SOA based Adaptive Reliable Service Delivery Framework using QoS: A Fuzzy-Bayesian Network Approach by ijcsis


More Info
									                                                           (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                             Vol. 8, No. 5, August 2010

       SOA based Adaptive Reliable Service Delivery Framework
          using QoS: A Fuzzy-Bayesian Network Approach

                         R.Sivaraman                                              Dr.RM.Chandrasekeran
                        Dy.Director-CCT                                                Registrar
                  Anna University Tiruchirappalli                              Anna University Tiruchirappalli
                      Email:                                   Email:

Abstract                                                              associated with that particular category and the run-time
Service Oriented Architecture has gained a universal                  evaluation of QoS metrics is not available To bridge the gap
acceptance as a strategy for developing new applications in           between the Web Services layer and the underlying QoS
dynamic environments through self contained, reusable and             aware transport methodologies, a suitable framework may
configurable services. By adherence to the web services               be developed which enables the efficient service selection.
standards stack, these services can interoperate with other
services and can also be transformed into a composition of                 This framework should allow both the clients and
services. The behavior and operation of a service has to be           service providers to specify their requests and service offers
closely monitored for an efficient delivery to the client.            with QoS properties. It should also map the QoS
Ranking and selection of the services among various service           requirements from the Web service and application layer
providers have become an important factor for a successful            onto the underlying networking layer thereby achieving an
business solution. Quality of Service (QoS) determines the            overall boost in the QoS across different layers as that of the
quality and usability of a service and is determined after            Internet model. For this a well defined set of QoS indices
analyzing the QoS attributes collected from various sources.          are used which includes non-runtime data like cost and run-
Hence it has become our research motivation for transfer              time data as availability, execution time, mean response
and reception of QoS data. In this paper, we propose an               time, trustworthy, performance, capacity. The cost is a
adaptive and reliable service delivery framework, which               measure of the price involved in requesting the service,
adopts a Fuzzy-Bayesian Network (FBN) approach through                which is based on volume of data and execution time is the
which the web services can be ranked and selected                     guaranteed max (or average or min) time taken between the
automatically. Also this framework offers reliability and             arrival of service request and catering of the request. The
availability aspects of QoS demand.                                   capacity represents the limit of concurrent requests for
                                                                      guaranteed performance. For some of the QoS metrics it is
 Keywords : Web service selection, QoS, Fuzzy-Bayesian                not dependable to rely on the service provider’s
        Network (FBN), Web Services ranking                           advertisement as it may have a biased version favoring their
                                                                      own service. So the QoS data has to be adjudged in a neutral
1. Introduction                                                       way and this leads to the development of an independent
                                                                      framework which not only relies on service provider but
     The QoS specification in Web Services has become the             also considers other indices.
need of the day. The service consumers aim to have a good
service performance with relatively low waiting time, high                 The rest of the paper is organized as follows, the section
availability and reliability to successfully use the service.         2 discusses the survey of literature, section 3 describes the
Also the service providers are formulating their efforts to           design and architecture of the proposed framework and
achieve high throughput guarantees with low response time             section 4 deals with the Fuzzy-Bayesian Network. In section
using dynamic capacity allocation, resource allocation and            5 we provide the implementation and experimental results.
load balancing in order to cater to a large volume of clients         Section 6 discusses the possible future work and it
with assured QoS. In the present architecture each of the             concludes.
input requests are manually mapped to a pre-specified
category of the provider at the registry. The service
consumer has little knowledge about the QoS attributes

                                                                                                ISSN 1947-5500
                                                               (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                 Vol. 8, No. 5, August 2010
2. Related Work

