A computing power network resource allocation method and device

Abstract

The application provides a computing power network resource allocation method and device, a transaction reservation price is calculated by deploying a reservation price calculation model off-chain in a blockchain network, and a sealed two-price auction is executed on-chain; a virtual valuation fitting model fits a virtual valuation function in a Myerson auction mechanism through a neural network, fits a cumulative distribution function and a distribution density function of the valuation through the neural network, optimizes approximation and lookup of a transaction reservation price capable of bringing the highest income to a user, and encourages the user to sell idle computing power resources. By separately deploying transaction reservation price calculation and an auction process off-line and on-line, the design difficulty of a smart contract of the blockchain network can be reduced.

Description

Technical Field

[0001] This invention relates to the field of computing network technology, and in particular to a method and apparatus for allocating computing network resources. Background Technology

[0002] In the era of rapid development of the digital economy, with the gradual maturation of technologies such as 5G network technology, edge computing, and cloud computing, the integration of network and computing power is becoming increasingly close. A computing power network is a new type of information infrastructure that allocates and flexibly schedules computing, storage, and network resources on demand among the cloud, network, and edge based on business needs.

[0003] Current computing resource allocation schemes primarily aim to maximize resource utilization, i.e., to maximize social welfare, without considering efficiency maximization. Existing technologies only consider the static allocation process of resources, neglecting the utilization of idle user resources. This approach leads to overall inefficiency in computing resource usage. Furthermore, considering the inherent burstiness of traffic, static allocation based primarily on bandwidth or computing power results in very low average network utilization.

[0004] Therefore, a new method for allocating computing resources is urgently needed. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a computing power network resource allocation method and apparatus to eliminate or improve one or more defects existing in the prior art, so as to solve the problem that the existing computing power resource allocation method cannot incentivize users to sell idle resources and cannot maximize user benefits.

[0006] On one hand, the present invention provides a method for allocating computing power network resources, which is executed synchronously on-chain and off-chain in a blockchain network. The method includes the following steps:

[0007] The computing power leasing entity submits the computing power index parameters of the idle computing power resources to the client nodes provided by the computing network operator for uploading to the blockchain network;

[0008] Multiple bidders estimate the value of the idle computing power resources based on the computing power index parameters, determine the bid price, and call the chaincode of the blockchain network to submit the bid.

[0009] The blockchain network submits each bidder's bid price to an off-chain reserved price calculation model to output a transaction reserved price. This reserved price calculation model includes a virtual valuation fitting model, a softmax function, and a ReLU function. The virtual valuation fitting model uses a pre-defined neural network model to fit the virtual valuation function in the Miles auction mechanism. Each bidder's bid price is input into the virtual valuation fitting model, which outputs a first-type virtual valuation for each bidder. The ReLU function is used to construct payment rules, and based on the inverse transformation of the virtual valuation fitting model, the actual payment price for each bidder is calculated. A first virtual user is introduced, whose corresponding first-type virtual valuation is 0. The first virtual user's transaction reserved price is calculated through the inverse transformation of the virtual valuation fitting model. Each bidder's first virtual valuation is input into the softmax function to output the probability of each bidder winning. The seller's expected revenue is calculated based on each bidder's actual payment price and corresponding winning probability. A loss function is constructed using the negative of the seller's expected revenue. The reserved price calculation model updates its parameters to minimize the loss function and outputs the transaction reserved function that maximizes the seller's expected revenue.

[0010] The blockchain network obtains the transaction retention function and calls the chaincode to execute a sealed binary auction.

[0011] In some embodiments, the blockchain network includes multiple sub-blockchain networks created and established by multiple computing network operators, and the sub-blockchain networks are connected by establishing channels.

[0012] In some embodiments, the blockchain network includes a first sub-blockchain network provided by a first computing network operator and a second sub-blockchain network provided by a second computing network operator. The first sub-blockchain network and the second sub-blockchain network each contain two peer nodes. The first sub-blockchain network also includes an order node responsible for ordering transactions and a certificate authority component for registering the identities of each bidder.

[0013] In some embodiments, the preset neural network model includes a two-layer structure. The first layer performs M×J linear mappings on the bid prices of each bidder and divides the results into M groups, each containing J linear functions. The mapping function expression of the first layer is:

[0014]

[0015] Among them, weight parameter v i This represents the bid price of user i;

[0016] The second layer performs max and min calculations on the output of the first layer and introduces nonlinear factors to fit the virtual estimation function.

[0017] The mapping expression of the preset neural network model to the virtual estimation function is as follows:

[0018]

[0019] Among them, weight parameter v i This represents the bid price of user i;

[0020] The formula for the inverse transformation of the virtual estimation function is:

[0021]

[0022] Among them, weight parameter This represents the virtual valuation of the i-th bidder.

[0023] In some embodiments, the first virtual valuation of each bidder is input into a softmax function to output the probability of each bidder winning, calculated as follows:

[0024]

[0025] in, Let represent the probability that the i-th bidder wins. This represents the virtual valuation of the i-th bidder. The virtual valuation of the j-th bidder, where N represents the total number of bidders.

