A QOS DRIVEN APPROACH FOR PROBABILITY EVALUATION
OF WEB SERVICE COMPOSITIONS
Dessislava Petrova-Antonova and Aleksandar Dimov
Sofia University, Faculty of Mathematics and Informatics, Department of Software Engineering, 1164, Sofia, Bulgaria
Keywords: QoS, Composite web-services, QoS evaluation.
Abstract: Service oriented systems are becoming a major area in software engineering due to promise of low time and
cost of development efforts. However, in order to enable full benefit of service oriented architecture, some
technical aspects of service-level agreements, also known as Quality of Service (QoS) are important to be
evaluated in formal way. Currently evaluation of web service QoS is done mainly as underlying part of a
corresponding composition model and many models consider only limited number of QoS attributes. In this
paper we present an approach for evaluation of composite web-services, based on analysis of how they meet
their QoS, which is applicable at multiple levels of service composition as well as for wide range of QoS
attributes. Our approach is based on probability theory and is illustrated with a case-study.
1 INTRODUCTION
Currently Service Oriented Architecture (SOA)
based on web services is regarded as important pa-
radigm in many areas not only in software engineer-
ing. The promise of reduced cost and time for enabl-
ing a large range of company-wide business
processes shifts the research focus towards reason-
ing about web service compositions. On the other
hand, technical aspects of service-level agreements,
also known as Quality of Service (QoS) are impor-
tant to be considered as they define additional con-
straints and non-functional requirements toward the
web-services (Cardoso, 2004). Business workflows
and therefore – organizations success are directly
impacted by proper management of SLAs and keep-
ing them within terms of negotiated QoS metrics.
Various classifications (Tran et al., 2009);
(D’Mello et al., 2008) of QoS properties exist in the
literature. However, a widely accepted standard for
specification of web service quality does not exist.
The first efforts in this direction have started in the
context of Web Services Quality Model (WSQM)
prepared by OASIS (Kim and Lee, 2005). According
to WSQM, service level measurement quality is
subdivided into performance measurement sub-
factors including response time, throughput and
maximum throughput, and stability measurement
sub-factors such as availability, reliability and acces-
sibility. Unfortunately, only draft version of the
WSQM is available.
QoS properties may be distinguished on the basis
of their values. If the client requires the values of
QoS properties to be as higher as possible, then
these QoS properties can be classified as increasing
or positive. Otherwise, they are classified as decreas-
ing or negative. Examples for positive QoS proper-
ties are availability and throughput. In contrast,
response time and cost are regarded as negative QoS
properties.
In this context it is of paramount importance to
be able to reason about QoS of web services in for-
mal way. This will enable the activities of service
selection and evaluation when more than one service
satisfies the required functionality. Moreover, formal
approaches will make possible dynamic composition
and reconstruction of composite web-services.
The purpose of this paper is to present a syste-
matic QoS-based approach for evaluation of web
service compositions that applies for various kinds
of QoS attributes and multiple levels of service
composition. It is based on our previous work about
QoS evaluation of web services, assisting the search
process of web services (Petrova-Antonova, 2011).
The approach uses probability theory to quantify the
extent to which composite web services meet their
quality requirements. This may serve as a basis for
composite web service optimization based on quality
of its constituent services. Moreover in this way, one
321
Petrova-Antonova D. and Dimov A..
A QOS DRIVEN APPROACH FOR PROBABILITY EVALUATION OF WEB SERVICE COMPOSITIONS.
DOI: 10.5220/0003602303210326
In Proceedings of the 6th International Conference on Software and Database Technologies (ACT4SOC-2011), pages 321-326
ISBN: 978-989-8425-76-8
Copyright
c
2011 SCITEPRESS (Science and Technology Publications, Lda.)
may also be able to find bottlenecks in service com-
positions.
The remainder of the paper is structured as fol-
lows: Section 2 makes an overview of the related
work; Section 3 presents the theoretical foundation
of our approach; Section 4 describes the steps in the
approach for evaluation of composite web services;
Section 5 illustrates the approach with a typical case
study and finally section 6 concludes the paper and
states some directions for further research in the
area.
