A Computation Offloading Method Considering Task Priority in a Mobile Edge Computing Network

Abstract

The present invention relates to a computing offloading method considering task priorities in a mobile edge computing network, characterized in that the method comprises the steps of: S1, obtaining network configuration information of a system model; S2, generating a task set containing priority constraint relationships, and obtaining the total delay and energy consumption overhead of the task set for computing on each server; S3, randomly initializing and generating an initial particle swarm of the task set, allocating the tasks in the task set to each server respectively, and performing particle position encoding on each task to obtain a task scheduling sequence; S4, constructing a fitness evaluation function and calculating the fitness function values of each particle; S5, updating the particle velocity and position to obtain the historical optimal position; S6, iterating cyclically to obtain a globally optimal computing offloading method. The method of the present invention takes priorities as constraints and aims to minimize the total overhead, and reasonably offloads tasks with priority dependency relationships to different servers.

Description

Technical Field

[0001] The present invention belongs to the field of communication technologies, and particularly relates to a computing offloading method considering task priorities in a mobile edge computing network. Background Art

[0002] With the development of the Internet of Things, although the processing capabilities of current mobile devices are becoming increasingly powerful, mobile devices usually have limited battery capacity and cannot process huge application programs in a short time. Strict latency constraints have become an obstacle to running complex application programs on mobile devices. Mobile cloud computing can significantly reduce the processing latency of application programs because of its rich computing resources. Although the data processing speed is very fast, the network bandwidth is very limited. To address this key challenge, users can offload tasks to mobile edge computing servers to improve performance. Therefore, in the scenario of collaborative computing offloading among cloud servers, edge servers, and user local sides, formulating an offloading decision scheme has become a hot issue at present. Summary of the Invention

[0003] Based on the above-mentioned drawbacks and deficiencies existing in the prior art, one of the objectives of the present invention is to at least solve one or more of the above-mentioned problems existing in the prior art. In other words, one of the objectives of the present invention is to provide a task offloading and resource allocation method for a mobile edge computing network that meets one or more of the foregoing requirements.

[0004] To achieve the above-mentioned invention objective, the present invention adopts the following technical solutions:

[0005] A computing offloading method considering task priorities in a mobile edge computing network, the method comprising the steps of:

[0006] S1. Obtain the network configuration information of the system model, where the system model consists of several base station groups, a mobile cloud server, and a user local server belonging to a certain base station group. Each base station group includes a base station and an edge computing server;

[0007] S2. Generate a task set containing priority constraint relationships, obtain the computing delays and energy consumptions of the task set for computing on the user local server, the edge computing server, and the mobile cloud server according to the network configuration information, and obtain the transmission delays of each task in the task set during transmission between the user local server, the edge computing server, and the mobile cloud server;

[0008] S3. Randomly initialize and generate an initial particle swarm of the task set, allocate the tasks in the task set to the user local server, the edge computing server, or the mobile cloud server respectively. Each task is a particle, and perform particle position encoding on each task according to the server to which it is allocated to obtain a task scheduling sequence;

[0009] S4. Construct a fitness evaluation function based on the total overhead of computing delay, transmission delay, and energy consumption, and calculate the fitness function value of each particle according to the initial particle swarm and particle position encoding;

[0010] S5. Update the particle velocity and position to obtain the historical optimal positions of the particles and the particle swarm;

[0011] S6. Set the number of iterations N, and loop through steps S3 - S5 until the number of iterations to obtain the globally optimal computing offloading method.

[0012] As a preferred solution, the computing delay of each task in the task set for computing on the user's local server, edge computing server, and mobile cloud server is calculated by dividing the amount of calculation of each task by the number of CPU cycles that the user's local server, edge computing server, and mobile cloud server can provide per second.

[0013] As a preferred solution, if two tasks with a predecessor-successor relationship in the task set are on the same server, the transmission delay is 0.

[0014] As a preferred solution, in step S2, when determining the energy consumption of the task set for computing on the user's local server, edge computing server, and mobile cloud server, the energy consumption of the server is determined in combination with the hardware architecture of the server.

[0015] As a preferred solution, in step S5, the particle velocity and position are updated using the simulated annealing method.

[0016] As a preferred solution, the calculation method of the task scheduling sequence in step S3 includes the following steps:

[0017] S311. Sort the tasks in ascending order according to the layer values they are in to obtain the task set scheduling sequence;

[0018] S312. Calculate the server number to which the task is assigned from the particle position, that is, the server closest to the task.

[0019] S313. According to the task set scheduling sequence obtained in step S311 and the mapping relationship between the tasks and the servers obtained in S312, obtain the task scheduling sequence on each server.

[0020] As a further preferred solution, in step S311, for tasks with the same layer value, they are sorted according to the order of the individual gene values.

