IaaS模式“云训练”资源预测-调度方法

Resource prediction-scheduling method in IaaS mode“cloud training”

基础设施即服务(infrastructure as a service,IaaS)模式"云训练"是基于IaaS云计算提出的武器装备系统模拟训练的模式,根据用户需求对训练资源进行预测调度是提高训练效果的重要保证。分析了"云训练"中用户任务、资源需求特点,采用阈值法进行预处理,通过动态权值系综模型得到预处理结果。在此基础上,提出基于减法-模糊聚类的模糊神经网络的资源需求预测方法(subtractive-fuzzy clustering based fuzzy neural network,SFCFNN),并引入自适应学习率和动量项以提升收敛速度和稳定性。调度器根据预测结果实现用户需求与资源之间的动态匹配。实验表明该方法可精确预测用户资源需求,实现资源动态调度,有效提高资源利用率与训练效果。

Infrastructure as a service （IaaS） mode cloud training is a kind of equipment simulated training mode developed from IaaS mode cloud computing. In IaaS mode cloud training, resource scheduling based on users＇ demands is an important prerequisite for improving training efficiency and effect. Characters of resource requirements are analyzed and the threshold method is adopted to preprocess the data. Then a self-adaptive pre- diction method using subtractive-fuzzy clustering based fuzzy neural network （SFCFNN） prediction method is proposed. Self-adjusting learning rate and momentum weight are introduced into the method to improve the con- vergence and stability. Based on prediction results, the scheduler allocates resources for users dynamically. Sta- tistics indexes are used for validity check and results show that the proposed method can be used for accurate re- source demands predicting and resource dynamic scheduling. The resource utilization rate and training effect are improved.