2 RELATED WORK
Modeling of QoS of composite web-services has
been a matter of study from various researchers in
recent years. It has been mostly regarded as an un-
derlying part of a corresponding composition model.
А number of efforts that deal with the problem of
evaluation of web services with respect to their
composition exist in the literature (Bartalos and
Bielikova, 2009); (Lin et al., 2008); (May Shan and
Bishop, 2008); (Pistore et al., 2004); (Zheng and
Yan, 2008), but they do not take into account formal
modeling of QoS.
A number of interesting works present modeling
of QoS properties within their composition frame-
work, however, the models are restricted to several
properties only. Such approaches are (Cardellini et
al., 2007); (Ming et al., 2009); (Nam et al., 2009);
(Zheng et al., 2004).
For example, the methodology, described in
(Chang and Lee, 2010) evaluates the quality of ubi-
quitous web services in three dimensions: QoS
(quality of services), QoC (quality of contents), and
QoD (quality of devices). Further, based on Multi-
Criteria Decision Making (MCDM) with preference
functions, the quality factors are evaluated and prior-
ity ordering of web services is made. A heuristic
approach is presented in (Alrifai et al., 2008) that
achieve a close to optimal result while offering low
complexity and high speed of computations. In con-
trast to these approaches in our work we propose
more formal evaluation criteria, based on probability
theory instead of preference functions.
Interesting method for service composition is
proposed in (Che et al., 2009). Authors there use
XML schemas as a means for service composition.
The schema contains data about the conditions on
which to select the service, which include more than
just QoS criteria. However this approach does not
count for the indeterminism that a service may not
satisfy a quality requirement, i.e. they either say that
it will meet the requirement in any case or will not
do so at all. Instead, our approach takes into impre-
cision of quality attributes, by involving the mechan-
ism of probability.
Very formal technique for service composition
and composition evaluation is proposed in (Chakhar
et al.). In this work a specific method for evaluation
of the service upon each quality attribute is pro-
posed. While very exhaustive, such approach may
appear difficult to be applied in real practice. In our
work we try to overcome this issue by employing a
formal approach, applicable to any quality attribute
without modifications.
Another method for QoS evaluation which takes
into account imprecision in satisfaction of QoS
properties use the fuzzy set theory to formalize the
model (Xiong and Fan, 2008). In contrast we claim
that our approach is simpler, easier for implementa-
tion and hence faster, when considering dynamic
service evaluation and composition.
3 PRELIMINARIES
In this section we are presenting the background of
our approach – first we define web-service composi-
tions and then explain some basic elements of prob-
ability theory, which we use to formalize the ap-
proach for composite web service evaluation.
3.1 Composite Web Services
Software services represent abstract entities, subject
to reuse and in service-oriented architecture single
services may be organized into a more complicated
workflow that corresponds to a given business
process. Currently the most widespread methods for
design of composite services are choreography and
orchestration. Orchestration considers aggregation of
service operations and actually represents a
workflow, while choreography means aggregation of
services which results in another service. In our
work we make an evaluation of the composition
after it has been implemented, i.e. regardless of the
composition method.
For instance, several services WS
1
, WS
2
, etc.,
may be composed into a composite web-service
CWS
1
as shown on Fig. 1.
As seen from the figure, some services in the
composition may be complex services themselves.
Further in the paper, we denote such web services as
dependent, because their invocation and QoS proper-
ties depend on the results from execution and QoS
properties of the services that they compose. When
ICSOFT 2011 - 6th International Conference on Software and Data Technologies
322
executing composite of the kind shown on fig. 1, we
assume that of sub services WS
3
and WS
4
, are auto-
nomous in their execution, as they do not take input
from another web services in the composition. We
call such web services independent.
Figure 1: Composite web services.
In order to assess service qualities we use a so-
called service history log that contains data, obtained
during service execution, with concrete measures
about quality parameters. History log values of de-
pendent web services are obtained during their
monitoring in testing environment, where they are
isolated from web services on which they depend.
History log values of independent web services are
obtained during monitoring into a production envi-
ronment.
