A computing power resource allocation method based on combinatorial double auction and differential privacy

Abstract

A computing power resource allocation method based on combinatorial double auction and differential privacy, comprising: at the beginning of a time slot, a user device issues a task request to an edge server and generates bidding information; the edge server forwards the task request issued by the user device to an edge node, the edge node calculates the confusion distance between itself and the task request, generates query information and participates in the bidding; the edge server receives the bidding information sent by the user device and the query information returned by the edge node, sets a number of price vectors, and selects an edge node for each task request as the winning buyer and seller based on each price vector to obtain a resource allocation matrix corresponding to each price vector, and then selects a price vector as the final price vector, and the corresponding resource allocation matrix is the final resource allocation matrix. The present invention relates to the field of communications, and can realize the sharing pricing of different types of resources and ensure the effectiveness of resource allocation incentives, and protect privacy during the incentive process.

Description

Technical Field

[0001] The present invention relates to a computing power resource allocation method based on combined double auction and differential privacy, and relates to the field of communications. Background Art

[0002] In recent years, with the rapid development of the mobile Internet and intelligent terminals, more and more new applications have emerged in people's daily lives. However, due to physical distance limitations, the resources available to mobile devices are usually limited, so it is difficult for mobile devices to effectively process some new applications. To solve the above problems, the concept of mobile edge computing has emerged. Mobile edge computing deploys network device entities with storage and computing capabilities at the edge of the mobile network to provide IT services and computing capabilities for mobile users.

[0003] Regarding how to design an efficient computing power resource allocation mechanism to motivate edge nodes to provide the network resources they own for mobile users, there are currently the following main technical solutions:

[0004] Technical Solution 1: Patent Application No. CN 202110243640.5 (Application Name: An Edge Computing Cooperative Computing Resource Allocation Method Based on Online Incentive, Application Date: 2021-03-05, Applicant: University of Electronic Science and Technology) proposes an edge computing cooperative computing resource allocation method based on online incentive, including establishing a task cooperative computing model; establishing an auction model between the main edge computing server and task executors; establishing a utility model for the buyer and a utility model for the seller, thereby establishing a utility model for the system; combining the task cooperative computing model to construct a system utility maximization problem; when a task arrives at the main edge computing server, the main edge computing server provides a bid to the auctioneer; after the auctioneer obtains the bid information, it determines the optimal allocation plan for the task with the goal of maximizing the system utility; determining the payment price for the winning bidder; calculating the task with the optimal resource allocation plan;

[0005] Technical Solution 2: Patent Application No. CN 202110571751.9 (Application Name: A Resource Allocation Method Based on Auction Theory in a Mobile Edge Computing Environment, Application Date: 2021-05-25, Applicant: Northeastern University) proposes a resource allocation method based on auction theory in a mobile edge computing environment, which involves initializing the resource capacity of the edge server; mobile users submitting computing task requirements and valuations to the edge server; reordering the resource configuration combinations of the edge server according to the normalization processing results; converting the computing task requirements submitted by mobile users into CPU resource and channel resource requirements; determining the bid targets of mobile users and participating in the bidding competition; using the primal-dual approximation algorithm to determine the winners among the mobile users participating in the bidding; determining the price that each winner needs to pay through the VCG bidding mechanism;

[0006] Technical Solution 3: The patent application number is CN 201810360743.8 (application name: Mobile Crowdsensing Based on Double Auction and Its Resource Allocation and Incentive Mechanism Method, application date: April 20, 2018, applicant: Southeast University) which proposed an online combinatorial resource allocation and payment method based on bilateral auction. Each buyer (SP) provides bid information; each seller (MVNO) provides asking price information; the auctioneer (intermediary) calculates the bid density function and constructs a joint bid density matrix, and obtains the bid density sorting vector through ascending or descending order; then sequentially determines whether the resources owned by the seller associated with each element can fully meet the resources applied by the buyer. If it is satisfied, the corresponding seller and buyer are regarded as the winning institutions; then based on the critical minimum bid density as a reference, determine the fee qm to be charged by each winning seller; at the same time, based on the critical maximum bid density as a reference, determine the fee pn to be paid by each winning buyer; finally, select the buyers and sellers corresponding to pn≥qm for matching to achieve resource allocation.

