Task unloading method oriented to content and service cache joint optimization in edge computing

Abstract

The invention discloses a content and service cache joint optimization-oriented task unloading method in edge computing, which comprises the following steps of: 1, establishing a mathematical model for transmission time and energy consumption of content and service and computing time and energy consumption of a service task according to a three-layer architecture of a cloud server, an edge server and user equipment in the edge computing; 2, establishing a joint optimization problem of mixed integer nonlinear programming according to the mathematical model in the step 1; and step 3, solving a joint optimization problem of mixed integer nonlinear programming in the step 2. According to the method, a mixed integer nonlinear programming problem which is originally difficult to solve is approximately converted into a convex optimization problem which is easy to solve through a successive convex approximation method, so that the problem that a content and service cache decision and a task unloading decision in edge computing are difficult to solve is solved, and the communication time and the energy consumption in an edge computing system are reduced; and the utility value of the edge server is maximized.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of edge computing, and particularly relates to a task offloading method for content and service cache joint optimization, which is suitable for video streaming media, real-time data processing and other computationally intensive edge application scenarios. BACKGROUND

[0002] In an edge computing system, the storage capacity and computing resources of an edge server are limited. How to efficiently manage content cache (such as video resources) and service cache (such as target detection, subtitle generation and other computing services), and achieve optimization of task offloading, is a key challenge to improve system performance. In the prior art, content cache and service cache are usually optimized independently, and there is a lack of effective mechanism for joint decision-making of the two, resulting in high system latency, high energy consumption and insufficient resource utilization.

[0003] Chinese Patent Application No. 2021105134040 discloses an intelligent dynamic task computing offloading method based on an edge computing system. The application considers optimizing the performance of a mobile edge computing system from the level of a mobile edge computing server, builds an intelligent overclocking mobile edge computing system model, and solves the trade-off problem between system utility and system stability in dynamic task computing offloading. However, the dynamic nature of user requests and the correlation between content and services are not fully utilized, making it difficult to achieve optimal cache strategy under limited resources. SUMMARY

[0004] In view of the above problems existing in the prior art, the application provides a task offloading method for content and service cache joint optimization in edge computing.

[0005] The application adopts the following technical solutions:

[0006] A task offloading method for content and service cache joint optimization in edge computing includes the following steps:

[0007] Step 1: According to the three-layer architecture of cloud servers, edge servers and user devices in edge computing, a mathematical model is established for the transmission time and energy consumption of content and services, and the computing time and energy consumption of service tasks.

[0008] Step 2: A joint optimization problem of mixed integer nonlinear programming is established according to the above mathematical model.

[0009] Step 3: Solve the joint optimization problem of mixed integer nonlinear programming.

[0010] Preferably, in step 1, the specific process of establishing the mathematical model is as follows: according to the three-layer architecture in edge computing: cloud server, edge server, and user equipment, the time and energy consumption of transmitting content and services from the edge server to the user equipment, the time and energy consumption of transmitting content and services from the cloud server to the edge server, the time and energy consumption of transmitting content and services from the cloud server to the user equipment, and the computing time and energy consumption of service tasks on the edge server are established into a mathematical model.

[0011] Preferably, in step 2, the specific process of establishing the joint optimization problem of mixed integer nonlinear programming is as follows:

[0012] S21, establish a communication time expression in the transmission process;

[0013] S22, establish a computing time expression of service execution;

[0014] S23, establish an energy consumption expression generated by the communication and computing process;

[0015] S24, establish an optimization objective function expression;

[0016] S25, establish an optimization problem constraint expression;

[0017] S26, under the constraints in S25, establish the entire optimization problem with the goal of maximizing the expression of the objective function in S24.

[0018] Preferably, in step S21, the communication time expression in the transmission process includes the communication time of the edge server to the user equipment and the communication time of the cloud server to the edge server.

[0019] The communication time expression of the edge server to the user equipment can be represented as:

[0020]

[0021] wherein, represents the storage capacity occupied by the content, represents the bandwidth allocated by the edge server to the user equipment, p E represents the transmission power of the edge server, represents the channel gain of the edge server n to the user equipment D i , and 2 represents the noise power.