3.2 Elements of Probability Theory
Generally speaking probability theory is a mathe-
matical discipline that deals with analysis of random
events. More precisely in this paper we are interest-
ed on how to estimate the likeliness that a certain
event will occur. In this section we present the some
aspects of the probability theory that are used in next
section to leverage our approach for QoS estimation
of composite web services.
Each QoS property can be represented by a dis-
crete random variable X with several possible values
x that correspond to the recorded values in the histo-
ry log.
A random variable can be described by a proba-
bility mass function (PMF), which captures the
probabilities of the values that the random variable
can take. If x is any possible value of discrete ran-
dom variable X, the PMF of x, denoted p
X
(x) is the
probability of the event {X = x} consisting of all
outcomes that give rise to a value of X equal to x
(Bertsekas and Tsitsiklis, 2000):
=
(
{
=
}
)
.
(1)
For a given value x it is often necessary to compute
the probability that the value of the random variable
X will be at most x. This can be accomplished by a
cumulative distribution function (CDF). The CDF of
a discrete random variable provides the probability
P(X≤x) (Bertsekas and Tsitsiklis, 2000):
(
≤
)
=
()
.
(2)
Probabilistic models usually concern several random
variables. For example, in the context of web servic-
es, the QoS requirements of the web service compo-
sition consist of definitions for several QoS proper-
ties that can be seen as random variables belonging
to the same experiment. The joint CDF of two inde-
pendent random variables X and Y is as follows:
,
(
,
)
=
(
≤)(≤
)
.
(3)
Note that for negative QoS properties probability
that {X ≥ x} may be calculate as P(X ≥ x) = 1 -
P(X≤x).
The conditional PMF of a random variable X that
is conditioned on an event A with probability P(A) is
defined as follows (Bertsekas and Tsitsiklis, 2000):
|
(
)
=
(
=
|
)
=
(
{
=
}
)
∩()
(
)
. (4)
The conditional CDF of two random variables X and
Y is as follows (Bertsekas and Tsitsiklis, 2000):
|
(
|
)
=
(
≤,≤)
( ≤)
=
,
(
,
)
()
.
(5)
The conditional CDF can be used for calculation of
joint CDF as follows (Bertsekas and Tsitsiklis,
2000):
,
(
,
)
=p
(
y
)
p
|
(x|y).
(6)
4 AN APPROACH FOR
EVALUATION OF
COMPOSITE WEB SERVICES
This section presents a QoS driven approach for
probability evaluation of composite web services.
The approach consists of seven steps that are de-
scribed bellow in detail.
Let web services that satisfy functional require-
ments of the composition form the set S:
=
{
,
,…,
,…,
}
,=1÷
(7)
where n is the number of candidate web services.
Step 1: Identification of Independent and Depen-
dent Web Services in the Composition. As men-
tioned in Section 3.1 web services in the composi-
tion can be dependent or independent according to
whether a given web service uses the results from
the execution of another web service in the composi-
tion or not. Therefore, the first step of the algorithm
requires separation of the web services into two
A QOS DRIVEN APPROACH FOR PROBABILITY EVALUATION OF WEB SERVICE COMPOSITIONS
323
different sets – S
’
for independent web services and
S
’’
for dependent web services.
=
,
,…,
,…,
,=1 ÷
(8)
=
,
,…,
,…,
,= ÷
(9)
Step 2: Extraction of QoS Data. In Section 3.2 it
was shown that each QoS property of web services
can be represented with a discrete random variable
with several possible values that correspond to the
recorded values in the history log. Thus, the values
of QoS properties of a particular web service form
the following matrix:
=
⋯
⋯
⋯
⋮⋮⋱⋮
⋯
,=1÷
(10)
where x
ml
is the m-th value of QoS property q
l
ob-
tained from history log and m is the number of val-
ues obtained.
Step 3: Extraction of Events from Composition
Requirements. The QoS requirements of the web
service composition are considered as events. They
form the following matrix:
=
⋯
⋯
⋯
⋮⋮⋱⋮
⋯
(11)
where r
ij
={X ≤ x} for negative QoS properties and
r
ij
={X≥x} for positive QoS properties.