[0007] In the above existing technical solutions, the resource allocation method provides corresponding benefit compensation to resource providers through contribution evaluation. However, in the incentive process, users' personal information and behavior information will be leaked to other participating parties. Distributed idle nodes in the network still face risks of identity, location, resource capabilities, and decision-making information exposure when participating in collaborative computing. Even with certain benefit compensation, due to privacy concerns, distributed idle nodes in the network are still reluctant to participate in computing power sharing, resulting in the failure of the resource allocation method and the inability to fully mobilize the available computing power resources in the network.

[0008] Therefore, how to realize the pricing of different types of resource sharing and ensure the effectiveness of resource allocation incentives, and protect privacy during the incentive process, so as to promote distributed computing power nodes to actively contribute idle resources, has become a key technical issue that technicians focus on. Summary of the Invention

[0009] In view of this, the purpose of the present invention is to provide a computing power resource allocation method based on combinatorial bilateral auction and differential privacy, which can realize the pricing of different types of resource sharing and ensure the effectiveness of resource allocation incentives, and protect privacy during the incentive process, so as to promote distributed computing power nodes to actively contribute idle resources.

[0010] To achieve the above purpose, a computing power resource allocation method based on combinatorial bilateral auction and differential privacy of the present invention. Each computing power resource sharing area includes an edge server, several user devices, and several edge nodes providing idle resources, sharing k types of resource sets h1, h2,..., h k respectively represent the 1st, 2nd,..., kth types of resources shared, including:

[0011] Step 1. At the beginning of each time slot, all user devices publish their task requests: τ i =(q i , l i ), where τ i represents the i-th task request, is the resource set required for the task request τ i , respectively represent the quantities of the 1st, 2nd, …, k-th resources required by τ i , l i represents the location of the task request τ i , and generate their respective bidding information according to the task requests: where B i represents the bidding information generated corresponding to the task request τ i , b i is the highest budget price for the task request τ i , represents the time for which the task request τ i needs to occupy resources;

[0012] Step 2. The edge server forwards all the task requests published by the user devices to all edge nodes. Each edge node calculates the confusion distance between itself and each task request to generate its respective inquiry information and participate in the bidding competition. The generated inquiry information is: where A j is the inquiry information of the j-th edge node, is the inquiry vector, are respectively the lowest prices of the 1st, 2nd, …, k-th unit resources that the j-th edge node can provide, is the resource set that the j-th edge node can provide, are respectively the quantities of the 1st, 2nd, …, k-th resources that the j-th edge node can provide, is the distance information of the j-th edge node, is the confusion distance set of the j-th edge node, are respectively the confusion distances between the j-th edge node and the 1st, 2nd, …, I-th task requests. I is the total number of task requests. The confusion distance is calculated based on the location of each task request and ∈ j obtained by the Laplace mechanism, ∈ j is the personalized privacy budget set by the j-th edge node, is the resource idle time of the j-th edge node;

