CN111125752A - Privacy protection method in edge computing environment - Google Patents

Abstract

The invention discloses a privacy protection method in an edge computing environment, which is implemented by security coding based on orthogonal vectors and comprises the steps of obtaining parameters of a network where security coding edge computing based on the orthogonal vectors is located; partitioning input data of one side of a calculation task by using a task allocation algorithm, adding random block codes, and allocating the data to corresponding edge nodes; calculating a base vector of a null space vector of the coding matrix, and sending a part of the base vector of the null space to the user equipment according to the parameter A; generating an online computing requirement after a user has input data; and after receiving all the intermediate results returned by the edge nodes, the user utilizes the coefficient coding matrix to decode to obtain the required result. In the edge calculation based on the safety coding, the privacy of both sides of calculation is protected by adopting linear coding and adding random blocks, and meanwhile, the random blocks added by a user side are designed by utilizing the orthogonality of zero space vectors, so that the communication load and the decoding complexity are reduced, and the edge calculation privacy protection with low delay, high efficiency and safety is obtained.

Description

Privacy protection method in edge computing environment

Technical Field

The invention relates to the field of distributed edge computing, in particular to a privacy protection method realized by improving a security coding scheme in an edge computing environment.

Background

In recent years, with the rapid development of the internet of things and the popularization of 4G/5G networks, the era of everything interconnection has come, and the explosive increase of network edge devices makes traditional centralized big data processing taking cloud as a core unable to efficiently process data generated by the edge devices. Therefore, edge-based big data processing with an edge computing model as a core is produced and becomes a new research hotspot of the next generation network. Briefly, edge computing is the extension of cloud computing to the edge, and compared with the massive computing capability of a cloud end caused by classical cloud computing, the edge computing realizes the sinking of resources and services to the edge position, so that the interaction time delay can be reduced, the network burden is reduced, the service types are enriched, the service processing is optimized, and the service quality and the user experience are improved. In particular, edge computing utilizes network resources closer to the user or end device, such as the devices participating in the computing may be routers, servers, or even end devices. The edge computing is one of distributed computing, and for different computing tasks, edge nodes can realize mutual cooperation and parallel processing.

However, the computing resources in edge computing may have different kinds of devices, such as computers, internet of things devices, edge servers provided by different service providers, and the like. The heterogeneous nature of the computing resources in edge computing poses two challenges to edge computing. First, some edge nodes may become stragglers because edge compute nodes are typically not as capable as servers in a data center, and they may not be able to return compute results in a timely manner; secondly, in the edge computing, the process of migrating the computing process from the user or the cloud to the edge nodes brings great potential safety hazard to the security of the computing data, and the credibility of the edge nodes is uneven, so that how to ensure the privacy of the computing data becomes important.

Network coding is the concept first proposed in the paper Network Information Flow published on Information theory by IEEE Transactions on Information theory in r.a, n.c, et al, 2000. Network coding relates to the relevant fields of graph theory, information theory, coding theory and the like, and is mainly used in a multicast network. The network coding is an information exchange technology combining routing and coding, and the core idea is that information received on each channel is processed linearly or nonlinearly at each node in the network and then forwarded to downstream nodes, and intermediate nodes play the role of encoders or signal processors. According to the max-min cut theorem, the maximum transmission rate between the sender and the receiver of data cannot exceed the minimum cut (or the maximum flow) between the two parties, and the conventional multicast routing method generally cannot reach the upper bound. The Network Coding can be divided into two types according to the difference of the data Coding modes, one is Linear Network Coding (Linear Network Coding), and the other is Non-Linear Network Coding (Non-Linear Network Coding). Linear coding is simply a method of performing linear operation on a data set, i.e. coding, then transmitting, storing or calculating, and then decoding to obtain the required data. Compared with nonlinear network coding, linear coding is more common due to simplicity and high efficiency of coding and decoding.

