Font Size: a A A

Research On Cloud Service Composition And Its Key Problems

Posted on:2015-01-15Degree:DoctorType:Dissertation
Country:ChinaCandidate:S Q ZhangFull Text:PDF
GTID:1318330518991340Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
As a new computing model and commercial model, cloud computing achieves dynamic elasticity and scalability as the technical features and flexible service contracts for the commercial characteristics. By matching users' requirements and services' functionalities, cloud services can be reused, combined and verified, constituting loosely coupled applications on users' immediate needs. In recent years, with the continuous development of cloud computing technology and the more mature commercial models, some significant features emerge in cloud service composition scenarios, including personalized user requirements, massive candidate services, cross-cloud service scheduling and dynamic composition environment. In view of these challenges, how to efficiently discover and combine cloud services which are distributed across different cloud platforms into service applications, and ensure their stability is key to cloud service composition.For the past few years, QoS-aware service composition methods, as the mainstream approach to build distributed applications, have attracted wide attention from academia and industry, which also produced many results. However, in cloud computing environment, some challenges have emerged in providing users efficient, robust and personalized composite services. 1) Among competitive cloud platforms, a specific platform is unable to achieve one-stop service inquiry on other service platforms, which makes cross-cloud service composition difficult. 2) In cloud environment, facing massive cloud services with similar functions and different QoS values,providing personalized composite services to users is challenging. 3) To enrich user experiences,some packaged and coarse-grained services, which could perform multiple functions, are usually released. But in traditional service composition process, service evaluation is only performed on a given function, which usually ignores others' impacts on the whole service. 4) In dynamic cloud environment, cross-cloud service compositions tend to be unstable and prone to be failed due to some disabled component services. In facing these challenges, the main contributions of this paper are listed as follows:1) A framework for cloud service composition is first proposed in this paper. The framework consists of four levels, which are cloud service level, service network level, service discovery level and service composition level. Cloud service level is responsible for collecting service information from different cloud platforms through distributed service brokers. Then, a service network could be constructed based on the above service information in service network level. In service discovery level, community detection theories will be employed to discover services which can perform similar functions or collaborative functions in a service network. Finally, in service composition level, service selection, evaluation and composition will be performed, and service exceptions will also be captured and repaired.2) For achieving the one-stop service inquiry across different cloud platforms, a framework for cross-cloud service discovery is proposed, based on the service discovery module in the proposed system framework. In this service discovery framework, services are supposed to be managed by the independent cloud platforms, and service information can be extracted to construct a service network independently from any cloud platform. Then, a similarity-based community detection method for service networks is also presented, which takes service description information as services' node attributes, and service composition records(including QoS information) as their edge attributes. The two types of attributes can be assigned and adjusted with different weights, in community detection process, for achieving a good community division where services with similar functions or collaborative functions can be discovered. An experiment was also conducted for validating the proposed method.3) In order to improve the efficiency of personalized service composition when there are massive candidate services with similar functions and different QoS values, a User Preference-aware skyline Service (UPS) model is proposed. Based on the QoS attributes focused by a user, UPS could transform skyline service computation from all QoS attributes-based to the user preference-aware attributes-based. Then, based on the UPS model, a service selection method for Top-K composite services is also presented, which includes a multi-index based local service selection method and a service-lattice based optimization method for Top-K composite services. With an efficient computation for UPS services, the method is able to achieve a good performance on optimizing Top-K composite services. Finally, two datasets are employed for validating our method.4) In traditional service composition process, service evaluation is only performed on a given function, ignoring others' impacts on the whole service, which usually results in an inaccurate result for coarse-grained cloud services. For solving this problem, a Multi-Functional Specification model of service (MFS) model is proposed based on the multiple function specifications of services, and new QoS and service utility models are also presented. Then,using the MFS model, a service evaluation method is introduced for coarse-grained cloud services. The method first decomposes users' global constraints into local constraints, and extract common functions for each service pool according to their local constraints. Next,weights for each common functions will be calculated based on their invocation times and utility values for coarse-grained services will be aggregated using these weights and QoS values for each common function. The rank will be conducted for services based on their utility values. Finally, an experiment was conducted for validating the proposed method.5) For ensuring the stability of cloud serviced applications, a service network model and a service network construction method are proposed, based on a formal description of cloud services. With the partial service matching and complete service matching,the construction method connects cloud services according to their historical composition records. Then, based on the service network model, a service network-based exception handling method is proposed for two exception scenarios in cross-cloud service composition, where their solutions are also presented. Regarding the two scenarios, in the first scenario, only one component service is failed and, in the other one, multiple adjacent services are failed at the same time. At last, the performance of our method is analyzed, which is also compared with other related works.
Keywords/Search Tags:Cloud computing, cloud service, service composition, QoS, service network, user preference, Skyline, exception handling
PDF Full Text Request
Related items