[0022] The communication time expression of the cloud server to the edge server can be represented as:

[0023]

[0024] wherein, represents the storage capacity occupied by the content, Bc,n denotes the channel bandwidth between the cloud server and the edge server, p c denotes the transmission power of the cloud server, g c,n denotes the channel gain from the cloud server to the edge server, σ 2 denotes the noise power.

[0025] Preferably, in step S21, the communication time expression in the transmission process is specifically as follows:

[0026]

[0027] wherein, T i c denotes the total communication delay of the user i, x v,n denotes the content cache state.

[0028] Preferably, in step S22, the calculation time expression of the service execution is specifically as follows:

[0029]

[0030] wherein, y s,n denotes the service cache state, denotes the required storage capacity of the service s h . s,i denotes the computing capability of the edge server E n allocated to the user i, the computing capability of the edge server E n is denoted as F n (unit: FLOP / s), and the required computing amount of the service requested by the user i is denoted as f s .

[0031] The total delay of the user request includes the transmission delay of the request task and the calculation delay of the possible calculation task, so whether the content and the service are cached in the edge server has a great influence on the delay. In combination with the cache decision variable and the service request variable, the total delay expression of the user request is specifically as follows:

[0032]

[0033] Preferably, in step S23, the energy consumption expression generated in the communication process is specifically as follows:

[0034]

[0035] wherein, denotes the communication power from the cloud server to the user device, r s,i denotes the service request state, denotes the communication power from the edge server to the user device, represents the communication power of the cloud server to the edge server, and the energy consumption of the edge server mainly depends on the power allocation of the edge server and the task processing time.

[0036] In step S23, the energy consumption expression generated by the calculation process is specifically as follows:

[0037]

[0038] wherein the computing power of the edge server is represents the computing power of the edge server. The total energy consumption of the user request depends on the computing and communication of the edge server, and therefore whether the content and service are cached in the edge server has a greater impact on the energy consumption. The total energy consumption expression of the user i is specifically as follows:

[0039]

[0040] Preferably, in step S24, the optimization objective function expression is specifically as follows:

[0041]

[0042] wherein A i represents the income paid by the mobile user i, and λ1 and λ2 are weight coefficients of the delay and the energy consumption, respectively, for balancing the multi-objective optimization process.

[0043] Preferably, the constraints of the optimization problem in step S25 include content caching decision variable constraints, service caching decision variable constraints, service request decision variable constraints, cache capacity constraints of the edge server, and computing resource allocation constraints.

[0044] The content caching decision variable constraints are expressed as:

[0045]

[0046] The service caching decision variable constraints are expressed as:

[0047]

[0048] The service request decision variable constraints are expressed as:

[0049]

[0050] The cache capacity constraints of the edge server are expressed as:

[0051]

[0052] The computing resource allocation constraints are expressed as:

[0053]

[0054] where M is the total number of video contents provided by the cloud server, and is the set of video contents. H is the total number of services provided, and is the set of services. E n The storage and computing resources of the cloud server are denoted as C n and F n respectively, and the set of user devices is denoted as

[0055] Preferably, in step 3, the specific process of solving the mixed integer nonlinear programming problem is as follows:

[0056] S31, problem conversion, through binary variable relaxation, introducing a penalty term, and iteratively constructing a convex approximation, the non-convex problem is approximately converted into a sequence of convex optimization problems that can be efficiently solved.

[0057] S32, dynamic popularity prediction, the joint probability-based dynamic popularity prediction mechanism designs the cache priority to optimize the cache decision of the next time slot.

[0058] S33, according to the new converted objective function and cache decision in steps S31 and S32, the converted objective function is solved by a solver or a nonlinear integer programming algorithm.

