# Quality of Service Centric Web Service Composition: Assessing Composition Impact Scale towards Fault Proneness Introduction ervice-Oriented Architecture (SOA) simplifies information technology related operational tasks by consumption of ready-to-use services. Such SOA found to be realized currently in ecommerce domains such as B2B, B2C, C2B and C2C, in particular the web services are one that considered serving under this SOA. Web services are software components with native functionality that can be operable through web. Another important factor about this web services is that more than one service can be composed as one component by coupled together loosely. The standard WSDL is web service descriptive language that let the self exploration of the web services towards their functionality and UDDI is the registry that lets the devised web services to register and available to required functionality [1]. Since the task of composition is integrating divergent web services explored through different descriptors, it is the most fault prone activity. The functionality of service composition includes the activities such as (i) identify the tasks involved in a given business operation, (ii) trace related web services to fulfill the need of each task, (iii) couple these services by exploring the order of that services usage, which is based on the expected information flow, (iv) and resolve the given operation by ordering the responses of the web services that coupled loosely as one component. In order to achieve quality of service and secure transactions in web service composition and usage, the impact of the composition should be estimated before deploying those loosely coupled web services as one component. The Web service compositions used earlier that can be found in repositories and the services involved in those compositions helps to assess the impact of these web services towards fault proneness. The current composition strategies [2] [8] are error prone, since these State-of-theart techniques are not mature enough to guarantee the fault free operations. However, finding these compositions as fault prone after deployment is functionally very expensive and not significant towards end level solutions, also may leads to serious vulnerable. Hence the process of estimating the composition scope towards fault proneness is mandatory. [3] [4] [5] [6] [7] In this paper, we propose a novel statistical approach to estimate the impact scale of a service composition towards fault proneness. Our approach acts as an assessment strategy for any of existing web service composition approaches. The paper is structured as follows. Section 2 discusses related work. In section 3, the proposed statistical approach is explored, which followed by Section 4 that contains the results explored from empirical study. The conclusion of the proposal and future research directions were discussed in Section 5. # II. # Related Work Service compositions with malfunctioned web services lead to form the highly fault prone compositions. Henceforth the web service composition to serve as one component under SOA is complex and needs research domain attention to deliver effective strategies towards the QoS centric service. The model devised in [9] defined set of QoS factors to predict feasible services. Many of existing quality-aware service selection strategies aimed to select best service among multiple services available. The model devised in [8] considering the linear programming to find the linear combination of availability, successful execution rate, response time, execution cost and reputation, which is in regard to find the optimal service composition towards given business operation. The model devised in. [6] is considering the temporal validity of the service factors. The authors in [10] modeled a mixed integer linear program that considers both local and global constraints. The model devised in [7] is selecting services as a complex multi-choice multi-dimension rucksack problem that tends to define different quality levels to the services, which further taken into account towards service selection. All these solutions are depends strongly on the positive scores given by users to each parameter. However, it is not scalable to establish them in prospective order. Though the QoS strategies defined are used in service composition the factor fault proneness of the service composition is usual. In regard to this a model devised in [11] explored a mechanism for fault proliferation and resurgence in dynamically connected service compositions. Dynamically coupled architecture outcomes in further complexness in need of fault proliferation between service groups of a composition accomplished by not depending on other service groups. In a gist, it can be conclude that almost all of the benchmarking service quality assessment models are attribute specific, user rating specific or both. Hence importance of attributes is divergent from one composition requirement to other, and the user ratings are influenced by contextual factors, and another important factor is all of these bench mark models are assessing services based on their individual performance, but in practice the functionality of one service may influenced by the performance of other service. Henceforth here in this paper we devised a statistical approach that estimates the impact scale of service composition towards fault proneness, which is based on a devised metric called composition support of service compositions and service descriptors. The said statistical model works in two aspects. First, it estimates the impact of each web service descriptor to form a selected malfunctioned service composition. And then it estimates the higher and lower ranges of the impact scale o towards fault proneness, which is from the impact of each service descriptor and each malfunctioned service composition. Then these higher and lower ranges of the impact scale will be used to assess the impact of a newly composed service composition towards fault proneness. This strategy leads to estimate the problem of web service descriptor selection. The business solution expected might represented by several compositions, but selecting one of these compositions is strictly by their impact towards fault proneness. The proposed model is optimal in this regard. The detailed exploration of the proposed model is as follow: The approach of measuring Composition support ( ) metric is proposed in this paper. In regard to measure the composition support, we consider the bipartite graph that represents the composition weights. Build an undirected weighted graph UWG with web-service descriptors as vertices and edges between web-services descriptors. An edge between the two web-service descriptors will be weighted as follows The graph representation (fig. 1) indicates the bipartite relation between web-service descriptors and web service compositions. Composition weights of the different web service compositions represent their importance. Intuitively, a web service composition with high composition weight should contain many of the web-service descriptors with high composition