A method for verifiable computing on distributed cloud servers

By using systematic MDS code to encode and verify data in cloud computing, the problem of resource waste is solved, and the efficiency of identifying and excluding untrusted servers is improved.

CN114676415BInactive Publication Date: 2024-11-12SOUTHEAST UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210372592.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-04-11
Publication Date
2024-11-12
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

In cloud computing, users need to verify the calculation results in multiple cloud servers to identify untrustworthy servers. The existing methods lead to waste of resources because there are fewer untrustworthy servers and excessive verification times.

Method used

The data is encoded using (n+2m,n)-systematic MDS code, and the data is allocated to multiple servers by generating a matrix, and the systematic MDS code is used to detect and correct errors during the verification stage to identify untrustworthy ones. server.

Benefits of technology

It realizes the effective identification and exclusion of untrustworthy servers on distributed cloud servers, avoids resource waste, and improves the efficiency of computing verification.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114676415B_ABST
    Figure CN114676415B_ABST
Patent Text Reader

Abstract

The present invention discloses a verifiable computing solution on a distributed cloud server. The solution is mainly aimed at the situation where a user entrusts computing to multiple servers. If the result returned by each server is verified once, it will lead to a waste of resources when there are few untrustworthy servers. Therefore, the solution uses systematic MDS to encode data and uses the characteristics of MDS code to verify the results returned by multiple servers. The solution can detect whether there are errors in the returned results and identify the untrustworthy servers that cause errors.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of cloud computing, and mainly relates to a solution for performing verifiable computing on n cloud servers, wherein some of the n servers are untrustworthy servers. Background Art

[0002] In cloud computing, users will entrust the required calculations to cloud servers, but cloud servers may not calculate the results seriously due to economic factors, but return a random result. Therefore, users need to verify the results. The verified results will be accepted, otherwise the entrusted server will no longer be trusted. With the development of machine learning, the training set is getting larger and larger, and one server cannot process so much data. Therefore, it is necessary to entrust the calculation to n cloud servers. Assuming that some of the n cloud servers are not trustworthy, a more direct solution is to verify each server once, and those that pass the verification will be accepted. However, if the number of untrustworthy servers is much less than n, then most of these n verifications are useless, wasting resources. Therefore, in view of this situation, the present invention proposes a solution for verifiable calculations for n cloud servers. Summary of the invention

[0003] Technical problem: The technical problem to be solved by the present invention is to provide a solution for performing verifiable calculations on n cloud servers, some of which are untrustworthy servers, using MDS codes to encode the data to be calculated, and then using the encoded data to verify the returned results.

[0004] Technical solution: The technical solution adopted by the present invention is as follows:

[0005] An important problem in machine learning is the least squares linear regression problem. Suppose there is a training data set that contains K features. and the corresponding label y i ∈F q , i = 1, ..., K, the linear regression algorithm hopes to find the coefficients of the regression function These coefficients can make the labels and features fit perfectly. In order to minimize the empirical risk, the following optimization problem needs to be solved:

[0006]

[0007] Here as well as

[0008] In order to solve the above optimization problem, a commonly used method is the gradient descent algorithm. The gradient descent algorithm iteratively updates the weight vector w along the direction of gradient descent. In order to complete the above operation, the following equation needs to be updated:

[0009]

[0010] Here t represents the number of iterations.

[0011] like Figure 1 As shown, the user needs to calculate X T Xw, dataset The data set is too large to be processed by one server, so it is necessary to split the data set into n small data sets and delegate them to n servers, X = [X1 X2 … X n ] T , definition The calculation is delegated to multiple servers. Server i returns the calculation result A i w=X i X i T w, in fact, these servers are not necessarily trustworthy. The most common situation is that most servers will seriously process the calculations outsourced to them and return the results to the user. A small number of servers are dishonest. They may not calculate carefully and return wrong results. Let the number of servers be n, and the number of untrustworthy servers be m. Users need to verify the returned results and find out the untrustworthy servers. (n+2m,n)-MDS can correct up to m errors, so (n+2m,n)-systematic MDS code is used to encode the data set. The generator matrix of the systematic MDS code is:

[0012]

[0013] Here the number of columns N of the matrix satisfies N = n + 2m. The scheme includes the following stages:

[0014] 1) Preparation phase: The data required for each server calculation is A i The (t,u)th element of is represented by Constructing vectors And use the MDS shown above for all a tu Encoding gives:

[0015]

[0016] For all i = 1, 2, ..., N, use Construct a matrix of size s×s Because the MDS code used is systematic, for i = 1, 2, ..., n, For i=n+1,…,N,

[0017]

[0018] Users will store data Used in the verification phase.

[0019] 2) Calculation phase: The user provides w to n cloud servers, and server i calculates Λ i =A i w. Afterwards, the server will calculate the result Return to the user. The jth element of is represented by

[0020] 3) Verification phase: The user will perform the following operations to verify: The user randomly selects c numbers from {1,2,...,s} with equal probability, represented as J1,J2,...,J c For each j=J1,J2,...,J c , the user will calculate Λ i =A i w,i=n+1,...N the jth element is The result returned is is a (n+2m,n)-systematic MDS code. This codeword can detect 2m errors and correct m errors. If you just want to detect whether there are errors in the returned results, the number of untrustworthy servers can be up to 2m. On the other hand, if you do not just want to detect whether there are errors, but want to identify untrustworthy servers and no longer trust them, the number of untrustworthy servers can be up to m. Decoding, comparing the decoded result with the received result, can identify the untrustworthy server. The user has all j=J1,J2,...,J c Perform the above operations in sequence. If you want to detect whether there is an error, you can stop performing the operation when an error is identified. If you want to identify unreliable servers, you can stop subsequent verification when all unreliable servers are identified.