[0013] Step 3: The edge server receives the bidding information sent by all user devices and the query information returned by the edge nodes, sets a number of price vectors, and selects an edge node as the winning buyer and seller for each task request based on each price vector, so as to obtain a resource allocation matrix corresponding to each price vector. The resource allocation matrix is used to identify the winning buyer and seller, and then a price vector is selected as the final price vector according to the resource allocation matrix. The resource allocation matrix corresponding to the final price vector is the final resource allocation matrix.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: In the absence of appropriate interest compensation and privacy risks, idle devices in the network are reluctant to spend resources to participate in computing tasks. The present invention classifies the network nodes participating in the computing power resource sharing into three roles: resource providers, resource coordinators, and resource requesters, and designs a resource allocation method based on a combined bilateral auction model. It prices various types of resources separately, packages them for resource allocation, and designs a combined bilateral auction process, so as to realize the pricing of different types of resource sharing and the incentive for resource allocation; due to interest conflicts among distributed idle devices in the network, and at the same time, user personal information and behavior information will be leaked to other participating parties during the incentive process, and the computing power sharing faces the risk of information exposure, and it is difficult to establish trust among all participating parties. The present invention introduces two differential privacy mechanisms to provide personalized location privacy protection and decision privacy protection for the network nodes participating in the computing power sharing. In the design of the auction process, the Laplace mechanism is used to obfuscate the location information of the distributed maintenance terminal devices, and the exponential mechanism is used to obfuscate the pricing decision to protect the user decision privacy, thereby reducing the information exposure risk of the distributed computing power nodes in the network and fully mobilizing the available computing power resources in the network. Description of the Drawings

[0015] Figure 1 is a flowchart of a computing power resource allocation method based on a combined bilateral auction and differential privacy of the present invention.

[0016] Figure 2 is Figure 1 The specific step flowchart of Step 3.

[0017] Figure 3 is Figure 2 In Step 35, for task request τ f The specific step flowchart of selecting an edge node from the seller set as the winning seller, and then setting the resource allocation matrix corresponding to the price vector p according to the winning buyer and seller.

[0018] Figure 4 is an embodiment of a computing power resource sharing network applying the present invention.

[0019] Figure 5 is Figure 4Comparison graph of resource sharing benefits of embodiments under different numbers of resource requesters.

[0020] Figure 6 Yes Figure 4 Comparison graph of resource sharing benefits of embodiments under different numbers of resource providers. Detailed implementation manners

[0021] To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings.

[0022] Each computing power resource sharing network of the present invention includes a resource provider, a resource requester, and a resource coordinator. The resource requester consists of user devices, which are mobile devices / IoT devices that submit computing task requests to the edge server. The resource provider consists of edge nodes that provide idle resources, including intelligent devices with limited resource capabilities and edge servers that provide computing power resources. The resource coordinator consists of edge servers, which coordinate the heterogeneous user devices and edge nodes connected locally to complete resource allocation. Within each time slot, the set of resource providers and resource requesters within the coverage of the edge server is fixed, and k types of resources need to be shared. A typical auction includes three roles: a buyer, a seller, and an auctioneer. In the present invention, the buyer is the resource requester (i.e., the user device), the seller is the resource provider (i.e., the edge node), and the resource coordinator (i.e., the edge server) acts as the role of the auctioneer to manage the auction process.

[0023] As Figure 1 shown, a computing power resource allocation method based on combinatorial double auction and differential privacy of the present invention includes an edge server, several user devices, and several edge nodes that provide idle resources in each computing power resource sharing area, and shares k resource sets h1, h2,..., h k respectively represent the 1st, 2nd,..., kth types of resources shared, k is the number of shared resources, and the method includes:

[0024] Step 1: At the beginning of each time slot, all user devices publish their task requests: τ i =(q i , l i ), where τ i represents the ith task request, is the resource set required by the task request τ i , respectively represent the quantities of the 1st, 2nd,..., kth types of resources required by τ i , l i represents the position of the task request τ i , and generate their respective bidding information according to the task request: Among them, B i represents the task request τ i corresponding generated tender information, b i is the highest budget price for the task request τ i represents the time when the task request τ i needs to occupy resources;

[0025] Step 2: The edge server forwards all task requests sent by user devices to all edge nodes. Each edge node calculates the confusion distance between itself and each task request to generate its own inquiry information and participate in the tender bidding. The generated inquiry information is: Among them, A j is the inquiry information of the j-th edge node, is the inquiry vector, are respectively the lowest prices of the 1st, 2nd, …, k-th unit resources that the j-th edge node can provide, is the resource set that the j-th edge node can provide, are respectively the quantities of the 1st, 2nd, …, k-th resources that the j-th edge node can provide, is the distance information of the j-th edge node, is the confusion distance set of the j-th edge node, are respectively the confusion distances between the j-th edge node and the 1st, 2nd, …, I-th task requests. I is the total number of task requests. The confusion distance can be calculated based on the location of each task request (such as l i ) and ∈ j obtained by calculating based on the Laplace mechanism, ∈ j is the personalized privacy budget set by the j-th edge node, is the resource idle time of the j-th edge node;

