摘要
针对PaaS平台下资源离散性过强、数据量过大导致的分配不均问题,提出了一种基于多元回归的自动化分配算法实现有效解决。统计所有资源数据样本,利用多元回归算法计算观测序列中残差平方和最小的样本集,代入损失函数中求出损失差值,得到最小损失值,根据梯度下降规律寻找集合中梯度值最优样本,以该样本作为分配参照。设立一套内容为分配与参考样本值相关度最高的资源数据约束条件,不断迭代计算直至求得所有符合约束条件的资源,将符合条件的为一组分配,剩余为另一组分配。根据资源的时间和位数的分布序列,不断实施离散捕捉查找残留资源,再实施二次分配。仿真实验证明,所提方法分配后PaaS平台资源吞吐量增加,消耗代价减少,算法分配时间较低,整体实用能力强。
Aiming at the uneven distribution of resources caused by excessive discreteness of resources and large amount of data under the PaaS platform,an automatic allocation algorithm based on multiple regression is proposed to effectively solve the problem.Count all resource data samples,use the multiple regression algorithm to calculate the sample set with the smallest residual sum of squares in the observation sequence,and substitute it into the loss function to obtain the loss difference,obtain the minimum loss value,and find the optimal sample of the gradient value in the set according to the gradient descent law,using this sample as the allocation reference.Set up a set of resource data constraints for allocating the highest correlation with the reference sample value,and iteratively calculate until all the resources that meet the constraints are obtained.The qualified ones are allocated to one group,and the rest are allocated to another group.According to the time and bit distribution sequence of resources,discrete capture is continuously implemented to find residual resources,and then secondary allocation is implemented.Simulation experiments show that the PaaS platform resource throughput is increased after the proposed method is allocated,the consumption cost is reduced,the algorithm allocation time is low,and the overall practical ability is strong.
作者
刘鲤君
丁红
祁鸿燕
杜丽华
宋飞
LIU Li-jun;DING Hong;QI Hong-yan;DU Li-hua;SONG Fei(Qinghai Test and Calculation Center Co.,Ltd.,Xining,Qinghai 810001,China)
出处
《计算技术与自动化》
2023年第2期168-172,共5页
Computing Technology and Automation
基金
青海省科技大数据PaaS平台建设资助项目(2019-ZJ-T07)。
关键词
多元回归算法
梯度下降规律
损失函数
约束条件
分配参照样本
multiple regression algorithm
gradient descent law
loss function
constraints
allocate reference samples
