面向自主计算的存算传融合架构及技术挑战  

Cache-computation-transmission integration for automatic computing:architecture and technology challenges

作　　者：张珊 李响 李西烁 王志远 罗洪斌 Shan ZHANG;Xiang LI;Xishuo LI;Zhiyuan WANG;Hongbin LUO(School of Computer Science and Engineering,Beihang University,Beijing 100191,China;State Key Laboratory of Complex&Critical Software Environment,Beihang University,Beijing 100191,China;School of Cyber Science and Technology,Beihang University,Beijing 100191,China)

机构地区：[1]北京航空航天大学计算机学院,北京100191 [2]北京航空航天大学软件开发环境国家重点实验室,北京100191 [3]北京航空航天大学网络空间安全学院,北京100191

出　　处：《中国科学:信息科学》2025年第3期500-515,共16页Scientia Sinica(Informationis)

基　　金：国家重点研发计划(批准号:2022YFB4501000)资助项目。

摘　　要：传统云或边缘计算模式下,数据的存储、计算和传输分离:终端负责指定具体的计算和关联存储节点,网络仅在这些节点间提供传输路径而并不感知所承载的计算任务.这种模式不仅导致海量异构存算平台难以感知识别彼此的可用资源并形成协同合力、数据存储与计算孤岛化现象严重,还面临拓扑时变、计算节点失效等不确定性导致的任务执行时间长甚至中断等挑战.为此,本文提出一种面向自主计算的存算传融合网络架构,通过构建耦合但差异化管理存算传多维资源的控制面,以及支持形式化计算任务路由和调度的数据面,赋能自主计算的全流程实现.基于所提架构,提出了多维资源状态探测、任务联合调度与服务协同部署方法,实现任务需求拟合与环境适变的高效自主计算.此外,本文还探讨了该架构下的挑战以及可能的未来研究方向.Caching,computing,and communication are separately organized under the conventional cloud and edge computing.Specifically,end devices are responsible for specifying specific computation and associated storage nodes,while the network mainly provides transmission paths between these nodes without awareness of the computing tasks.In this regard,heterogeneous caching-computing platforms cannot discover and coordinate available resources,leading to prolonged or even interrupted task execution when dealing with dynamic network topologies and node failures.To address these challenges,we propose a caching-computing-communication integrated network architecture for automatic computing.The main design aspects include an integrated yet differentiated control plane to manage caching,computing,and communication resources,and a data plane supporting formalized computation task routing and scheduling.Furthermore,we explore methods for multidimensional resource state detection,joint task scheduling,and coordinated service deployment to enable efficient automatic computing that dynamically adapts to task demands and environmental changes.The related open research issues are also discussed.

关 键 词：自主计算 存算传融合 网络架构 

分 类 号：TP30[自动化与计算机技术—计算机系统结构]