基于核密度估计的软实时任务QoS控制

针对软实时任务的服务质量（QoS）控制问题, 提出一种采用核密度估计预测控制的资源自适应调整方法。该方法首先结合资源预留策略建立软实时任务的服务质量模型, 并采用核密度估计方法, 在无法预知和假设具体作业执行时间分布的前提下通过滑动窗口样本预测当前作业的执行时间概率分布, 反馈控制器基于预测值和上一个作业完成时获得的服务质量动态调整资源以保证当前作业服务质量。仿真实验结果表明, 采用核密度估计的预测控制能有效保证任务服务质量, 在大的系统扰动下能稳定收敛。

QoS control based on kernel density estimation in soft real-time task

This paper proposed an adaptive resource predictive control based on kernel density estimation to deal with soft real-time tasks quality of service (QoS) control problem. First, the method combined reservation-based (RB) policy to establish mathematical model for evaluating QoS of soft real-time tasks. Then, kernel density estimation method predicted the probability distribution of the current job execution time through the sliding window samples. It replaced taking job execution time probability distribution as a prior. The controller took the QoS of last job as input, and adjusted the resource allocation dynamically by the predictive value of the current job execution time. Simulation results show that the predictive control based on kernel density estimation can effectively guarantee the QoS of tasks, and it remains stable convergence in large system disturbances.