[0026] In some embodiments, the ReLU function is used to construct payment rules, and the actual payment price for each bidder is calculated based on the inverse transformation of the virtual valuation fitting model, including:

[0027] Let the first virtual user be b. N+1 When the first virtual user wins, it indicates that the auction has failed; the payment rules are constructed using the ReLU function, and the calculation formula is as follows:

[0028]

[0029] Where, p i This represents the final payment price when the i-th bidder wins. This represents the virtual valuation of the i-th bidder. This represents the highest virtual valuation among all bidders. This represents the second highest virtual valuation among all bidders, and N represents the total number of bidders.

[0030] In some embodiments, the loss function is calculated as follows:

[0031]

[0032] in, Let p represent the probability that the i-th bidder wins. i Let N represent the final payment price when the i-th bidder wins, N represent the total number of bidders, and L represent the number of auction rounds.

[0033] On the other hand, the present invention also provides a method for allocating computing power network resources, the method being executed synchronously on-chain and off-chain in a blockchain network, the method comprising the following steps:

[0034] The computing power leasing entity submits the computing power index parameters of the idle computing power resources to the client nodes provided by the computing network operator for uploading to the blockchain network;

[0035] Multiple bidders estimate the value of the idle computing power resources based on the computing power index parameters, determine the bid price, and call the chaincode of the blockchain network to submit the bid.

[0036] The blockchain network submits each bidder's bid price to an off-chain reservation price calculation model to output a transaction reservation price. The reservation price calculation model includes a clustering model, multiple virtual valuation fitting models, a softmax function, and a ReLU function. The clustering model, based on each bidder's historical bid probability distribution, divides each bidder into multiple clusters according to the similarity of the bid probability distribution using a preset clustering algorithm. Each cluster corresponds to a virtual valuation fitting model. The virtual valuation fitting model uses a preset neural network model to fit the virtual valuation function of the corresponding cluster in the Miles auction mechanism. The bid price of each bidder in each cluster is input into the corresponding virtual valuation fitting model, and a second virtual valuation for each bidder is output. The bidder with the highest second virtual valuation in each cluster is selected as a candidate bidder, and its corresponding second virtual valuation is output.

[0037] The payment rules are constructed using the ReLU function, and the actual payment prices of the remaining candidate bidders are calculated based on the inverse transformation of the virtual valuation fitting model. A first virtual user is introduced, whose corresponding second-type virtual valuation is 0. The transaction reservation price is calculated for the first virtual user through the inverse transformation of the virtual valuation fitting model. The second-type virtual valuation of each bidder is input into the softmax function to output the probability of the corresponding candidate bidder winning. The seller's expected revenue is calculated based on the actual payment price of each candidate bidder and the corresponding probability of winning. A loss function is constructed using the negative of the seller's expected revenue. The reservation price calculation model updates its parameters by minimizing the loss function and outputs the transaction reservation function that maximizes the seller's expected revenue.

[0038] The blockchain network obtains the transaction retention function and calls the chaincode to execute a sealed binary auction.

[0039] In some embodiments, the preset clustering algorithm uses KL divergence to measure the difference in bid distribution between two bidders;

[0040] Each cluster is obtained by clustering based on the historical bid distribution of bidders and is updated according to set conditions; the set conditions include updating according to set time intervals or set auction rounds.

[0041] On the other hand, the present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the steps of the above-described method.

[0042] The beneficial effects of the present invention are at least as follows:

[0043] The computing power network resource allocation method and apparatus of this invention deploys a reservation price calculation model off-chain to calculate the transaction reservation price and executes a sealed binary auction on-chain. A virtual valuation fitting model fits the virtual valuation function in the Myers auction mechanism using a neural network, and fits the cumulative distribution function and distribution density function of the valuation using a neural network to optimize and find the transaction reservation price that brings the highest return to users, encouraging users to sell idle computing power resources. By deploying the transaction reservation price calculation and auction process offline and online respectively, the design difficulty of blockchain network smart contracts can be reduced.

[0044] Furthermore, in scenarios where bidders exhibit differences in their bid distribution, a clustering process is introduced to construct virtual valuation fitting models for each category under mixed bid distribution scenarios. This aims to better fit the cumulative distribution function and distribution density function of valuations in different bid distribution scenarios, thereby reducing the impact of differences in bid distribution among different bidders on returns and improving profitability.

[0045] Additional advantages, objects, and features of the invention will be set forth in part in the description which follows, and will also become apparent in part to those skilled in the art upon studying the description, or may be learned by practice of the invention. The objects and other advantages of the invention can be realized and obtained by means of the structures specifically pointed out in the specification and drawings.

[0046] Those skilled in the art will understand that the objectives and advantages achievable with the present invention are not limited to those specifically described above, and that the above and other objectives achievable with the present invention will become clearer from the following detailed description. Attached Figure Description

[0047] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, are not intended to limit the scope of the invention. In the drawings:

[0048] Figure 1 This is a diagram of the blockchain network structure used in the computing power network resource allocation method according to an embodiment of the present invention.

[0049] Figure 2 This is a schematic diagram of the virtual valuation fitting model in the computing power network resource allocation method according to an embodiment of the present invention.

[0050] Figure 3 This is a schematic diagram illustrating the execution logic of the reserved price calculation model in the computing power network resource allocation method when all bidders have the same bid distribution, according to an embodiment of the present invention.