Linear Network Coding (LNC) is a promising technology for protecting data confidentiality and reducing transmission delay. Based on linear coding, Wang et al propose using coding to guarantee Information Theoretic Security (ITS) and Weak Security (WS) in data transmission. Li et al propose a Distributed Computing coding scheme (CDC) to achieve a balance of Computing and communication loads.

Secure Coded Distributed Computing (SCDC) is studied in many documents of computer science and machine learning. P. Mohassel et al propose a security outsourcing computation scheme based on homomorphic encryption, and a. Castiglione et al propose a security key generation scheme. Bitar et al propose the use of a ladder-type coding scheme in the field of edge-computed coding to reduce the computation delay. However, most documents only consider the security of one party, m, J, et al consider the secure distributed matrix multiplication of confidential input, but do not consider the effect of dequeuing, and h, Yang, et al consider the use of Polynomial Codes (PCs) to solve the confidentiality and dequeuing effects of both parties of input, but bring a lot of extra communication load and decoding complexity.

Therefore, there is a need to provide a new method of privacy protection in edge computing environments.

Disclosure of Invention

The invention aims to provide a privacy protection method in an edge computing environment, which optimizes a coding scheme to reduce the total cost on the premise of ensuring the safety of data input by two computing tasks, thereby realizing the optimization of the safe coding edge computing based on orthogonal vectors which is not proposed at present.

In order to achieve the purpose of the invention, the technical scheme adopted by the invention is as follows: a privacy protection method in an edge computing environment is realized by security coding based on orthogonal vectors, and comprises the following steps:

a: obtaining parameters of a network where a security coding edge calculation based on an orthogonal vector is located, wherein the parameters comprise:

a set of edge device parameters;

calculating data input by one party of the task;

an adjustable safety parameter;

b: and (4) carrying out pretreatment operation according to the parameters in the A: b, partitioning input data of one side of the calculation task into blocks by using a task allocation algorithm according to the parameters obtained in the step A, adding random block codes into the blocks, and allocating the blocks to corresponding edge nodes;

calculating a base vector of a null space vector of the coding matrix, and sending a part of the base vector of the null space to the user equipment according to the parameter A;

c: the user generates an online computing requirement after inputting data: after receiving the corresponding parameters, the user equipment adds random data and a zero space basis vector stored by the user equipment for coding, and then sends the coded data to the edge equipment; after receiving input parameters of a user, the edge equipment participating in calculation carries out operation on the input parameters and the stored coded data and sends an intermediate calculation result to the user;

d: and after receiving all the intermediate results returned by the edge node, the user decodes the intermediate results by using the coefficient coding matrix generated in the B to obtain the required results.

In a preferred technical scheme, the parameters in the step A are as follows:

a1: edge device set W = { W ₁ ，w ₂ ，…,w _p }；

A2; number of edge devicesp；

A3: calculating task sizem,n；

A4: number of blocks of data matrixk；

A5: data matrix blockA；

A6: input vectorx；

A7: computing tasks for a userAx；

A8: safety regulation parameters；

A9: a finite field size q;

wherein the data matrix blockARepresenting an original matrix block, which is an m × n matrix; safety regulation parametersThe safety degree of the system to the user input vector is adjusted under the control of the cloud; all operations of the system are carried out within the range of the finite field q.

In the above technical solution, the preprocessing operation in step B specifically is:

b1: the cloud divides the input data into data blocks with corresponding quantity according to corresponding input parameters, generates corresponding coefficient coding matrixes according to the number of edge nodes and the number of original calculation data blocks, then codes the original calculation data, and finally sends the coded data to edge equipment participating in calculation;

b2: and calculating the base vector of the null space vector of the coding matrix, and sending part of the base vector of the null space to the user equipment according to the calculation task scale.