[0059] Preferably, in step S31, the non-convex problem is approximately converted into a sequence of convex optimization problems that can be efficiently solved by iteratively constructing a convex approximation. The specific steps are as follows:

[0060] 1) Binary variable relaxation: the cache decision variable and the service request decision variable are relaxed from binary variables with values of 0 or 1 to continuous variables and with values in the range [0, 1], and a penalty term is introduced to encourage integer solutions, and the penalty term expression is represented as:

[0061]

[0062] The penalty term reaches a minimum value of 0 when the variable takes a value of 0 or 1, which encourages the relaxed variable to tend to a binary value. The non-convexity of the original problem mainly comes from the hyperbolic constraint and the variable product

[0063] For the nonlinear term in the objective function, an auxiliary variable τ s,i,n is introduced, and the auxiliary variable expression can be represented as:

[0064]

[0065] This hyperbolic constraint is non-convex, and at the kth iteration point By constructing convex approximation through first-order Taylor expansion, the expression can be represented as:

[0066]

[0067] The approximation is linearly constrained, satisfies convexity, and is equivalent to the original constraint at the current iteration point.

[0068] 2) Handling the coupling term of cache decision variable and latency-energy

[0069] For the non-convex term of at the kth iteration point , by constructing convex approximation through first-order Taylor expansion, the expression can be represented as:

[0070]

[0071] Similarly, for the term of constructing convex approximation, the expression can be represented as:

[0072]

[0073] 3) Handling the coupling term of service request variable and computation latency

[0074] For the non-convex term of the form r s,i τ s,i,n at the kth iteration point , constructing convex approximation, the expression can be represented as:

[0075]

[0076] In combination with the above variable transformation and convex approximation design, the McCormick envelope is constructed to construct a convex upper bound, ensuring that the approximation is strictly greater than the original function (since the objective is to maximize). In [0, 1] × [τ min , τ max ], the expression of the McCormick envelope can be represented as:

[0077]

[0078] At the kth iteration, the approximation expression can be represented as:

[0079]

[0080] In the iteration, Δr, Δτ is constrained, and the McCormick envelope ensures that the upper bound holds.

[0081] In combination with the above approximation, the expression of the objective function at the kth iteration can be represented as:

[0082]

[0083] The convex approximation expression of the objective function can be expressed as:

[0084]

[0085] wherein, and respectively represent the convex approximation expression of the delay and the energy consumption at the current iteration point, and ρ represents the penalty term coefficient.

[0086] Preferably, in step S32, the dynamic popularity prediction is based on a joint probability-based dynamic popularity prediction mechanism, which fuses the potential characteristics of the video and the service to optimize the cache decision in the next time slot. In time slot t, the probability expression of the video being requested by the user can be expressed as:

[0087]

[0088] wherein, the number of times of the video being requested is represented as N v (t), and the total number of requests in the time slot is represented as N total (t).

[0089] Based on the historical data and the current observation, the video request probability is obtained, and the predicted probability of the video being requested is obtained through prediction The predicted value of the next time slot is represented as:

[0090]

[0091] wherein, The smoothing factor β ∈ (0, 1). For each pair of video and service Let the number of times of being requested simultaneously be represented as N v,s (t), the observation joint probability expression can be expressed as:

[0092]

[0093] Meanwhile, the joint request probability is defined as the probability of being requested simultaneously by the user, and the predicted value is expressed as:

[0094]

[0095] The correlation between the video content and the computing service can be quantified through the conditional probability, and the conditional probability expression can be expressed as:

[0096]

[0097] Based on the defined cache priority expression, it can be expressed as:

[0098]

[0099] wherein w1, w2 [0, 1] and w1 + w2 = 1, denotes the video v obtained by prediction m requested prediction probability, denotes the video v obtained by prediction m and service s h simultaneously requested prediction probability, joint probability is affected by priority v,s (t+1) influence cache variables, priority v,s (t+1) higher set of videos and services represent that they are often requested simultaneously.

