A Bayesian learning-based end-to-end network resource allocation method

Abstract

The present invention discloses an end-to-end network resource allocation method based on Bayesian learning, which belongs to the technical field of calculation, inference or counting. The method adopts a solution based on a game framework, proposes a network resource pricing strategy that reflects the service provider's demand information, and realizes more reasonable network resource pricing and allocation by analyzing the network resource status and service provider behavior. The method first establishes a game model between multiple service providers and operators. The operator selects the unit network slice price in the first stage based on limited demand information; in the second stage, the demand information is updated to determine the price. The simulation results show the existence of the Nash equilibrium point of the system model, and proposes a responsive network resource slice allocation scheme to avoid resource shortages and network congestion caused by service providers' concentrated resource requests, thereby realizing reasonable allocation of network resources.

Description

Technical Field

[0001] The present invention relates to communication technology, and in particular discloses an end-to-end network resource allocation method based on Bayesian learning, belonging to the technical field of calculation, inference or counting. Background Art

[0002] In recent years, existing networks have struggled to support the application scenarios of 5G networks. Deploying multiple different communication networks for different application scenarios is unrealistic, and the resulting huge costs are unaffordable for operators. Therefore, the Next Generation Mobile Network Alliance proposed the concept of network slicing, which involves building different network slices on demand on a single physical infrastructure using software-defined networking and technologies such as network function virtualization. Various resource allocation schemes aim to isolate resources between slices, ensuring that a failure in one slice in a dynamic network environment does not affect communications in other slices. However, most current resource scheduling research continues to follow the principles of traditional network resource scheduling algorithms, such as allocating virtual network resources through network mapping based on load balancing, which ignores the actual needs of service providers.

[0003] In their article, "Strategies for Network Slicing Negotiation in a Dynamic Resource Market," Alessandro Lieto et al. propose a technical-economic game to study user strategies under a Nash equilibrium, based on a game-theoretic market mechanism for allocating network resource slices. This mechanism dynamically regulates negotiations between different tenants and proves that at least one Nash equilibrium exists in this type of game, and that in multiple scenarios, the Nash equilibrium is always reached when implementing the best-response dynamic algorithm.

[0004] The present invention aims to analyze the network resource allocation problem of operators by adopting a game economic model, and convert it into a non-cooperative game problem between multiple service providers and operators to achieve the goal of optimal resource allocation. Summary of the Invention

[0005] The purpose of the present invention is to address the deficiencies of the above-mentioned background technology and provide an end-to-end network resource allocation method based on Bayesian learning. By adopting a solution based on a game framework, a network resource pricing strategy that reflects the service provider's demand information is proposed. By analyzing the service provider's needs, the invention achieves the purpose of reasonably pricing network resources and optimizing resource allocation schemes, and solves the technical problem that the existing resource scheduling scheme ignores the actual needs of the service provider.

[0006] The present invention adopts the following technical solutions to achieve the above-mentioned purpose:

[0007] In this invention, an operator is considered to provide Q units of network resource slices within a limited time, and the service provision time is divided into two phases: a normal phase (the first phase T1) and an incentive phase (the second phase T2). Multiple service providers request network resources from the operator in the system, and the number of service providers is a Poisson process with an arrival rate Λ, which obeys a priori gamma distribution with parameters α and β. Service providers need to choose a phase to request network resources. The service provider's decision mainly depends on the valuation v of the network resources, which obeys a uniform distribution between [0, 1]. If the service provider chooses to request resources in phase T2, its valuation v will decrease by δ times with the period, for example, by 0.8 times. In the decision-making process, the service provider adopts a threshold strategy θ: when v ≥ θ, the service provider requests resources in phase T1; when v < θ, the service provider requests resources in phase T2. For operators, in phase T1, an initial price p1 is set for network resources based on the initial number Q of network resource slices and the prior gamma distribution. In phase T2, operators change p1 to p2 based on the updated network resource status and the predicted scale of service providers. This paper demonstrates the existence of a Nash equilibrium point in this system model and optimizes the allocation of operator network resources by analyzing a game model. Simulation results from this paper demonstrate the existence of a Nash equilibrium point in this system model.

[0008] The present invention solves the technical problem by adopting an end-to-end network resource allocation method based on Bayesian learning, which includes the following steps:

[0009] Step 1: Assume that the probability of service provider j obtaining one unit of network resources in the normal stage T1 is for:

[0010]

[0011] In formula (1), λ1 is the Poisson rate at which service providers arrive at the system during the regular phase T1, λ1 = Λ(1-θ), It represents the expected value of the probability that each unit of network resources is allocated to y+1 service providers on average during the normal phase T1.

[0012] Step 2: Assume that the expected utility of service provider j in the normal stage T1 is

[0013]

[0014] In formula (2), v j is the valuation of one unit of network resources by service provider j, and p1 is the price of one unit of network resources in the regular stage T1.

[0015] Step 3: Assume that the probability of service provider j obtaining one unit of network resources in the incentive stage T2 is for:

[0016]

[0017] In formula (3), λ2 is the Poisson rate at which the service provider arrives at the system in the incentive phase T2. p2 is the price of one unit of network resource after the update in the incentive phase T2.