Due to the application of the technical scheme, compared with the prior art, the invention has the following advantages:

1. the invention considers the common safety problem of both sides of the calculation task in the edge calculation, and simultaneously utilizes the orthogonal characteristic of the null space to particularly design the coding block of the user-side data, and aims to reduce the calculation delay, the communication delay and the decoding complexity on the premise of ensuring the data safety privacy and completing the calculation task, thereby finally achieving the purpose of submitting the completion speed of the calculation task.

2. The method provided by the invention provides a privacy protection strategy based on the safe coding edge calculation of the orthogonal vector, and can effectively reduce the communication load and the decoding complexity.

Drawings

FIG. 1 is a flow chart of a privacy preserving strategy for orthogonal vector based secure code edge computation according to an embodiment of the present invention;

FIG. 2 is an example of a Secure Coded Edge Computing (SCEC) model;

fig. 3 to 6 are graphs showing the results of data evaluation by changing different parameters according to the embodiment of the present invention.

Detailed Description

The invention is further described with reference to the following figures and examples:

example (b): a computing framework of safe coding edge computing based on orthogonal vectors is provided, and the method reduces communication load and complexity of user equipment decoding by specially designing a coding scheme of input data at a user terminal on the basis of ensuring data privacy and completing computing tasks.

Referring to fig. 1, a flow chart for constructing a security code based edge calculation is shown.

The specific method comprises the following steps:

a: obtaining parameters of a network where the security code-based edge calculation is located,

the parameters include:

a1: edge device set W = { W ₁ ，w ₂ ，…,w _p }；

A2; number of edge devicesp；

A3: calculating task sizem,n；

A4: division of data matrixNumber of blocksk；

A5: data matrix blockA；

A6: input vectorx；

A7: computing tasks for a userAx；

A8: safety regulation parameters；

A9: a finite field size q;

where A1 and A2 are parameters known to the system and indicate that there are p edge devices in the system, for example, the ith edge device is called w _i A3 represents the scale of calculation task, and the number of rows and columns of matrix A in the parameters shown in A5 under the scene, A4 represents the number of blocks of the original matrix and is set by cloud, A5 represents the original matrix block which is an m × n matrix and is completely stored in the cloud in advance, and if m is assumed to be<<n, A6 is data input by user equipment and is an n multiplied by 1 input vector, A7 represents a calculation task target of the system, and a user hopes to obtain a return result Ax after inputting an x vector; a8 represents an adjustable security parameter, which can be controlled by the cloud, and adjusts the security degree of the system to the user input vector; a9 represents the size of finite field, that is, all operations of the system are performed within the range of the finite field q.

B: and performing pretreatment operation according to the parameters in the A. And B, partitioning the input data of one side of the calculation task by using a task allocation algorithm according to the parameters obtained in the step A, adding random block codes into the data, and allocating the data to corresponding edge nodes. Specifically, the cloud divides an input data matrix into a corresponding number of data blocks according to corresponding input parameters, generates a corresponding coefficient coding matrix according to the number of edge nodes and the number of original calculation data blocks, codes the original calculation data, and finally sends the coded data to edge devices participating in calculation. And calculating the base vector of the null space vector of the coding matrix, and sending part of the base vector of the null space to the user equipment according to the parameter A. The specific method comprises the following steps:

b1: fig. 3 shows the impact of the number of random data blocks participating in encoding and the overall cost of different task allocation schemes. In a system with six edge devices, the data matrix is divided into 2 blocks according to row average{A ₁ ,A ₂ }，If the average score cannot be obtained, the last block can adopt a zero padding mode. The cloud end encodes the random block of the A code to be 1, and the number of vectors sent to the user end after encoding is 1.

Firstly, a data matrix block A is stored in the cloud in advance, and the cloud firstly equally divides the matrix A into 2 blocks according to rows{A ₁ ,A ₂ }，If the average division cannot be realized, the last block can adopt a zero padding mode, so that the method can be used for solving the problem of low cost

And

is of a size of

Second, to protect the security of data A, the cloud generates a key over the finite field q

Random coding block

The cloud encodes and distributes A to the system according to the corresponding encoding coefficient

-

An edge device, an ith edge device

Data stored thereon is

E.g. of

Upper memory

，

Upper memory

The specific encoding process is shown in the upper right corner of fig. 2.