[0100] The present application has the following beneficial effects:

[0101] The present application solves the problem that the content and service cache decision and task offloading decision in edge computing are difficult to solve by converting the originally difficult mixed integer nonlinear programming problem into a convex optimization problem that is easy to solve through the successive convex approximation method, reduces the communication time and energy consumption in the edge computing system, and maximizes the utility value of the edge server. BRIEF DESCRIPTION OF DRAWINGS

[0102] Figure 1 An edge computing system architecture diagram related to a task offloading method for content and service cache joint optimization in edge computing of a preferred embodiment of the present application;

[0103] Figure 2 A cache logical relationship diagram in the preferred embodiment of the present application;

[0104] Figure 3 A comparison result graph between the preferred embodiment of the present application and the prior art. DETAILED DESCRIPTION

[0105] The following detailed description of the preferred embodiment of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but only represents selected embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative labor are within the scope of protection of the present application.

[0106] The task offloading method for content and service cache joint optimization in edge computing of the present embodiment includes the following steps:

[0107] Step 1, a mathematical model is established according to the edge computing environment, specifically, according to the three-layer architecture in edge computing: cloud server, edge server and user equipment, the time and energy consumption of content and service transmission from the edge server to the user equipment, the time and energy consumption of content and service transmission from the cloud server to the edge server, the time and energy consumption of content and service transmission from the cloud server to the user equipment, and the computing time and energy consumption of service tasks in the edge server are established into a mathematical model;

[0108] Step 2, a joint optimization problem of mixed integer nonlinear programming is established according to the mathematical model of step 1, the specific process is as follows:

[0109] S21, a communication time expression in the transmission process is established, the communication time expression in the transmission process includes the communication time of the edge server to the user equipment and the communication time of the cloud server to the edge server;

[0110] The communication time of the edge server to the user equipment is represented as:

[0111]

[0112] Wherein, represents the storage capacity occupied by the content, represents the bandwidth allocated by the edge server to the user equipment, p E represents the transmission power of the edge server, represents the channel gain of the edge server n to the user equipment D i , and σ 2 represents the noise power.

[0113] The communication time of the cloud server to the edge server is represented as:

[0114]

[0115] Wherein, represents the storage capacity occupied by the content, B c,n represents the channel bandwidth between the cloud server and the edge server, p c represents the transmission power of the cloud server, g c,n represents the channel gain of the cloud server to the edge server, σ 2 represents the noise power.

[0116] In step S21, the communication time expression in the transmission process is specifically as follows:

[0117]

[0118] Wherein, T i c represents the total communication delay of user i, x v,n∈{0, 1} represents the content cache state.

[0119] S22, establish the computing time expression of service execution, the specific expression is as follows:

[0120]

[0121] Wherein, y s,n ∈{0, 1} represents the service cache state, represents the service s h The required storage capacity, ψ s,i represents the edge server E n The proportion of computing resources allocated to user i, the edge server E n The computing capacity of the edge server is F n (unit: FLOP / s), the required computing amount of the service requested by user i is f s .

[0122] The total delay of user request includes the transmission delay of request task and the possible computing delay of computing task, so whether the content and service are cached in the edge server has a great influence on the delay. Combined with the cache decision variable and the service request variable, the total delay expression of user request is as follows:

[0123]

[0124] S23, establish the energy consumption expression generated by the communication and computing process, the communication energy consumption expression is as follows:

[0125]

[0126] Wherein, represents the communication power from the cloud server to the user equipment, r s,i ∈{0, 1} represents the service request state, represents the communication power from the edge server to the user equipment, represents the communication power from the cloud server to the edge server, and the energy consumption of the edge server mainly depends on the power distribution and task processing time of the edge server.

[0127] The energy consumption expression generated by the computing process is as follows:

[0128]

[0129] Wherein, the computing power of the edge server is represents the computing power of the edge server. The total energy consumption of user request depends on the computing and communication of the edge server, so whether the content and service are cached in the edge server has a great influence on the energy consumption. The total energy consumption expression of user i is as follows:

[0130]

[0131] S24, establish an optimization objective function expression, the specific expression is as follows:

[0132]

[0133] Where, A i represents the income paid by the mobile user i, λ1 and λ2 are the weight coefficients of delay and energy consumption, respectively, for balancing the multi-objective optimization process.

