【作者】 周伟；

【导师】 王进；

【作者基本信息】 苏州大学 ， 软件工程（专业学位）， 2020， 硕士

【摘要】 近年来,随着物联网和5G通信技术的飞速发展,世界进入万物互联时代,网络边缘的数据量呈爆炸式增长。以云计算为核心的集中式数据处理方式不能有效满足新时代下的数据增长与服务需求,利用靠近数据产生终端的计算资源对数据进行就近处理的边缘计算应运而生。然而,由于边缘计算的内容感知、多源异构、实时计算、终端资源受限等特性,使得数据安全和隐私保护问题日益严峻。本文主要研究了基于线性编码的边缘计算场景下的隐私保护策略。在“云-边-端”三层体系边缘计算模型中,以矩阵乘法作为代表性计算任务,设计基于线性编码的存储分配和编码方案,以保护计算双方数据的安全性和隐私性。主要工作有如下:1、在边缘计算模型中,针对宽矩阵的运算,本文研究了基于正交向量的高效安全的编码方案。通过添加随机编码块,保证了计算双方数据的信息论安全需求;随机块的添加,增加了通信负载和解码复杂度,为了减少通信负载和解码复杂度,本文提出了基于正交向量的安全编码方案,利用编码矩阵的零空间中向量的正交特性对用户端添加的随机块进行一定的设计;而所添加随机块的随机性下降导致安全性的降低,在基础方案上,本文提出了两个改进方案以解决用户数据的安全性降低问题。2、在边缘计算模型中,针对窄矩阵的运算,本文研究了基于本地多次解码的最小通信编码方案。一方面,通过线性编码保证数据矩阵的弱安全;另一方面,基于方案的按列编码的特性,本文提出了本地多次解码方案以找到有最小通信的解以削减通信负载,由于本地数据无需上传,直接保证了数据的隐私性,且本方案有效降低了通信量和边缘节点的计算量,进而能够有效降低服务时间。实验结果表明,本文提出的两种方案解决了不同类型的矩阵乘法运算的数据安全和隐私性问题,减少了通信负载,提高了计算效率,对边缘计算的隐私保护策略研究具有重要意义。更多还原

【Abstract】 In recent years,with the rapid development of Internet of things(IoT)and 5G network,the world has stepped in the era of Internet of everything(IoE).The amount of data on the edge of the network is increasing explosively,so that the centralized data processing method of cloud computing cannot effectively meet the data growth rate and service demand in the new era.Edge computing(EC)is generated in this context,it uses the computing resources close to the data generating terminal to process the data nearby.However,due to the characteristics of edge computing such as content awareness,multi-source heterogeneity,real-time computing and limited terminal resources,data security and privacy protection issues in edge computing are becoming increasingly serious.The thesis focuses on the study of the privacy protection strategy in edge computing scenarios based on linear coding.Specifically,in the "cloud-edge-user" three-tier edge computing model,matrix multiplication is used as a representative computing task,linear coding-based storage allocation and coding schemes are designed to protect the security and privacy of the both sides in the computing.The main tasks are as follows:1、In EC model,for the operation of a wide matrix,the efficient and secure coding scheme based on orthogonal vectors is studied.By adding random coding blocks,the data can meet the requirement of information theoretical security(ITS).The addition of random blocks increases the communication overhead and decoding complexity.In order to figure out this issue,an orthogonal vector based secure code(OVSC)scheme is proposed,which uses the orthogonality of the vectors in the nullspace of the coded matrix to design the random block added on the user side.The decrease in randomness of the added random block leads to a decrease in security,two improved schemes are proposed to solve the problem.2、In EC model,for the operation of a narrow matrix,the minimum communication secure coded edge computing scheme based on local multiple decoding is studied.On the one hand,weakly security(WS)of data matrix is guaranteed by linear coding;on the other hand,based on the characteristics of the scheme coded by column,a multiple decoding without upload scheme(MDw/oU)is proposed to find the solution with minimal communication to reduce the communication overhead.Since the local data,i.e.,the input vector does not need to be uploaded,the privacy of user data is directly guaranteed.Additionally,the scheme can effectively reduce the communication overhead and the computation amount of edge nodes,it can effectively reduce the service time.The experimental results show that the two schemes proposed in the thesis solve the security and privacy problem in matrix multiplication with different types.Additionally,they reduce the communication overhead and improve the computation efficiency.Our work has important significance for the research on privacy protection strategies in edge computing.更多还原