[0051] Figure 4 This is a schematic diagram illustrating the execution logic of the reserved price calculation model in the computing power network resource allocation method when all bidders have the same bid distribution, according to another embodiment of the present invention.

[0052] Figure 5 This invention presents a comparison chart showing the effectiveness of the computing power network resource allocation method described in one embodiment of the invention with the calculation of the transaction reservation price using a real virtual valuation function when all bidders have the same bid distribution.

[0053] Figure 6 To address scenarios where bidders exhibit varying bid distributions, this invention presents a benefit analysis diagram comparing the computing power network resource allocation method described in one embodiment with and without classification, using the classification method. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.

[0055] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.

[0056] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0057] It should also be noted that, unless otherwise specified, the term "connection" in this article can refer not only to a direct connection, but also to an indirect connection involving an intermediary.

[0058] In the following description, embodiments of the invention will be illustrated with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts, or the same or similar steps.

[0059] To optimize resource allocation in computing networks, existing technologies primarily aim to maximize bandwidth utilization and social welfare, without considering how to maximize efficiency. Furthermore, current computing resource allocation schemes do not address how to restrict resource usage by users, leading to overall inefficiency in resource utilization.

[0060] Therefore, this invention provides a method and apparatus for allocating computing network resources, and designs a computing network resource auction mechanism based on the Myerson auction. In scenarios where user bids are distributed similarly, the invention introduces a reserve price to combine the Myerson auction with Hyperledger for the first time, maximizing resource allocation revenue while creating conditions for the application of blockchain in auction scenarios. In scenarios with different user bid distributions, a Myerson auction network model based on a clustering algorithm is proposed, enabling resource allocation to maximize revenue in more complex scenarios.

[0061] This application provides corresponding solutions for two application scenarios, specifically including scenarios where the bids of all bidders are distributed in the same way and scenarios where the bids of all bidders are distributed in a mixed way.

[0062] Specifically, on one hand, the present invention provides a method for allocating computing power network resources. For scenarios where bidders have the same bid distribution, the method is executed synchronously on-chain and off-chain within the blockchain network. The method includes the following steps S101–S104:

[0063] Step S101: The computing power leasing entity submits the computing power index parameters of the idle computing power resources to the client node provided by the computing network operator for uploading to the blockchain network.

[0064] Step S102: Multiple bidders estimate the value of the available computing power resources based on the computing power index parameters, determine the bid price, and call the blockchain network's chaincode to submit the bid.

[0065] Step S103: Refer to Figure 3 The blockchain network submits each bidder's bid price to an off-chain reservation price calculation model to output a transaction reservation price. This model includes a virtual valuation fitting model, a softmax function, and a ReLU function. The virtual valuation fitting model uses a pre-defined neural network model to fit the virtual valuation function in the Myers auction mechanism. Each bidder's bid price is input into the virtual valuation fitting model, which outputs the first type of virtual valuation for each bidder. The ReLU function is used to construct payment rules, and based on the inverse transformation of the virtual valuation fitting model, the actual payment price for each bidder is calculated. A first virtual user is introduced, whose corresponding first type of virtual valuation is 0. The actual payment price of the first virtual user after the inverse transformation of the virtual valuation fitting model is used as the transaction reservation price. Each bidder's first virtual valuation is input into the softmax function to output the probability of each bidder winning. The seller's expected revenue is calculated based on each bidder's actual payment price and corresponding winning probability. A loss function is constructed using the inverse of the seller's expected revenue. The reservation price calculation model updates its parameters by minimizing the loss function and outputs the transaction reservation function that maximizes the seller's expected revenue.

[0066] Step S104: The blockchain network obtains the transaction retention function and calls the chaincode to execute a sealed binary auction.

[0067] It should be noted beforehand that the blockchain network is deployed by a computing network operator to execute the auction process for computing resources. This embodiment introduces the Miles auction scheme to maximize user returns in scenarios with a large number of bidders. This embodiment introduces a transaction reserve price, separating the calculation of the transaction reserve price from the actual auction process, executing them offline and online respectively. The calculation of the transaction reserve price is performed offline, while only the auction portion is deployed and executed online. This saves computing power on the blockchain and simplifies the deployment of smart contracts.

[0068] In some embodiments, the blockchain network includes multiple sub-blockchain networks created and established by multiple computing network operators, and the sub-blockchain networks are connected by establishing channels.

[0069] In some embodiments, the blockchain network includes a first sub-blockchain network provided by a first computing network operator and a second sub-blockchain network provided by a second computing network operator. The first and second sub-blockchain networks each contain two peer nodes. The first sub-blockchain network also includes an order node responsible for ordering transactions and a certificate authority component for registering the identities of each bidder.

[0070] In some embodiments, this application uses a simple auction system framework built on Hyperledger Fabric, such as... Figure 1 As shown. The organizations were created by different computing network operators. Figure 1 It contains two organizations, Org1 and Org2. Each organization contains two peer nodes, and Org1 additionally provides an order node responsible for transaction sorting.