     With the widespread use of Web Services the QoS have                 service composition as a fuzzy constraint satisfaction
became the significant factor in selecting a best service                 problem. Each QoS criteria has five fuzzy sets describing its
among the providers and it constitutes the differentiating                constraint levels: Poorly Acceptable (PA), Almost
point for a group of services. QoS generally refers to                    Acceptable (AA), Acceptable (A), Very Acceptable (VA),
nonfunctional properties of Web services such as price,                   and Extremely Acceptable (EA). The overall QoS
performance, reliability, integrity, availability, accessibility,         preference of a user is then represented by a fuzzy
interoperability, security, etc. [1]. Certain traditional                 expression which is composed of a group of fuzzy sets
approaches [2] make use of QoS policies and a weighing                    connected by the and logical operator. They have introduced
system to judge through the attributes. This approach is                  necessary definitions for mapping a Web service dynamic
applicable only when there is a relatively small set of                   composition to a fuzzy constraint satisfaction problem
services and attributes. Since all the services cannot be                 (FCSP) and finding its solution through NP-hard [12].
weighed through a common attribute thereby limiting this                  Xiong [13] modeled QoS based service selection as a
approach to a smaller set of services. Also rule based model              fuzzy multiple criteria decision making problem (FMCDM).
proposed by Patel [3] is too restrictive and provide a              QoS criteria are evaluated by a set of linguistic expressions
condensed set of alternatives for service execution. Menasce              L1 = {Very Poor (VP), Medium Poor (MP), Poor (P),
[4] in his research has highlighted the need for QoS                      Medium (M), Good (G), Medium Good (MG), Very Good
definition, specification and evaluation in WS from the                   (VG)}. Similarly, the weights of QoS criteria are expressed
perspective of both service provider and service user. W3C                by a set of linguistic expression L2 = {Very Low (VL), Low
[5] has summarized the key requirements of QoS for Web                    (L), Medium (M), High (H), Very High (VH)}. These
services. Recently, several research works have dealt with                linguistic expressions are mapped to fuzzy set membership
the definition of QoS languages for WS-based applications.                functions to yield the results. Agarwal [14] suggested
Ludwig [6] in his work have discussed about the QoS                       that user preferences can be defined as a set of fuzzy If-
languages and the service level agreements(SLA) that are to               Then rules. The If part contains membership functions of
be initiated between the service consumer and service                     various QoS criteria of a Web service and the Then part is
providers for proper operation of the system after satisfying             one of the membership functions of a special concept
the QoS requirements. Hewlett-Packard (HP) has proposed a                 representing the rank of the Web service. In each rule, fuzzy
Web Services Management Language (WSML) and                               concepts like fast, medium, slow or cheap, medium,
framework and IBM have developed a Web Service Level                      expensive are used to express imprecise values of QoS
Agreement (WSLA) language.                                                properties and these fuzzy concepts are modeled as
                                                                          membership functions.

     In a SOA environment the QoS data of the service                     3. Proposed Framework
providers continuously change during the lifetime of a
service. There is no mechanism that periodically updates the                   In this paper we propose to create a reliable
QoS data into the registries. Al-Masri et al.[7] have stated              architectural framework that can manage faults and can
that about 53% URL’s of the Universal Business                            adapt itself to provide a reliable service. This framework
Registries(UBR) are invalid and they cannot be validated                  decouples the clients from the service providers and takes
with a valid WSDL document. This means that the metadata                  care of the service execution after due consideration of all
information collected at the time of registration of a service            QOS issues. Presently the web service clients are typically
in the registry, has been modified and not yet been reported              hardwired with the service providers and does not adapt
to the registry for proper updation. Since the information                dynamically in case of service failures. Unavailability of a
contained in the UBR are not accurate, it utilized more                   service is handled only with a manual intervention during
resources when performing the binding of web services and                 failures and fixing them accordingly. This involves service
wasted a considerable time in trying to communicate with                  stoppage which directly hinders the performance of the
either non functional or poorly functioning web services.                 system. The proposed framework addresses this issue by
The clients should know the revised QoS information to                    acting as a layer in between the web service clients and the
choose the appropriate service accordingly.                               web service providers and dynamically reconfigures itself to
                                                                          respond to the faults so that the business processes can
          Various query based and monitoring based                        continue in the event of service failures. By virtue of using
methods exist [8-10] for obtaining the QoS values of the                  SOA and XML based web services, this framework is
services. The former periodically queries the service and                 multiplatform enabled along with multi protocol support
actively request for the QoS information whereas the later                that can bridge various interfaces, thereby removing the
utilizes a monitoring engine that runs on the service                     difficulty of integrating mobile and wireless sensor network
provider side as a middleware or acts as a central proxy. Lin             clients with this framework. The clients submit their[11] have formalized the service selection for a Web                requests to the framework and during runtime the

