OPTIMIZING QOS-BASED SERVICE SELECTION

IN SERVICE-ORIENTED ARCHITECTURES

Dieter Schuller

Multimedia Communications Lab (KOM), Technische Universität Darmstadt,

Rundeturmstr. 10, 64283 Darmstadt, Germany

Keywords: Service-Selection-Problem, Quality of Service, Optimization, Linear Programming, SOA-Security.

Abstract: In Service-oriented Architectures, services can be composed in a loosely coupled manner to realize business

processes. Thereby, mentioned services are not necessarily located only within the borders of the own en-

terprise. In the Internet of Services, multiple service providers offer various services on several service mar-

ketplaces. In case services with comparable functionalities but varying quality levels are available at differ-

ent costs, service requesters can decide, which services from which service providers to select. My research

focuses on this service-selection-problem for complex workflows by formulating a linear optimization prob-

lem, which can be solved optimally using linear programming techniques. As the actual execution path is

probably not known at planning time (e.g., for conditional braches), a worst-case, and an average-case anal-

ysis is performed. In addition to considering non-functional, quantitative service properties, I am working in

my research towards integrating qualitative service features (as, e.g., security) for different, complex

workflow patterns.

1 RESEARCH PROBLEM

In present times, enterprises in highly competitive

markets require to execute their business processes

and workflows (as IT supported business processes)

efficiently. Due to globalization and deregulation of

markets, the business processes have to be flexible

in addition. This is necessary in order to enable en-

terprises to react quickly to market driven environ-

mental changes by adapting their business processes.

As enterprises often have several legacy systems and

a multitude of applications, running on different

platforms and operating systems, implemented with

various programming languages, the required flex-

ibility is hard to achieve. So, due to this heterogenei-

ty of enterprises’ IT architectures, an approach for

integrating the applications and legacy systems of

the enterprises is necessary to realize the mandatory

flexibility and efficiency regarding the enterprises’

business processes.

Considering Service-oriented Architectures

(SOA), these business processes take center stage.

They are realized by composing autonomous, loose-

ly coupled services. Depending on their granularity,

they can provide a more or less complex functionali-

ty (cf. Krafzig, Banke, Slama, 2004). Thereby, the

enterprises do not need to implement all required

services on their own. It is possible to rely on the

expertise of specialists by invoking services from

dedicated, external service providers, enabling en-

terprises to concentrate on their core business com-

petencies. In fact, cross-organizational business

processes gain more and more importance (Ley-

mann, Roller, 2000). In the Internet of Services, var-

ious services are offered by different service provid-

ers on service marketplaces. In case that functionally

equivalent services with varying Quality of Service

(QoS) at different cost levels are available, enter-

prises (as service requesters) can decide, which ser-

vices from which service providers to select. A cer-

tain degree of flexibility and efficiency can be

achieved by selecting those (external) services,

which meet the enterprise’s requirements best with

respect to functional and (especially) non-functional

service properties. The service selection decision

based on QoS properties is addressed in the service-

selection-problem.

The service-selection-problem is well known in

the literature and discussed by several authors, re-

cently (Yu, Lin, 2005; Anselmi, Ardagna, Cremone-

si, 2007; Huang, Lan, Yang, 2009). In this problem,

a couple of services (WS – Web service) has to be

532

Schuller D. (2010).

OPTIMIZING QOS-BASED SERVICE SELECTION IN SERVICE-ORIENTED ARCHITECTURES.

In Proceedings of the 5th International Conference on Software and Data Technologies, pages 532-538

DOI: 10.5220/0003047905320538

 SciTePress

selected based on their QoS values to realize a giv-

en, requested workflow respectively process (PS –

Process steps). It is assumed that a set of services,

which are able to realize the given workflow with

respect to functional requirements, is already identi-

fied. This situation is depicted exemplarily for a se-

quential workflow in Figure 1.

WS 1,m

WS 1,3

WS 1,2

WS 1,1

WS 2,m

WS 2,3

WS 2,2

WS 2,1

WS 3,m

WS 3,3

WS 3,2

WS 3,1

WS n,m

WS n,3

WS n,2

WS n,1

PS1 PS2 PS3 PSn

Figure 1: Service-Selection-Problem.

