Classic-quantum hybrid architecture cloud-side server cluster computing power distribution method

Abstract

The invention discloses a classical-quantum hybrid architecture cloud-side server cluster computing power distribution method, which comprises the following steps: collecting computing power distribution requests of community users, constructing computing power demand information on a quantum circuit on quantum simulation software, and converting a parameter part into a quadratic unconstrained binary optimization model, so as to improve the computing power distribution efficiency. And the model is optimized by using a classical optimization method to obtain updated parameters, the parameters are introduced into the quantum circuit for calculation, and an optimal solution is gradually approached by iteratively optimizing the parameters in the quantum circuit in combination with a variable component sub-algorithm. Judging whether the result meets the expectation or not, and if yes, outputting; and if not, optimizing the quadratic unconstrained binary optimization model on a classic computer by using a classic optimization method to obtain updated parameters. And adjusting the quantum circuit according to the updated parameters. And repeating the steps until a judgment condition is met or a preset number of times is reached. The method has the advantages of improving calculation efficiency, reducing hardware dependence, enhancing algorithm adaptability, improving calculation precision and the like.

Description

Technical Field

[0001] The present invention relates to the field of classical-quantum hybrid computing and optimization, and in particular to a cloud-edge server cluster computing power allocation method for a classical-quantum hybrid architecture. Background Art

[0002] With the continuous advancement of computing technology, quantum computing, as an emerging computing paradigm, has shown tremendous potential. In an era of increasing information users, computing power has become a resource. This method addresses the question of how to use quantum computing to rationally allocate computing power from cloud-edge servers to community users. However, quantum hardware is still in its developmental stages and faces numerous limitations, such as a limited number of qubits and insufficient quantum state stability. Meanwhile, classical computers offer significant advantages in hardware maturity and software ecosystems. Therefore, hybrid architectures combining quantum and classical computing have become a hot topic of research. Existing quantum computing methods mostly rely on direct computations on quantum hardware, but due to hardware limitations, computational efficiency and accuracy often fall short of practical requirements. On the other hand, while classical computers face performance bottlenecks when processing large-scale problems, they offer extensive experience and mature tools for algorithm optimization and parameter tuning. Therefore, this invention focuses on leveraging the advantages of both classical and quantum computers to design an efficient classical-quantum hybrid algorithm for cloud-edge server computing distribution. Summary of the Invention

[0003] In order to improve the efficiency of classical computers in processing cloud-edge server computing power allocation and adapt to quantum algorithms in advance, the purpose of the present invention is to provide a cloud-edge server cluster computing power allocation method with a classical-quantum hybrid architecture. Through the collaborative work of classical computers and quantum computing, the parameters in the quantum circuit are optimized to improve computing efficiency.

[0004] The technical solutions of the present invention are as follows:

[0005] A method for allocating computing power to a cloud-edge server cluster in a classical-quantum hybrid architecture, comprising the following steps:

[0006] 1) The cloud-edge server cluster includes:

[0007] A cloud server;

[0008] a plurality of edge servers connected to the cloud server;

[0009] 2) A classical-quantum hybrid architecture is deployed on the cloud server. Community users send computing power demand information to the cloud server, which then uses the classical-quantum hybrid architecture to calculate the computing power allocation coefficient. The computing power allocation result is then sent to the cloud server and edge server, which then process the demand according to the computing power allocation result.

[0010] The method of the present invention designs a classical-quantum hybrid architecture. By collecting computing power allocation requests from community users, the computing power demand information is constructed into a quantum circuit on the quantum simulation software, and the parameter part is converted into a quadratic unconstrained binary optimization model. The model is optimized using classical optimization methods to obtain updated parameters, and the parameters are introduced into the quantum circuit calculation. Combined with the variational quantum algorithm, the parameters in the quantum circuit are iteratively optimized to gradually approach the optimal solution. The result is judged whether it meets the expectations. If it does, it is output; if it does not, the quadratic unconstrained binary optimization model is optimized using classical optimization methods on a classical computer to obtain updated parameters. The quantum circuit is adjusted according to the updated parameters. The above steps are repeated until the judgment conditions are met or the preset number of iterations is reached. The present invention has the advantages of improving computing efficiency, reducing hardware dependence, enhancing algorithm adaptability, and improving computing accuracy.

