-
oriented architecture. Our main goal is to determine which
properties are implied in adaptive systems, with a special
focus on service architectures – i.e., to be able to evaluate
adaptivity in services.
For this purpose, this paper presents an initial study of
the field, which will be used to delimit the boundaries of the
area and to check the reliability of the hypothesis about the
service-oriented approach and its applicability to evaluation.
The core of this study is structured as a systematic review:
after defining a set of goals and the corresponding research
questions, and discussing the background on the field, the
review makes an extensive bibliography review, which is
carefully examined and analyzed in order to achieve the
corresponding conclusions.
The paper is structured as follows: first, we present the
context of our study, including the definition of four primary
goals and the method of our systematic review. Then, we
provide some background justifying the interest of this study,
as well as the implicit connections between its areas. After
that, we characterize the revised information, and outline the
method we have used to locate and classify this information,
describing the performed searches and their results. We end
by summarizing the conclusions from several perspectives.
331
ICSEA 2011 : The Sixth International Conference on Software Engineering Advances
Copyright (c) IARIA, 2011. ISBN: 978-1-61208-165-6
II. CONTEXT OF THE STUDY
Though it might seem a secondary issue, the relevance of
adaptivity is such that even well-known authors as Kramer &
Magee have claimed [7] that “a significant advance in the
techniques which are required for the effective development
of adaptive systems would imply an advance of an order of
magnitude in every fundamental aspect of Software
Engineering”.
Having this relevance in mind, the main goal which has
driven the conception of this study is focused in finding a
model which makes possible, by using a set of attributes, to
define and assess adaptivity in the context of services. This
model could alternatively take the form of a framework, or
even a methodology.
Then, this paper intends to provide a characterization of
the field of adaptivity. For this purpose, it lays out a set of
specific goals to drive the study, which should make possible
to measure and digest the breadth of the field, and to confirm
the need of the study itself. Moreover, it also intends to value
the most important contributions in the process.
From these goals, the paper follows a methodological
approach based on [8] with the purpose to achieve a greater
soundness than a traditional narrative description. The more
important limitations of such a study are also considered:
publication limitations (publishing bias), and selection
limitations (selection bias). The first one refers to the relative
impact of negative studies –i.e. which have not significant
differences with previous proposals–, when these have not
been published, or are only rarely referenced in the literature.
Also, there could be interesting studies written in another
language, or even duplicate references which could later
influence the metanalysis. The second one is related to the
definition of inclusion and exclusion criteria: the purpose is
to avoid to neglect the inclusion of relevant work, and also to
include misleading papers, which hinder dealing with the
relevant topics in an objective way.
The systematic review will be developed in the next
sections. First, we will describe the main goals of the study,
and outline our methodological approach. Next, we will
briefly explore the background on service-oriented
architectures and their relationship to adaptivity, focusing on
the need for evaluation and the influence of dynamic service
composition models. Then the systematic review itself is
unfolded: after presenting the main data sources, the search
strategies and selection criteria are described – to later
present the results of the review and discuss the conclusions.
A. Goals of the Study
Therefore, the goals of our systematic study are: (1) To
confirm the breadth and applicability range of adaptivity. (2)
To verify the novelty of this field of study within the context
of Software Engineering. (3a) Related to the previous one, to
evaluate adaptivity in service-oriented architectures. (3b) To
assess interesting contributions which could be applied to the
study’s primary goal (i.e., to determine the properties which
characterize adaptive systems). (4) By exploring the previous
four points, to identify the used terminology.
B. Methodological Approach in the Study
This (systematic) study begins by planning the review,
then conducting the review, and finally reporting the review.
The first activity of this process is a bibliographic search.
Based on a set of research questions related to context
definition, modelling, and management, we defined a list of
keywords and search strings used for our investigation.
The defined searches will be oriented to cover the goals
of the study, as proposed in section II.A. In this part of the
process, the selected keywords and their synonyms are of a
great relevance: the obtained results strongly depend on a
good selection of these terms.
For this reason, we also designed and realized an specific
search, focusing on articles and papers which tried to provide
a wider vision of the field, such as (other) research reviews,
overviews, state-of-the-art articles, etc.