Secondly, in order to protect the security of the input vector x, a random vector needs to be added

To encode x. In order to not increase the traffic and decoding complexity of on-line computation and further reduce the total service time, make it and the encoded data

Is linearly related to the null space basis vector of (1), thus

Is characterized by being

It possesses a special property:

is an orthogonal vector of any edge device memory data block, i.e.

. When x is input, x is encoded

Then will be

Broadcast to all edge devices, each edge device receiving

Starting parallel computation later, e.g. ith edge device

Receive to

Post-calculation

. Since any 3 rows of the coding coefficients B are linearly independent, after the calculation results of any 3 edge nodes are returned to the user equipment, the coding coefficients of the three nodes can form a full rank matrix, and after the user combines all intermediate results, the desired result Ax is obtained by left-multiplying the inverse of the coding coefficients, the specific process is shown in the lower right corner of fig. 2.

Defining the upload traffic Lu as the number of different packets sent to the edge device Lu =1, the download traffic Ld as the number of the smallest packets to be downloaded to obtain the final result Ld =3, the decoding complexity Dc as the number of multiplications to be performed for decoding, specifically as the operation of inverting a (k +1) -dimensional matrix and the operation of multiplying a (k +1) -dimensional matrix and an m-dimensional vector, so Dc =3³+3m。

The method shown in fig. 2 is named as Orthogonal Vector Secure Code (OVSC) scheme, which mainly includes three steps, and the algorithm flow is as follows:

1) pretreatment process (cloud): input device

(the data matrix) of the data,

p (number of edge nodes)

k (number of blocks of A, k +1< p),

s (adjustable safety parameters),

q (size of finite field)

A is divided into k blocks according to row average, and each block has the size of

If the two blocks cannot be equally divided, the last block is filled with 0 and a block is generated

Random coding block of size

Thereby obtaining

；

Generating a coefficient matrix in the finite field q according to the values of p and k

(B is a Van der Waals matrix, where any k +1 row of B is linearly independent)

Cloud to ith edge device

Distributed coded blocks are

， (

Line i of B, the whole coding block is

) Storing one block on each edge, for p blocks;

computing

Of (2) null space

Transmitting s number of

Base vector

To the user equipment

(the null space is defined for a set as follows:

) Therefore, it is

；

2) And (3) calculating on line: user input x, expectation calculation result Ax

The user first constructs

I.e. by

In that

In (1), can obtain

；

User computation coding block

Broadcast to the edge devices;

edge device i calculation

And returned to the user equipment. (p edge devices parallel computation)

3) And (3) decoding calculation: the user receives (k +1) intermediate results.

Wherein

Is aA matrix of k +1 rows in B represents the coding coefficients of the earliest returned k +1 results. Since any k +1 row of B is linearly independent, it is thus possible to use a linear array

Is a full rank matrix.

The user decodes Ax as follows:

the edge computing network in this example is a local area network or a wireless network. In this example, a comparison was made of the performance of privacy protection policies in the computation of the security coding edge based on orthogonal vectors. The protocols involved in the experiment were as follows.

1.CECw/oSThis scheme does not take into account the safe time, total service time.

2. And the PC adopts polynomial coding to jointly consider and calculate the total service time when the data safety of the two parties is calculated.

3. OVSC, total service time when employing a secure coding scheme based on orthogonal vectors.

In the implementation experiment process, a control variable method is used, other variables are kept unchanged, a certain variable is changed, results are collected, data comparison is carried out on the total service time reached by three distributed coding schemes under the influence of different variables, and the main variables of the control variable method are as follows:

1, m: the number of rows in matrix a, the size of the computational tasks,

。

n: the number of columns in matrix a, the size of the calculation task,

。

3, k: the number of blocks of the matrix a,

the edge calculation load L =1/k is determined.

