算力网络中高效算力资源度量方法  

Efficient Computing Resource Metric Method in Computing-First Network

作　　者：柴若楠 郜帅[1] 兰江雨 刘宁春 Chai Ruonan;Gao Shuai;Lan Jiangyu;Liu Ningchun(School of Electronic and Information Engineering,Beijing Jiaotong University,Beijing 100044)

机构地区：[1]北京交通大学电子信息工程学院,北京100044

出　　处：《计算机研究与发展》2023年第4期763-771,共9页Journal of Computer Research and Development

基　　金：国家重点研发计划项目(2022YFB2901900);国家自然科学基金项目(61972026,61802014)。

摘　　要：随着新型网络业务的不断发展和对算力需求的不断提高,算力网络技术逐渐走进人们的视野并不断发展壮大.而算力度量,作为度量各类算力平台中计算和存储能力的方法,在算力网络业务感知和算力资源高效调度中扮演着重要的角色.目前算力度量的研究尚处于起步阶段,已有的度量方法相对单一,只考虑了部分静态或动态指标,难以保证算力资源利用率和算力资源匹配准确率.设计了一种先静后动的混合式度量方法(hybrid metric method,HMM),该方法结合静态和动态指标来度量算力资源,考虑了算力节点的基础性能及其动态工作状态的变化,并且在静动态的度量指标的选取上也进行了全面的考量.通过实验和数据分析证明,所提度量方法HMM能有效提升算力资源利用率和算力资源匹配准确率.With the continuous development of new network services and the increasing demand for computing,computing-first network(CFN)has attracted people's attention and is gradually developing.As a method to measure the computing and storage capacity of various computing platforms,the computing resource metric plays an important role in achieving user awareness and efficient scheduling of computing resources in CFN.At present,the research on computing resource metrics is in its infancy.Most of those that only consider some static or dynamic indicators are relatively simple,which cannot guarantee the utilization of computing resources and the precision of matching computing resources.In this study,we design a hybrid metric method(HMM),which combines static and dynamic indicators to measure computing resources.This method takes the basic performance of the computing nodes and the dynamic changes in their working state into account.In addition,we also consider lots of static and dynamic indicators to enhance the comprehensiveness of HMM.The experiments and a large number of data analyses show that the metric method we propose has good improvement in the utilization of computing resources and the precision of matching computing resources.

关 键 词：算力网络 算力资源 算力度量 混合式度量方法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]