[0071] Taking Org1 as an example, this organization registers... The node submits transactions to linked clients and helps the order node distribute transactions. Additionally, the organization needs to register an endorsing node. Used to verify the legitimacy of transactions. During the auction process, Org1 and Org2 can create a channel C1 for joint auctions. Furthermore, Org1 has a built-in Certificate Authority (CA1) component. Bidder user1 needs to register with the organization's Fabric CA to obtain an identity certificate issued by the organization. The certified user can complete the bidding by calling the chaincode through the client node client1.

[0072] In step S101, the computing power leasing entity is the user with idle computing power resources in the computing power network. The computing power leasing entity submits the indicator parameters of the idle computing power resources to the computing network operator for on-chain storage and auction. The computing power leasing entity provides the computing power indicator parameters of the idle computing power resources, which may include parameters related to the average execution speed of fixed-point instructions per second (MIPS) and floating-point operations per second (FLOPS).

[0073] In step S102, multiple bidders can estimate the value of the available computing power resources based on the computing power index parameters and submit a bid price. Bidders can access and query the computing power resources that are being offered for sale through client nodes in the blockchain network and bid on them.

[0074] In step S103, in the scenario of independent and identically distributed user bids for computing network resources auctions, the resource suppliers consist of tenants selling idle resources. Based on the Myerson auction, the goal of maximizing the expected revenue of the computing network operator can be achieved. Furthermore, the Myerson auction can be replaced by a sealed-bid binary auction with a transaction reservation price t, denoted as SPA0. The specific rules are as follows:

[0075] Assume the highest bidder in the auction is b. i The bid value is v i The second highest bidder was b. j The bid value is v j SPA0 allocates the item to the highest bidder. A reserve price t is introduced in the payment rule p. When v i When v < t, the phenomenon of unsold tickets will occur; when v i ≥t, v j When <t, the final actual transaction price is p. i =t; when v j When ≥t, then use price v j The transaction is completed. Then, in a single-item auction scenario, bidder b is defined. n The benefit function is as follows:

[0076]

[0077] Among them, u n Indicates benefit, v n Indicates bidder b n The bid price, p n Indicates bidder b n The actual transaction price.

[0078] In SPA0, the highest benefit for a bidder is achieved only by submitting an honest bid, unaffected by the behavior of other bidders. SPA0 satisfies the IC property, which means incentive compatibility, i.e., under the mechanism rules, each participant can achieve their optimal strategy without considering the behavior of other participants. Participants are not influenced by the mechanism rules to take different actions to obtain better results. Satisfying IC means that users can only achieve optimal benefits in the scenario of submitting honest bids. At the same time, since the winner's payment will not exceed its valuation, and the non-winner's utility is 0, there is no negative benefit. SPA0 satisfies the IR property, which means individual rationality. Based on the above allocation and payment rules, the expected return model satisfying the IC and IR properties can be defined as follows:

[0079]

[0080] In the formula, r i Indicates bidder b iFor the estimated value of the auctioned item, v represents the bidder's actual offer. j This represents the second-highest bid value in round t of the transaction, and g is the allocation rule in the auction mechanism.

[0081] This embodiment uses an offline-deployed reservation price calculation model to fit the virtual valuation function in the Myerson auction process. By maximizing user returns, it calculates the transaction reservation price to obtain the maximum benefit. In a Myerson auction, the optimal transaction reservation price can be calculated using the virtual valuation function. In a Myerson auction, each bidder's bid price is first transformed into a virtual valuation using the following formula:

[0082]

[0083] Among them, v i Let F represent the bid price of the i-th bidder. i (v i Let f represent the cumulative distribution function of bid prices. i (v i Let represent the probability density function of the bid price. In real-world auction scenarios, the distribution of user bids is quite complex, making it difficult to directly calculate the virtual valuation function. This application utilizes a neural network with a reasonable structural design to fully approximate arbitrarily complex nonlinear relationships. Specifically, the reserved price calculation model includes a virtual valuation fitting model. This model uses a pre-defined neural network model to fit the virtual valuation function in the Myers auction mechanism. It takes each bidder's bid price as input and, after mining the characteristics of the bid distribution, outputs the first-type virtual valuation for each bidder.

[0084] In some embodiments, Figure 2 As shown, the preset neural network model includes a two-layer structure. The first layer performs M×J linear mappings on the bid prices of each bidder and divides the results into M groups, each containing J linear functions. The mapping function expression for the first layer is:

[0085]

[0086] Among them, weight parameter v i This represents the bid price of user i;

[0087] The second layer performs max and min calculations on the output of the first layer and introduces nonlinear factors to fit the virtual estimation function.

[0088] The mapping expression of the preset neural network model to the virtual estimation function is as follows:

[0089]

[0090] Among them, weight parameter v i This represents the bid price of user i;

[0091] A significant advantage of this two-layer network structure is its ease of inverse transformation. Based on the inverse transformation of the virtual estimation fitting model, the formula for the inverse transformation is:

[0092]

[0093] Among them, weight parameter Let represent the virtual valuation of the i-th bidder. Based on this, by introducing a first virtual user with a virtual valuation of 0, the reserve price for the transaction can be calculated.

[0094] To ensure the auction satisfies the IC property, the allocation rule design must be implementable. According to Myerson's lemma, an allocation rule g is implementable if and only if it is monotonically increasing. To ensure the mapping function from actual valuation to virtual valuation is monotonically increasing, this patent controls the weights during network design by setting the weight w to be greater than 0. Because the softmax function is monotonically increasing, the allocation rule g is also monotonically increasing, meaning bidders with higher virtual valuations have a higher probability of winning.

