Web服务器系统中基于反馈控制的比例延迟保证

随着Web应用在商业领域的广泛使用，Web服务器系统需要在高负载下提供区分服务，以满足用户的不同需求。为实现以延迟作为评价指标的区分服务，本文在Web服务器系统的连接管理和请求处理两个层次建立了基于反馈控制的比例延迟保证模型。模型中的反馈控制器通过动态计算和调节不同类别客户占用的资源：（服务线程和数据库连接），能保证高优先级的客户较快得到服务而不同类别客户的平均延迟比保持不变。为测试闭环系统的性能，设计了两种分别服从均匀分布和重尾分布的动态负载。仿真结果表明，即使并发客户连接的数目剧烈变化，控制器作用下的服务器系统仍然能够达到较好的比例延迟保证，可靠地为用户提供区分服务。

Feedback Control-based Proportional Delay Guarantees in Web Server Systems

As more business applications become Web enabled, Web server systems evolve to provide service differenti- ation to satisfy different requirements of clients under heavy load conditions. To implement service differentiation with respect to delays in the Web server system, in this paper, proportional delay guarantee models based on feedback con- trol are constructed at connection management level and request processing level. Controllers in the models can dynami- cally calculate and adjust the amount of resources, the worker threads and the database connections, for different types of clients according to their proportional delay requirements. Two kinds of workloads, which follow uniform and heavy- tailed distributions respectively, are generated for simulation of the close-loop system. Experiment results indicate that, proportional delay guarantees can be achieved in the Web server system successfully even if the number of the concurrent clients is changed abruptly under different kinds of workloads.