【作者】 刘安丰；

【导师】 陈志刚；

【作者基本信息】 中南大学 ， 计算机应用技术， 2005， 博士

【摘要】 Web服务占据了当前网络70%以上的通信流量,并呈明显上升的趋势,在下一代互联网中Web服务普遍被认为代表了网格的发展方向。如何建立合理的Web服务结构模型,并实现有效QoS(Quality of Service)控制的机制与策略,从而满足不断增长的客户需求,为用户提供有效的Web服务,并为不同用户提供区分服务和性能保证,这是当今Web服务发展所面临的巨大挑战,也是实现下一代网络不可缺少的关键技术。 本文针对基于局域网的Web服务集群和网格Web服务二个研究对象,深入研究了Web服务的结构模型、Web QoS控制的策略和算法。针对Web服务集群,首先提出了一种基于QoS-aware的Web集群区分服务模型,然后在此基础上针对系统负载均衡、区分服务、克服自相似访问业务对QoS的影响、高可用性等几个方面的QoS控制指标提出了一系列解决策略与算法。采取的策略是:先进行某一方面的QoS优化,然后逐步在随后提出的解决方案中综合前面提出的QoS指标一起优化,最后,达到使系统总体回报率最大化的同时,使各项QoS指标得到优化。针对网格Web服务还没有较完善的结构模型,提出了一种多级多层的Web服务结构模型,并对其进行了改进使其具有从协议低层支持负载均衡。总结来说,本文的主要研究工作如下: (1)提出了一种Web集群区分服务结构模型,它较好地解决了以往在Web集群系统中前后台区分服务分离的不足。在此基础上,将基于事务的区分服务接纳控制策略运用到此模型。理论分析与实验结果表明:在系统过载时能有效保证系统的稳定性,高优先级任务得到较好的服务,比基于会话的接纳控制效果要好。 (2)提出了一种结合网络自相似访问特征的接纳控制算法,它综合考虑了请求访问特征以及负载情况等多方面的因素自适应调整,综合优化Web集群系统中多维调度目标,实验结果证明了算法的有效性,它既能适应网络自相似访问特征,又可支持多维QoS的调度优化。 (3)提出了一种资源优化的QoS控制算法,针对不同的服务有不同的资源需求,提出了一种既能使系统负载均衡又能充分利用系统资源的启发式算法。实验证明,该启发式算法能显著地降低资源优化分配中的计算复杂度,使其能满足实时调度的需要。在此基础上,进一步提出了一种资源优化的双最小均衡区分服务调度算法,该算法综合了多维QoS目标优化以及区分服务。在与其它调度策略如分离式调度算法的对比结果显示:双最小均衡调更多还原

【Abstract】 Web service, widely considered as the future trend of grid in the next generation of Internet, takes up more than 70% of communication traffic and goes into its boom period. One of the major challenges Web service evolution faces, which is also a fundamental technology for the next generation networks, is how to build a well-defined architecture model and implement effective QoS control mechanisms and policies, so to fulfill increasing users’ demand and moreover, to provide varied services and guarantee high performance for different users.The paper, focused on 2 research objects consisting of the LAN Web server clusters and the Grid Web service, carries on an intensive study about Web service architecture model, Web QoS control policies and, algorithms. At first, a Web cluster differentiate service model based on QoS-aware is put forward aiming at the Web server clusters. And then on such a basis a series of solution policies and algorithms is brought up with the guideline of system load balancing, differentiate service, integral self-similar access transaction characteristics and high availability. The policy suggests that QoS optimization at the beginning is taken on some aspect, and then gradually incorporated with the QoS guidelines mentioned above in the following solution, at last brought down on the maximal of system total payoff ratio as well as each QoS guidelines. Given the incomplete structure model of Grid Web service,a multi-level multi-tier architecture model for Web service is held out with the enhancement of QoS features in the capability of load balancing support in the lower part of the protocol. To sum up, Major research contributions are as follows:(1) A differentiate service architecture model over Web cluster is put forward to take a better effect on the solution to the shortcoming of differentiate service detachment in the front-back end over previous Web cluster system. On such a basis, the admission control policy for differentiate service based on the transaction is applied in the model. The experiment result and theory analysis reveal that the system stability is effectively guaranteed in the case of system over-loading and a better service quality is allocated to the high-priority task. it is better than the ones based on the session.(2) A admission control algorithm combined with the self-similar access characteristics over networks is brought forward. It is likely capable of overcoming the defect of self-similar request with the adjustment to the multiple factor regarding the consideration into the request access characteristics and load condition as well as optimizing the multi-dimension objective in the Web cluster system. The experiments result provesthe availability of the algorithm that can fit the self-similar access characteristics over networks as well as the schedule optimization in multi-dimension QoS.(3) A QoS control algorithm of resource optimization is set forth with the essence of multi-dimension QoS objective control. According to different service demands different resources, a heuristic algorithm (LBBA for shore) that makes full use of system resources also balances system load is raised. The experiment proves that the algorithm can take a apparent decline in the computation complexity in the resource optimization allocation to fulfill the requirement of real-time scheduling. Given the foundation, a dual minimal balance differentiate service scheduling algorithm (DMBA for shore) of resource optimization is further brought up in the combination of multi-dimension resource QoS optimization and differentiate service as well.(4) A QoS scheduling policy based on the inaccurate state aiming at the inaccuracy of the Web cluster system status is held forth. It makes a modeling of system inaccurate status via probability analysis in the purpose of maximizing the value ratio implemented by system with the total consideration into the differentiate service as well as other QoS guidelines. The simulation shows that the algorithm is likely to overcome the inaccuracy of system status and is better than EDF counterpart in the performance. Also taking account the self-similar access transaction flow and differentiate service directing towards the high availability, a error-tolerated scheduling algorithm is held out with the guarantee of high availability for high priority task.(5) A Grid Web service structure model over WAN is proposed. On a basis of the OGSA framework, it separates the logic frame of Web service resource from physical resource structure. And it carries on the searching and locating for Web service resource via the Web service naming system(WSNS for short)in the form of hierarchy organization and maintaining for Web service resource by the area autonomy system (AAS for short). Then it makes an improvement on the structure model. On a basis of the maintaining the advantage of original structure, a resource organization tree (WSROT for short) is appended with the support of good characteristic for QoS guarantee. In conclusion, it is a much better organization structure model over the current Computational Grid.更多还原