The initial terminology search should just be considered
as an approximation, and it will be later tuned and adjusted,
to be refined by means of the obtained results during all the
process. To some extent, the process itself serves as the main
control in this initial search phase, and it could cause an
additional iteration within the systematic review process – it
just depends on the actual extension and variability of the
terminology in the field.
After the search, we proceed to select and evaluate the
obtained information. For this purpose, as already noted, the
study defines acceptance and rejection criteria related to its
specific goals, and in particular to the main goal – which was
the reason to do the study, in the first place.
To finish, the obtained results will be analyzed and
interpreted. In this phase, the process will make possible to
synthesize the results with regard to the proposed goals.
III. BACKGROUND: ADAPTIVITY IN SERVICE-ORIENTED
ARCHITECTURES
Nowadays, the notions of service orientation (or service-
oriented computing, SOC) [9] and service-oriented
architectures (SOA) have been totally integrated in the
current conception of software. This is the reason why they
define a perfect workbench to assess adaptive properties in
generic software systems.
Therefore, this section reviews the context of work in
both fields, focusing in the assessment of adaptivity, and
service-oriented architectures.
A. Adaptivity Assessment & Evaluation
Adaptive systems can be defined as “systems able to
react to automatically adapt themselves to changes in their
environment”. This reaction can be specifically programmed,
or could rise from an emergent behavior. This category of
systems has been globally designated with the name of self-*
systems [10], which explicitly refers to the variability of the
concrete aspect to consider. However, in recent times most
authors prefer to designate them with the generic name of
adaptive (or self-adaptive) systems, like this paper did also in
the Introduction.
332
ICSEA 2011 : The Sixth International Conference on Software Engineering Advances
Copyright (c) IARIA, 2011. ISBN: 978-1-61208-165-6
Adaptivity is a complex field of study. First, because the
term has explicitly been conceived to be generic, so we must
first decide which specific feature (“attribute”) are we going
to consider every time. And second, because too often we
lack a clear reference model which could serve as the basis
to compare to the system’s degree of adaptivity. Therefore, it
is necessary to have some kind of model to make possible to
assess those capabilities, either quantitatively (in the ideal
case) or at least roughly, by approximation.
In general, the development of adaptive systems, as well
as the more concrete development of adaptive services, lacks
a clear set of methods and metrics able to assess the actual
capabilities of a specific implementation. That is, it is really
difficult to even decide if a given system is “adaptive” or not.
In fact, this particular distinction is almost intuitive; however
the increasing importance of this features, and the difficulties
in their implementation, highlight the relevance of achieving
the definition of a quantitative approach, able to deal with
concrete values. For this reason, one of the main goals of this
study focuses in checking existing references, which enable
or guide the process to obtain either the model or relevant
metrics, to be able to assess the level of adaptivity, even in a
qualitative way.
B. Service-Oriented Architecture: Service Composition
Mechanisms
A well-known definition of service-oriented architecture
(SOA), as given by Michael Papazoglou [11] states that it is
“a meta-architectural style, based in loosely coupled
services, which provides flexibility to business processes in
an interoperable way, and independently from the
technology”. Therefore, its main goal is interoperability,
which is itself a consequence of loose coupling.
However, a standard definition of SOA is still debated, in
spite of the popularity of the term – probably because it has
been used with different meanings in different contexts, and
referring to different technological aspects. Beyond those
details which distinguish the many variants of the concept of
service (web services, RESTful services, grid systems, etc.),
there are still several intrinsic features in its definition. These
features imply that service-oriented architectures are a priori
more dynamic and flexible than many “traditional” ones, in
particular component-based architectures – and this can be
considered inherent to its own nature.
From this point of view, it is interesting to note at least
two of these features, which suggest this kind of architecture
as a good evaluation workbench for adaptivity:
1) External Composition Mechanisms. First, services are
always part of a modular system – they are conceived to be
used as part of a larger structure. However, there is a subtle
difference to more traditional approaches: service systems
are designed to be composed at runtime.
