“东数西算”下的高效数据流通策略研究

doi:10.11871/jfdc.issn.2096-742X.2023.05.001

数据与计算发展前沿 ›› 2023, Vol. 5 ›› Issue (5): 3-12.

CSTR: 32002.14.jfdc.CN10-1649/TP.2023.05.001

doi: 10.11871/jfdc.issn.2096-742X.2023.05.001

• 专刊:数据要素安全高效流通的关键技术 • 上一篇下一篇

“东数西算”下的高效数据流通策略研究

沈林江^*(),仇树卿,崔超,许俊东,李兆滨,耿晓巧

浪潮通信信息系统有限公司，算力网络研究院，山东济南 250100

收稿日期:2023-04-26 出版日期:2023-10-20 发布日期:2023-10-31
通讯作者: *沈林江（E-mail: shenlj@inspur.com）
作者简介:作者简介：沈林江，浪潮通信信息系统有限公司，算力网络研究院院长，公司副总经理，主要从事算力网络相关前沿理论分析、技术研究和产品设计。
负责系统建设、优化方案设计以及论文总体方向与框架。
SHEN Linjiang is director of Computing Power Network Research Institute and deputy general manager of Inspur Communication Information System Co., Ltd. Currently, he is mainly engaged in cutting-edge theoretical analysis, technical research and product design related to computing power network.
In this paper, he is responsible for system construction, optimization scheme design, and the overall direction and framework of the paper.
E-mail: shenlj@inspur.com

Research on Efficient Data Flow Management under “East-West Computing Resource Transfer”

SHEN Linjiang^*(),QIU Shuqing,CUI Chao,XU Jundong,LI Zhaobin,GENG Xiaoqiao

Computing Power Network Research Institute, Inspur Communication Information System Co., Ltd., Jinan, Shandong 250100, China

Received:2023-04-26 Online:2023-10-20 Published:2023-10-31

摘要/Abstract

摘要：

【目的】 东数西算场景下的数据流通策略需要综合考虑能耗、成本、时延等各类系统优化目标，本文采用深度强化学习算法实现高效的数据分级和流转策略。【方法】 首先对东数西算场景下的数据要素流通系统架构和关键业务逻辑进行分析，并基于数据分类分级、数据传输、数据服务等关键流程的控制因素，融合多类业务优化目标，构建系统优化的通用数学模型，最后通过深度强化学习实现问题求解和策略优化。【结果】 通过系统仿真，与多种基线算法进行对比，验证了本文方法在收敛性、系统长期收益、目标均衡等方面的优势。【局限】 本文中方法对相关系统进行了合理的简化建模，在生产中需要基于现有方法，结合实际系统进行策略分析和回报函数定义，以进一步提升方法的推广能力和应用效果。【结论】 东数西算等复杂场景下，综合考虑多种业务优化目标，并通过深度强化学习等算法对数据流通策略进行优化，能够在保障服务质量的基础上，有效提升系统自身性能。

关键词: 东数西算, 数据流通, 深度强化学习

Abstract:

[Objective] The data element circulation strategy under the East-West Computing Resource Transfer” project needs to consider comprehensively various system optimization objectives, such as energy consumption, cost, and latency. This paper uses the deep reinforcement learning algorithm to implement an efficient data classification and circulation strategy. [Methods] Firstly, the data element circulation system architecture and key business logic under the “East-West Computing Resource Transfer” project are analyzed, and based on the control factors of key processes such as data classification and grading, data transmission, and data services, multiple types of business optimization goals are integrated to construct a general mathematical model for system optimization, and finally problem-solving and strategy optimization are realized through deep reinforcement learning. [Results] Through system simulation and comparison with multiple baseline algorithms, the advantages of the proposed method in convergence, long-term system benefit, and target equilibrium are verified. [Limitations] The proposed method in this paper simplifies the modeling of the relevant system. Further improvement can be achieved by combining the existing methods with practical system analysis and defining reward functions based on the actual system. [Conclusions] In complex scenarios like “East-West Computing Resource Transfer”, comprehensive consideration of multiple business optimization goals and optimizing data element circulation strategies through algorithms such as deep reinforcement learning can effectively improve the performance of the system while ensuring service quality.

Key words: East-West Computing Resource Transfer, data flow management, deep reinforcement learning

沈林江, 仇树卿, 崔超, 许俊东, 李兆滨, 耿晓巧. “东数西算”下的高效数据流通策略研究[J]. 数据与计算发展前沿, 2023, 5(5): 3-12.