                                                                                                    ISSN 1947-5500
                                                                   (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                     Vol. 8, No. 5, August 2010
framework will select on behalf of the clients the most                       regular Internet link. In case of failure of any of the link the
appropriate service available or its equivalent compatible                    other takes control over thereby offering the client a reliable
service depending on the QOS parameters. As shown in Fig.                     connection path. The route redirector acts as a cache and
1, the modules of the proposed framework include a Service                    redirection router in case of any link failure to re-route the
Controller (SC),                                                              reply through any of the available clients. Since all the
                                                                              clients are interlinked together, the reply for Client-2 can be
                                                                              redirected and pushed through Client-4 in case of failure of
                                                                              connectivity between Client-2 and the server.

                                         Service                        Service               Service                Service
                    Client -1                                        Processor (SP)        Dispatcher(SD)           Provider -1
                                      Controller (SC)
                                                                                               FBN Agent
                 Client -2             Connection Manager                                                             Provider -2
                                                                                            QoS / QoE Monitor             Service
              Client -3                Admission Authority             Service                                           Provider -3
                                                                     Registry (SR)                                           Service
            Client -n                   Route Redirector
                                                                                                                           Provider -4
                                                                                           QoS          QoE

                                                             Fig.1. Proposed Framework

     Service Processor (SP), Service Registry (SR) and a                      The service dispatcher (SD) is responsible for the transfer of
Service dispatcher (SD) which holds a QoS and QoE                             request from the service consumer to the service provider. It
repository. All the requests from the clients are routed                      makes use of the Fuzzy-Bayesian Network algorithm for
through this framework to the services located in the                         efficient judging of the QoS and QoE values. These values
servers. The clients search for the available services through                are collected from the repositories and continuously
this framework, get the WSDL and start communicating                          monitored for the dynamic changes. The services are ranked
with the desired service. The role of SP is for registering and               accordingly and the service requests from the clients will be
configuring all the available and compatible services from                    directed to the available provider for service invocation.
various service providers. Contractual relationships are
established with the service providers and after due testing,                 4. Fuzzy-Bayesian Network Approach
the services are registered and configured into this
framework. All the available services are stored in SR that                        Fuzzy sets were derived from the concepts proposed by
acts as a private UDDI registry. The SR also holds                            L.A. Zadeh. Bayesian network is a model of uncertain
additional information like the number of retries in case of a                knowledge representation and reasoning based on the
service access failure and the list of equivalent services                    probability and graph theory and was proposed by J. Pearl
available with other service providers.                                       [15].Fuzzy Bayesian networks (FBN) represent a machine
                                                                              learning model comprising of variables which are
     The SC acts as an admission authority that blocks the                    simultaneously fuzzy and uncertain. A FBN is a Bayesian
unauthorized SOAP messages from the clients. The clients                      network which has fuzzy variables [16]. In the proposed
authenticate themselves by their physical hardware address                    work, Y represents the services and X represents the QoS
which has been previously registered with the SC. The list                    metrics. By the Bayesian equation [17], if X and Y are
of all registered clients and their respective hardware                       events, and P(Y)>0, then the conditional probability of X ,
addresses are stored in the SC which are validated during                     given Y is,
service requests and only authorized clients are permitted to                                                 P ( XY )
invoke service requests. The SC also holds a connection                                          P( X ¦ Y) =
manager which holds a multiple connection links with the                                                        P (Y )
client routed through different Internet Service Providers                                P ( XY ) = P (Y ) P ( X ¦ Y) = P(X)P(Y ¦ X)
(ISPs). In our experimental study each client is connected                    Assume that {Yi, i € n}is a countable unit of events. Let Y
with three different links, a Virtual Private Network(VPN)                    be another event and suppose that we know P(Yi) and
link, a Multi Protocol Labeled Switching (MPLS) link and a