Services cannot assume anything about the rest of the
elements in the composition. First, their interface is separated
from the rest of the service, and therefore services never
interact directly to the rest of the system. Second, they are
not designed as part of a concrete compound: instead of that,
once they are implemented and deployed, they are included
in some composite system, which was later conceived.
These are the reasons why the well-known composition
models for services (choreography and orchestration) must
be conceived as external compositions. Thus a service does
not even need to know if it is contained in a composite: the
business logic (the “intelligence” of the system) belongs in
the structure itself, not in its individual components. Within
an orchestration, it is in the orchestrator; but choreographies
are even more complex, as the composition schema is
distributed along the composite – i.e. it is decentralized.
Every individual service receives just a “local” subset of
instructions, without a perspective of the global plan. Even
service mashups, a promising approach, are again an external
composition model – in fact, essentially an orchestration.
Another consequence is crosscutting. Unlike traditional
composition, service models do not preclude that the same
service is simultaneously a part of more than one composite.
This implies that every service composition is orthogonal to
any other which is performed later [6].
2) Instrinsically Open Architecture. Of course, many
existing systems, and distributed systems in particular, have
claimed to define an open architecture. In practice, an open
system is every system which, by defining or using an
standard interface, is able to compose any external element
defined as a client of that interface. However, if constraints
imposed by this interface are too strict, the limits they define
hinder the capture of information about the different clients
– i.e. it would present an homogeneous architecture,which is
exactly the opposite of our goal.
Services use a different approach: the interface is defined
at the beginning, to offer a concrete functionality (a service),
and to guarantee a certain quality level (i.e. QoS). But apart
from that, services are conceived, even at the technical level,
to be composed to any other element able to interact to them.
Therefore, they are presented as the ultimate open system: in
the specific case of RESTful web services, for instance, the
only actual constraint is the use of the HTTP protocol, which
was conceived using the REST architectural style itself - and
this is not an actual constraint, nowadays.
Also, we have to consider that the current evolution of
service systems has a clear trend towards a significant rise of
the scale. The original “XML web services” were in general
small modules, of a scale comparable to that of objects, or
even smaller. Currently, the concept is clearly shifting to be
equivalent to so-called Software as a Service (SaaS) – where
the scale of a service is similar to that of a complete
application. In fact, the approach itself is evolving from the
potential provided by a concrete technology which focused
on interoperability, to the design of a new, generic software
distribution model (shifting from “product” to “service”).
In any case, current service-oriented architectures, when
this term is understood in the wider sense [12], present the
same features of flexible and open composition we have
already noted – and this makes them adequate as a
workbench for adaptivity evaluation in software systems.
IV. REVISED INFORMATION & METHODOLOGY
This study is based on information obtained from several
digital bibliography search engines. Specifically, we have
333
ICSEA 2011 : The Sixth International Conference on Software Engineering Advances
Copyright (c) IARIA, 2011. ISBN: 978-1-61208-165-6
used search engines from the best known and most widely
recognized publishers in the fields of Computer Science &
Information Technology, as well as Google Scholar.
To have a preliminary structuring of the area, we first
considered the results provided by Google Scholar. The goal
was to assess the research activity on adaptive systems in the
period 2000-2011, including every potential environment,
and comparing these results to those in the specific subarea
of service-oriented architectures in section A .
The remaining searches followed a more systematic
approach, guided by specific goals (in the form of questions),
specifically those which were proposed in section II.A.
Throughout all this search process, we have considered
the possibility of evaluating the used terminology, with the
purpose of extending the search to a wider scope – but still
within the parameters of the study. This evaluation has made
possible to change and evolve the initial searches, to the final
form we will describe in the following.
After performing those search processes, our inclusion
and exclusion criteria were used to select the most relevant
articles. Then we also examined the references cited in these
papers, with the purpose to select other relevant papers,
which were not located previously due to their publication
stage, or which have been published by some additional
publisher. This way, the publishing bias we mentioned in
section II.B.
A. Search Strategy and Selection of Areas
The search strategy has been guided by our goals, by
answering to a set of questions.