[0134] S25, establish an optimization problem constraint expression, including content cache decision variable constraint, service cache decision variable constraint, service request decision variable constraint, edge server cache capacity constraint, and computing resource allocation constraint;

[0135] The content cache decision variable constraint is expressed as:

[0136]

[0137] The service cache decision variable constraint is expressed as:

[0138]

[0139] The service request decision variable constraint is expressed as:

[0140]

[0141] The edge server cache capacity constraint is expressed as:

[0142]

[0143] The computing resource allocation constraint is expressed as:

[0144]

[0145] Where, the video content provided by the cloud server has M types, and the set is expressed as The services provided have H types, and the set is expressed as E n The storage and computing resources of the edge server are respectively denoted as C n and F n , and the user equipment set is denoted as

[0146] Step 3, solve the joint optimization problem of the mixed integer nonlinear programming in step 2, the specific process is as follows:

[0147] S31, problem transformation, through binary relaxation, introduce penalty term, iterative construction of convex approximation, the non-convex problem is approximated to a sequence of convex optimization problems that can be efficiently solved, the specific steps are as follows:

[0148] 1) binary variable relaxation: simplify optimization solution, relax the binary variable to continuous variable And The value range is [0, 1], and a penalty term is introduced to encourage integer solution, and the penalty term expression is expressed as:

[0149]

[0150] The penalty term reaches the minimum value 0 when the variable takes the value 0 or 1, which promotes the relaxation variable to tend to binary value. The non-convexity of the original problem mainly comes from the hyperbolic constraint in the delay calculation of the objective function And variable product

[0151] For the nonlinear term in the objective function, an auxiliary variable τ s,i,n The auxiliary variable expression can be expressed as:

[0152]

[0153] This constraint is a hyperbolic constraint and is non-convex, which needs to be further processed. At the kth iteration point A first-order Taylor expansion is constructed to construct a convex approximation, which can be expressed as:

[0154]

[0155] The approximation is a linear constraint, which satisfies the convexity, and is equivalent to the original constraint at the current iteration point.

[0156] 2) coupling term of cache decision variable and delay energy consumption

[0157] For The non-convex term at the kth iteration point , a first-order Taylor expansion is constructed to construct a convex approximation, which can be expressed as:

[0158]

[0159] Similarly, for The term is constructed to construct a convex approximation, which can be expressed as:

[0160]

[0161] 3) coupling term of service request variable and calculation delay

[0162] For rs,i τ s,i,n non-convex term at the kth iteration point Convex approximation is constructed at the kth iteration point, the expression can be represented as:

[0163]

[0164] Combined with the above variable conversion and convex approximation design, McCormick envelope is used to construct the convex upper bound, which ensures that the approximation is strictly greater than the original function (since the goal is to maximize). In [0, 1] x [τ min ,τ max ], the expression of the McCormick envelope can be represented as:

[0165]

[0166] At the kth iteration, the approximation expression can be represented as:

[0167]

[0168] In the iteration, Δr, Δτ is constrained to limit, and the McCormick envelope ensures that the upper bound holds.

[0169] Combined with the above approximation, the objective function expression of the kth iteration can be represented as:

[0170]

[0171] The convex approximation expression of the objective function can be represented as:

[0172]

[0173] wherein, and represent the convex approximation expression of the delay and energy consumption at the current iteration point, and ρ represents the penalty term coefficient.

[0174] S32, dynamic popularity prediction, the dynamic popularity prediction mechanism based on joint probability designs cache priority, optimizes the cache decision of the next time slot, in time slot t, the probability expression that the video is requested by the user is represented as:

[0175]

[0176] wherein, the number of times that the video is requested is represented as N v (t), and the total number of requests in the time slot is represented as N total (t).

