Multi-stage Event-driven Architecture Tuning Mechanism

With the rapid development of the Internet, more and more Internet business services have being deployed on the Web. This growing dependence upon Internet services underscores the importance of their availability and ability to handle large loads. At the same time, exponential growth of the Internet population is placing unprecedented demands upon the scalability and robustness of these services. Staged Event-Driven Architecture (SEDA) presents an architecture for handling the massive concurrency and load conditioning demands of busy Internet services, which can support the extreme concurrency and burst applications. Therefore SEDA has become one of the hot fields in Web server performance optimization.In this dissertation, we review the progress and important achievements in the research of Web server performance optimization, and illustrate the significance of SEDA for promoting the development of Web techniques. We point out the main problems of SEDA in practical application. On the one hand, SEDA applies stages to execute the blocking operation, but the asynchronous API is insufficient to synchronized Java I/O, which can block all the stages easily and cause the bottleneck of system performance; on the other hand, SEDA, without a powerful global management role in its performance adjustment mechanism, can not effectively coordinate the control policy between each stage. This dissertation presents a self-adaptive optimization mechanism which improves SEDA on performance adjustment. Related efforts result in following achievements:1) The design principle, systematic architecture and performance management of SEDA are deeply analyzed. On basis of the comparison of other server software architecture, we point out the advantage and disadvantage of SEDA concerning server performance optimization.2) A self-adaptive optimization mechanism is proposed. It adopts load-balance and feed-back control as the core techniques. From the aspect of stage performance adjustment, heuristic optimization algorithm and initiative queue management algorithm are combined to optimize the stage management policies, including batch control for adjusting the number of events carried by working thread, service degradation to avoid the block between stages, self-feedback performance parameter adjustment thread resources of stages. From the aspect of global management, admittance control is applied to avoid performance overload and load-balance clustering techniques are applied to coordinated thread resources and throughput performance between stage requests. The two optimization processes can be effectively combined by feed-back mechanism.3) The self-adaptive mechanism is applied to the SEDA server architecture. By the way of improving the SEDA systematic control model and stage factor model resource distribution algorithm, the system control and stage adjustment are greatly enhanced.4) Based on the self-adaptive mechanism of the SEDA server architecture, we develop a web server prototype system—OptiStage Server. Compared with other current common Web servers, the performance testing results show that, on specific circumstances—high concurrency and high load, the OptiStage Server is more efficient on self-adapting and has high throughput.