摘要
为解决AI并发数据流中计算节点负载失衡问题,本文利用数据流的周期性特征,提出了一种基于负载阈值的双模态负载均衡算法。在低负载场景下,采用基于优劣解距离法(Technique for Order Perference by similarity to Ideal Solution,TOPSIS)的静态分配算法,通过整合CPU利用率、内存占用率、网络带宽等关键指标,量化计算节点的静态性能,并依据此量化结果进行数据流分配,同时还确定了触发动态策略的负载转换阈值;在高负载场景下,启用基于改进遗传算法的动态优化算法:通过引入迭代状态自适应的概率操作机制,动态调整交叉概率与变异概率,构建了以计算节点集群的负载均衡度和平均总时延为优化目标的效用函数模型。实验结果表明,基于负载阈值的双模态负载均衡算法显著提升了计算节点间的负载均衡性,并优化了系统整体性能。
To address the problem of load imbalance among computing nodes in AI concurrent data streams,this paper proposes a dual-modal load balancing algorithm based on load thresholds,utilizing the periodic characteristics of data streams.The core design of this algorithm is as follows:in low-load scenarios,a static allocation algorithm based on TOPSIS(Technique for Order Preference by Similarity to Ideal Solution)is employed.This means integrating key indicators such as CPU utilization,memory usage,and network bandwidth to quantify the static performance of computing nodes,and data stream allocation is conducted based on this quantification.Thresholds for triggering dynamic strategies are also identified.In high-load scenarios,a dynamic optimization algorithm based on an improved genetic algorithm is activated.By introducing an iterative state adaptive probability operation mechanism,the crossover probability and mutation probability are dynamically adjusted,creating a utility function model with load balancing degree of computing node clusters and average total delay as optimization objectives.Experimental results indicate that the dual-modal load balancing algorithm based on load thresholds significantly enhances the load balancing among computing nodes and optimizes the overall system performance.
作者
吴宗泽
陈庆奎
WU Zongze;CHEN Qingkui(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)
出处
《智能计算机与应用》
2025年第12期1-8,共8页
Intelligent Computer and Applications
基金
国家自然科学基金(61572325)
上海重点科技攻关项目(19DZ1208903)。
关键词
AI数据流
负载均衡
静态性能
遗传算法
双模态算法
AI data flow
load balancing
static performance
genetic algorithm
dual-modal algorithm