[0018] Step 4: Assume that the expected utility of service provider j in the incentive stage T2 is

[0019]

[0020] In formula (4), [0.8v j -p2(θ,x)] + Indicates [0.8v j -p2(θ, x)]≥0, where p2(θ, x) is a function of the price of one unit of network resources in the incentive stage T2 with respect to θ and x.

[0021] Step 5: According to the implicit function theorem and Brouwer’s fixed point theorem, when all service providers adopt the same threshold θ, there is an equilibrium in the sub-game among the service providers.

[0022] Step 6: Assume that the operator’s profit function in the normal stage T1 is:

[0023]

[0024] In formula (5), π1 is the benefit value of the operator in the normal stage T1, α and β are the parameters of the prior gamma distribution of the service provider arrival rate Λ in the normal stage T1 of the system, and the probability density function of the gamma distribution is in,

[0025] Step 7: Assume that the operator’s profit function in the incentive phase T2 is:

[0026]

[0027] In formula (6), π2 is the operator's benefit value in the incentive phase T2, and x is the number of network resources requested in the regular phase T1. The operator updates the distribution parameters of the service provider arrival rate Λ according to the value of x. The parameters of the posterior gamma distribution of the service provider arrival rate Λ in the incentive phase T2 of the system are: and Its probability density function is in,

[0028] Step 8: From steps 6 and 7, we can see that the operator’s overall optimal benefit is π R (Q):

[0029]

[0030] Step 9: Find the optimal price for the regular phase T1 Take the first-order derivative of π2 with respect to p2 and set it equal to 0 to find the optimal pricing for the incentive period T2 Then and Substituting into formula (7), we can obtain the optimal benefit of the operator.

[0031] Step 10: From step 9, we can know that θ when the operator obtains the optimal benefit * value, θ * =arg max θ π R (Q), and substituting it into λ1 and λ2, we can find the potential demand of the service provider in the two stages:

[0032]

[0033]

[0034] Formula (8) is a negative binomial distribution Among them, N1 represents the network resource request amount of the service provider in the normal stage T1, and its expected value is P(N1=k) represents the probability that the number of network resource requests by the service provider in the normal stage T1 is k; Formula (9) is also a negative binomial distribution Among them, N2 represents the network resource request amount of the service provider in the incentive stage T2, and its expected value is P(N2=k) represents the probability that a service provider's network resource request volume during the incentive phase T2 is k. This allows us to derive the expected value of service provider demand at each phase, i.e., the potential demand volume. This effectively avoids resource shortages and network congestion caused by concentrated resource requests from service providers, thereby achieving a rational allocation of network resources.

[0035] The present invention adopts the above-mentioned technical solution and has the following beneficial effects: based on the analysis of the service provider's decision to request network resources, the present invention maps the network resource allocation problem into the operator's benefit maximization problem, and analyzes the benefit maximization problem of the service provider and the operator by adopting a game theory economic model. In the process of analyzing the benefit maximization problem of the service provider and the operator, the operator determines the posterior probability density function of the Poisson speed of the service provider reaching the system in the regular phase based on the prior probability density function of the Poisson speed of the service provider reaching the system in the regular phase and the observed prior gamma distribution parameters. The optimal pricing of the operator in the regular phase and the incentive phase is determined by the constraint that the expected utility of the service provider in the regular phase and the incentive phase are equal and the constraint that the service provider has the maximum benefit in the incentive phase. Finally, the optimal threshold for the service provider to decide at which phase to request resources is determined by the optimal benefit when the operator responds to the optimal pricing, and then the potential resource demand of the service provider in each phase is updated, effectively studying the optimal decision of the service provider to request resources and the optimal pricing scheme of the operator, thereby obtaining the optimal resource allocation scheme, and effectively avoiding resource shortages and network congestion caused by the concentrated resource requests of service providers. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 Schematic diagram of the system model of the present invention.

[0037] Figure 2 Schematic diagram of service provider decision-making in an embodiment.

[0038] Figure 3 Schematic diagram of the existence of Nash equilibrium in the embodiment.

[0039] Figure 4 An allocation map of network resource slices in an instance. DETAILED DESCRIPTION

[0040] The technical solution of the invention is described in detail below with reference to the accompanying drawings: This embodiment is implemented on the premise of the technical solution of the present invention, and a detailed implementation method and specific operation process are given, but the protection scope of the present invention is not limited to the following embodiments.

[0041] The specific steps are as follows:

[0042] This first embodiment provides an end-to-end network resource allocation based on Bayesian learning.

[0043] like Figure 1 As shown, the first embodiment provides an end-to-end network resource allocation solution based on Bayesian learning, including: benefit functions of service providers and operators; decision thresholds of service providers; benefits of operators when equilibrium is reached; and arrival rates of service providers.