[0021] Beneficial effects: The present invention proposes a verifiable computing solution on a distributed cloud server, which can be applied to perform verification calculations on multiple servers and can identify up to m untrustworthy servers. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is the structure of distributed cloud computing. DETAILED DESCRIPTION

[0023] The technical solution of the present invention is described in detail below, but the protection scope of the present invention is not limited to the implementation examples.

[0024] The present invention proposes a verifiable computing solution performed on a distributed cloud server.

[0025] An implementation example is given below:

[0026] Assume that there are 6 servers, among which there are less than 2 untrustworthy servers, and the user wants to identify the untrustworthy servers and no longer trust these servers in subsequent calculations.

[0027]

[0028] The user uses the generator matrix of the systematic MDS code above and sets the number of columns N of the generator matrix to satisfy N = 10. The verification calculation scheme includes the following stages:

[0029] 1) Preparation phase: The data required for each server calculation is A i The (t,u)th element of is represented by t,u=1,2,...,s,construct vector And use the (10,6)-systematic MDS as shown above for all a tu Encoding gives:

[0030]

[0031] For all i=1,2,…,10, use Construct a matrix of size s×s Because the MDS code used is systematic, for i=1,2,3,4,5,6, For i=7,8,9,10,

[0032]

[0033] Users will store data Used in the verification phase.

[0034] 2) Calculation phase: The user provides w to n cloud servers, and server i calculates Λ i =A i w. Afterwards, the server will calculate the result Return to the user. The jth element of is represented by

[0035] 3) Verification phase: The user will perform the following operations to verify: The user randomly selects c numbers from {1,2,...,s} with equal probability, represented as J1,J2,...,J c For each j=J1,J2,...,J c , the user will calculate Λ i =A i w,i=7,8,9,10The jth element is The result returned is is a (10,6)-systematic MDS code. This codeword can detect 4 errors and correct 2 errors. If you just want to detect whether there are errors in the returned results, the number of untrustworthy servers can be up to 4. On the other hand, if you do not just want to detect whether there are errors, but want to identify the untrustworthy servers and no longer trust them, the number of untrustworthy servers can be up to 2. Decoding, comparing the decoded result with the received result, can identify the untrustworthy server. The user has all j=J1,J2,...,J c Perform the above operations in sequence. If you want to detect whether there is an error, you can stop performing the operation when an error is identified. If you want to identify unreliable servers, you can stop subsequent verification when all unreliable servers are identified.

Claims

1. A method for performing verifiable computing on a distributed cloud server, characterized in that: The steps include: Step 1: The user encodes the data set to be calculated and entrusts it to n cloud servers; Step 2: Each server performs calculations separately and returns the calculation results to the user; Step 3: The user verifies the returned result; The specific steps of step 1 include: constructing a training data set, the training data set contains K features and the corresponding label y i ∈F q ,i=1,…,K; Create the following optimization problem to minimize the empirical risk: in, is the coefficient of the regression function of the linear regression algorithm; Use the gradient descent algorithm to iteratively update the weight vector w in the direction of gradient descent, and update the following equation: Where t represents the number of iterations; Divide the training set X into n sub-matrices evenly, that is, X = [X1 X2…X n ] T ,at this time definition The computation is delegated to n servers, where the number of untrustworthy servers is m. The dataset is encoded using the (n+2m,n)-systematic MDS code. The generator matrix of the (n+2m,n)-systematic MDS code is: The number of columns N of the matrix G satisfies N = n + 2m; The data required for each server calculation is A i The (t,u)th element of is represented by Constructing vectors And use the MDS shown above for all a tu Encoding gives: For all i = 1, 2, ..., N, use Construct a matrix of size s×s For i = 1, 2, ..., n, For i=n+1,…,N, Users will store data Used in the verification phase; The specific steps of step 2 include: the user provides w to n cloud servers, and server i calculates Λ i =A i w; after that, the server will calculate the result Return to the user; The jth element of is represented by The specific steps of step 3 include: the user performs the following operations to verify: the user randomly selects c numbers from {1,2,...,s} with equal probability and represents them as J1,J2,...,J c ; For each j = J1, J2, ..., J c , the user calculates Λ i =A i w,i=n+1,...N the jth element is The product of; the result returned is a (n+2m,n)-systematic MDS code that can detect 2m errors and correct m errors. If you just want to detect whether there are errors in the returned results, the number of untrustworthy servers can be up to 2m; if you do not just want to detect whether there are errors, but want to identify the untrustworthy servers and no longer trust them, the number of untrustworthy servers can be up to m; the result Decoding and comparing the decoded result with the received result can identify untrustworthy servers.

2. A method for performing verifiable computing on a distributed cloud server according to claim 1, characterized in that: For all j=J1,J2,...,J c The above operations are performed in sequence. If you want to detect whether there is an error, stop the process when an error is identified. If you want to identify an unreliable server, stop subsequent verification when all unreliable servers are identified.