[0021] Compared with the prior art, the beneficial effects of the present invention are:

[0022] The method of the present invention can unload the subtask set with priority dependency at the user side to different servers under the collaborative computing scenario of the cloud server, the edge server and the user local side, with the priority of the user side subtask set as the constraint and the goal of minimizing the total task completion delay and the total system energy consumption. Description of the Drawings

[0023] Figure 1 is a flowchart of a computing offloading method considering task priority in a mobile edge computing network according to an embodiment of the present invention;

[0024] Figure 2 is a schematic structural diagram of a system model according to an embodiment of the present invention;

[0025] Figure 3 is a schematic diagram of a task model diagram according to an embodiment of the present invention. Detailed Embodiments

[0026] In order to more clearly illustrate the embodiments of the present invention, the specific embodiments of the present invention will be described below with reference to the accompanying drawings. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts, and other embodiments can also be obtained.

[0027] Embodiment: This embodiment provides a computing offloading method considering task priority in a mobile edge computing network, and its flowchart is as Figure 1 shown: First, perform step S1, obtain the network configuration information of the system model. The structure of the system model is as Figure 2 shown, which consists of several base station groups, a mobile cloud server, and a user local server belonging to a certain base station group. Each base station group includes a base station and an edge computing server; a server can only process one task at the same time. When the user side task set is too large, the base station group to which the user belongs can communicate with other base station groups in the system model for collaborative computing, and the scheduling scheme is centrally decided by the mobile cloud server.

[0028] The network configuration information of the system model consists of the information of each processor. P = {p i |1 ≤ i ≤ M} is the set of processors in the system model, where p1 is the user local processor, p2 is the edge computing server of the base station group where the user is located, p M is the mobile cloud server, and the rest are other mobile cloud servers in the same system model as the base station group to which the user belongs. f L , f E , f Crespectively represent the computing capabilities of the user's local server, the edge computing server, and the mobile cloud server, that is, the number of CPU cycles that can be provided per second. The network configuration information also includes transmission parameters such as the fixed transmission power of the user's local server, the channel gain between the user's local server and the base station, and the channel bandwidth.

[0029] After obtaining the network configuration information, perform step S2: generate a task set including priority constraint relationships, obtain the computing delay and energy consumption of the task set when computing on the user's local server, the edge computing server, and the mobile cloud server according to the network configuration information, and obtain the transmission delay of each task in the task set when transmitting between the user's local server, the edge computing server, and the mobile cloud server;

[0030] Specifically, the priority constraint relationship between task sets is represented by a directed acyclic graph model such as Figure 3 the task model graph, that is, G = <V, E>. Where V is the task set, V = {v i | 1 ≤ i ≤ N}. E is the set of directed edges between tasks, representing the priority constraint relationship between tasks. e ij ∈ E means that task v i can be executed only after task v j is executed. Each task v i = {d i , g i}, where d i is the number of CPU cycles required to complete the task, and g i is the size of the task input data.

[0031] For the known DAG model, the set of predecessor nodes of task v i is pre(v i ), and the set of successor nodes is suc(v i ). The layer value of each task in the task set can be determined by the following formula:

[0032]

[0033] h(i) represents that if task v i has no predecessor nodes, it is used as the first node. If there are predecessor nodes, it is used as the next node.

[0034] According to the network configuration information obtained in S1, the computing delay of tasks on each server can be obtained. This embodiment provides a specific calculation method for the computing delay:

[0035] The computing delay of the i-th task on the user's local server is:

[0036] The computing delay of the i-th task on the edge computing server is:

[0037] The computing delay of the \(i\)-th task on the mobile cloud server is:

[0038] Combining them, the computing delay of the \(i\)-th task on the server is: where \(\alpha\), \(\beta\), \(\gamma\) are 0-1 variables and \(\alpha+\beta+\gamma = 1\).

[0039] For the transmission delay of tasks during transmission among the user's local server, the edge computing server, and the mobile cloud server, the following method is used for calculation:

[0040] Considering the quasi-static channel model, the upload rate of the user to offload tasks to the affiliated base station is: where \(P_0\) represents the fixed transmit power of the user terminal, \(h_0\) represents the channel gain between the user and the BS, \(\sigma\) 2 represents the power of additive white Gaussian noise, and \(w\) represents the channel bandwidth.

[0041] The downlink rate of the base station to the user terminals belonging to it is: where \(P\) E represents the fixed transmit power of the base station.

[0042] Then, combining them, in the precedence constraint relationship, the transmission delay from a predecessor task \(v\) j to a successor task \(v\) i is:

[0043]

[0044] The above formula includes ten cases, and the arrow direction represents the offloading platform direction. For example: \(c\) i,α→β represents that the predecessor task \(j\) is processed on the local server, and the successor task \(i\) is processed on the edge computing server of the affiliated base station.