support. The underpinning association of web service compositions and web-service descriptors is that of association between hubs and authorities in the HITS model [13]. { } foreach wsds wsds SWSS ? ? | | 1 ( ) {1 [( , ) ]} | | i j SWSS k i j The devised process of identifying web service composition weights using bipartite graph is explored below: Let consider a matrix format of the connection weights of the bipartite edges between web-service descriptors and web-service compositions in given bipartite graph. The weight of the each web service composition as a hub in a bipartite graph is initialized as 1, which we represented as matrix (table 1). Table 1: Initializing the weight of the each web service composition as hub in bipartite graph with 1 and represented them as a matrix u as follows. Let the weights between descriptors and compositions of the given bipartite graph (see fig 1) and form a matrix such that rows represent descriptors (authorities) and columns represent compositions (hubs) and refer that matrix as A, As referred in HITS [13] algorithm, find each web service descriptor (authority) weight, which is can be done as follows: ' v A Xu = Here in the above equation v is the matrix representation of the web service descriptor weights as authorities, # ' A is the transpose matrix of the matrix A , which is the matrix representation of connection weights between web service compositions as hubs and web service descriptors as authorities in bipartite graph. Then the actual weights of the web service compositions (hubs) can be measured as follows: u AXv = The matrix multiplication between matrix A and matrix v results the actual weights of the service compositions as hubs. Then the composition support cs of webservice descriptor wsd can be measured as follows Then the standard deviations of the ? each service composition from ? will be measured further, which is as follows: ( ) The service composition is said to be fault prone if | | 2 1 (| | 1) i SWSS wsc i sdv SWSS ? ? ? = ? ? ? ? ? ? ? =wsc h ? ? > IV. # Empirical Analysisand of Proposed Model This work explored the credibility of the proposed model on set of 296 service compositions. The above said data set contains 294 samples, out of that 250 samples were used to devise the Degree of fault prone threshold and its upper and lower bounds. Further we used the rest 44 records to predict the fault proneness scope. Interestingly, the empirical study delivered promising results. The statistics explored in table 10 We used accuracy estimation (the percentage of valid predictions by the proposed) as the main performance measure. In addition to measuring accuracy, the precision, recall, and F-measure were used to analyze the performance; these are defined using following equations. As per the empirical study conducted the ?? + found here are 41 and ð??"ð??" + are 0, henceforth precision is 1. t rc t f + + ? = + Here in above Equation, the ' rc ' indicates the recall, ð??"ð??" ? indicates the false negative. As per the results explored in empirical study ð??"ð??" ? are 11, hence the ???? value is 0.788. Here in the above Equation, ??indicates the Fmeasure. And the F-measure found from the results of the empirical study is 0.88143 As per the results explored, the proposed model is accurate to the level of 79%. The failure percentage is 21%, which is not negligible but considerably performed well. V. # Conclusion The model devised in this paper is a method of estimating web service composition impact scale the towards fault proneness. This approach is a statistical analysis that derives lower and higher range of service composition impact scale towards fault proneness. In regard to this initially an undirected graph that connects the involved web service descriptors as vertices with weighted edges. The edge weight of to vertices is the ratio of service compositions contains services from both descriptors act as vertices to a selected edge. Further a bipartite graph build between web service compositions as hubs and web service descriptors used to compose those compositions as authorities. Further hub and authority weights were calculated as explored in section 3, and further these weights were used to estimate the service composition impact scale towards fault proneness. The estimated service composition impact scale higher and lower range values can be used further to estimate the impact of any service composition towards fault proneness. The empirical analysis was conducted on dataset with 296 divergent web service compositions. The explored results are indicating the significance of the proposed model. In future to improve the accuracy of the devised model, the correlation of the service descriptors will be estimated, which is done by considering the web-services of each descriptor as categorical value set. Further, web-service reputation can also be considered to estimate the impact of a service composition towards fault proneness. ![Composition of web services is loosely interconnected set of Web service operations that acts as a single component, which offers solutions for divergent tasks of an operation.](image-2.png "C") 1![Figure 1: bipartite graph between web service compositions and web-service descriptors If a web-service descriptor](image-3.png "CFigure 1 :?") 10Total Number of web service composites296Total number web service descriptors used140Total number of edges determined1560Total number of bipartite edges found27776Service impact scale threshold ? 0.46795646260519363 composition © 2014 Global Journals Inc. (US) * Serviceoriented computing MPPapazoglou DGeorgakopoulos Communications of the ACM 46 10 2003 * Constraint-driven Web Service Composition in METEOR-S RAggarwal IEEE Conference on Service Computing 2004 * Planning and monitoring the execution of web service requests ALazovik MAiello MPapazoglou International Conference on Service-Oriented Computing 2003 * Semi-automatic Composition of Web Services Using Semantic Descriptions ESirin JHendler BParsia Web Services: Modeling, Architecture and Infrastructure workshop in ICEIS 2003 * Web Service Composition -Current Solutions and Open Problems BSrivastava JKoehler Proceedings of ICAPS Workshop on Planning for Web Services ICAPS Workshop on Planning for Web Services 2003 * An Approach to Temporal-Aware Procurement of Web Services OMartin-Diaz ARuize-Cortes ADuran CMuller International Conference on Service-Oriented Computing 2005 * Service Selection Algorithms for Composing Complex Services with Multiple QoSConstraints TYu KJLin International Conference on Service-Oriented Computing 2005 * QoS-aware Middleware for Web Services Composition LZeng BBenatallah IEEE Transactions on Software Engineering 30 5 2004 * Quality of service for workflows and web service processes JCardoso ASheth JMiller JArnold KKochut Journal of Web Semantics 1 3 2004 * Global and Local QoS Constraints Guarantee in Web Service Selection DArdagna BPernici IEEE International Conference on Web Services 2005 * Handling Faults in Decentralized Orchestration of Composite Web Services GChafl SChandra PKankar VMann International Conference on Service-Oriented Computing 2005