Step 4: Calculation of PMF for each Unique Value
of each QoS Property for all Web Service in the
Composition. The probabilities of the values that
given QoS property takes are calculated according to
equation (1).
Step 5: Calculation of CDF for each QoS Property
of Independent Web Services in the Composition.
The CDFs calculated for each QoS property of all
independent web services form the following matrix:
=
…
′
′
⋮
′
⋯
⋯
⋮⋮⋱⋮
⋯
(12)
CDFs P
ij
=P(r
ij
) for that case are calculated accord-
ing to equation 2.
Step 6: Calculation of Conditional the CDF for
Each QoS Property of Dependent Web Services in
the Composition. The conditional CDFs calculated
for each QoS property of all dependent web services
form the following matrix:
|
÷
"
=
…
"
"
⋮
"
⋯
⋯
⋮⋮⋱⋮
⋯
(13)
where z is the number of web services, from which a
given web service S
i
takes input. Here, the condi-
tional CDFs P
ij
are calculated based on QoS data
obtained during execution of web service in testing
environment, according to equation (5). Note that in
this way we consider that dependent web-services
are executed given that all other services they invoke
have met their quality requirements.
Step 7: Calculation of Joint CDF for QoS Proper-
ties of the Web Service Composition. Finally, we
calculate the probability the QoS properties of the
composition to satisfy preliminary defined require-
ments using equation 6. The result is as follows:
⋯
=
⋯
(14)
Pseudo code of the algorithm that may serve as a
basis for implementation of the proposed approach is
presented on Listing 1.
Listing 1. Pseudo code of the QoS driven algorithm for
probability evaluation of web service composition
FUNCTION Evaluate (log, requirements)
FOR EACH web service in S DO
Extract data from history log
IF web service is independent THEN
Insert web service in the set S
’
ELSE
Insert web service in the set S
’’
END IF
Create matrix Q
ws
END FOR
FOR EACH QoS requirement of the composi-
tion
Extract an event
Insert an event into matrix R
END FOR
FOR EACH web service in S DO
FOR EACH QoS property DO
FOR EACH unique value of QoS prop-
erty DO
Calculate PMF
END FOR
IF web service is independent THEN
Calculate CDF
ELSE
Calculate conditional CDF
END IF
END FOR
END FOR
FOR EACH QoS of the composition
Calculate joint CDF
END FOR
END FUNCTION
5 CASE STUDY
This section shows how to apply our approach with
a case-study that represents a Travel Booking sam-
ICSOFT 2011 - 6th International Conference on Software and Data Technologies
324
ple, taken from (Travel Booking Sample, 2011). A
business process of the sample invokes four web
services as shown in Table 1.
Table 1: Web services of the Travel Booking business
process.
Web service name Symbol Type
checkCreditCard WS
1
Independent
checkFlightReservation WS
2
Dependent
checkHotelReservation WS
3
Dependent
checkCarReservation WS
4
Dependent
The composition diagram, corresponding to the
business process is shown on Fig. 2. The QoS prop-
erties that will be used in the example are Successful
execution rate (q1), Reputation (q2), Availability
(q3) and Response time (q4). Here, the execution
rate is successful when its value is 1.
Figure 2: Travel booking process.