[0095] In some embodiments, the first virtual valuation of each bidder is input into a softmax function to output the probability of each bidder winning, calculated as follows:

[0096]

[0097] in, Let represent the probability that the i-th bidder wins. This represents the virtual valuation of the i-th bidder. The virtual valuation of the j-th bidder, where N represents the total number of bidders.

[0098] After obtaining the win rate of each bidder, it is also necessary to calculate the actual payment price for each winning bidder. Specifically, pricing is based on the second-highest price method on the virtual bid, and then converted back to the real bid domain for deduction. In some embodiments, the payment rules are constructed using the ReLU function, and the actual payment price corresponding to each bidder and the first virtual user is calculated based on the inverse transformation of the virtual valuation fitting model, including:

[0099] Let the first virtual user be b. N+1When the first virtual user wins, the auction fails. The payment rules are constructed using the ReLU function, and the calculation formula is as follows:

[0100]

[0101] Where, p i This represents the final payment price when the i-th bidder wins. This represents the virtual valuation of the i-th bidder. This represents the highest virtual valuation among all bidders. This represents the second highest virtual valuation among all bidders, and N represents the total number of bidders.

[0102] Furthermore, based on each bidder's probability of winning and the actual payment price upon winning, the seller's expected revenue can be calculated, and a loss function can be constructed. Specifically, the formula for the loss function is:

[0103]

[0104] in, Let p represent the probability that the i-th bidder wins. i Let N represent the final payment price when the i-th bidder wins, N represent the total number of bidders, and L represent the number of auction rounds.

[0105] During iterative training, the profit is maximized by minimizing the loss function, and the transaction retention function generated based on this is used for online auctions.

[0106] In step S104, on the chain, a binary auction is performed based on the obtained transaction reservation price by calling the chaincode.

[0107] On the other hand, the present invention also provides a method for allocating computing power network resources. For scenarios where bids from various bidders are distributed in a mixed manner, the method is executed simultaneously on-chain and off-chain within the blockchain network. The method includes the following steps S201–S204:

[0108] Step S201: The computing power leasing entity submits the computing power index parameters of the idle computing power resources to the client node provided by the computing network operator for uploading to the blockchain network.

[0109] Step S202: Multiple bidders estimate the value of the available computing power resources based on the computing power index parameters, determine the bid price, and call the blockchain network's chaincode to submit the bid.

[0110] Step S203: Refer to Figure 4The blockchain network submits each bidder's bid price to an off-chain reservation price calculation model to output the transaction reservation price. The reservation price calculation model includes a clustering model, multiple virtual valuation fitting models, a softmax function, and a ReLU function. The clustering model, based on each bidder's historical bidding probability distribution, divides each bidder into multiple clusters according to the similarity of their bidding probability distributions using a preset clustering algorithm. Each cluster corresponds to a virtual valuation fitting model. The virtual valuation fitting model uses a preset neural network model to fit the virtual valuation function of the corresponding cluster in the Miles auction mechanism. The bid price of each bidder in each cluster is input into the corresponding virtual valuation fitting model, which outputs a second virtual valuation for each bidder. The bidder with the highest second virtual valuation in each cluster is selected as a candidate bidder, and its corresponding second virtual valuation is output.

[0111] The payment rules are constructed using the ReLU function, and the actual payment prices of the remaining candidate bidders are calculated based on the inverse transformation of the virtual valuation fitting model. A first virtual user is introduced, whose corresponding second-type virtual valuation is 0. The transaction reservation price is calculated for the first virtual user through the inverse transformation of the virtual valuation fitting model. The second-type virtual valuation of each bidder is input into the softmax function to output the probability of the corresponding candidate bidder winning. The seller's expected revenue is calculated based on the actual payment price and the corresponding probability of winning of each candidate bidder. A loss function is constructed using the negative of the seller's expected revenue. The reservation price calculation model updates its parameters by minimizing the loss function and outputs the transaction reservation function that maximizes the seller's expected revenue.

[0112] Step S204: The blockchain network obtains the transaction retention function and calls the chaincode to execute a sealed binary auction.

[0113] Steps S201, S202, and S204 can refer to the contents of steps S101, S102, and S104 above.

[0114] In step S203, for scenarios where users have different bidding distributions during the auction of computing network resources, since the virtual valuation functions corresponding to different distributions are also different, this embodiment first clusters each bidder in the scenario of mixed distribution bidding, and then constructs a virtual valuation fitting model for each cluster to fit the virtual valuation function of different distributions.

[0115] In some embodiments, the preset clustering algorithm uses KL divergence to measure the difference in bid distribution between two bidders; each cluster is obtained by clustering based on the historical bid distribution of bidders and is updated according to set conditions; the set conditions include updating according to set time intervals or set auction rounds.

[0116] Specifically, the clustering model uses KL divergence to classify bidding users, calculating the distance between different users based on their historical bidding distribution. Assuming two users are p and q, the KL divergence calculation formula is as follows:

[0117]

[0118] Where p(x) and q(x) represent two user bidding probability distributions, and x represents the user's bidding price.