SHEN Linjiang, QIU Shuqing, CUI Chao, XU Jundong, LI Zhaobin, GENG Xiaoqiao. Research on Efficient Data Flow Management under “East-West Computing Resource Transfer”[J]. Frontiers of Data and Computing, 2023, 5(5): 3-12, https://cstr.cn/32002.14.jfdc.CN10-1649/TP.2023.05.001.

图/表 10

图1

图2

表1

DQN算法流程"

DQN算法流程

1、初始化回放池、神经网络

Q

参数

θ

，定义环境空间

s

以及行为空间

a

、模型迭代周期

n

、模型更新步长

η

、折扣因子

υ

2、for each episode do:
3、初始化状态

s

4、 for each step do：
5、采集当前系统状态

s o l d

6、根据

ϵ

-greedy方法，生成策略

a

7、与系统交互，获得

r, s n e w

8、采用

< s o l d, a, r, s n e w >

更新回放池
9、 if step %

n

== 0 then
10、获取批量数据

< s o l d, a, r, s n e w >

11、更新模型参数：
12、

y = r + υ m a x a Q s n e w, a; θ

13、

θ ← θ + η ? y - Q s o l d, a; θ 2 / ? θ

14、 end for
15、end for

表1

表2

表3

仿真系统时延参数"

	$d s$ (MB)	东部	西部	传输
时延(s)	[0,16]	0.01	0.01	0.001
	(16,64]		[0.02,0.03]	0.001
	(64,100]		[0.15,0.25]	0.001