[0177] Based on the historical data and the current observation, the video request probability is obtained, and the predicted probability that the video is requested is obtained through prediction The predicted value of the next time slot is represented as:

[0178]

[0179] where, The smoothing factor β ∈ (0, 1). For each pair of video and service Let the number of times they are requested simultaneously be denoted as N v,s (t), the observed joint probability expression can be expressed as:

[0180]

[0181] Meanwhile, the joint request probability defined as the probability of both being requested simultaneously by a user, the exponential smoothing predicted value is expressed as:

[0182]

[0183] The correlation between video content and computing services can be quantified by the conditional probability, the conditional probability expression can be expressed as:

[0184]

[0185] Based on the defined cache priority expression can be expressed as:

[0186]

[0187] where, w1, w2 ∈ [0, 1] and w1 + w2 = 1, denotes the predicted probability of video v m being requested, denotes the predicted probability of video v m and service s h being requested simultaneously, the joint probability is affected by the priority ξ v,s (t+1), the group of video and service with higher ξ v,s (t+1) indicates that they are often requested simultaneously.

[0188] S33, according to the new conversion target function and cache decision in steps S31 and S32, the transformed target function is solved by a solver or a nonlinear integer programming algorithm.

[0189] As Figure 1As shown, it is a three-layer framework based on content-service caching mechanism. The overall architecture of the system is composed of three layers of cloud server (CS), edge servers (ESs) and user devices (UDs). Among them, the cloud server is responsible for providing all video content and computing services. The video content provided by the cloud server has M kinds, and the set is represented as The services provided have H kinds, and the set is represented as When the edge has not cached or cannot handle some requests, the request is transmitted to the cloud server for processing. The set of base stations is denoted as Each edge server n is equipped with a server to undertake the caching and computing tasks of content and service programs, and the set of edge servers is denoted as E n The storage and computing resources of the cloud server are denoted as C n and F n , respectively. The set of user devices is denoted as User devices are distributed in the coverage range of each base station and interact with the nearest edge server through a wireless network.

[0190] As shown in Figure 2 , each system running time is discretized into equal length time slots User device D i submits a task request to the system in each time slot, defined as a binary tuple R i = (v m , s h ), and then the system receives the return result according to the user's request information and after the task is obtained and completed. In video streaming applications, users send video content requests to the system, and after obtaining the video content, they may require services related to the video content (such as subtitle generation, video recognition analysis, etc.). Storing high request rate and high popularity video content and its corresponding services in edge servers close to the user side can greatly reduce the transmission delay and computing load of the cloud, reduce the communication and computing overhead caused by redundant transmission, and improve the overall performance of the system.

[0191] To evaluate the effectiveness of the task offloading method for joint optimization of content and service caching in edge computing proposed in the preferred embodiments of the present application on cache hit rate, a comparative experiment on cache hit rate under different cache capacities is selected. Based on python3.9, the ECSR (Edge Content-Service Relaxation Optimization algorithm) simulation scene is built, and the SCIP (Solving Constraint Integer Programs) solver is used to realize the proposed algorithm. When the solution is completed, ECSR can be applied to the joint optimization of content caching, service caching and computation offloading in dynamic and variable environment. The following four benchmark methods are selected for comprehensive performance comparison: LECS (Learning-based Cooperative Caching Strategy): This method establishes a model for the content caching problem, and uses an improved algorithm based on dynamic programming combined with the least squares method to solve the mixed integer nonlinear programming problem; COSC (Computation Offloading and Service Caching): This method models the joint optimization problem of service caching and computation offloading, and solves it alternately based on Lagrange multiplier iteration; LFU (Least Frequently Used): A cache replacement strategy based on the access frequency of content or service; LRU (Least Recently Used): A cache replacement strategy that replaces the content that has not been accessed in the latest time slot.