[0026] Step 3: The edge server receives the tender information sent by all user devices and the inquiry information returned by edge nodes, sets several price vectors, and selects an edge node as the winning buyer and seller for each task request based on each price vector, so as to obtain the resource allocation matrix corresponding to each price vector. The resource allocation matrix is used to identify the winning buyer and seller, and then selects a price vector as the final price vector according to the resource allocation matrix. The resource allocation matrix corresponding to the final price vector is the final resource allocation matrix.

[0027] Such as Figure 2 shown, Figure 1 Step 3 can further include:

[0028] Step 31: Preset several price vectors p = {p1, p2, …, p​k}, where p1, p2, …, p k are the prices of the 1st, 2nd, …, kth resources respectively. All the set price vectors form a price vector set P, and then a price vector p is selected from the price vector set P;

[0029] Step 32: Form a seller set with all the edge nodes that return query information;

[0030] Step 33: Select several task requests from all the task requests to form a buyer task subset, and then, according to the selected price vector p, calculate the payment price of each task request in the buyer task subset one by one: b i ′ is the payment price of the task request τ i ; p z is the price of the zth resource. Then, determine whether the maximum budget price of each task request is greater than or equal to the payment price. If so, it means that the maximum budget price of this task request is sufficient to pay for the required resources, and continue to calculate the payment price of the next task request. If not, it means that the maximum budget price of this task request is not enough to pay for the required resources, delete this task request from the buyer task subset, and continue to calculate the payment price of the next task request;

[0031] Step 34: Calculate the bid density of each task request in the buyer task subset: where bd i is the bid density of the task request τ i ; M i is the total resource requirement of the task request τ i ; ω z is the price weight of the zth resource, and its value can be determined according to the historical auction price of this resource. Then, all the task requests in the buyer task subset are sorted in descending order according to their bid densities;

[0032] Since the bid of a task request may imply the urgency of the task (an urgent task bids higher), and since the types and quantities of resources required by the buyer are different, it is impossible to measure the price of the buyer's unit resource by a simple bid sorting and use this as the sorting of the winning buyer. Therefore, the present invention defines the bid density based on an approximation model to measure the bid of the buyer's unit resource;

[0033] Step 35: Extract each task request in the buyer task subset in order, select an edge node from the seller set as the winning seller for it, and then set the resource allocation matrix X corresponding to the price vector p according to the winning buyer and seller I×J= {x ij}, where X I×J is a binary matrix, J is the total number of edge nodes, x ij ∈ X I×J , x ij = 1 indicates that the task request τ i is executed by the j-th edge node, otherwise x ij = 0;

[0034] Step 36: Determine whether all price vectors p in the price vector set P have been selected. If so, continue to Step 37; if not, continue to select an unselected price vector p from the price vector set P and turn to Step 32;

[0035] Step 37: Calculate the score function corresponding to each price vector p in the price vector set P: is the unit cost for the j-th edge node to provide the z-th resource, and its value can be determined according to historical cost values. Then, calculate the decision probability distribution of the price vector according to the exponential mechanism: ε is the privacy budget, ΔQ is the sensitivity of the score function, ΔQ = Q max - Q min , Q max , Q min are the maximum and minimum values of the score functions corresponding to all price vectors respectively, is the cumulative distribution of all price vectors. Finally, randomly select a price vector from the price vector set P as the final price vector according to the probability distribution, and the resource allocation matrix corresponding to the final price vector is the final resource allocation matrix;

[0036] Step 38: Calculate the payment prices of all winning buyers and the rewards obtained by the seller according to the final price vector: r j is the reward obtained by the j-th edge node.