[0045] Specifically, the following are the calculation formulas for the ten cases:

[0046]

[0047] where \(\pi\) E is the transmission delay between base stations in the system model, and \(\pi\) C is the transmission delay from the base station to the mobile cloud server. \(y\) i,α , \(y\) j,β are both 0-1 variables. For example: when the predecessor task \(v\) j is offloaded to the edge computing server of the user's affiliated base station, \(y\) j,β = 1.

[0048] For the energy consumption of task sets during computing on the user's local server, edge computing server, and mobile cloud server, the following method can be used for calculation:

[0049] The energy consumption of tasks on the user's local server, edge computing server, and mobile cloud server are respectively:

[0050]

[0051]

[0052]

[0053] Among them, hc is the server hardware coefficient.

[0054] The total energy consumption of the system is:

[0055] Combining the computing delay, transmission delay, and energy consumption when the above task set is computed on the user's local server, edge computing server, and mobile cloud server, the model of the total optimization problem is:

[0056]

[0057] s.t. α + β + γ = 1, α, β, γ ∈ {0, 1};

[0058]

[0059]

[0060] 0 ≤ λ ≤ 1.

[0061] Then perform step S3, randomly initialize to generate the initial particle swarm of the task set, allocate the tasks in the task set to the user's local server, edge computing server, or mobile cloud server respectively, each task is a particle, and perform particle position encoding for each task according to the server it is allocated to;

[0062] Specifically, in the initialization stage, randomly initialize the particle positions. In this embodiment, the total number of processors is M, the number of user tasks is N, and let the maximum initial velocity be VP max = 2 × M, the maximum initial position be XP max = M, the search space dimension D = N. The initialized velocity VP of the particle is randomly generated within [-VP max / 2, VP max / 2]. If the serial number of the processor to which task v i is allocated is b i , then the initial particle position xp i of task v i = (b i-1) + β, where β ∈ U[0,1] is a pseudo-random number uniformly distributed within the range of [0,1].

[0063] After obtaining the particle position encoding, the specific steps of step S31 for calculating the task scheduling sequence are

[0064] S311. First, sort the tasks according to their layer values from small to large to obtain the task set scheduling sequence; further, for tasks with the same layer value, sort them according to their individual gene values from large to small.

[0065] S312. Calculate the server number to which the task is assigned from the particle position xp i That is, the server closest to the task, b is the server number to which task v i is assigned. i

[0066] S313. According to the task set scheduling sequence obtained in step S311 and the mapping relationship between tasks and processors obtained in S312, the scheduling sequence on each processor can finally be obtained.

[0067] Then perform step S4: Construct a fitness evaluation function, and calculate the fitness function values of each particle according to the initial particle swarm and particle position encoding;

[0068] Construct an adaptive function where λ is the gravity factor, used to balance the two optimization objectives of the total user delay and the total system energy consumption, is the normalization factor, which keeps the two optimization objectives of the total user delay and the total system energy consumption at the same order of magnitude.

[0069] where the total delay T of the task set total = max{s i + t i}, t i is the execution delay of task i, s i is the start execution time of task v i ; s i is determined by the formula and is the start execution time under resource constraints where task v k is the predecessor task of task v i in the scheduling sequence of the assigned server; is the start execution time under dependency constraints

[0070] ​After calculating the adaptive function of each particle using the adaptive function, perform step S5 to update the particle velocity and position to obtain the historical optimal positions of the particles and the particle swarm;

[0071] Update the velocity of the particle according to the following formula: where c1 and c2 are learning factors; t is the number of iterations; w is the inertia weight. The inertia weight determines the inheritance value of the particle's current velocity and can be used to balance the global nature of convergence and the convergence speed.

[0072] The calculation formula for the inertia weight w is where w max and w min represent the maximum and minimum values of the inertia weight respectively; maxt is the total number of iterations. The historical best point experienced by the particle is denoted as AP i =(ap i1 , ap i2 , …, ap iD ), and the historical best point experienced by all particles in the population is denoted as AG i =(ag i1 , ag i2 , …, ag iD ).

[0073] Update the position of the particle according to the following equation:

[0074]

[0075] Furthermore, the particle position and velocity update use the simulated annealing method, and the specific steps are as follows:

[0076] A1: Select particle A j , A j from the particle swarm in sequence, set the fitness value of A as the initial solution, and initialize the optimal task allocation scheme A best =A j ; set the initial temperature R0, the initial iteration temperature is R it =R0, the termination temperature R f , the temperature reduction value ΔR, and the number of inner loop times N inner .

