A Study On Continuously Optimization With Two-Stage Adatpation For Adaptive Service Based Software Systems

With the development and mature of Web services, service-based software system (SBS) has already been a popular way to build large-scale distributed applications in the open environment based on Internet. As an SBS is always running on an open environment, where the number of available services is changing dynamically and their performance may also change constantly, modifying an SBS manually to improve its performance becomes more difficult. On the contrary, in order to change with the environment autonomously, an SBS need to be able to adjust itself to improve the system performance against the changing environment, which is called service-based adaptive software system (ASBS) and has been put forward as an important paradigm for constructing distributed software systems in the open environment.Since an SBS is usually regarded as a composite Web service, most of current studies on optimizing ASBS are still focused on dynamic service composition and dynamic service selection. However, different from composite Web services which are proposed from a theoretical point, SBS is proposed from a system view and emphasizes the whole life cycle from business analysis, to business modeling, environment modeling, system initialization and its constant evolution as well as the adaptive execution, until the system is abolished finally. This requires different optimization mechanism and optimization strategies for continuously optimizing the system performance.In order to realize the continuous performance optimization of ASBS, this thesis focuses on how to build a system optimization mechanism which is able to meet requirements of different instances on different levels, how to achieve the balance between the effectiveness and efficiency of optimization, and how to provide systems which can both satisfy customers’ requirement and maximize providers’ revenue in providing such system, tries to provide a supporting environment for all of the either existing or future system instances. It divides the adaptation into two stages, online reactive adaptation for reliable execution and offline predictive adaptation for optimal execution, which could be helpful to combine reactive adaptation and predictive adaptation, as well as local optimization and global optimization to form a uniform framework for ASBS, which make both providers and consumers able to get their own value in providing and using an SBS. The main works in this thesis includes:(1) To realize continuous optimization in ASBS, propose the idea to divide the adaptation process into online reactive adaptation and offline predictive adaptation, which results in a two-stage adaptation mechanism supporting the combination of local optimization and global optimization, as well as reactive adaptation and predictive adaptation. Moreover, a uniform framework which is able to improve both the effectiveness and efficiency of optimization and ensure the revenue of both provider and consumer is formed.(2) To support the information feed back from off-line adaptation to online adaptation, a Pareto based supporting environment for the SBS optimal execution is proposed. Based on the principle of Pareto dominance, services could be selected and managed dynamically on multiple dimensions directive, which breaks the limitation that only generate single best solution, and is able to provide a group eclectic solutions to support various requirements of different instances and on difference degrees.(3) To cope with the uncertainty of the impact of a given change on the system execution, an offline predictive adaptation approach based on dynamic change impact analysis is proposed. Based on current approaches which gauge the change impact by rangeability, this thesis tries to extend on analysis from a dynamic point, and propose a system execution context model for analyzing the change impact dynamically. A corresponding proactive optimization strategy is given to ensure the necessity and validity of the predictive adaptation.(4) In order to optimize the revenue of system provider in the ASBS optimization, the long-term revenue driven online reactive optimization approach is proposed. By analyzing the formation of system long-term revenue, a state-active-revenue (SAR) model is designed to capture the dynamic relationship among system state, online adaptation activities and their revenue, which could further be used to form the online adaptation process as a partially observed Markov decision process to support optimize providers’revenue as well as satisfying consumers’requirements.(5) In order to validate the proposed two-stage adaptation based continuous optimization of the ASBS, a reflection framework is proposed and a prototype system for supporting the optimal execution of an SBS is designed and implemented based on ServiceMix (which is an open project belongs to Apache). On this prototype system, a simple example application of bank loan business is used to show the effectiveness and application value of the proposed two-stage based optimization approach.