[0044] The first step is to assume that each service provider has a different valuation of network resources, denoted by v, which follows a uniform distribution in the range [0, 1]. Operators cannot determine this parameter. The probability of a service provider obtaining one unit of network resources is combined to derive the service provider's benefit function. For operators in the system, the impact of their benefit function primarily includes network resource pricing, system congestion levels, and the service provider's strategy. This yields the operator's benefit function.

[0045] The second step is to consider the sub-game between service providers. In the sub-game model of the service provider requesting network resources in the present invention, it is assumed that when other service providers adopt corresponding thresholds, the behavior of a service provider that can maximize its own benefit value by changing its own threshold is called a strategy, and all possible strategies constitute a decision set. If the threshold strategy adopted is θ, when v>θ, s1(v|θ)>s2(v|θ); when v<θ, s1(v|θ)<s2(v|θ), such as Figure 2 According to the implicit function theorem and Brouwer fixed point theorem, when all service providers adopt the same threshold θ, the subgame has a Nash equilibrium.

[0046] In the third step, the present invention uses a reverse deduction method to solve the game model. First, based on the number of network resources requested in the first phase, Bayesian learning is used to update the scale of service providers in the system to obtain the expressions of p1(θ) and p2(θ) under Nash equilibrium. In the present invention, other parameters are reasonably set: E[Λ] = 10; the initial number of network resources is Q = 10. Under Nash equilibrium, the operator benefit function curve is obtained using MATLAB software simulation, as shown in the figure below. Figure 3 shown.

[0047] Finally, according to the maximum benefit of the operator under Nash equilibrium in the third step, we can find θ at this time, and use reverse deduction to solve the service provider distribution of T1 and T2, thereby inferring the number of resources requested by service providers in each stage, such as Figure 4 As shown, resource shortages and network congestion caused by concentrated resource requests from service providers can be effectively avoided, thereby achieving reasonable allocation of network resources.

[0048] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent replacements, and transformations made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A Bayesian learning-based end-to-end network resource allocation method, characterized in that: The optimal pricing of the operator in the regular phase is determined under the constraint that the expected utility value of the service provider in the regular phase and the expected utility value of the service provider in the incentive phase are equal. The expected utility value of the service provider in the regular phase is expressed as The expected utility value of the service provider in the incentive phase is calculated according to the expression Calculate, where is the expected utility value of service provider j in the regular stage T1, v j is the price of one unit of network resources by service provider j, p1 is the price of one unit of network resources in the normal stage T1, is the probability that service provider j can obtain one unit of network resources in the regular stage T1, λ1 is the Poisson rate at which service providers arrive at the system during the regular phase T1, λ1 = Λ(1-θ), where Λ obeys a priori gamma distribution with parameters α and β. is the expected value of the probability that each unit of network resources is allocated to y+1 service providers in the normal stage T1, Q is the initial number of network resources, is the expected utility value of service provider j in the incentive stage T2, [0.8v j -p2(θ,x)] + Indicates [0.8v j -p2(θ,x)]≥0, where p2(θ,x) is the price of one unit of network resources in the incentive phase T2 as a function of θ and x, θ is the threshold strategy adopted by the service provider, and x is the number of network resources requested by the service provider in the regular phase T1. is the probability that service provider j can obtain one unit of network resources in the incentive stage T2, λ2 is the Poisson rate at which service providers arrive at the system during the incentive phase T2, p2 is the price of one unit of network resources after the update in the incentive phase T2; The pricing at the maximum expected utility of the service provider in the incentive stage is taken as the operator's optimal pricing in the incentive stage. The optimal threshold strategy adopted by the service provider when the operator responds to the optimal pricing and obtains the optimal benefit is solved. The method for solving the optimal threshold strategy adopted by the service provider when the operator responds to the optimal pricing and obtains the optimal benefit is: Based on the expression Solve the optimal benefit when the operator responds to the optimal pricing, The best pricing for operators in the regular stage, Optimal pricing for operators during the incentive phase, According to the expression θ * =argmaxθπ R (Q) Solve the optimal threshold strategy adopted by the service provider when the operator responds to the optimal pricing and obtains the optimal benefit, θ * The optimal threshold strategy adopted by the service provider when the operator responds to the optimal pricing and obtains the best benefits; Based on the optimal threshold strategy adopted by the service provider, the Poisson rate at which the service provider reaches the system in the normal phase and the Poisson rate at which the service provider reaches the system in the incentive phase are updated. The network resource request volume of the service provider in the normal phase is calculated based on the updated Poisson rate at which the service provider reaches the system in the normal phase. The network resource request volume of the service provider in the incentive phase is calculated based on the updated Poisson rate at which the service provider reaches the system in the incentive phase. Among them, P(N1=k) is the probability that the network resource request amount N1 of the service provider in the normal stage T1 is k, and P(N2=k) is the probability that the network resource request amount N2 of the service provider in the incentive stage T2 is k.

2. The end-to-end network resource allocation method based on Bayesian learning according to claim 1, characterized in that: The method for solving the maximum expected utility of the service provider in the incentive stage is: setting the first-order derivative of the operator's benefit function in the incentive stage T2 to zero.

Images