In my research, the service-selection-problem for

complex, cross-organizational workflows is focused.

For this, a linear optimization problem considering

different workflow patterns is formulated, which can

be solved optimally using mixed integer linear pro-

gramming (MILP) techniques from the field of oper-

ations research (cf. Domschke, Drexl, 2007), if such

a solution exists.

In addition, as the invoked services are probably

offered by external service providers, the exchanged

(SOAP-) messages (for the communication between

the external service provider and the requester) leave

the borders of the own enterprise. These messages

could be intercepted and manipulated by an attacker

(cf. Miede, Nedyalkov, Schuller, Repp, Steinmetz,

2009), potentially leading to failures with respect to

the execution of the business processes

(cf. Figure 2).

Therefore, security requires to be considered dur-

ing service selection in order to avoid or at least to

recognize service manipulations before the probably

modified messages (the results of the service execu-

tion) are used and applied in the enterprise’s process.

Figure 2: Message Interception.

2 OUTLINE OF OBJECTIVES

My research aims at the development of a smart

workflow broker, which accepts (complex)

workflow requests in service-based environments,

computes an optimal execution plan, performs its

execution, and delivers the results back to the initial

requester. Thereby, the service selection decision is

focused. The question is which services – realizing

the requested functionality – from which service

providers to select in order to minimize the total

costs for the invoked services and to satisfy prede-

fined constraints with respect to QoS (specified by

the requester). Further, how can the execution of an

optimal execution plan be recovered in case of ser-

vice failures? Therefore, my research concentrates

on the following topics:

 Aggregation functions aggregating the non-

functional service properties for (combina-

tions of) complex workflow patterns.

 Linear optimization problem formulation with

respect to non-functional service properties for

complex workflows, which can be solved by

applying MILP, considering a worst-case and

an average-case analysis.

 Heuristic algorithms in order to compute almost

optimal solutions with acceptable computa-

tional efforts, if the number of process steps in

the workflow is increasing.

 Quantification of qualitative service properties

as, e.g., security in order to be able to consider

such properties as quantitative QoS parame-

ters for the optimization.

 Re-planning strategies at design and execution

time to (rapidly) react to positive and negative

deviations of actual QoS values from the ex-

pected ones.

Thereby, a scenario of distributed workflows with a

large amount of service providers and service /

workflow requesters is considered. My research fo-

cuses on the formulation of a linear optimization

problem to compute the optimal solution to the ser-

vice-selection-problem as well as on the develop-

Service Requester

Service Provider A

Service Provider B

Service Marketplace

OPTIMIZING QOS-BASED SERVICE SELECTION IN SERVICE-ORIENTED ARCHITECTURES

533

ment of (efficient) algorithms providing nearly op-

timal solutions. Here, not only sequential but com-

plex workflow patterns are regarded.

3 EXPECTED OUTCOME

In order to develop the mentioned workflow broker

accepting workflow requests and computing –

dependent on the workflow length (number of

process steps) – (nearly) optimal solutions, the fol-

lowing contributions are planed or already realized.

To start with, an aggregation model for the ag-

gregation of different (aggregation) types of QoS

parameters addressing the service-selection-problem

for several complex workflow patterns was devel-

oped (an extension for further complex workflow

patterns is currently ongoing). This model is trans-

formed in a linear optimization problem. Applying

MILP techniques, an optimal solution can be com-

puted, if such a solution exists. The respective im-

plementation is currently ongoing. As the considered

complex workflow patterns include conditional

branches, the ultimately executed path (through the

workflow) is not known at planning time. So, it has

to be decided, which of the paths to consider for the

optimization. If all possible paths require satisfying

the predefined constraints, a worst-case analysis is

performed. An average-case analysis leads to an

execution plan, where the constraints are satisfied in

average. In a best-case analysis, only one of the po-

tential paths would have to satisfy the restriction for

a certain QoS parameter, which is not of practical

interest and will therefore not be considered further.