According to Step 1 of the proposed approach,
the web services in the composition form two sets
depending on whether the web service is indepen-
dent or not. These sets are defined as follows:
=
{
}
(15)
=
,
,
(16)
On the Step 2 we have to extract data from history
log in order to define the QoS matrices for each web
service:
=
1 0.90 0.80 5.00
1 0.80 0.95 4.10
1 0.75 0.90 4.50
1 0.80 0.85 4.50
1 0.80 0.85 4.80
1 0.95 0.75 4.00
1 0.80 0.80 4.60
=
1 0.60 0.75 5.00
1 0.80 0.95 5.10
1 0.75 0.91 5.20
1 0.80 0.85 4.80
1 0.80 0.95 5.00
0 0.75 0.92 8.00
1 0.80 0.80 4.60
=
1 0.90 0.80 5.00
1 0.80 0.95 5.00
1 0.85 0.90 5.20
1 0.85 0.85 4.40
1 0.88 0.85 4.50
1 0.95 0.90 4.00
1 0.92 0.90 4.60
1 0.90 0.85 5.00
1 0.80 0.75 5.00
1 0.70 0.70 5.20
1 0.80 0.85 4.50
1 0.80 0.85 4.60
1 0.75 0.80 4.80
0 0.80 0.80 8.60
(17)
On the Step 3 the composition requirements are
extracted. The events that are obtained from the
requirements form the following matrix:
=
{
=1
}
{
≥0.75}
{
≥0.80
}{
≤5.0
}
{
=1
}{
≥0.65
}{
≥0.85
}{
≤6.0
}
{
=1
}{
≥0.85
}{
≥0.85
}{
≤5.0
}
{
=1
}{
≥0.75
}{
≥0.75
}{
≤7.0
}
(18)
According to Step 4, the probabilities of the values
that the QoS property of each web service takes are
calculated. They are used to compute the CDF of
each QoS property of independent web services on
the Step 5 and the conditional CDF of each QoS
property of dependent web services on the Step 6.
The result is as follows:
=
110.8571
(19)
=
0.857 0.857 0.857 0.857
1 0.857 0.857 0.857
0.857 0.857 1 0.857
(20)
Finally, on Step 7 the joint CDF for QoS properties
of the web service composition is calculated.
=
0.735 0.63 0.463 0.63
(21)
Given the relatively small number of input data, the
fact that we get probability values less than 0.80
may be considered as good result. Given the re-
quirements for the composite web-service, the re-
sults may be optimized by selecting a better candi-
date web services.
6 CONCLUSIONS AND FUTURE
WORK
There is a growing need for establishment of sound,
industry-wide techniques and methodologies for
service composition, not only in design time, but
also in run-time. The later compositions are usually
called dynamic web-services. For this purpose an
important assessment criteria for candidate services
is fulfillment of their quality attributes. This paper
proposes an approach for evaluation of composite
web services, based on probability analysis of how
do they meet their QoS requirements. This approach
is applicable for evaluation of any set of QoS by just
calculating the probability if a given QoS would be
satisfied or not, based on service execution history
log.
Another benefit of our approach is that it uses
probabilities for evaluation of QoS properties and
many of them like reliability and availability are
actually measured with the same metrics.
WS
2
WS
3
CO
WS
4
CWS
WS
1
A QOS DRIVEN APPROACH FOR PROBABILITY EVALUATION OF WEB SERVICE COMPOSITIONS
325
Main directions for future research include
enabling a framework for dynamic service composi-
tion, based on proposed approach. As short-term
future work, we plan to implement the proposed
algorithm and experiment with various industrial
examples of web service compositions. Another
direction for further research is to enrich our ap-
proach with ability to reason about timing and se-
quence of service executions.
ACKNOWLEDGEMENTS
The work presented in this paper was partially sup-
ported by grants from the National Science Fund, in
Bulgaria under the MU-01-143 (ADEESS) project
and the SISTER project, funded by the EC in FP7-
SP4 Capacities via agreement no.: 205030.
REFERENCES
Alrifai, M., Risse, T., Dolog, P., Nejdl, W.: A Scalable
Approach for QoS-based Web Service Selection. In
Proceedings of ICSOC Workshops (2008) 190-199.
Bartalos, P., Bielikova, M.: Fast and Scalable Semantic
Web Service Composition Approach Considering
Complex Pre/Postconditions. In: International Work-
shop on Web Service Composition and Adaptation,
(2009) 414–421.
Bertsekas, D., Tsitsiklis J.: Introduction to Probability.
Cambridge, Massachusetts, USA (2000).
Cardellini, V., Casalicchio, E., Grassi, V., Lo Presti, F.:
Flow-Based Service Selection for Web Service Com-
position Supporting Multiple QoS Classes. In: IEEE
International Conference on Web Services (2007) 743-
750.
Cardoso, J., Miller, J., Sheth, A., Arnold, J.: Modeling
Quality of Service for Workflows and Web Service
Processes. Web Semantics Journal: Science, Services
and Agents on the World Wide Web, Vol. 1, Issue 3,
Elsevier Inc, MA, USA (2004)281-308.