[0011] In step 2), the classical-quantum hybrid architecture includes:

[0012] Parameter input module, used to initialize the parameter part;

[0013] A quantum circuit simulation module is used to input the parameters into the module and iteratively optimize the constructed demand matrix to obtain the computing power allocation coefficient;

[0014] The parameter optimization output module is used to determine whether the computing power allocation coefficient meets the expected value of the target. If it does, the computing power allocation coefficient is output as a result; if it does not, the computing power allocation coefficient is returned to the quantum circuit simulation module.

[0015] In step 2), the parameters are iteratively optimized to obtain the computing power allocation coefficient, specifically including:

[0016] 2.1) Construct the basic quantum gates of the quantum circuit, convert the community user location, computing power requirements, and edge server location information into the quantum circuit, convert the parameter portion of the quantum circuit into a quadratic unconstrained binary optimization model, use classical optimization algorithms to optimize the model, obtain initial parameters, introduce the initial parameters into the quantum circuit, run the quantum circuit on a computer, simulate the quantum computing process, and obtain measurement results;

[0017] 2.2) Determine whether the measurement result meets expectations. If so, output the result. If not, use a classical optimization algorithm on a computer to optimize the quadratic unconstrained binary optimization model again to obtain updated parameters. Adjust the quantum circuit based on the updated parameters. Repeat the steps until expectations are met or the preset number of iterations is reached. The parameters at this point are used as the computing power allocation coefficient.

[0018] In steps 2.1) and 2.2), the classical optimization algorithm adopts a simulated annealing algorithm or a genetic algorithm.

[0019] In step 2.2), the expectation is:

[0020]

[0021] Among them, Q(σ) is the expectation. When Q(σ) < 0, it is determined that it meets the expectation. σ is a variable with a value of {-1, 1}. i, j represent the row and column numbers of the matrix in the quadratic unconstrained binary optimization model. A ij represents the coefficient of the corresponding matrix position, which is composed of the edge server and the user's computing power requirements. μ is the penalty coefficient, which ranges from [2,5]. ij Represents the bias of σ at the corresponding matrix position.

[0022] In step 2), the computing power allocation result is obtained according to the computing power allocation coefficient, which specifically includes:

[0023] The computing power allocation coefficient is the final calculated parameter, according to the parameter function:

[0024]

[0025] The computing power allocation results and total cost can be obtained.

[0026] Minimize C: Minimize cost.

[0027] n: n edge servers.

[0028] m: m users.

[0029] y j : The startup status of the j-th edge server.

[0030] x ij : The connection status of the i-th user to the j-th edge server.

[0031] FixedCost j : The fixed cost of the j-th edge server.

[0032] CompCost ij : The computational cost of the i-th user connecting to the j-th edge server.

[0033] TransCost ij : The transmission cost of the i-th user connecting to the j-th edge server.

[0034] Compared with the prior art, the present invention has the following advantages:

[0035] 1. Reduced hardware dependency: No quantum hardware required. This invention can run on ordinary classical computers, without relying on quantum hardware. This makes quantum computing technology easier to promote and apply, lowering the hardware threshold for quantum computing.

[0036] 2. High versatility. The quadratic unconstrained binary optimization model is a universal optimization model applicable to a wide range of combinatorial optimization problems. By converting the parameters into a quadratic unconstrained binary optimization model and solving it using a variational quantum algorithm, the present invention is better suited to different types of optimization problems and exhibits high versatility and adaptability.

[0037] 3. Improve computational efficiency. Quantum computing leverages the superposition and entanglement properties of quantum bits to process multiple possible solutions simultaneously, significantly improving computational efficiency. Although simulating quantum circuits on classical computers cannot fully achieve the parallelism of quantum hardware, by optimizing quantum circuit design and simulation algorithms, higher efficiency can still be achieved on classical computers than with traditional classical algorithms. Variational quantum algorithms can quickly approach the optimal solution by iteratively optimizing the parameters of quantum circuits. Compared to traditional classical optimization algorithms, this method typically requires fewer iterations to find a near-optimal solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 Cloud server-edge server and computing demand distribution diagram. The lower red triangles indicate optional edge server deployment locations, and the upper red triangles indicate cloud computing server deployment locations. The numbers indicate the computing demand at the corresponding locations.