As the calculation of the optimal solution re-

quires strong computational efforts, heuristic algo-

rithms are currently being developed and imple-

mented as well.

Aside from these contributions regarding the op-

timization techniques, a first approach for the quan-

tification of qualitative service properties (as, e.g.,

for security) has been developed. The aim is to be

able to consider also qualitative service properties

for the service selection by integrating the quantified

qualitative values as additional quantitative QoS

parameters into the optimization problem. Regarding

the projected smart workflow broker, security as

qualitative service property is focused. Referring to

Figure 3, the Security Module makes sure that (po-

tentially) requested security requirements (e.g., inte-

grity and confidentiality) are satisfied.

To increase the robustness of the broker towards

service failures, a re-planning mechanism has been

developed and implemented prototypically. Here, a

service monitor triggers the interruption of the cur-

rently executed service and its replacement – by an

alternative (functionally equivalent) service based on

semantic annotations – if a negative deviation of the

measured form the expected (in Service-Level-

Agreements (SLAs) contracted) execution time is

detected. This is also done, if problems are encoun-

tered with the initial service invocation. The re-

planning mechanism – working on execution time as

QoS parameter – is planed to be extended for further

QoS parameters, for additionally recognizing posi-

tive SLA deviations, and for operating supplementa-

ry at design time. Here, experienced data about the

QoS values of particular services come into play by

(rather) considering these values for the optimization

instead the ones committed in SLAs.

As already mentioned, the re-planning mechan-

ism is based on semantic annotations. This leads to

the presupposition respectively requirement that the

service descriptions considered in the smart

workflow broker scenario are enriched with seman-

tic information about the services’ functionalities.

Here, an extension for the prototypically imple-

mented semantic matchmaker is planned.

An architectural blueprint of the targeted smart

workflow broker is depicted in Figure 3.

Internet

Figure 3: Smart Workflow Broker.

4 METHODOLOGY

For the development of the smart workflow broker,

more than one research methodologies are applied.

An overview of the contribution and the research

methodology used is given in the following list:

 Conceptual work to analyze the research prob-

lem and to design the architecture

 Measurements of Web services in a test-bed to

record actual QoS behavior for re-planning

mechanisms

ICSOFT 2010 - 5th International Conference on Software and Data Technologies

534

 Analytical evaluation and optimization of

worst-case and average-case service selection

as well as analytical models for QoS aggrega-

tion

 Simulation to evaluate worst-case and average-

case optimization and the re-planning mechan-

ism

 Prototypical implementation of the architecture

and the re-planning mechanism

Combining these methodologies allows treating the

addressed research questions in an integrated man-

ner.

5 STAGE OF THE RESEARCH

In this section, the contributions of my hitherto re-

search are presented in summary. Here, the service-

selection-problem is addressed performing an aver-

age-case and a worst-case analysis. In addition, ini-

tial concepts have been developed to be able to inte-

grate security and other qualitative QoS parameters

into the optimization, and a basic re-planning me-

chanism has been implemented for sequential

workflow patterns.

5.1 QoS Aggregation

In order to solve the service-selection-problem, it is

necessary to be able to aggregate the QoS values of

the eligible services, i.e. the services capable of rea-

lizing the process steps of the considered business

process respectively workflow. Otherwise, it is not

possible to compare different service selections for

the workflow, which is necessary to compute the

optimal solution.

How the QoS values have to be aggregated de-

pends on the considered workflow pattern and on the

aggregation type of the regarded QoS parameter.

E.g., the costs for the whole workflow are calculated

by summing up the costs of the selected services,

whereas the workflow reliability is determined by

multiplying the reliabilities of these services. In my

research, the aggregation types summation, multipli-

cation and the Min/Max-operator are considered

covering quite all kinds of QoS parameters. In

(Schuller, Papageorgiou, Schulte, Eckert, Repp,

Steinmetz, 2009), generic aggregation functions for

these aggregation types have been developed consi-

dering the workflow pattern sequence. In (Schuller,

Eckert, Miede, Schulte, Steinmetz, 2010), aggrega-

tion functions for the workflow patterns parallel split

(AND-split), synchronization (AND-join), exclusive

choice (XOR-split), simple merge (XOR-join), and

arbitrary cycles (loops) are shown for a worst-case

and an average-case analysis.