[0077] A2: Perform a mutation operation on individual A j , and the mutated individual is A' j . Calculate the fitness function ΔF = fitness(A' j' ) - fitness(A j' );

[0078] A3: If the fitness function value of A' j is less than that of Abest the fitness function value of, then A best = A' j ;

[0079] A4: If ΔF < 0, A j = A' j ; If ΔF > 0, then judge where ξ = Rand(0,1), if it holds then A j = A j , otherwise A j = A' j ;

[0080] A5: Judge whether the number of loop iterations reaches N inner , if not, return to step A2, if so, execute step A6;

[0081] A6: Calculate the iterative temperature R it = R it -ΔR, when R it < R f , output the allocation scheme A best , otherwise execute step A2.

[0082] A7: Repeat steps A1 - A6 until all particles in the particle swarm are updated.

[0083] After all particles are updated, obtain the historical optimal positions of each particle and the historical optimal position of the particle swarm according to the update history of the particle swarm.

[0084] Perform step S6, set an iteration number, in this embodiment, the iteration number is set to 200, loop and iterate steps S4 - S5 until the iteration number 200 is reached. Then use the finally determined historical optimal positions of the particles to determine the optimal server for each task allocation, and obtain the globally optimal computing offloading method.

[0085] It should be noted that the above embodiments only elaborate in detail the preferred embodiments and principles of the present invention. For those of ordinary skill in the art, according to the idea provided by the present invention, there will be changes in the specific implementation manners, and these changes should also be regarded as the protection scope of the present invention.

Claims

1. A computing offloading method considering task priority in a mobile edge computing network, characterized in that The method includes the following steps: S1. Obtain the network configuration information of the system model, where the system model consists of several base station groups, a mobile cloud server, and a user local server belonging to a certain base station group. Each base station group includes a base station and an edge computing server; S2. Generate a task set containing priority constraint relationships, obtain the computing delay and energy consumption of the task set for computing on the user local server, the edge computing server, and the mobile cloud server according to the network configuration information, and obtain the transmission delay of each task in the task set during transmission between the user local server, the edge computing server, and the mobile cloud server; S3. Randomly initialize and generate an initial particle swarm of the task set, allocate the tasks in the task set to the user local server, the edge computing server, or the mobile cloud server respectively. Each task is a particle, and perform particle position encoding on each task according to the server to which it is allocated to obtain a task scheduling sequence; S4. Construct a fitness evaluation function according to the total overhead of the computing delay, the transmission delay, and the energy consumption, and calculate the fitness function values of each particle according to the initial particle swarm and the particle position encoding; S5. Update the particle velocity and position to obtain the historical optimal positions of the particles and the particle swarm; S6. Set the number of iterations N, loop steps S3 - S5 until the number of iterations to obtain the globally optimal computing offloading method; The construction of the fitness evaluation function according to the total overhead of the computing delay, the transmission delay, and the energy consumption specifically includes: Construct the adaptive function where λ is the gravity factor, which is used to balance the two optimization objectives of the total user delay and the total system energy consumption, is the normalization factor, which keeps the two optimization objectives of the total user delay and the total system energy consumption at the same order of magnitude; Among them, the total delay T of the task set total = max{s i + t i}, where t i is the execution delay of task i, and s i is the start execution time of task v i ; s i Determined by the formula , is the start execution time under resource constraint conditions where task v k is the predecessor task of task v i in the scheduling sequence of the assigned server; is the start execution time under dependency constraint conditions 2. The computing offloading method considering task priority in a mobile edge computing network according to claim 1, characterized in that, The computing delay of each task in the task set for computing on the user local server, the edge computing server, and the mobile cloud server is calculated by dividing the amount of computation of each task by the number of CPU cycles that the user local server, the edge computing server, and the mobile cloud server can provide per second.

3. The computing offloading method considering task priority in a mobile edge computing network according to claim 1, wherein, If two tasks with a predecessor-successor relationship in the task set are on the same server, the transmission delay is 0.

4. The calculation offloading method considering task priority in a mobile edge computing network according to claim 1, characterized in that In step S2, when determining the energy consumption of the task set for computing on the user local server, the edge computing server, and the mobile cloud server, the energy consumption of the server is determined in combination with the hardware architecture of the server.

5. The computing offloading method considering task priority in a mobile edge computing network according to claim 1, wherein In step S5, the particle velocity and position are updated using the simulated annealing method.

6. The calculation offloading method considering task priority in a mobile edge computing network according to claim 1, characterized in that The calculation method of the task scheduling sequence in step S3 includes the following steps: S311. Sort according to the layer value of the task from small to large to obtain a task set scheduling sequence; S312. Calculate the server serial number to which the task is assigned based on the particle position xp, that is, the server closest to the task. i ​ S313. According to the task set scheduling sequence obtained in step S311 and the mapping relationship between the tasks and the servers obtained in S312, obtain the task scheduling sequences on each server.

7. The computing offloading method considering task priority in a mobile edge computing network according to claim 6, characterized in that In step S311, for tasks with the same layer value, sort them according to the size order of the individual gene values.

Images