S, s: the number of null space basis vectors stored by the user equipment,

。

in fig. 3 to 6, OVSC is superior to PC scheme, close to CECw/oS scheme, and as the calculation scale increases, in fig. 3 and 4, the service time of OVSC scheme is always close to CECw/oS scheme, smaller than PC scheme, and when n reaches 15000 and m reaches 1500, the service time of OVSC scheme is reduced by 15% compared to PC scheme. In fig. 5, as the value of the number k of blocks increases, the download traffic of the three schemes also increases, but the OVSC scheme is still close to the CECw/oS scheme and smaller than the PC scheme, and when k reaches 10, the service time of the OVSC scheme is reduced by 34% compared with the PC scheme. In fig. 6, since only the OVSC scheme has the parameter s involved, the PC and CECw/oS schemes are not affected, however, a small increase in s does not result in a large increase in service time, since s is involved only in the encoding phase of the user, which can be typically completed within 20 ms. By s reaching 100, the OVSC scheme has 8% less service time than the PC scheme and about 8% higher service time than the CECw/oS scheme.

The method provided by the embodiment of the invention considers the privacy protection strategy based on the orthogonal vector in the secure coding edge calculation, and reduces a large amount of communication load and decoding complexity.

In addition, the method provided by the invention has good application value, can be used for guiding a privacy protection strategy based on safe linear coding in an edge computing environment, reduces communication load and decoding complexity, reduces total service time and brings lower delay response.

Claims (3)

1. A method of privacy protection in an edge computing environment, characterized by: the secure coding implementation based on orthogonal vectors comprises the following steps:

a: obtaining parameters of a network where a security coding edge calculation based on an orthogonal vector is located, wherein the parameters comprise:

a set of edge device parameters;

calculating data input by one party of the task;

an adjustable safety parameter;

b: and (4) carrying out pretreatment operation according to the parameters in the A: b, partitioning input data of one side of the calculation task into blocks by using a task allocation algorithm according to the parameters obtained in the step A, adding random block codes into the blocks, and allocating the blocks to corresponding edge nodes;

calculating a base vector of a null space vector of the coding matrix, and sending a part of the base vector of the null space to the user equipment according to the parameter A;

c: the user generates an online computing requirement after inputting data: after receiving the corresponding parameters, the user equipment adds random data and a zero space basis vector stored by the user equipment for coding, and then sends the coded data to the edge equipment; after receiving input parameters of a user, the edge equipment participating in calculation carries out operation on the input parameters and the stored coded data and sends an intermediate calculation result to the user;

d: and after receiving all the intermediate results returned by the edge node, the user decodes the intermediate results by using the coefficient coding matrix generated in the B to obtain the required results.

2. A method of privacy protection in an edge computing environment according to claim 1, wherein: the parameters in the step A are as follows:

a1: edge device set W = { W ₁ ，w ₂ ，…,w _p }；

A2; number of edge devicesp；

A3: calculating task sizem,n；

A4: number of blocks of data matrixk；

A5: data matrix blockA；

A6: input vectorx；

A7: computing tasks for a userAx；

A8: safety regulation parameters；

A9: a finite field size q;

wherein the data matrix blockARepresenting an original matrix block, which is an m × n matrix; safety regulation parametersThe safety degree of the system to the user input vector is adjusted under the control of the cloud; all operations of the system are carried out within the range of the finite field q.

3. A method of privacy protection in an edge computing environment according to claim 1, wherein: the pretreatment operation in the step B is specifically as follows:

b1: the cloud divides the input data into data blocks with corresponding quantity according to corresponding input parameters, generates corresponding coefficient coding matrixes according to the number of edge nodes and the number of original calculation data blocks, then codes the original calculation data, and finally sends the coded data to edge equipment participating in calculation;

b2: and calculating the base vector of the null space vector of the coding matrix, and sending part of the base vector of the null space to the user equipment according to the calculation task scale.

Images