A specification of further – in my research (re-

garding the QoS aggregation) yet unconsidered –

workflow patterns is available in (Aalst, Hofstede,

Kiepuszewski, Barros, 2003).

5.2 Optimization

Based on the abovementioned aggregation functions,

a linear optimization problem is formulated in

(Schuller et al., 2009) for sequential workflow pat-

terns. In (Schuller et al., 2010), combinations of the

workflow patterns parallel split, synchronization,

exclusive choice, simple merge, and arbitrary cycles

are considered for this purpose. In Figure 4, a (com-

plex) example workflow for a potential combination

of these patterns is shown (in Business Process

Modeling Notation).

Figure 4: Example Workflow.

The challenge for the formulation of the linear opti-

mization problem lies in the combination of the ag-

gregation functions for the regarded complex

workflows and in the linearization of the partly non-

linear aggregation functions with respect to the deci-

sion variables. As, for instance, the reliability of the

eligible services has to be multiplied, the respective

decision variables – stating whether the currently

considered service is selected or not – are multiplied,

too. Having such conditions in the problem formula-

tion, MILP techniques to compute an optimal solu-

tion to the service-selection-problem are no longer

applicable. Therefore, a linearization step for the

mentioned non-linear aggregation functions is man-

datory in order to generate linear constraints. For

this purpose, an approximation, proposed by

(Heckmann, 2008), is applied to the multiplicative

non-linear functions. Conditions with the Min/Max-

operator are transformed into a set of equivalent

linear constraints. As here, the amount of added

conditions grows exponentially with the number of

process steps within a conditional branch, increasing

the restriction strength is proposed in (Schuller et al.,

2010).

OPTIMIZING QOS-BASED SERVICE SELECTION IN SERVICE-ORIENTED ARCHITECTURES

535

Having computed the optimal execution plan,

aggregating the QoS values of the selected services

to calculate the workflow QoS does not lead to ex-

actly one value per QoS parameter, but to a set of

values, depending on the executed path. So, due to

multiple execution paths, (only) a range of values

representing the workflow QoS can be calculated.

5.3 Quantification of Qualitative

Service Properties

As the exchanged messages (between service pro-

vider and requester) could be intercepted and mani-

pulated by an attacker (cf. Figure 2) if services from

external services are invoked, security mechanisms

as, e.g., encryption, digital signatures and checksums

require to be considered. To integrate these qualita-

tive service features into the optimization, a quanti-

fication of these properties is necessary. Therefore,

the security features have to be ascertained and con-

sistently assigned to numeric values, to consider

them as quantitative QoS parameters for the optimi-

zation. By the term “consistently” it is meant that the

same value is always assigned to a certain ascer-

tained property. In this context, Jaquith defines a

good metric in (Jaquith, 2007) as a “consistent stan-

dard for the measurement”. It should be possible to

express the metric as a numeric value rather than

having qualitative labels as “high”, “medium” or

“low” (Jaquith, 2007).

A first approach to a consistent concept for such

a quantification of qualitative QoS parameters –

which is at a very early stage – is presented in

(Schuller, 2010).

5.4 Re-planning

When invoking one of the services in the execution

plan, a service monitor is started, too. As imple-

mented so far, the mentioned monitor measures the

time the invoked service takes to provide its functio-

nality. If the measured time exceeds the – in SLAs –

contracted execution time (negative SLA deviation),

the execution of this service is stopped and a search

for alternative services (with respect to functionali-

ty) is started. Referring to Figure 3, the service di-

rectory is queried to provide alternative services

based on semantic annotations. For this purpose, the

semantic matchmaker is used. With this set of alter-

native services, a new execution plan is compiled by

solving the service-selection-problem for the re-

maining (yet unexecuted) process steps. In other

words, the initial execution plan is re-planned – a

dynamic re-selection of services is performed – tak-

ing the already executed services respectively their

QoS values into consideration for the adaptation of

the restrictions of the (new) optimization problem.