[0039] Figure 2 : The objective function changes with the number of iterations.

[0040] Figure 3 : Running time changes with the number of iterations.

[0041] Figure 4 : Comparison of objective functions and running times.

[0042] Figure 5 : Comparison of convergence speed between classical optimization methods and classical-quantum hybrid architecture methods.

[0043] Figure 6 : Comparison of running times between classical optimization methods and classical-quantum hybrid architecture methods.

[0044] Figure 7 :Flowchart of the cloud-edge server cluster computing power allocation method for the classical-quantum hybrid architecture. DETAILED DESCRIPTION

[0045] like Figure 7 As shown in FIG, a method for allocating computing power of a cloud-edge server cluster in a classical-quantum hybrid architecture includes the following steps:

[0046] Step 1: Build a classical-quantum hybrid architecture, turn on the classical computer, and install quantum simulation software.

[0047] Step 2: Design corresponding quantum circuits and quantum gate operations based on the characteristics of the problem.

[0048] Step 3: Convert the quantum circuit parameters in step 2 into a quadratic unconstrained binary optimization model. First, set the original parameters with constraints as:

[0049] f Q (x)=∑ i≤j q ij x i x j The classical form of , where x takes values of 0 and 1, transforms the value form of QUBO so that σ=2x-1, σ is the transformed variable, taking values of -1 and 1. It is transformed into:

[0050] f Q (x) = X T QX.

[0051] In quadratic matrix form, X is a space vector composed of σ. The expectation is:

[0052] Q(σ)=-∑ ij A iJ σ i σ j -μ∑ i h i σ i .

[0053] When expectation < 0, the target result is met.

[0054] Step 4: Use classical optimization algorithms (such as simulated annealing, genetic algorithm, etc.) to solve the quadratic unconstrained binary optimization model, obtain the optimized parameters, and introduce them into the quantum circuit to improve the efficiency and accuracy of quantum computing.

[0055] Step 5: Iteratively optimize the quantum circuit to gradually approach the optimal solution. In each iteration, the classical computer updates the parameters of the quadratic unconstrained binary optimization model based on the current quantum circuit output. It also continuously adjusts the parameters of the quantum circuit, achieving collaborative optimization of quantum computing and classical optimization until the desired number of iterations is reached.

[0056] To illustrate this method in detail, let’s solve an optimization problem on computing power allocation in a cloud server-edge server cluster. Specifically,

[0057] 1. Determine the optimal location and number of edge servers.

[0058] 2. Assign users to edge servers or cloud servers to meet computing needs.

[0059] 3. Minimize total costs while ensuring that all user needs are met.

[0060] The distribution of computing requirements is as follows Figure 1 As shown, the red lower triangle is the optional deployment point of the edge server, and the red upper triangle is the point of the cloud computing server.

[0061] Since the maximum amount of computation that an edge server can provide is limited to 12, the demand value within some services may exceed this threshold. Therefore, we need to not only ensure comprehensive coverage but also adopt a selective strategy in resource allocation to avoid wasting resources due to over-provisioning. This leads to a logical framework for handling excess computation demand, with three processing modes:

[0062] 1. Direct cloud processing: Some computing needs can be directly processed by cloud computing machines. This is usually suitable for tasks that are not sensitive to latency or have high computational requirements.

[0063] 2. Edge server processing: Computing requirements that are closer to the edge server should be processed by the edge server first to reduce data transmission delay and improve response speed.

[0064] 3. Collaborative processing between cloud computing and edge servers: When edge servers' computing power is insufficient to meet specific needs, an efficient collaborative mechanism must be designed. Computational tasks beyond the reach of edge servers are divided, with some still handled by the edge servers and others transferred over the network to cloud computers for processing. This process involves complex resource scheduling and load balancing, requiring fine-grained algorithmic design to ensure a balance between efficiency and fairness.

[0065] Table 1 shows the details of the computing capabilities of the two servers:

[0066] Table 1

[0067] Cloud Server Edge Server radius ∞ 3 Calculation upper limit ∞ 12 Fixed Fee - 10 Calculate Fees n 2n Shipping costs 2x x

[0068] Step 1: Build a classical-quantum hybrid architecture. Hardware environment: Run on an ordinary classical computer and install the quantum simulation software Qiskit.

