基于渐进约束违背删除策略的服务合成执行路径优化

Optimizing the Service Composition Execution Paths Based on a Gradual Constraint Violation Deleting Scheme

在面向服务计算领域,服务合成正成为分布异构资源整合、业务集成的一个新型计算模式。在开放互联的因特网环境下,由于存在多个功能相似但服务质量不同的服务提供商,从而导致一个抽象的服务合成方案存在多条可执行路径。然而,在满足用户综合QoS约束需求的情况下,如何有效协同单个服务的QoS从多条候选路径选出一条优化的执行路径,便成为服务合成QoS研究中的一个关键问题。遗憾的是,该问题还没有得到很好的解决。为此,我们提出一个采取渐进约束违背删除策略建立优化执行路径的方法。我们的方法能够避免探测多条执行路径,而是在靠本地最优化策略建立的执行路径上,通过全局QoS计算和QoS约束检查模型找出QoS约束违背,并进而采取渐进约束违背删除策略逐渐删除这些约束违规,并建立起一条近优的执行路径。实验表明,我们的方法很好地降低了建立优化执行路径的时间复杂性。

In service-oriented computing systems, service composition is emerging as a new computing paradigm in integrating distributed and heterogeneous resources and applications. A critical issue in the service composition area is how to achieve an optimized overall end-to-end quality of service （QoS） composition＇s requirements by effectively coordinating individual QoS constraints for a single service. Unfortunately, this issue has not been well addressed. In this paper, we propose a QoS constraint correction approach by gradually deleting QoS constraint violations for optimizing the service composition execution paths. Our methods avoid directly detecting multiple execution paths for a service composition, but first employ the local optimization policy to build an initial execution path with- out considering user-expressed end-to-end QoS composition constraints. Based on this path, global QoS computing medals can be used to calculate the global QoS values for each quality attribute Then, the QoS constraint violations can be detected by comparing the global QoS values with the end-to-end QoS composition requirements expressed by a user. Furthermore, a correction algorithm by gradually deleting the QoS constraint violations is proposed to recursively correct such violations by resdecting critical service providers. As a result, an optimized execution path can be rebuilt to meet the overall end-to-end QoS composition requirements An experiment and evaluation on our SwinDew-G（Swinburne Decentralized Workflow for Grid） simulation platform have been made to demonstrate the feasibility and effectiveness of our approach.