Dependable Service Models And Their Optimization Algorithms Supporting Cross-Organizational Workflow Applications

With the rapid development of cloud computing,service computing and social computing,the complexity and volume services are explosively increasing in distributed service-oriented cloud systems.The paradigm shift from closed business information systems to open,dynamic and loosely coupled service-oriented workflows requires more effective approaches to support interactions.However,the inherent uncertainty and unreliability of large-scale service-oriented cloud systems often pose threats to the operations of cross-organizational workflow applications.Furthermore,in such failure-prone large-scale scenarios,it is still a research challenge to integrate dependable cloud services into a cross-organizational workflow application in a prompt and efficient way.Therefore,in this dissertation,we discuss cross-organizational workflow evaluation problems,including service reliability,service performance,social service relationship,and service reputation.The traditional service evaluating approaches are improved by the adoption of process integration framework,multi-swarm particle optimization,mini-batch online learning,social collaborative computing,and online clustering.Furthermore,dependable service computing models and their optimization algorithms are proposed for the integration of cross-organizational workflows in large-scale service-oriented cloud systems.The contributions of this research are listed as follows.1)Existing workflow scheduling models need to traverse every service candidate in large-scale service-oriented cloud systems until the optimal scheduling solution is found,which would lead to high computational costs.In this dissertation,a new framework based on process integration technology is proposed to support workflow scheduling in large-scale service oriented applications.The framework integrates several computing models,including service skyline filtering model,deadline bottom level model,and weight adjusted optimization model.Firstly,a skyline filtering notion is introduced to construct service skyline set;secondly,a deadline top level model is employed to find the initial solution in service skyline set;finally,an iteratively adjusted weight optimization model is proposed to search the optimal scheduling solution.Theoretical analysis and extensive experiments on synthetically generated dataset demonstrate the effectiveness and scalability of the proposed algorithms.2)As the dynamic cross-organizational process-based collaborations prevail,the reliability of collaborative workflows becomes more and more important than ever.To support such complex collaboration scenarios,a novel population co-evolutionary reliability model is proposed for cloud service-oriented workflow systems.Tree structure-based reliability measures are employed to evaluate the reliability of the cloud service-oriented workflows.Moreover,the constrained reliability optimization is converted into a bi-objective optimization problem;and a novel discrete particle swarm optimization model is presented.In the proposed model,firstly,discrete particle codes and disturbance updating operators are designed;secondly,a new uniform diversity fitness measure is defined to search the distributed Pareto solutions;finally,the global non-dominated set of optimal solutions is constructed based on the fitness measures.Experimental results illustrate the effectiveness of the proposed model.3)The disqualification of third-party services in large-scale service-oriented cloud systems has posed great threat to cross-organizational workflow applications.Service failures and performance violations in some critical business scenarios need to be proactively prevented instead of recovery triggered by the occurrence of the failures.In most existing performance evaluation approaches,one single evaluation value is computed to evaluate service performance.Such a simple performance evaluation value may not be able to reflect the real performance level very well.However,to depict the status of workflows exactly,a good workflow application requires a more comprehensive performance evaluation;and to proactively prevent the performance violation of workflows,it is a critical task in such complex scenarios to predict the service performance;moreover,traditional batch machine learning techniques could not satisfy the requirement of real-time predicting in large-scale service-oriented applications.Therefore,in this dissertation,a comprehensive evaluation approach is proposed to analyze service performance supporting cross-organizational workflows by providing several evaluation features.Furthermore,an online mini-batch prediction model is proposed to prevent the performance violation of workflows in an efficient way.Theoretical analysis and experimental results indicate that the proposed approach is feasible and effective.4)Recently,Human-interactions are substantial part of web service-oriented collaborations and cross-organizational business processes.Social networks can help to process crowdsourced workflow tasks among humans in a more effective manner.However,it is challenging to identify a group of prosperous collaborative partners with a leader to work on joint cross-organizational workflow tasks in a prompt and efficient way,especially when the number of alternative candidates is large in social networks.Therefore,in this dissertation,a new and efficient approach has been proposed to identify collaborative partners optimally based on existing social relations.Three innovations in the proposal are: firstly,a set of new concepts has been defined to remodeling the social graph;then,a subgraph connector-based betweenness centrality algorithm has been enhanced to efficiently find the leader who serves as the host manager of the joint workflow tasks;finally,to improve the efficiency of computation,an innovative algorithm is proposed to identify the collaborative partners of the leader by confining the searching space in the set of connector nodes.For validation purpose,theoretical analysis and extensive experiments are conducted;and the experimental results have demonstrated that the proposed algorithms outperform several existing algorithms in terms of computation time in dealing with the increasing number of workflow task executing candidates.5)With the increasing emergence of mass cloud services in large-scale cloud systems,it has become more and more important than ever to effectively make web service recommendation by taking advantages of past usage experiences of cross-organizational workflows.However,in the cross-organizational workflow event log,there exist many noise ratings given by malicious service users;furthermore,it is still a research challenge in such large-scale service-oriented scenarios to recommend dependable cloud services in an effective way.Therefore,in this dissertation,an effective reputation-based integrating framework of service recommendation is proposed to support cross-organizational workflow applications.The algorithm framework consists of three stages: firstly,a new online clustering method based on modified update rule is proposed to tackle the challenges of mining large scale service data for a collaborative filtering recommendation algorithm;secondly,the feature of clustering is used to check the collusion;finally,a personnel trust evaluation model is integrated in the framework to iteratively identify the bias and prestige of service candidates in recommendation listing.Extensive experimentation and comparison to related work indicate that the proposed approach is feasible and effective.