[0037] Repeating Steps 32 to 35, the resource allocation matrices corresponding to all price vectors in the price vector set P can be obtained. Through the above solution process, given the price vector p, the winning buyers and sellers can be obtained, that is, a resource allocation scheme X I×J is determined for each price vector p. Next, select one of the prices p as the final price with the goal of maximizing the revenue of the resource provider. To ensure the decision privacy of the participants and prevent inference attacks, the present invention designs the pricing process by combining the average price mechanism and the exponential mechanism. The average price mechanism means that all winning resource requesters pay the same price for the same type of unit resource.

[0038] Such as Figure 3As shown, taking the task request τ f as an example, in step 35, for the task request τ f select an edge node from the seller set as the winning seller, and then set the resource allocation matrix corresponding to the price vector p according to the winning buyer and seller, which may further include:

[0039] Step 351: Construct a candidate set of sellers for the task request τ f in the buyer task subset, and extract the inquiry information of an edge node from the seller set;

[0040] Step 352: Determine whether the amount of each resource required by the task request τ f is less than or equal to the amount of resources that can be provided by the extracted edge node, and the resource idle time of the extracted edge node is greater than or equal to the time when the task request τ f needs to occupy resources. If so, continue to step 353; if not, go to step 354;

[0041] In step 352, the amount of each resource required by the task request τ f is less than or equal to the amount of resources that can be provided by the extracted edge node. Taking the extraction of the l-th edge node as an example, that is, when z ∈ [1, k], is less than or equal to which is the amount of the z-th resource that can be provided by the extracted l-th edge node;

[0042] Step 353: Calculate the total price of the task request τ f under the price vector p and the lowest total price of the extracted edge node: where c f is the total price of the task request τ f under the price vector p, d l is the lowest total price of the l-th edge node, and determine whether c f is greater than or equal to d l . If so, add the extracted edge node to the candidate set of sellers for the task request τ f , and then continue to step 354; if not, continue to step 354;

[0043] Step 354: Determine whether there are still unextracted edge nodes in the seller set. If so, continue to extract the inquiry information of the unextracted edge nodes and go to step 352; if not, continue to step 355;

[0044] Step 355: Determine the task request τ fWhether the number of edge nodes in the seller candidate set of [[τ]] is greater than 1. If so, calculate the probability of the comparison result of the true distances of every two edge nodes in the seller candidate set, and delete the edge node with the larger true distance until the number of edge nodes in the seller candidate set is 1, and then continue with step 356; if not, continue with step 356;

[0045] Step 356: Read the task request τ f For the edge node x in the seller candidate set of [[τ]], then set the element value corresponding to the task request τ I×J and the edge node x in the resource allocation matrix X f to 1, and set the element values corresponding to the task request τ f and other edge nodes to 0. Finally, delete the edge node x from the seller set.

[0046] In step 355, since there may be more than one eligible edge node, there is a situation where one buyer candidate corresponds to multiple seller candidates, thus resulting in the problem of seller winning candidate conflict. The present invention can adopt the nearest location strategy to eliminate the problem of seller winning candidate conflict. Step 355 further includes:

[0047] Step 1: Determine whether the number of edge nodes in the seller candidate set of the task request τ f is greater than 1. If so, continue to the next step; if not, continue with step 356;