[0119] In clustering algorithms, KL divergence helps this patent group users with similar bidding distributions. Because KL divergence is asymmetric, the formula for calculating the distance D between clustering algorithms in this embodiment is as follows:

[0120]

[0121] The KL divergence-based clustering algorithm first selects one bidding user as the initial cluster center. Then, it calculates the minimum distance D from each of the remaining users to all cluster centers; a larger D indicates a higher probability of becoming a cluster center. A roulette wheel algorithm is used to select the next cluster center, and this process is repeated to generate k cluster centers. After initialization, the distance D from each bidding user to a cluster center is calculated, and the bidding users are assigned to the nearest cluster center. The cluster centers are then updated based on the user bidding distribution within the cluster. Finally, the algorithm checks if the clustering has converged. If not, the distances from users to cluster centers are recalculated until convergence or a specified number of training rounds are reached.

[0122] After the clustering is completed, the bid price of each bidder in each cluster can be denoted as a. k k represents the number of clusters. A virtual estimation fitting model is set for each cluster, and a... k Input the corresponding virtual valuation fitting model, fit the virtual valuation in the Myerson auction, and denote it as the second virtual valuation. The highest second virtual valuation in the k-th aggregation is denoted as... The construction method of the virtual valuation fitting model can refer to step S103.

[0123] Assume the second virtual valuation corresponding to the k-th type of bidding distribution is denoted as... S represents the number of users in the k-th bidding distribution, where the maximum value is denoted as maxV. k Therefore, the probability of each candidate bidder ultimately winning can be defined using the following allocation rule:

[0124]

[0125] in, Let e ​​represent the probability that the k-th candidate bidder wins, where K represents the number of clusters and e is the natural base.

[0126] Based on the same principle as step S103, each virtual valuation fitting model in each step S203 calculates the true bid of each candidate bidder through inverse transformation. In this process, a virtual user is introduced, whose corresponding second-type virtual valuation is 0, and the transaction reservation price is calculated through the inverse transformation of the virtual valuation fitting model.

[0127] In the process of constructing payment rules using the ReLU function and calculating the actual payment price for the remaining candidate bidders based on the inverse transformation of the virtual valuation fitting model, it is assumed that user b of type k... ks The maximum virtual valuation corresponding to the bid is denoted as . User b j The virtual valuation corresponding to the bid is second only to b. ki , recorded as The actual payment price for each candidate bidder can then be expressed as:

[0128]

[0129] Furthermore, based on the winning probability of each candidate bidder and the actual payment price upon winning, the seller's expected revenue can be calculated, and a loss function can be constructed. Specifically, the formula for the loss function is:

[0130]

[0131] Among them, g k p represents the probability that the k-th candidate bidder will win. k Let L represent the actual price paid by the k-th candidate bidder, and L represent the auction round number.

[0132] During iterative training, the profit is maximized by minimizing the loss function, and the transaction retention function generated based on this is used for online auctions.

[0133] On the other hand, the present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the steps of the above-described method.

[0134] The present invention will now be described with reference to a specific embodiment:

[0135] This embodiment evaluates the performance of the auction model through a series of simulation experiments. First, a computing resource auction network for different scenarios was built based on TensorFlow. The general parameters of the network model are as follows: the number of items auctioned in each round is 1, the number of training sets and test sets are both 10,000, the learning rate lr = 0.001, each training batch requires 64 rounds of complete user bidding information, the number of iteration rounds is 200,000, the total number of linear functions in the model is N = 100, the number of groups is M = 10, and each group contains J = 10 linear functions.

[0136] First, we analyzed auction scenarios with the same resource valuation distribution. In Experiment 1, we tested the network's accuracy in calculating the reservation price and its effect on improving seller revenue. During training, we used user data with three different distribution types: uniform distribution X~U(0,1), exponential distribution X~E(3), and random distribution rand. In the random distribution, a random number between 0 and 1 was introduced. When the random number was below 0.75, the user's bid distribution was X~U(0,2); when the random number was greater than or equal to 0.75, the user's bid distribution was X~U(2,5). All three distributions can directly calculate the virtual valuation function, thus providing a more intuitive understanding of the accuracy of the reservation price calculation. The final experimental results are as follows: Figure 5 As shown, Figure 5 The results show the virtual estimation function fitted by a neural network for bids under three different distributions, and the virtual estimation function calculated according to formula (3). When the points intersect, the reserve price for the transaction can be accurately found according to formula (6).

[0137] The above experiments were all conducted under the scenario of independent and identically distributed user bids. Considering that there are different scenarios of user bid distribution in the actual auction process, Experiment 2 analyzes the auction scenario with mixed bid distribution. The user bid distribution includes three types: uniform distribution X~U(0,1), uniform distribution X~U(0,2), and exponential distribution X~E(3).

[0138] In the mixed bidding distribution scenario, five samples are generated for each of the three distributions mentioned above. For these samples, users are first classified using a KL divergence-based clustering algorithm, with three types. The number of input nodes in the model equals the number of types, and users of the same type are input from the same node. As a comparison, an existing auction model is introduced, with a similar structure. Figure 3 Similarly, without classifying users, all user bids are used as input to train a Myerson auction model. The expected auction revenues produced by the two models are as follows: Figure 6As shown. The auction model proposed in this embodiment has a higher expected return than the existing Myerson auction model, and is closer to the expected return under a real Myerson auction.