Chakhar, S., Youcef, S., Mousseau, V., Mokdad, L., Had-
dad, S.: Multicriteria Evaluation-Based Conceptual
Framework for Composite Web Service Selection.
Available at http://www.lipn.univ-paris13.fr/~youcef/
BookQoS.
Chang H., Lee K.: A Quality-Driven Web Service Com-
position Methodology for Ubiquitous Services. In:
Jornal of Information Science and Engineering 26(6)
(2010) 1957-1971.
Che, H., Li, Y., Oberweis, A., Stucky, W.: Web Service
Composition Based on XML Nets. In: 42st Hawaii In-
ternational Conference on Systems (2009) 1-10.
D’Mello, D. et al.: A QoS Broker Based Architecture for
Dynamic Web Service Selection. Proceeding of the
Second Asia International Conference on Modelling &
Simulation. Kuala Lumpur, Malaysia, (2008) 101-106.
Farhan, H. K., Javed, M. Y., Saba, B. and Sikandar, H.
K.,: QoS Based Dynamic Web Services Composition
& Execution. In: International Journal of Computer
Science and Information Security, Vol. 7, Issue 2,
USA (2010) 147-152
Kim, E., Lee, Y.: OASIS: Quality Model for Web Servic-
es. 2005, http://www.oasis-open.org/committees/
download.php/15910/WSQM-ver-2.0.doc.
Lin, N., Kuter, U., Sirin, E.: Web service composition
with user preferences. In: 5
th
European semantic web
conference on the semantic web: research and appli-
cations, Lecture Notes in Computer Science, Vol.
5021, Spain (2008) 629-643.
May Chan K. S., Bishop J.: A Self-healing Cycle for Web
Service Composition. In: 17
th
International World
Wide Web Conference (2008).
Ming, Q., Khendek, F., Serhani, A., Dsouli, R., Roch, G.:
An Architecture for Automatic QoS Adaptation for
Composite Web Services. In: International Journal of
Web Services Practices, Vol. 4, No.1 (2009) 18-27.
Nam, W., Kil, H., Lee, J.: QoS-Driven Web Service Com-
position Using Learning-Based Depth First Search. In:
IEEE Conference on Commerce and Enterprise Com-
puting (2009) 507-510.
Pistore, M., Barbon, F., Bertoli, P., Shaparau, D., Traver-
so, P.: Planning and Monitoring Web Service Compo-
sition. Lecture Notes in Computer Science Vol. 3192
(2004), 106-115
Petrova-Antonova D.: А QoS Aware Approach for Web
Service Selection Based on Probability Evaluation.
IADIS International Conference Informatics 2011,
Rome, Italy, 20 - 22 July (2011) (to appear).
Tran, V. et al.: A new QoS ontology and its QoS-based
ranking algorithm for Web services. In Simulation
Modelling Practice and Theory. Vol. 17, No. 8, (2009)
1378-1398.
Zheng, X., Yan, Y.: An Efficient Syntactic Web Service
Composition Algorithm Based on the Planning Graph
Model. In: 8
th
IEEE International Conference on Web
Services (2008) 691-699.
Xiong P., Fan, Y.: QoS-aware Web Service Selection by a
Synthetic Weight. In: IEEE Transactions on Systems,
Man, and Cybernetics, Part A, Vol. 38, Issue 4, (2008)
888-895.
Zeng, L., Benatallah, B., Ngu, A., Dumas, M., Kalagna-
nam, J. T., Chang, H.: In: QoS-aware Middleware for
Web Services Composition. IEEE Transactions on
Software Engineering, Vol. 30 Issue 5 (2004) 311-
327.
Travel Booking Sample, (IBM WebSphere), last visited 05
April 2011. http://publib.boulder.ibm.com/infocenter/
adiehelp/v5r1m1/index.jsp?topic=/com.ibm.etools.ctc.
bpel.doc/samples/travelbooking/travelBooking.html.
ICSOFT 2011 - 6th International Conference on Software and Data Technologies
326