[0048] Step 2: Select the query information of any two edge nodes u, v from the seller candidate set: Calculate the probability P(d fu <d fv ) that the true distance of edge node u is less than that of v: where d fu , d fv are the true distances between the task request τ f and the edge nodes u, v respectively, are the corresponding obfuscated distances of d fu , d fv respectively, ∈ u , ∈ v are the personalized privacy budgets set by the edge nodes u, v respectively. The obfuscated distance is obtained by adding noise to the true distance. η u , η v are the noises between the true distances and the obfuscated distances of the edge nodes u, v respectively, that is and η u ~Lap(0,1 / ∈ u ), η v ~Lap(0,1 / ∈ v ), ηu and η v are variables subject to the Laplace distribution. The smaller the personalized privacy budget (such as ∈ u , ∈ v ), the greater the noise, and the higher the level of privacy protection. Then, it is determined whether P(d fu < d fv ) is greater than 1 / 2. If so, it means that the true distance of u is less than the true distance of v, and the marginal node v is deleted from the seller candidate set, and the process turns to step 1; if not, it means that the true distance of u is greater than the true distance of v, and the marginal node u is deleted from the seller candidate set, and the process turns to step 1.

[0049] In step 2, where is the known information of the auctioneer. Therefore, the above problem can be transformed into a probability problem of two-dimensional continuous variables (η u , η v ) on the plane . η u and η v are independent of each other, and double integral can be used for calculation. Among them, f(η u )f(η v ) can be expressed as

[0050] Figure 4 This is an embodiment of a computing power resource sharing network applying the present invention. In order to simulate the resource sharing domain in the computing power resource sharing network scenario, Figure 4 the coverage radius r of the resource coordinator in i is 500. Each resource sharing domain includes J resource providers (denoted as RP), I resource requesters (denoted as RR), and a resource coordinator (denoted as RC). According to the number of participants, points (x i ) are randomly generated in this area as the positions of the resource requesters, and points (x j , y j ) are the positions of the resource providers. Assuming that the task position is the same as the resource requester position, the actual distance can be calculated according to the positions The price of a unit resource is normalized to [0, 1]. In this embodiment, two types of computing power resources, namely CPU and memory, are considered, so k = 2 is set. The price vector value granularity is 0.05. That is

[0051] Since differential privacy is introduced in the auction process of the computing power resource allocation method of the present invention, the auction result is probabilistic and uncertain. The performance is evaluated based on the results of 1000 trials. The unified pricing method is extended to a combinatorial double auction as the baseline algorithm in the first simulation scenario, also known as the deterministic scheme without the exponential mechanism. The auction result of the baseline algorithm is obtained by the maximum value Q(p).

[0052] Figure 5 compared Figure 4 the resource sharing benefits of the embodiments under different numbers of resource requesters (the number of resource providers is set to 60). First, the benefits under the computing power resource allocation method proposed by the present invention are lower than those of the deterministic scheme (without differential privacy). Because the computing power resource allocation method proposed by the present invention determines the pricing scheme with the probability of the differential privacy method, while the deterministic scheme selects the pricing corresponding to the maximum benefit. It can also be found that the benefits under the deterministic scheme are the upper limit of the computing power resource allocation method proposed by the present invention. The gap between them can be regarded as a trade-off between revenue and decision privacy. In addition, we set ε = 100, 200, 300 in Figure 5 (a), (b) and (c) to compare the impact of the privacy budget. As the privacy budget becomes larger, the gap between them will become smaller. This is consistent with the concept of the privacy budget. Second, as more resource providers participate, the revenue will increase. Because as the demand grows, resource providers will execute more tasks and obtain more rewards.

[0053] Figure 6 contrasted Figure 4 the resource sharing benefits of the embodiments under different numbers of resource providers (the number of resource requesters is set to 250). First, similar to Figure 5 the same, there is a gap between the computing power resource allocation method proposed by the present invention and the deterministic scheme in terms of achieving privacy protection. And in Figure 5 it, the gap becomes larger as the number of resource providers increases. Because in addition to protecting decision privacy, location privacy is also related to the number of resource providers. As there are more and more resource providers, the cumulative uncertainty of the locations of all resource providers will increase.

[0054] The above results show that the method of the present invention can achieve resource allocation results and pricing results with the same trend as the unified pricing scheme, thus ensuring the effectiveness of the resource allocation method. At the same time, the incentive result in the present invention is based on the confused distance calculation, and the real location of the participants cannot be obtained. Due to the uncertainty of the incentive result in the present invention, the participants cannot affect the incentive result by changing a single bid price, thus achieving the effects of location privacy protection and user decision privacy protection.