                                                                                                           ISSN 1947-5500
                                                                                   (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                                     Vol. 8, No. 5, August 2010
P(X│Yi) for each i ∈ n. Accordingly the total probability
formula is,                                                                                  6. Conclusion
                   P(Y)=         ∑ P(Y ) P( X ¦ Y )
                                 i =1
                                               i             i                               The quality and eventually the performance of a SOA
                                                                                             system depends upon the selection of a suitable service from
                            Bayesian equation is                                             the available services after giving due consideration on the
                                                   P ( X jYi )                               QoS requirements. Since the components of a typical SOA
                      P (Yi ¦ X j ) =                                                        application are built by integration from various service
                                                P( X j )                                     providers in a distributed environment, the QoS
                                          P (Yi ) P ( X j ¦ Yi )                             requirements change in runtime. In fact the quality
             P (Yi ¦ X j ) =             n

                                        ∑ P(Y ) P( X
                                        i −1
                                                    i        j   ¦ Yi )
                        for i=1 to n, j=1 to m
Assuming that E1, E2, ….. En are events in the experiment,
the multiplication rule of probability is given by
 P ( E1 ) P ( E2 ¦ E1 ).....P ( E N ¦ E1 , E2 ,....E N −1 ) .
If the prior probability P(Yi ) and the prior conditional
probability P(Xj │Yi) is known then, P(Yi │Xj)can be
derived from the Bayesian Equation. According to the
derivations of the fuzzy subset [18,19,20,21], any function
µ Ã which maps the domain U to the closed interval [0,1],
(µ Ã:U→[0,1]) ascertains a fuzzy subset Ã. The function µ Ã
is a membership function of the fuzzy subset and so µ Ã (u)
is the membership grade of u for à and its scalar probability                                                    Fig. 2. FBN Rules
                     ∼                                                                       also relies on the underlying support systems and on the
                P ( A) = ∑ x ∈ xµ A ( x) P ( x) .
                                  ∼                                                          network resources. In this paper we analyzed the Web
Using the above equation the fuzzy Bayesian equation can                                     services selection based on the Fuzzy-Bayesian Network
be expressed as,                                                                             model and the experimental results showed a considerable
        ∼                                                                    ∼
                                                                                             increase in performance and reliability. This work may open
    P(Y ¦ X ) = ∑∑ µ ∼ (Yi )µ ∼ ( X j ) P( X j ¦ Yi ) P ( X i )           P( X )             up the doors towards a broader analysis which may include
                 i∈I j ∈J    Y           X
                                                                                             many other approaches along with FBN model to improve
                                                                                             the selection of Web services.
5. Implementation and Experimental Results
The proposed framework was implemented in C# and
performed on 4 servers acting as service providers. The                                      1.   Understanding Quality of Services for Web Services
servers had an Intel Quad Core processor with 8GB RAM                                             and       Web      Services      OoS      Requirements,
running on a Windows2008 operating system. One of them                                  
acted as the Service Dispatcher that monitors and hosts the                                       html
QoS and QoE repositories. 16 clients were connected to this                                  2.   Using Policy-Based QoS to Enable and Manage WMM
framework each located over a distance of 200Kms from the                                         in        Enterprise        Wireless      Deployments,
servers. The clients were composed of Intel Core2Duo                                    
processor with 2GB RAM having Windows XP operating                                                7.aspx
system. The clients and servers were connected on a MPLS-                                    3.   C. Patel, K. Supekar, Y. Lee, 2003, A QoS Oriented
VPN framework with a backup VPN connectivity and a                                                Framework for Adaptive Management of Web Service
redundant Internet line from different ISPs. This                                                 based Workflows, Lecture Notes in Computer Science
experimental setup is created to simulate an e-business                                           (LNCS), Volume 2736, Springer-Verlag, ISBN 3-540-
environment with the clients and offices spaced in different                                      40806-1, pp. 826 – 835.
locations. The QoS metrics of cost, availability, execution                                  4.   D. A. Menasce. QoS Issues in Web Services. IEEE
time, mean response time, trustworthy, performance and                                            Internet Computing, 6(6):72–75,Nov./Dec.2002.
capacity are measured and they represent the inter-                                          5.   Qos for web services: Requirements and possible
organizational transactions only. Suitable Fuzzy Bayesian                                         approaches.W3C Working Group Note, 25 November
rules were created and the model was tested in the Matlab                                         2003,      Available     at
was shown in Fig.2                                                                                office/TR/2003/ws-qos/.

