QoS Modeling, Prediction And Assurance In SOA

Service-Oriented Architecture (SOA), as the new generation software archi-tecture, serves the purpose of responding quickly to the changing environmen-tal conditions in multiple fields, such as system integration. E-Commerce and E-Administration. Therefore, SOA is gaining increasing attention in both academic and practice areas. In SOA, services with standard interfaces provided by different organizations can be rapidly composed and reused, effectively supporting resource sharing and application integration under a distributed and heterogeneous environ-ment.Autonomous.loosely coupled and highly interoperable in nature. web services become the preferred standards-based way to realize SOA. With the widespread application of SOA. a great many web services with the same, similar or overlapping functionalities are available on Internet. Consequently, the non-functional feature of a service, in particular Quality of Service (QoS), becomes the key criterion for service selection and composition. Due to the complex and dynamic nature of the distributed environment, the management of QoS should be incorporated into the entire lifecycle of a service in order to satisfy the requirements of service customers. Several key issues of QoS management, including QoS prediction, QoS-based service selection, QoS monitoring and QoS assurance, are intensively under investigation.To build an appropriate QoS model lays the groundwork for further studies. Existing solutions model service QoSs either as deterministic values or probabilistic distributions. And different QoS metrics are modeled independently. Although probability-based QoS models reflect certain dynamic features of QoS, all these work overlooked an important aspect in QoS modeling, time. Most QoS metrics, such as response time, availability, are time dependent, e.g., they can change with time dramatically. Moreover, different QoS metrics always correlate when considering the influence of time. Existing works underestimate the importance of time on QoS, leading to the lose of accuracy on QoS modeling. Therefore, it is highly desirable to establish QoS models that, reveal the timing relationship of QoS and to carry out the corresponding QoS researches based on the new QoS model. Another attractive issue is service substitution, which is an effective way to cope with service failure or QoS degradation. Since it is not realistic to find totally equiva-lent substitutive service in practice, the mechanism of adaptive service substitution gains much attention recently. Specifically, the adaptive substitution method for conversational services, i.e., how to rapidly develop adaptors to resolve mismatches both at interface level and protocol level, has become the focus of intense research.In this thesis, we mainly concentrate on QoS modeling, QoS prediction and QoS assurance. The main contributions include:1. We demonstrate the necessity of dynamic QoS modeling for web services. A Cycle QoS Model(CQM) is proposed to capture the periodically changing QoS patterns. We also define a service survivability model for complex service systems that adopt agile QoS management mechanisms.2. We propose a dynamic QoS modeling method for web services based on Hidden Markov Model(HMM). A QoS observation vector sequence is acquired by analyzing the service log. Then we apply traditional HMM algorithms to get the reasonable time segmentation for the changing QoS, and based on which, we can set up the Cycle QoS Model or service survivability model eventually.3. We investigate the QoS prediction algorithms for composite services based on the proposed CQM. The composition patterns we focus on include sequential, synchronized parallel, asynchronized parallel, conditional and loop patterns. Three problems can be solved:(1)determining the time cycle of a composite service;(2) estimating the QoS of the composite service if it is invoked at a certain time point;(3)obtaining the CQM of the composite service. Specifically, according to the two kinds of constraint relationship between the service response time and the length of QoS cycle, two QoS prediction algorithms are proposed respectively. including the segmental QoS aggregation algorithm and the one-way QoS reduction algorithm.4. We address the run-time QoS assurance mechanisms. In detail, we propose an automatic approach to generate adaptors to support substitution for conversational services. The approach is based on graph planning techniques in which service sub-stitution is encoded as a graph planning problem. Then a solution can be achieved by constructing a leveled graph followed by a backward chaining graph exploring. The interface and protocol incompatibility between the conversational services can be solved consequently.