[0055] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A computing power resource allocation method based on combined double-sided auction and differential privacy, characterized in that Each computing power resource sharing area includes an edge server, several user devices, and several edge nodes providing idle resources, sharing k resource sets h1, h2, …, h k respectively represent the 1st, 2nd, …, kth shared resources, including: Step 1: At the beginning of each time slot, all user devices publish their task requests: τ i =(q i , l i ), where τ i represents the i-th task request, is the resource set required for the task request τ i , respectively represent the quantities of the 1st, 2nd, …, k-th resources required by τ i , l i represents the location of the task request τ i , and generate their respective bidding information according to the task requests: where B i represents the bidding information generated corresponding to the task request τ i , b i is the highest budget price for the task request τ i , represents the time when the task request τ i needs to occupy resources; Step 2: The edge server forwards all task requests issued by user devices to all edge nodes. Each edge node calculates the confusion distance between itself and each task request to generate its own query information and participate in the bidding. The generated query information is: where A j is the query information of the j-th edge node, is the query vector, are respectively the minimum prices of the 1st, 2nd, …, k-th unit resources that the j-th edge node can provide, is the resource set that the j-th edge node can provide, are respectively the quantities of the 1st, 2nd, …, k-th resources that the j-th edge node can provide, is the distance information of the j-th edge node, is the confusion distance set of the j-th edge node, are respectively the confusion distances between the j-th edge node and the 1st, 2nd, …, I-th task requests. I is the total number of task requests. The confusion distance is calculated based on the location of each task request and ∈ j obtained by calculation based on the Laplace mechanism, ∈ j is the personalized privacy budget set by the j-th edge node, is the resource idle time of the j-th edge node; Step 3. The edge server receives the bidding information sent by all user devices and the inquiry information returned by the edge nodes, sets a number of price vectors, and selects an edge node as the winning buyer and seller for each task request based on each price vector, so as to obtain a resource allocation matrix corresponding to each price vector. The resource allocation matrix is used to identify the winning buyer and seller, and then a price vector is selected as the final price vector according to the resource allocation matrix. The resource allocation matrix corresponding to the final price vector is the final resource allocation matrix.

2. The method according to claim 1, wherein Step 3 further includes: Step 31. Preset a number of price vectors p = {p1, p2,..., p k}, where p1, p2,..., p k are the prices of the 1st, 2nd,..., kth resources respectively, and form a price vector set P with all the preset price vectors, then select a price vector p from the price vector set P; Step 32. Form a seller set consisting of all edge nodes that return inquiry information; Step 33: Select several task requests from all task requests to form a buyer task subset, and then, according to the selected price vector p, calculate the payment price of each task request in the buyer task subset one by one: is the payment price of task request τ i where p z is the price of the z-th type of resource, and then determine whether the maximum budget price of each task request is greater than or equal to the payment price. If so, it means that the maximum budget price of the task request is sufficient to pay for the required resources, and continue to calculate the payment price of the next task request. If not, it means that the maximum budget price of the task request is not enough to pay for the required resources, delete the task request from the buyer task subset, and continue to calculate the payment price of the next task request; Step 34: Calculate the bid density of each task request in the buyer task subset: where bd i is the bid density of task request τ i , M i is the total resource requirement of task request τ i , ω z is the price weight of the z-th resource, and then arrange all task requests in the buyer task subset in descending order according to their bid densities; Step 35: Extract each task request in the buyer task subset in sequence, select an edge node from the seller set as the winning seller for it, and then set the resource allocation matrix X corresponding to the price vector p according to the winning buyer and seller I×J ={x ij}, where X I×J is a binary matrix, J is the total number of edge nodes, x ij ∈X I×J , x ij =1 indicates that the task request τ i is executed by the j-th edge node, otherwise x ij =0; Step 36. Determine whether all price vectors p in the price vector set P have been selected. If so, continue to Step 37; if not, continue to select an unselected price vector p from the price vector set P and turn to Step 32; Step 37. Calculate the scoring function corresponding to each price vector p in the price vector set P: is the unit cost for the j-th edge node to provide the z-th type of resource, and then the decision probability distribution of the price vector is calculated according to the exponential mechanism: ε is the privacy budget, and ΔQ is the sensitivity of the scoring function, where ΔQ = Q max - Q min , Q max , Q min are the maximum and minimum values of the scoring function corresponding to all price vectors, respectively. is the cumulative distribution of all price vectors. Finally, a price vector is randomly selected from the price vector set P according to the probability distribution as the final price vector, and the resource allocation matrix corresponding to the final price vector is the final resource allocation matrix.