The questions were bound to specific terminology. The
variety of meanings of some of the terms used in our search
made necessary to apply an iterative, evolutionary approach,
in which those search terms were finely tuned. At the end of
the process, our study has made possible to obtain a specific
terminology summary, which covers goal (4). This specific
terminology, obtained from multiple sources in the revised
information, has been represented using a pyramidal mesh,
which will be detailed in section IV.C. Therefore the most
significant terms and notions related to our field of study
have been collected, also emphasizing their similarities and
differences, something which is not always completely clear.
This way, in our iterative process we have refined concepts
such as autonomic vs. autonomous, adaptation vs. self-
adaptation, adaptive, self-organization, self-monitoring, etc.
The definition of these terms, as part of the results for our
goal (4), is briefly explained in section IV.C, where it also
explains the aforementioned pyramidal structure.
Within these terms, we should emphasize those which
were considered for our search, namely:
Adaptation, adaptive, adaptivity, self-*
“Software service”, service-oriented, SOA
Evaluation, “quality model”
In order to fulfill our first and second goals, we
performed a series of searches on Google Scholar, as well as
other databases. In the final search on Scholar, the questions
related to these goals were the following:
Assessment of the number of articles dealing with
adaptivity, against the number of those doing the
same in the service-oriented architecture area.
Which disciplines (research areas) are dealing with
and applying adaptation?
The first search, which intends to identify the different fields
of study related to adaptivity, is driven by the following
queries, referring to the compared subsets:
Query #1: (("autonomic" OR adaptive OR
adaptation OR autonomous OR adaptivity OR self)
AND (evaluation OR quality))
Query #2: (("autonomic computing" OR adaptive
OR adaptation OR autonomous OR adaptivity OR
self) AND (evaluation OR quality)) AND
(("software service") OR ("service-oriented
Architecture") OR ("Service Oriented
Architecture"))
These queries, on the Google Scholar engine, resulted in
about 7.806.600 references for query #1 and a total of 17.565
for query #2. The refined search provides roughly about
1500 results every year, from 2000 to 2011. The scope of the
study is very wide, covering almost any scientific area –
which is not surprising and confirms our intuition.
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
2001
2000
770
2980
3180
3050
2630
1980
1410
668
379
243
189
86
Figure 1. References for Query #2 from 2000 to 2011 (1-12)
The most representative areas for query #1 are: Life Science,
Engineering, Social Science and Law, Mathematics and
Statistics, Medicine and Computer Science (e.g. Ubiquitous
Computing, Grid Environments [13], mobile systems and
services [14] [15], Domotics [16], etc.)
Goals numbered as (3) are essential in the context of this
study – i.e. the evaluation of adaptivity in service-oriented
architectures. Related searches have been more specific, and
they have already been performed in bibliography databases
from the publishers themselves. The purpose was to obtain a
more accurate list of articles, trying to reach all the relevant
information – without any accidental loss. We also have used
2; 2980; 18%
3; 3180; 19%
4; 3050; 17%
5; 2630; 15%
6; 1980; 11%
7; 1410; 8%
8; 668; 4%
12; 86; 0%
11; 189; 1%
9; 379; 2%
10; 243; 1%
1; 770; 4%
1
2
3
4
5
6
7
8
9
10
11
12
334
ICSEA 2011 : The Sixth International Conference on Software Engineering Advances
Copyright (c) IARIA, 2011. ISBN: 978-1-61208-165-6
references from the selected articles, and in relevant cases we
have also searched for the corresponding citations.
For instance, a representative query could be:
Query #3: ((("autonomic computing" or self) and
(adaptive or adaptation or adaptivity)) and
(evaluation or quality)) and (("software service") or
("service-oriented Architecture") or ("Service
Oriented Architecture"))
B. Inclusion and Exclusion Criteria.
1) Related to goals (1) and (2): Neither inclusion nor
exclusion criteria were defined – this search was delimited
just by query clauses themselves, i.e. queries #1 and #2. This
could seem less “systematic” than the remainder of the study.
However, we did not intend to do a detailed classification of
areas and fields of study, but to assess if our suggestion (to
focus on service architectures) was reasonable. This goal
alone could be used to justify a specific systematic study,
which would be even more complex than the one presented
here. The reason to include this goal is to perform a shallow
examination of some of the areas suggested by many search
engines, with the purpose of perceiving the actual extension
of the field, as well as its growing rate. A systematic study
on this specific aspect would be of great interest to detect
methods or tools (from other fields) which could be applied
in the context of adaptive software.
