Research On Component Service Resource Dynamic Adjustment-based Self-Adaptation Performance Optimization Approach For SBS Cloud Application

The importance of self-optimization and adaption ability emerges with the growth of the software system's complexity and the opening,the dynamic and the control difficulty of its runtime environment.With the extensive application of cloud computing technology,it is a very important problem how to ensure the performance requirements of cloud service users adaptively and minimize the cost of resources.The pay-on-demand mode and the characteristics of utility computing of cloud computing paradigm requires that cloud environment can not only satisfy the dynamic resource allocation for application system with minimum cost but also dynamically adjusting resources adaptively when the application system deviates from the expected behavior,based on which the application system can provide services that meet the user's expectations continuously.In view of this,it has become an important development direction to make the service system under the cloud environment adaptable.Since the service-based software system(SBS)has the characters of flexible configuration and dynamic refactoring.It has become the main form of building large scale,distributed application system in Internet environment and plays a more and more important role in various application fields,such as scientific research,communication and finance.However,most of the existing software adaptive technology regards the software system as a whole entity.When it is applied directly to an SBS built with multiple service components,the following conditions may occur:The performance of a component service is very good after the optimization,but the performance of the whole system has not been significantly improved,or all the service components has been adjusted and the overall system performance has been improved,but the resources costis too much greater.In this dissertation,the idea of dynamically adjusting the resources of some component services to influence the overall performance of SBS is proposed.Based on this idea,an adaptive optimization framework for SBS cloud application performance based on dynamic adjustment of component service resources is established with the MAPE-K(Monitor-Analyze-Plan-Execute-Knowledge)model as the prototype.Based on the framework,this dissertation proposes an adaptive optimization method for different optimization requirements,and verifies the effectiveness of the proposed adaptive method on the experimental cloud optimization platform.These adaptive optimization methods provide an effective solution for the performance optimization of SBS cloud applications.The contribution is reflected in the following aspects:(1)A general framework for adaptive optimization of SBS cloud application performance is proposed,the idea is to adjust the resources of some component services to influence the overall performance of SBS,and the process of adaptive optimization is divided into off-line phase and on-line phase.The adaptive process reaction time during the on-line phase is improved effectively by forming the adaptation rule database during the off-line phase.(2)A hybrid genetic algorithm based adaptive optimization method for SBS cloud application performance is proposed to cope with the situation that existing adaptive methods usually treat the application system as a whole entity,which may lead optimization to ineffectiveness or high cost when applied in SBS cloud application.This adaptive method is oriented to global optimization,it optimizes the SBS performance to the constraints of the service-level agreement(SLA)with minimum costs.The adaptation adjustment generation is completed in two phases.Constructing an adaptation template set related to a specific target in the off-line phase can effectively reduce the amount of computation in adaptation decision making,so as to ensure the time-validity of adaptive decision.(3)A reinforcement learning based adaptive optimization method for SBS cloud application performance is proposed to cope with two issues:the adaptive optimization method based on the performance model is applied to SBS cloud application can not adapt to the dynamic changes of cloud environment,and the efficiency of adaptive optimization method based on intelligent optimization algorithm is low.This adaptive method is oriented to gradually optimization with the idea of"Execute-accumulate-learn-decide" and a model free online learning algorithm.The result of adaptive decision making is continuously optimized through online learning,so that the adaptive method can cope with the dynamic changing of cloud environment and SBS.The shaping operators are composed of trigger event target and action target,which reduce the scope of candidate adaptive actions,and improve the learning efficiency of the algorithm.(4)A continuous double auction based adaptive optimization method for SBS cloud application performance is proposed to cope with the issue that most of the current adaptive methods only focus on how to prevent the occurrence of SLA violation from the individual level,without considering the overall revenue of cloud environment providers and cloud resource users(SBS cloud service providers)from the system point of view.This adaptive method is oriented to overall revenue maximization.It promotes the trade-off of resource prices,the respective revenues and quality of service(QoS)requirements between resource providers and users,and maximize the respective income of both side relatively and ensure the SBS cloud application performance meets with the requirements of SLA.The fitness between component service and resources is added to the auction process,to allocate resources for component services that fit their characteristics as to achieve the goal of increasing revenue.(5)This dissertation designs and implements a cloud application performance optimization support platform based on kernel-based virtual machine(KVM),and takes an server with scenic voice guide system as an example.The optimization effects of the proposed adaptive methods under various scenarios are analyzed and compared in this platform to verify the effectiveness of the proposed adaptive method and its practical application value.