                                                                                                                        ISSN 1947-5500
                                                             (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                               Vol. 8, No. 5, August 2010
6.    H. Ludwig. Web services qos: External SLAs and                    21. Shiyong Li, Engineering Fuzzy Mathematics with
      internal policies or: How do we deliver what we                       Applications, Harbin Institute of Technology Press,
                                                                            Harbin, 2004.
7.    promise? In Proceedings of the Fourth International               22. Hao Tang, Shi Liu, Basic Theory of Fuzzy Bayesian
      Conference on Web Information Systems Engineering                     Networks and Its Application in Machinery Fault
      Workshops (WISE’03). IEEE, 2004.                                      Diagnosis, Fourth International Conference on Fuzzy
8.    E. Al-Masri and Q. H. Mahmoud, “Investigating Web                     Systems and Knowledge Discovery (FSKD 2007).
      Services on the World Wide Web,” Int’l World Wide
      Web Conf., April, 2008,Beijing, China,pp.795-803.
9.    L. Zeng, H. Lei, and H. Chang, “Monitoring the QoS
      for Web Services” Proc. Int’l Conference on Service-
      Oriented Computing, LNCS 4749, 2007, pp. 132–144.
10.   Z. Du and J. Huai, “Research and Implementation of an
      Active Distributed Web Service Registry,” Journal of
      Software, 2006,17(3),454-462
11.   F. Li, F. Yang, et al., “A Policy-driven Distributed
      Framework for Monitoring Quality of Web Services,”
      Proc. IEEE Int’l Conf. on Web Services, 2008, pp.708-
12.   M. Lin, J. Xie, H. Guo, and H. Wang. Solving Qos-
      Driven Web Service Dynamic Composition as Fuzzy
      Constraint Satisfaction. In Proc. of the IEEE
      International Conference on E-Technology, E-
      Commerce and EService, IEEE Computer Society, pp.
      9–14, 2005.
13.   D. Dubois, H. Fargier, and H. Prade. Possibility Theory
      in Constraint Satisfaction Problems: Handling Priority,
      Preference and Uncertainty. In Journal of Applied
      Intelligence, Vol. 6(4), pp. 287–309, 1996.
14.   P. Xiong and Y. Fan. QoS-Aware Web
      ServiceSelection by a Synthetic Weight. In Proc. of the
      4th International Conference on Fuzzy Systems and
      Knowledge Discovery, pp.632–637, 2007.
15.   S. Agarwal and S. Lamparter. User Preference Based
      Automated Selection of Web Service Compositions. In
      Proc. of the ICSOC Workshop on Dynamic Web
      Processes, Amsterdam, Netherlands, 2005.
16.   J. Pearl, Probabilistic Reasoning in Intelligent Systems:
      Networks of Plausible Inference, Morgan Kaufmann,
      San Mateo, CA, 1988
17.   Christopher Fogelberg, Vasile Palade, and Phil
      Assheton. Belief propagation in fuzzy Bayesian
      networks. In Ioannis Hatzilygeroudis, editor, 1st
      International Workshop on Combinations of Intelligent
      Methods and Applications(CIMA) at ECAI’08, pages
      19–24, University of Patras, Greece, 22 July 2008.
18.   Guangzu Liu, Probability Theory and Applied
      Mathematical, Higher Education Press, Beijing, 2000
19.   Giulianella Coletti and Romano Scozzafava,
      “Conditional probability, fuzzy sets, and possibility: a
      unifying view”,Fuzzy Sets and Systems, 2004, volume
      144, issue 1, pp.227-249.
20.   Xingui He, Fuzzy Theories and Fuzzy Techniques in
      Knowledge Processing, National Defence Industry
      Press,Beijing, 1998.

                                                                                                  ISSN 1947-5500

To top