2) Related to goals numbered as (3). In this search
process, queries are quite more specific, and they mainly
focus in evaluating adaptivity by means of self-properties.
Therefore it considers papers including models, frameworks,
metrics and evaluations on the topic. This study excluded
papers not dealing with self-properites, and those which did
not focus on assessing adaptivity/autonomic features.
3) Related to goal (4): The resulting terminology has
been extracted from papers selected in the previous phase.
Therefore their inclusion and exclusion criteria are the same.
However, some additional selection criteria are also added;
specifically, articles which define or clarify terminological
aspects, or which perform reviews in which terminological
features are also clarified.
C. Results
1) Related to goals (1) and (2): The range and scope of
the many fields of study which apply adaptivity is too wide
to be considered in this paper – in fact, it would require an
specific study itself. Therefore, for this purpose we refer to
the results outlined in section IV.A, and to the conclusions
summarized in sections V.B and V.C, which expose a global
vision for this part of our study.
2) Related to goal (3): This goal, together with results
about terminology from goal (4), provides a characterization
of adaptivity. The following table summarizes briefly this
part of the study. It describes representative categories of
existing work, indicating for each one of them references,
goals, projects, metrics and their organization.
TABLE I. EVALUATION OF ADAPTIVITY
Ref
Evaluation of Adaptivity
Goal/ Project/ Context
Metrics/ Organization
[28]
[29]
Metrics to evaluate Self-*
systems criteria
/ -- /
Web-based C/S, E-learning
(AHA!), Videoconference,
Multiagent Systems
The many metrics for each
Propierties (reuse, genericity…)
/methodological, architectural,
intrinsic characteristic and
runtime
[30]
Metrics for restarting
strategies in WS Reliable
Messaging (WSRM)
/ -- / WSRM
Effective Transmission Time
(ETTi), Unnecessary Resource
Consumption (URCi),
Savings (SAVi)
/ Adaptation parameters
(structures, payoff,
environments, time)
[34]
Quality Model for the
software architecture of
self-healing applications
(based on ISO 9126)
/Attribute-based
architectural styles
(ABAS)
/ --
Traditional quality attributes
(Maintainability –Modifiability,
Extensibility-, Reliability –Fault
tolerance, Robustness-)
Specific Autonomic Quality
attributes (Support for detecting
anomalous system behavior,
Failure Diagnosis, Simulation of
expected behavior, Differencing
between expected and actual
behavior, Testing of correct
behavior). Autonomic Metrics:
Detection ratio, Detection time,
Fault Model Observability,
Awareness, Coupling
/ Traditional and Autonomic
attributes
[35]
User-level Quality of
Service (QoS)
(Context awareness)
/ PLASTIC, model PFM
/ Pervasive Networking
Environment
Performance evaluation
[36]
Quality model to evaluate
Self-* attributes (adopts 6
features of ISO 9126:
Reliability, Efficiency,
Maintainability, Usability,
Functionality, Portability)
/ --
/ --
The autonomic maturity of each
level in complex software
(Complexity of development,
business domain and
management)
/ Three-level Autonomic
Evaluation Model
(Software Complexity, Relative
Quality Factor, Autonomic
features).
Fuzzy comprehensive evaluation
(qualitative factors)
3) Related to goal (4): These results are summarized in
Fig. 2, which shows the wide spectrum of so-called self-
properties, ranging from very generic properties which can
be applied in many systems (such as context-awareness) to
specific attributes which are only found in some approaches
(like emergence). Apart from these, there are several other,
less frequent, properties – also, many of them are referred to
using different names and variants (self-managing vs. self-
management). All these issues have been considered in the
study, and they are implicitly included in this paper.
Fig. 2 represents three pyramids rather than one – they
are conceptually related, but they must be studied separately.
Pyramid #1 represents environmental adaptation, i.e. the
335
ICSEA 2011 : The Sixth International Conference on Software Engineering Advances
Copyright (c) IARIA, 2011. ISBN: 978-1-61208-165-6