The execution of the initial workflow then continues

by invoking the (re-planned) services. This re-

planning mechanism is presented for sequential

workflow patterns in (Schuller et al., 2009).

In case, the described mechanism is repeated

several times during workflow execution, applying

heuristics is probably advisable due to the mentioned

scalability issues when computing an optimal solu-

tion.

6 STATE OF THE ART

The main topic of my research, the service-

selection-problem, is widely studied in the literature.

Heuristic solutions to this problem are proposed and

implemented by several authors (Anselmi et al.,

2007; Jaeger, Muehl, Golze, 2005; Mabrouk, Geor-

gantas, Issarny, 2009). Tree-based algorithms, where

the complex workflow is transformed into a tree-

based execution structure (BPTree) based on the

alignment of the considered workflow patterns are

proposed in (Menascé, Casalicchio, Dubey, 2008).

Unfolding arbitrary cycles and computing optimal

execution plans for each possible path in the com-

plex workflow is proposed in (Ardagna, Pernici,

2007; Zeng, Benatallah, Ngu, Dumas, Kalagnamam,

Chang, 2004). The authors solve the service-

selection-problem for complex workflows by formu-

lating and solving the linear optimization problem

using MILP techniques solely for (the unfolded)

sequential workflows. So, all possible execution

paths have to be known and considered consecutive-

ly for the computation of an optimal solution.

These shortcomings are overcome in my re-

search, where arbitrary and recursive combinations

of (the in section 5.1 mentioned) workflow patterns

can be considered for the optimization without a

preliminary step for identifying all possible execu-

tion paths. The multiple execution paths are directly

taken into account when formulating the optimiza-

tion problem. In other words, an optimal solution to

the service-selection-problem is computed consider-

ing all execution paths conjunctly. So the execution

paths are not considered independently from each

other. This is an improvement regarding the men-

tioned approaches (in the literature) as a consecutive

computation for all paths considers the paths inde-

pendently from each other which probably leads to a

case where different services are assigned to the

ICSOFT 2010 - 5th International Conference on Software and Data Technologies

536

same process step when computing solutions for

different execution paths.

7 CONCLUSION AND OUTLOOK

In highly competitive markets, flexible and efficient

business process execution is indispensable. Follow-

ing the paradigm of SOA, using services from (ex-

ternal) service providers (probably having large ex-

pertise in a certain business field), enables a flexible

and efficient process execution. In case, multiple

service providers offer (quite) equivalent services at

different quality and cost levels on various service

marketplaces (as is probably the case in the Internet

of Services), enterprises have the opportunity to de-

cide which services from which service providers to

select accomplishing the required functionalities

based on their QoS preferences, increasing the

achieved efficiency regarding the business process

execution.

Therefore, the service-selection-problem for

complex workflows is addressed in my research. In

this context, I contributed or contribute the follow-

ing:

 An approach for aggregating QoS values for

(combinations of) complex workflow patterns.

 An approach for solving the service-selection-

problem optimally (if an optimal solution ex-

ists) without preliminary identifying all execu-

tion paths.

 A heuristic algorithm for a nearly optimal solu-

tion of the service-selection-problem due to

scalability issues considering the optimal solu-

tion.

 A concept for a quantification of qualitative

service properties in order to integrate security

into the service selection decision.

 A re-planning strategy to react upon positive

and negative SLA deviations for complex

workflows.

In particular, the integration of further complex

workflow patterns as multi-choice (OR-split) and

multi-merge (OR-join) into the optimization prob-

lem – yet insufficiently considered (even in the men-

tioned heuristic approaches) in the literature – and

the development of a reasonable concept for quanti-

fication of security is currently investigated. The

extension of the re-planning mechanism for complex

workflows in combination with the progression of

the semantic matchmaker (and the therefore required

semantic annotations) is addressed as well in my

future research.