[0069] Step 2: Design a quantum circuit. There are 5 edge servers and 11 users. Whether each edge server is activated can be represented by a qubit, and whether each user is connected to a certain edge server can also be represented by a qubit. First, determine the activation status of the edge servers and build a 16-qubit quantum circuit containing 5 edge servers and 11 users. Initialize the qubits, initializing all qubits to |0>. Use the Hadamard gate H to put all qubits into a superposition state, use the CNOT gate to represent the connection between the user and the edge server, and use the single-bit rotation gate R to represent the connection between the user and the edge server. x and R y Adjust the state of the qubits to represent different configurations.

[0070] Step 3: Convert the parameter part of the quantum circuit in step 2 into the form of a quadratic unconstrained binary optimization model. The parameter function can be expressed as:

[0071]

[0072] in:

[0073] Minimize C: Minimize cost.

[0074] y j : The startup status of the j-th edge server.

[0075] x ij : The connection status of the i-th user to the j-th edge server.

[0076] FixedCost j : The fixed cost of the j-th edge server.

[0077] CompCost ij : The computational cost of the i-th user connecting to the j-th edge server.

[0078] TransCost ij : The transmission cost of the i-th user connecting to the j-th edge server.

[0079] Convert the parameter function into a quadratic unconstrained binary optimization model:

[0080]

[0081] Where X is a vector of binary variables and Q is a matrix of quadratic coefficients.

[0082] Step 4: Use simulated annealing algorithm to optimize the quadratic unconstrained binary optimization model. Initialize the parameters to initial temperature T = 100, cooling rate α = 0.99, and the number of iterations to 1000. Randomly generate the initial solution:

[0083] X=[1,1,0,1,0]

[0084] Generate a neighboring solution X by flipping a qubit ′ :

[0085] X ′ =[1,0,0,1,0]

[0086] Compute the cost of the new solution:

[0087] C ′ =X ′T QX ′

[0088] And calculate the expectation at this time:

[0089]

[0090] At this time, it is expected that Q(σ)>0. If the target result is not met, the temperature T is reduced, and T=99. When the temperature T=1, the iteration is stopped. At this time, the objective function gradually decreases with the increase of the number of iterations. Figure 2 , and the running time increases with the number of iterations, see Figure 3 In order to find a balance point that takes both into account, the intersection of the two curves is generally selected, such as Figure 4 .

[0091] Step 5: Iteratively optimize the parameters in the quantum circuit and gradually approach the optimal solution. Introduce the initial parameters into the quantum circuit and convert the initial parameters x obtained by the simulated annealing algorithm into the parameters θ and φ of the quantum circuit, which are:

[0092] θ=[0.5,1.2,0.8,1.5,0.3,0.7,0.9,1.1]

[0093] φ=[0.4,1.3,0.6,1.4,0.2,0.8,1.0,1.2]

[0094] Run the quantum circuit on a classical computer and obtain the quantum state under the current parameter θ. Measure the quantum state and the measurement result is:

[0095] [1,1,0,1,0]

[0096] Calculate the expected Q(σ). At this time, Q(σ) < 0, which satisfies the target result. Output this parameter. Calculate the total cost and record the optimal solution and optimal cost in the current iteration: C = X T QX,

[0097] We obtain C = 371.61, which is the final cost. The specific distribution is output as shown in Table 2-6. At the same time, we selected a classical optimization method without quantum circuits for comparison, which demonstrates the advantages of faster convergence and shorter running time of the classical-quantum hybrid architecture. The data visualization comparison is shown in Figure 5-6 .

[0098] Table 2 Calculation results of quadratic unconstrained binary optimization model

[0099]

[0100] Table 3 Edge activation status table

[0101] Serial number Edge coordinates Activation Status 1 [2,1] 1 2 [2.3] 0 3 [4.5] 0 4 [6.5] 1 5 [6.1] 1

[0102] Table 4 Computational requirements for user-to-edge transmission

[0103] Serial number Edge coordinates User coordinates Computational requirements at the user transmission edge 1 [2,1] [3,1] 3 2 [2,1] [3,3] 4 3 [6,5] [6,6] 5 4 [6,5] [5,5] 3 5 [6,5] [3,5] 4 6 [6,1] [6,3] 7 7 [6,1] [5,2] 5