[0192] Taking cache hit rate as the core indicator, the experimental results are shown in Figure 3 The influence of the cache capacity of the edge server on the performance of different methods is evaluated by the present application. The experimental results show that with the increase of cache capacity, the cache hit rate of all methods shows an upward trend. This shows that cache capacity is a key factor affecting hit rate, that is, a larger cache space allows more content and services to be stored, thereby increasing the probability that user requests will be satisfied by the edge server. Among them, the ECSR method of the present application captures the correlation of videos and services by obtaining cache priority based on joint request probability, and preferentially caches high-popularity combinations, so that ECSR can still efficiently utilize limited space when the cache capacity is low. It shows that the task offloading method for joint optimization of content and service caching in edge computing proposed in the preferred embodiments of the present application can effectively alleviate the influence of high resource constraints on joint cache hit rate.

[0193] To sum up, the application discloses a task offloading method for content and service cache joint optimization in edge computing, which is suitable for video streaming, real-time data processing and other computing-intensive scenarios. The method first establishes a mathematical model containing transmission time, energy consumption and computing time according to the three-layer architecture of edge computing (cloud server, edge server and user equipment); secondly, a mixed integer nonlinear programming joint optimization problem is constructed, including the expressions of communication time, computing time and energy consumption, and the function with the maximum utility value as the target and the constraint conditions such as cache capacity and resource allocation; finally, through binary relaxation, convex approximation and dynamic popularity prediction, the non-convex problem is approximately converted into a sequence of convex optimization problems and solved. The method solves the coupled optimization problem of content and service cache decision and task offloading, reduces the communication delay and energy consumption of the edge computing system, and thus improves the utility value of the edge server.

[0194] Therefore, the application approximates the originally difficult mixed integer nonlinear programming problem to a convex optimization problem that is easy to solve through the successive convex approximation method, solves the problem that the decision of content and service cache decision and task offloading in edge computing is difficult to solve, reduces the communication time and energy consumption in the edge computing system, and maximizes the utility value of the edge server.

[0195] The above embodiments are only used to illustrate the technical solutions of the application rather than limit it. Although the application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the application can be modified or replaced by equivalents, and these modifications or replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the application.

Claims

1. A task offloading method for content and service cache joint optimization in edge computing, characterized in that, The method comprises the following steps: Step 1, according to the three-layer architecture of cloud server, edge server and user equipment in edge computing, a mathematical model is established for transmission time and energy consumption of content and service, and computing time and energy consumption of service task; Step 2, a joint optimization problem of mixed integer nonlinear programming is established according to the mathematical model of step 1; Step 3, the joint optimization problem of mixed integer nonlinear programming in step 2 is solved.

2. The task offloading method of claim 1, wherein, In step 2, the joint optimization problem of mixed integer nonlinear programming comprises: Expressions of communication time in the transmission process, expressions of computing time of service execution, expressions of energy consumption of communication and computing process, expressions of optimization objective function, and expressions of optimization problem constraints are established; The joint optimization problem is established with the maximum objective function as the target under the condition of meeting the optimization problem constraint.

3. The task offloading method of claim 2, wherein, The communication schedule expression includes an edge server to user device time and a cloud server to edge server time The edge server to user device communication time is expressed as: wherein, denotes the storage capacity occupied by the content, denotes the bandwidth allocated by the edge server to the user device, p E denotes the transmission power of the edge server, denotes the channel gain from the edge server n to the user device D i , σ 2 denotes the noise power; The communication time from the cloud server to the edge server is represented as: wherein, denotes the storage capacity occupied by the content, B c,n denotes the channel bandwidth between the cloud server and the edge server, p c denotes the transmission power of the cloud server, g c,n denotes the channel gain from the cloud server to the edge server, σ 2 denotes the noise power; The communication time expression in the transmission process is specifically as follows: where T i c denotes the total communication latency of user i, x v,n ∈{0,1} denotes the content cache status.