ACKNOWLEDGEMENTS

This work is supported in part by the E-Finance Lab

e. V., Frankfurt am Main, Germany

(http://www.efinancelab.com).

REFERENCES

Van der Aalst, W., Ter Hofstede, A., Kiepuszewski, B.,

Barros, A., 2003. Workflow Patterns, Distributed Pa-

rallel Databases, 14, 1, pp. 5-51.

Anselmi, J., Ardagna, D., Cremonesi, P., 2007. A QoS-

based Selection Approach of Autonomic Grid Servic-

es. In International Conference on Service-oriented

Computing. Vienna, Austria.

Ardagna, D., Pernici, B., 2007. Adaptive Service Composi-

tion in Flexible Processes, Transactions on Software

Engineering, 33, 6, pp. 369-384.

Domschke, W., Drexl, A., 2007. Einführung in Operations

Research, Springer Verlag. Heidelberg.

Heckmann, O., 2004. A System-oriented Approach to Effi-

ciency and Quality of Service for Internet Service Pro-

viders, PhD thesis. TU Darmstadt, Fachbereich Infor-

matik.

Huang, A., Lan, C.-W., Yang, S., 2009. An optimal QoS-

based Web service selection scheme, Information Sci-

ences, 179, pp. 3309-3322.

Jaeger, M., Rojec-Goldmann, G., 2005. SENECA-

Simulation of Algorithms for Selection of Web Ser-

vices for Composition. In Technologies for E-Services.

Trondheim, Norway.

Jaquith, A., 2007. Security Metrics: Replacing Fear, Un-

certainty, and Doubt, Addison-Wesley Professional.

Krafzig, D., Banke, K., Slama, D., 2004. Enterprise SOA:

Service-Oriented Architecture Best Practices, Prentice

Hall PTR. New Jersey, USA.

Leymann, F., Roller, D., 2000. Production Workflow:

Concepts and Techniques, Prentice Hall PTR. New

Jersey, USA.

Mabrouk, N., Georgantas, N., Issarny, V., 2009. A Seman-

tic End-to-End QoS Model for Dynamic Service-

oriented Environments. In Principles of Engineering

Service Oriented Systems. Vancouver, Canada.

Menascé, D., Casalicchio, E., Dubey, V., 2008. A Heuris-

tic Approach to optimal Service Selection in Service-

oriented Architectures. In Workshop on Software and

Performance. New Jersey, USA.

Miede, A., Nedyalkov, N., Schuller, D., Repp, N.,

Steinmetz, R., 2009. Cross-organizational Security –

The Service-oriented Difference. In International Con-

ference on Service-Oriented Computing (ICSOC 2009

Workshops). Stockholm, Sweden.

Schuller, D., Papageorgiou, A., Schulte, S., Eckert, J.,

Repp, N., Steinmetz, R., 2009. Process Reliability in

Service-oriented Architectures. In Digital Ecosystems

and Technologies. Istanbul, Turkey.

Schuller, D., 2010. Sichere und Zuverlässige

Prozessausführung in Serviceorientierten

OPTIMIZING QOS-BASED SERVICE SELECTION IN SERVICE-ORIENTED ARCHITECTURES

537

Architekturen. In Central-European Workshop on

Services and their Composition. Berlin, Germany.

Schuller, D., Eckert, J., Miede, A., Schulte, S., Steinmetz,

R., 2010. QoS-Aware Service Composition for Com-

plex Workflows. Forthcoming in International Con-

ference on Internet and Web Applications and Ser-

vices. Barcelona, Spain.

Yu, T., Lin, K.-J., 2005. Service Selection Algorithms for

Composing Complex Services with Multiple QoS

Constraints. In International Conference on Service-

oriented Computing. Amsterdam, Netherlands.

Zeng, L., Benatallah, B., Ngu, A., Dumas, M.,

Kalagnanam, J., Chang, H., 2004. QoS-Aware Mid-

dleware for Web Services Composition, Transactions

on Software Engineering, 30, 5, pp. 311-327.

ICSOFT 2010 - 5th International Conference on Software and Data Technologies

538