[0139] This invention introduces Myerson auctions into computing network resource trading for the first time, designing a computing network resource auction mechanism for complex bidding distribution scenarios. The auction aims to maximize the expected revenue of resource suppliers, making it more closely aligned with real-world scenarios. In scenarios with uniform user bidding distribution, this invention combines Myerson auctions with smart contracts for the first time. By introducing a reserve price, the auction network model and the auction transaction model are separated, creating conditions for the introduction of Hyperledger into the auction mechanism. In scenarios with mixed user bidding distribution, this invention proposes an auction network model based on KL divergence classification, allowing the auction mechanism to be extended to mixed bidding distribution scenarios. Compared to existing auction network models, the model proposed in this invention is less affected by the number of users and bidding distribution, resulting in a more significant improvement in revenue.

[0140] This invention, in scenarios where user bid distribution is the same, introduces a reserve price to separate the auction network model from the auction transaction model, and successfully incorporates blockchain into resource allocation to maximize the expected returns for resource suppliers. In scenarios with different user bid distributions, this invention proposes an auction model based on KL divergence classification, which exhibits better training performance compared to existing Myerson auction models.

[0141] In summary, the computing power network resource allocation method and apparatus of this invention deploys a reservation price calculation model off-chain to calculate the transaction reservation price and executes a sealed binary auction on-chain. The virtual valuation fitting model uses a neural network to fit the virtual valuation function in the Myers auction mechanism, and uses a neural network to fit the cumulative distribution function and distribution density function of the valuation, optimizing and finding the transaction reservation price that brings the highest return to the user, thus encouraging users to sell idle computing power resources. By deploying the transaction reservation price calculation and auction process offline and online respectively, the design difficulty of blockchain network smart contracts can be reduced.

[0142] Furthermore, in scenarios where bidders exhibit differences in their bid distribution, a clustering process is introduced to construct virtual valuation fitting models for each category under mixed bid distribution scenarios. This aims to better fit the cumulative distribution function and distribution density function of valuations in different bid distribution scenarios, thereby reducing the impact of differences in bid distribution among different bidders on returns and improving profitability.

[0143] Corresponding to the above method, the present invention also provides an apparatus / system including a computer device, the computer device including a processor and a memory, the memory storing computer instructions, the processor executing the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the apparatus / system performs the steps of the method as described above.

[0144] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the aforementioned edge computing server deployment method. The computer-readable storage medium can be a tangible storage medium, such as random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, floppy disks, hard disks, removable storage disks, CD-ROMs, or any other form of storage medium known in the art.

[0145] Those skilled in the art will understand that the exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention. When implemented in hardware, it can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the desired tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave.

[0146] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0147] In this invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or in place of features of other embodiments.

[0148] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations of the embodiments of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for allocating computing power network resources, characterized in that, The method is executed synchronously on-chain and off-chain in the blockchain network, and includes the following steps: The computing power leasing entity submits the computing power index parameters of the idle computing power resources to the client nodes provided by the computing network operator for uploading to the blockchain network; the computing power index parameters include parameters related to the average execution speed of single-word fixed-point instructions and the number of floating-point operations per second; Multiple bidders estimate the value of the idle computing power resources based on the computing power index parameters, determine the bid price, and call the chaincode of the blockchain network to submit the bid. The blockchain network submits each bidder's bid price to an off-chain reserved price calculation model to output a transaction reserved price. This reserved price calculation model includes a virtual valuation fitting model, a softmax function, and a ReLU function. The virtual valuation fitting model uses a pre-defined neural network model to fit the virtual valuation function in the Miles auction mechanism. Each bidder's bid price is input into the virtual valuation fitting model, which outputs a first-type virtual valuation for each bidder. The ReLU function is used to construct payment rules, and based on the inverse transformation of the virtual valuation fitting model, the actual payment price for each bidder is calculated. A first virtual user is introduced, whose corresponding first-type virtual valuation is 0. The first virtual user's transaction reserved price is calculated through the inverse transformation of the virtual valuation fitting model. Each bidder's first virtual valuation is input into the softmax function to output the probability of each bidder winning. The seller's expected revenue is calculated based on each bidder's actual payment price and corresponding winning probability. A loss function is constructed using the negative of the seller's expected revenue. The reserved price calculation model updates its parameters to minimize the loss function and outputs the transaction reserved function that maximizes the seller's expected revenue. The blockchain network obtains the transaction retention function and calls the chaincode to execute a sealed binary auction; The preset neural network model includes a two-layer structure. The first layer performs M×J linear mappings on the bid prices of each bidder and divides the results into M groups, each containing J linear functions. The mapping function expression of the first layer is: Among them, weight ,parameter , m=1,2,…,M, j=1,2,…,J; v i This represents the bid price of user i; The second layer performs max and min calculations on the output of the first layer and introduces nonlinear factors to fit the virtual estimation function. The mapping expression of the preset neural network model to the virtual estimation function is as follows: Among them, weight ,parameter , m=1,2,…,M, j=1,2,…,J; v i This represents the bid price of user i; The formula for the inverse transformation of the virtual estimation function is: Among them, weight ,parameter , m=1,2,…,M, j=1,2,…,J; This represents the virtual valuation of the i-th bidder.