表3

图3

图4

图5

图6

图7

参考文献 27

[1]	梁芳, 佟恬, 马贺荣, 等. 东数西算下算力网络发展分析[J]. 信息通信技术与政策, 2022, 48(11): 79-83.
[2]	郭亮. “东数西算”之五大热点问题探讨[J]. 通信世界, 2022, 909(23): 40-42.
[3]	林成创, 檀童和, 曾桓涛, 等. “东数西算”赋能智慧交通行业发展[J]. 数据通信, 2023, 212(1): 24-27.
[4]	解云鹏, 马思聪, 田毅, 等. 从“东数西算”甘肃节点看中国电信的算力调度探索与实践[J]. 通信世界, 2022, 908(22): 34-37.
[5]	周勇, 郭转转, 曾沁, 等. 论“东数西算”对气象行业的影响[J]. 数据与计算发展前沿, 2022, 4(5): 42-49.
[6]	钟瑛. 推动“东数西算”实现数据要素跨区域融通[J]. 北京观察, 2023, 389(3): 26.
[7]	周成祖, 吴文, 蔡晓强. 基于分类分级的数据安全防控策略研究[J]. 数据与计算发展前沿, 2023, 5(1): 128-135.
[8]	陈其宇, 许海兰. 大数据环境下数据安全传输与存储分析与研究[J]. 信息系统工程, 2023, 353(5):129-131.
[9]	陈良臣. 大数据存储安全的关键技术研究[J]. 集成电路应用, 2021, 38(11): 46-47.
[10]	郭丽. 工业互联网数据传输安全问题及改进策略研究[J]. 现代工业经济和信息化, 2022, 12(7): 117-119.
[11]	CETOLA S. A Method for Comparative Analysis of Trusted Execution Environments[D]. Portland: Portland State University, 2021.
[12]	ZHANG X, KANG Y, CHEN K, et al. Trading Off Privacy, Utility and Efficiency in Federated Learning[J/OL]. [2022-09-01]. ArXiv, abs/2209. 00230. https://arxiv.org/abs/2209.00230.
[13]	张燕, 杨一帆, 伊人, 等. 隐私计算场景下数据质量治理探索与实践[J]. 大数据, 2022, 8(5): 55-73. doi: 10.11959/j.issn.2096-0271.2022073
[14]	ZHAN Y, ZHANG J, HONG Z, et al. A Survey of Incentive Mechanism Design for Federated Learning[J]. IEEE Transactions on Emerging Topics in Computing, 2022, 10(2): 1035-1044.
[15]	LIM W Y B, LUONG N C, HOANG D T, et al. Federated Learning in Mobile Edge Networks: A Comprehensive Survey[J]. IEEE Communications Surveys & Tutorials, 2020, 22(3): 2031-2063.
[16]	SERGEI A, BOHDAN T, FRANZ G, et al. SCONE: secure Linux containers with Intel SGX[C]. Proceedings of the 12th USENIX conference on Operating Systems Design and Implementation (OSDI'16), USENIX Association, USA, 2016: 689-703.
[17]	SUZAKI K, NAKAJIMA K, OI T, et al. TS-Perf: General Performance Measurement of Trusted Execution Environment and Rich Execution Environment on Intel SGX, Arm TrustZone, and RISC-V Keystone[J]. IEEE Access, 2021, 9(1): 133520-133530. doi: 10.1109/ACCESS.2021.3112202
[18]	左亚兵, 王凯, 杨帆, 等. 基于用户偏好的协作内容缓存策略[J]. 计算机应用研究, 2022, 39(1): 123-127.
[19]	CAI Y, LLORCA J, TULINO M A, et al. Joint Compute-Caching-Communication Control for Online Data-Intensive Service Delivery[J/OL]. [2022-09-01].ArXiv, abs/2205.01944. https://arxiv.org/abs/2205.01944.
[20]	BEN-AMEUR A, ARALDO A, CHAHED T. Cache Allocation in Multi-Tenant Edge Computing via online Reinforcement Learning[C]. ICC 2022-IEEE International Conference on Communications, Seoul, Korea, 2022: 859-864.
[21]	范灵俊, 郑卫城, 彭亚松, 等. “东数西算”工程落地的关键问题分析与对策建议[J]. 中国信息安全, 2023, 159(2): 100-104.
[22]	ALE L, ZHANG N, FANG X, et al. Delay-Aware and Energy-Efficient Computation Offloading in Mobile-Edge Computing Using Deep Reinforcement Learning[J]. IEEE Transactions on Cognitive Communications and Networking, 2021, 7(3): 881-892. doi: 10.1109/TCCN.2021.3066619
[23]	ZHANG W, YU F, WANG X, et al. R²Fed: Resilient Reinforcement Federated Learning for Industrial Applications[J/OL]. IEEE Transactions on Industrial Informatics, 2022. doi:10.1109/TII.2022.3222369. https://ieeexplore.ieee.org/document/9950718.
[24]	MAO Y, ZHANG J, SONG S H, et al. Stochastic Joint Radio and Computational Resource Management for Multi-User Mobile-Edge Computing Systems[J]. IEEE Transactions on Wireless Communications, 2017, 16(9): 5994-6009. doi: 10.1109/TWC.2017.2717986
[25]	隆云滔, 王磊, 刘海波. 跨境数据流动治理规则研究[J]. 数据与计算发展前沿, 2023, 5(1): 74-84.
[26]	LILLICRAP T P, JONATHAN J H, ALEXANDER P, et al. Continuous control with deep reinforcement learning[J/OL]. [2015-09-09]. ArXiv,abs/1509.02971. https://arxiv.org/abs/1509.02971.
[27]	JOHN S, SERGEY L, PHILIPP M, et al. Trust region policy optimization[C]. Proceedings of the 32nd International Conference on International Conference on Machine Learning, 2015, 37(ICML'15): 1889-1897.

	东部	西部	传输
能耗(10^-7J)	[0.04,0.06]	[0.015,0.025]	0.01
成本(元)	[0.05,0.08]	[0.01,0.02]	0.01

“东数西算”下的高效数据流通策略研究

Research on Efficient Data Flow Management under “East-West Computing Resource Transfer”

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 10

参考文献 27

相关文章 5

编辑推荐

Metrics

本文评价

[1]	方肖, 胡正银, 韩锐, 郑亮. 科学数据“东数西算”组织机制与传输模式研究[J]. 数据与计算发展前沿, 2023, 5(5): 13-22.
[2]	贾晓丰, 高嵩, 周琰, 薛钦亮, 范举. 一种面向超大城市治理的数据高效跨域流通技术框架[J]. 数据与计算发展前沿, 2023, 5(5): 35-45.
[3]	栗蔚,王雨萌,立言,赵伟博,苏越. “东数西算”背景下算力服务对算力经济发展影响分析[J]. 数据与计算发展前沿, 2022, 4(6): 13-19.
[4]	周勇,郭转转,曾沁,郝伊一. 论“东数西算”对气象行业的影响[J]. 数据与计算发展前沿, 2022, 4(5): 42-49.
[5]	杨昕,沈文海. “东数西算”趋势下的气象算力网络和算力服务体系架构[J]. 数据与计算发展前沿, 2022, 4(5): 50-59.