3. The method according to claim 2, wherein After Step 37, it further includes: Step 38: Calculate the payment prices of all winning buyers and the remuneration obtained by the seller according to the final price vector: r j is the remuneration obtained by the j-th edge node.

4. The method according to claim 2, wherein Taking the task request τ f as an example, in step 35, for the task request τ f select an edge node from the seller set as the winning seller, and then set the resource allocation matrix corresponding to the price vector p according to the winning buyer and seller, further including: Step 351: For the task request τ in the buyer task subset f Construct a seller candidate set and extract the query information of an edge node from the seller set; Step 352: Determine whether it is task request τ f The amount of each required resource is less than or equal to the resources that can be provided by the extracted edge node, and the resource idle time of the extracted edge node is greater than or equal to the task request τ f The time for resource occupation. If so, continue to step 353; if not, go to step 354; Step 353, calculate the task request τ f The total price under the price vector p and the lowest total price of the extracted edge nodes: where c f is the total price of the task request τ f under the price vector p, d l is the lowest total price of the I-th edge node, and determine whether c f is greater than or equal to d l , if so, add the extracted edge node to the seller candidate set of the task request τ f , and then continue to step 354; if not, continue to step 354; Step 354. Determine whether there are still unextracted edge nodes in the seller set. If so, continue to extract the inquiry information of the unextracted edge nodes and turn to Step 352; if not, continue to Step 355; Step 355: Determine the task request τ f whether the number of edge nodes in the seller candidate set is greater than 1. If so, calculate the probability of the comparison results of the true distances between every two edge nodes in the seller candidate set, and delete the edge node with the larger true distance until the number of edge nodes in the seller candidate set is 1, and then continue with Step 356; if not, continue with Step 356; Step 356, read the task request τ f among the edge nodes x in the seller candidate set of I×J , and then set the element value corresponding to the task request τ f and the edge node x in the resource allocation matrix X f to 1, and set the element values corresponding to the task request τ and other edge nodes to 0. Finally, delete the edge node x from the seller set.

5. The method according to claim 4, wherein Step 355 further includes: Step 1, determine the task request τ f whether the number of edge nodes in the seller candidate set is greater than 1. If so, proceed to the next step; if not, proceed to step 356; Step 2: Select the query information of any two edge nodes u and v from the seller candidate set: Calculate the probability P(d fu <d fv ) that the true distance of edge node u is less than that of v: where d fu and d fv are the true distances between the task request τ f and edge nodes u and v respectively, are the corresponding obfuscated distances of d fu and d fv respectively, ∈ u and ∈ v are the personalized privacy budgets set by edge nodes u and v respectively. The obfuscated distance is obtained by adding noise to the true distance. η u and η v are the noises between the true distances and obfuscated distances of edge nodes u and v respectively, that is and η u ~Lap(0, 1 / ∈ u ), η v ~Lap(0, 1 / ∈ v ). η u and η v are variables subject to the Laplace distribution. Then judge whether P(d fu <d fv ) is greater than 1 / 2. If so, it means that the true distance of u is less than that of v. Delete edge node v from the seller candidate set and go to Step 1; if not, it means that the true distance of u is greater than that of v. Delete edge node u from the seller candidate set and go to Step 1.

Images