2. The computing power network resource allocation method according to claim 1, characterized in that, The blockchain network comprises multiple sub-blockchain networks created and established by multiple computing network operators, and these sub-blockchain networks are connected through the establishment of channels.

3. The computing power network resource allocation method according to claim 2, characterized in that, The blockchain network includes a first sub-blockchain network provided by a first computing network operator and a second sub-blockchain network provided by a second computing network operator. The first sub-blockchain network and the second sub-blockchain network each contain two peer nodes. The first sub-blockchain network also includes an order node responsible for transaction sorting and a certificate authority component for registering the identities of each bidder.

4. The computing power network resource allocation method according to claim 1, characterized in that, The first virtual estimate of each bidder is input into the softmax function to output the probability of each bidder winning. The calculation formula is as follows: in, Let represent the probability that the i-th bidder wins. This represents the virtual valuation of the i-th bidder. The virtual valuation of the j-th bidder, where N represents the total number of bidders.

5. The computing power network resource allocation method according to claim 4, characterized in that, Payment rules are constructed using the ReLU function, and the actual payment price for each bidder is calculated based on the inverse transformation of the virtual valuation fitting model, including: Let the first virtual user be b. N+1 When the first virtual user wins, it indicates that the auction has failed; the payment rules are constructed using the ReLU function, and the calculation formula is as follows: in, This represents the final payment price when the i-th bidder wins. This represents the virtual valuation of the i-th bidder. This represents the highest virtual valuation among all bidders. This represents the second highest virtual valuation among all bidders, and N represents the total number of bidders.

6. The computing power network resource allocation method according to claim 1, characterized in that, The formula for calculating the loss function is: in, Let represent the probability that the i-th bidder wins. Let N represent the final payment price when the i-th bidder wins, and let N represent the total number of bidders. Indicates the number of auction rounds.

7. A method for allocating computing power network resources, characterized in that, The method is executed synchronously on-chain and off-chain in the blockchain network, and includes the following steps: The computing power leasing entity submits the computing power index parameters of the idle computing power resources to the client nodes provided by the computing network operator for uploading to the blockchain network; the computing power index parameters include parameters related to the average execution speed of single-word fixed-point instructions and the number of floating-point operations per second; Multiple bidders estimate the value of the idle computing power resources based on the computing power index parameters, determine the bid price, and call the chaincode of the blockchain network to submit the bid. The blockchain network submits each bidder's bid price to an off-chain reservation price calculation model to output a transaction reservation price. This reservation price calculation model includes a clustering model, multiple virtual valuation fitting models, a softmax function, and a ReLU function. The clustering model, based on each bidder's historical bid probability distribution, divides each bidder into multiple clusters according to the similarity of their bid probability distributions using a preset clustering algorithm. Each cluster corresponds to a virtual valuation fitting model. The virtual valuation fitting model uses a preset neural network model to fit the virtual valuation function of the corresponding cluster in the Miles auction mechanism. The bid price of each bidder in each cluster is input into the corresponding virtual valuation fitting model, which outputs a second virtual valuation for each bidder. The highest second virtual valuation in each cluster is selected. The following steps are taken: First, bidders are selected as candidate bidders, and their corresponding second virtual valuations are output. A payment rule is constructed using the ReLU function, and the actual payment price for the remaining candidate bidders is calculated based on the inverse transformation of the virtual valuation fitting model. A first virtual user is introduced, whose corresponding second-type virtual valuation is 0. The first virtual user's transaction reservation price is calculated using the inverse transformation of the virtual valuation fitting model. The second-type virtual valuations of each bidder are input into the softmax function to output the probability of the corresponding candidate bidder winning. The seller's expected revenue is calculated based on the actual payment price and the corresponding probability of winning for each candidate bidder. A loss function is constructed using the negative of the seller's expected revenue. The reservation price calculation model updates its parameters by minimizing the loss function and outputs the transaction reservation function that maximizes the seller's expected revenue. The blockchain network obtains the transaction retention function and calls the chaincode to execute a sealed binary auction; The preset neural network model includes a two-layer structure. The first layer performs M×J linear mappings on the bid prices of each bidder and divides the results into M groups, each containing J linear functions. The mapping function expression of the first layer is: Among them, weight ,parameter , m=1,2,…,M, j=1,2,…,J; v i This represents the bid price of user i; The second layer performs max and min calculations on the output of the first layer and introduces nonlinear factors to fit the virtual estimation function. The mapping expression of the preset neural network model to the virtual estimation function is as follows: Among them, weight ,parameter , m=1,2,…,M, j=1,2,…,J; v i This represents the bid price of user i; The formula for the inverse transformation of the virtual estimation function is: Among them, weight ,parameter , m=1,2,…,M, j=1,2,…,J; This represents the virtual valuation of the i-th bidder.

8. The computing power network resource allocation method according to claim 7, characterized in that, The preset clustering algorithm uses KL divergence to measure the difference in the bid distribution between two bidders; Each cluster is obtained by clustering based on the historical bid distribution of bidders and is updated according to set conditions; the set conditions include updating according to set time intervals or set auction rounds.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1 to 8.

Images