资源约束网络中基于LSSVM的预测反馈调度

Least squares support vector machines based predictive feedback scheduling for resource-constrained networks

针对资源约束网络负载的动态变化,设计了一个基于最小二乘支持向量机(LSSVM)的反馈调度器.它可以周期性地监测网络资源,在线预测下一周期的可适用网络利用率,并根据预测值采用插值法得到控制回路的下一个采样周期,从而实现系统资源的动态分配.对采用固定带宽分配、基于LSSVM以及基于Elman神经网络的反馈调度进行了比较,结果表明,所提出的策略能使系统在可变负载情况下稳定运行,并在控制质量和网络服务质量之间取得平衡.

A feedback scheduler based on least squares support vector machines （LSSVM） is designed in order to attack the resource-constrained networks workload variations,which periodically monitors the network resources,predicates the available utilization for the next period,and adopts interpolated method to calculate the next sampling period from predicative value. Consequently,the system＇s resources are dynamically allocated by using this feedback scheduling mechanism. Three different strategies,which are fixed bandwidth allocation,LSSVM based feedback scheduling technique and Elman neural network based feedback scheduling technique,are compared respectively. The results of simulation show that the predictive feedback scheduling strategy can guarantee the stability of the system under flexible workload,and prove that the proposed strategy is an effective tradeoff method between the quality of control and service.