多核Web服务器自适应动态请求调度算法

Self-adaptive Dynamic Requests Scheduling Algorithm in Multi-core Web Server

为了提高Web服务器处理用户请求的能力,大多数Web服务器已经采用多核处理器.但传统的动态请求调度算法,例如先到先服务算法,既没有考虑多核处理器的结构特征也没有考虑动态请求服务时间的分布,因此不能充分利用多核处理器的运算能力.为解决此问题,在之前工作的基础上提出一种新的动态请求调度算法.新算法可周期性地预测表征负载状态的动态请求队列的权值,根据预测的权值确定动态请求分配方案,并依据此方案对到达Web服务器的动态请求进行调度.为评估此算法的性能,以此算法为基础进行了仿真实验.实验结果表明,新的算法不仅能解决多核Web服务器中存在的"乒乓"效应等影响性能发挥的问题,而且能在较长时间内保持处理器核心间的负载均衡.

In order to improve performance of handling user requests,most of web servers have adopted multi-core processors in past years. However, traditional dynamic request scheduling algorithms, such as First-Come-First-Served algorithm, don＇t consider the archi- tecture characteristics of multi-core processors and the distribution of the dynamic requests＇ service time. Therefore, the scheduling algorithms couldn＇t fully exploit the performance of multi-core processors. To address this problem,a new algorithm is proposed in this paper which is basis of our previous works. The new algorithm could periodically predict the weight of dynamic request queue which indicates the impact on CPU load. According to the predicted weight, the new dynamic request assignment solution is determined, by which dynamic requests are scheduled. Simulation experiments based on the new algorithm have been done to evaluate the new algorithm. The simulation experiment results show that the proposed algorithm could not only avoid the problem of ping-pong effect in multi-core systems, but also keep the load balance between cores in long time period.