[0104] Table 5 Computational requirements for user-to-cloud transmission

[0105] Serial number Cloud Coordinates User coordinates The computing demand of user-transferred cloud 1 [4,0] [1,4] 4 2 [4,0] [2,6] 5 3 [4,0] [4,2] 4 4 [4,0] [4,4] 11

[0106] Table 6 Computational requirements for edge-to-cloud transmission

[0107] Serial number Edge coordinates Cloud Coordinates Computing requirements of edge transport clouds 1 [6,5] [4,0] 4 2 [6,1] [4,0] 1

Claims

1. A cloud-edge server cluster computing power allocation method for a classical-quantum hybrid architecture, characterized by: The following steps are involved: 1) The cloud-edge server cluster includes: A cloud server; a plurality of edge servers connected to the cloud server; 2) A classical-quantum hybrid architecture is deployed on the cloud server. Community users send computing power demand information to the cloud server, which then uses the classical-quantum hybrid architecture to calculate the computing power allocation coefficient. The computing power allocation result is then sent to the cloud server and edge server, which then process the demand according to the computing power allocation result.

2. The cloud-edge server cluster computing power allocation method of the classical-quantum hybrid architecture according to claim 1 is characterized in that: In step 2), the classical-quantum hybrid architecture includes: Parameter input module, used to initialize the parameter part; A quantum circuit simulation module, configured to iteratively optimize the parameters of the parameter input module and the constructed demand matrix to obtain a computing power allocation coefficient; The parameter optimization output module is used to determine whether the computing power allocation coefficient meets the expected value of the target. If it does, the computing power allocation coefficient is output as a result; if it does not, the computing power allocation coefficient is returned to the quantum circuit simulation module.

3. The cloud-edge server cluster computing power allocation method of the classical-quantum hybrid architecture according to claim 2 is characterized in that: In step 2), the computing power distribution coefficient is calculated using the classical-quantum hybrid architecture, specifically including: 2.1) Construct the basic quantum gates of the quantum circuit, convert the community user location, computing power requirements, and edge server location information into the quantum circuit, convert the parameter portion of the quantum circuit into a quadratic unconstrained binary optimization model, use classical optimization algorithms to optimize the model, obtain initial parameters, introduce the initial parameters into the quantum circuit, run the quantum circuit on a computer, simulate the quantum computing process, and obtain measurement results; 2.2) Determine whether the measurement result meets expectations. If so, output the result. If not, use a classical optimization algorithm on a computer to optimize the quadratic unconstrained binary optimization model again to obtain updated parameters. Adjust the quantum circuit based on the updated parameters. Repeat the steps until expectations are met or the preset number of iterations is reached. The parameters at this point are used as the computing power allocation coefficient.

4. The cloud-edge server cluster computing power allocation method of the classical-quantum hybrid architecture according to claim 3 is characterized in that: In steps 2.1) and 2.2), the classical optimization algorithm adopts a simulated annealing algorithm or a genetic algorithm.

5. The cloud-edge server cluster computing power allocation method of the classical-quantum hybrid architecture according to claim 2 is characterized in that: In step 2.2), the expectation is: Among them, Q(σ) is the expectation. When Q(σ) < 0, it is determined that it meets the expectation. σ is a variable. i, j represent the row and column numbers of the matrix in the quadratic unconstrained binary optimization model. A ij Represents the coefficient of the corresponding matrix position, μ is the penalty coefficient, h ij Represents the bias of σ at the corresponding matrix position.

6. The cloud-edge server cluster computing power allocation method of the classical-quantum hybrid architecture according to claim 5 is characterized in that: In step 2.2), the value of σ is {-1, 1}; The value of μ is in the range of [2,5].

7. The cloud-edge server cluster computing power allocation method of the classical-quantum hybrid architecture according to claim 1 is characterized in that: In step 2), the computing power allocation result is obtained according to the computing power allocation coefficient, which specifically includes: The computing power distribution coefficient is the parameter finally calculated by the parameter function: Among them, Minimize C is the minimum cost, n is the number of edge servers, m is the number of users, y j is the opening status of the jth edge server, x ij is the connection between the i-th user and the j-th edge server, FixedCost j is the fixed cost of the j-th edge server.

Images