4. The task offloading method of claim 3, wherein, The service performs a computation time expression for: where y s,n ∈ {0,1} denotes the service cache status, denotes the service s h required storage capacity, ψ s,i denotes the edge server E n allocated to the user i, the computing resource proportion of the edge server E n is denoted as F n , and the computing amount required by the service requested by the user i is denoted as f s ; The total time delay expression of the user request is:

5. The task offloading method of claim 4, wherein, The energy consumption expression generated by the communication process is as follows: wherein, represents the communication power of the cloud server to the user device, r s,i ∈{0,1} represents the service request state, represents the communication power of the edge server to the user device, represents the communication power of the cloud server to the edge server, and the energy consumption of the edge server depends on the power allocation of the edge server and the task processing time; The energy consumption expression generated by the calculation process is as follows: wherein the computing power of the edge server denotes the computing power of the edge server; the total energy consumption expression of the user i is specifically as follows:

6. The task offloading method of claim 5, wherein, The optimization objective function is: where A i represents the revenue paid by mobile user i, and λ1 and λ2 are the weight coefficients of delay and energy consumption, respectively.

7. The task offloading method of claim 6, wherein, The optimization problem constraint conditions comprise: The content cache decision variable constraint is represented as: x v,n ∈ {0, 1}, The service cache decision variable constraint is represented as: y s,n ∈{0,1}, The service request decision variable constraint is expressed as: r s,i ∈{0,1}, The cache capacity constraint of the edge server is expressed as: The computing resource allocation constraints are expressed as: Wherein, the video content provided by the cloud server has M kinds, and the set is represented as The services provided have H kinds, and the set is represented as E n The storage and computing resources of the cloud server are respectively denoted as C n and F n , and the user equipment set is denoted as 8. The task offloading method of any one of claims 1-7, wherein, Step 3 is specifically as follows: S31, the non-convex problem is converted into a sequence of convex optimization problems through binary variable relaxation and convex approximation; S32, based on joint probability prediction dynamic popularity, the cache priority is optimized; S33, the converted objective function is solved by a solver or a nonlinear integer programming algorithm.

9. The task offloading method of claim 8, wherein, Step S31 is specifically as follows: 1) Binary variable relaxation: relaxing the binary decision variables of cache placement and service request placement from 0 or 1 to continuous variables and with the range of [0, 1], a penalty term is introduced to encourage integer solutions, the penalty term expression is represented as: The non-convexity of the problem comes from the hyperbolic constraint in the latency of the objective function computation and variable products For the nonlinear term in the objective function, an auxiliary variable τ s,i,n is introduced, and the expression of the auxiliary variable is represented as The bi-convex constraint is non-convex, at the kth iteration point By constructing a convex approximation via first order Taylor expansion, the expression is represented as: 2) processing the coupling term of cache decision variable and time delay energy consumption For the non-convex term of at the kth iteration point , a convex approximation is constructed by a first-order Taylor expansion, expressed as: For The item builds a convex approximation, expressed as: 3) processing the coupling term of service request variable and computing time delay For non-convex terms, a convex approximation is constructed at the kth iteration point The expression is expressed as: In [0, 1] x [τ min ,τ max ], the McCormick envelope expression is constructed as: At the kth iteration, the approximate expression is represented as: The objective function expression of the kth iteration is represented as: The convex approximation expression of the objective function is represented as: where, and denote the convex approximation expression of the delay and energy consumption at the current iteration point, respectively, and ρ denotes the penalty term coefficient.

10. The task offloading method of claim 9, wherein, Step S32, at time slot t, the probability expression that the video is requested by the user is represented as: where the number of times a video is requested is denoted as N v (t), the total number of requests within a time slot is denoted as N total (t); Based on historical data and current observation, the video request probability is obtained, and the prediction probability of the video being requested is obtained through prediction The predicted value of the next time slot is represented as: where P v (t), a smoothing factor β ∈ (0, 1); for each pair of videos and services Let the number of times they are requested simultaneously be denoted as N v,s (t), then the observed joint probability expression is given by: Joint request probability defined as the probability that both are requested simultaneously by a user, is predicted using exponential smoothing: The relevance between the video content and the computing service is quantified by a conditional probability, which is expressed in the conditional probability expression: The defined cache priority expression is represented as: where w1, w2∈ [0, 1] and w1 + w2 = 1, denotes the video v obtained by prediction m requested prediction probability, denotes the video v obtained by prediction m and service s h requested prediction probability.

Images