A joint offloading decision and resource allocation optimization method in green edge computing system

Abstract

A joint offloading decision and resource allocation optimization method in a green edge computing system takes advantage of the advantages of MEC, allowing users to obtain higher service quality in applications with low latency requirements. First, based on Lyapunov optimization and the system's long-term energy consumption cost budget, the long-term cost constraint is converted into a deterministic problem for each time slot; then, for this problem, the independence and monotonicity of its variables are analyzed, and the optimal user device transmission power and MEC server CPU frequency are first calculated during the optimal offloading. After completing the optimal resource allocation, a minimum cost greedy algorithm based on pseudo-matrix theory is designed for the offloading problem to solve the optimal task offloading decision for the user device.

Description

Technical Field

[0001] The present invention belongs to the field of 5G network edge computing offloading technology and resource allocation, and in particular to task offloading and computing and communication resource allocation of a green MEC system with multiple user devices supporting energy capture. Background Art

[0002] Mobile edge computing, as an emerging computing paradigm in recent years, provides an effective solution to the problems of high latency, fast battery loss of user devices, and low service quality in latency-sensitive applications under the traditional cloud computing framework. However, compared with traditional centralized cloud computing servers, the computing and storage resources of edge servers are relatively limited. How to jointly optimize offloading decisions and resource allocation plans under a limited long-term energy cost budget so that users in the system can enjoy low-latency applications and reduce the amount of unfinished tasks of devices has become a very important research issue. Summary of the invention

[0003] The present invention aims to overcome the above-mentioned shortcomings of the prior art and provide a method for joint offloading decision and resource allocation optimization in a green edge computing system.

[0004] The present invention takes advantage of mobile edge computing and designs a computing task offloading model that supports multiple energy capture devices in a green MEC system. That is, bandwidth resources and edge server CPU cycles are allocated to user devices in the system online, and offloading decisions are designed to reduce system latency and reduce the number of unfinished tasks for users in each time slot.

[0005] In order to achieve the above object, the present invention provides the following inventive steps:

[0006] A method for joint offloading decision and resource allocation optimization in a green edge computing system, the method comprising the following steps:

[0007] Step 1: Divide the time of the edge computing system into several time slots of length τ. The number of N user devices supporting energy capture in the system is represented by k,

[0008] Step 2: Ignore the time and energy consumption of the calculation results, and take the green energy E captured by the user device in time slot t as in (t) Task transfer for local computing or offloading;

[0009] Step 3: Generate the k-th user equipment buffer update formula in time slot t: S k (t+1)=min(A k (t)+S k (t)-C k (t), Smax ), where S max The task buffer capacity of each user device, S k (t) is the task size of the buffer at the initial stage of the time slot, A k (t) is the size of the random arriving task at the end of the time slot, C k (t) is the amount of tasks completed in the current time slot, Put the task that arrives at the end of the time slot into the task buffer so that it can be calculated in the next time slot;

[0010] Step 4: When the task is offloaded, the task is first transmitted to the MEC server. According to the Shannon formula, the transmission rate expression of the task when the kth user equipment is unloaded in time slot t is generated:

[0011]

[0012] in is the transmission power of the kth user equipment in time slot t, and its maximum value is B is the channel bandwidth available to the user equipment, σ is the noise power, h k (t) is the channel power gain, and the offloading strategy of the kth user equipment in time slot t is represented as x k (t), y k (t):

[0013]

[0014]

[0015] Then the number of uninstall users is calculated

[0016] Step 5: Based on step 4, generate the transmission time expression of the kth user equipment in time slot t: Transmission energy consumption expression

[0017] Step 6: Calculate the CPU frequency value that should be taken by the local calculation of the k-th user equipment in time slot t:

[0018]

[0019] in is the maximum CPU frequency value of the user device, μ is the CPU cycle required for the user device and the MEC server CPU to process 1 bit of data, and κ L Effective switched capacitor for the CPU of the user device,

[0020] Step 7: Generate the power expression calculated locally by the kth user equipment in time slot t Energy consumption expression in For the calculation time, the time of all user devices performing local calculations in time slot t and the expression

[0021] Step 8: Generate the computation time expression for offloading the k-th user equipment to the MEC server in time slot t in is the CPU core frequency of the MEC server, the maximum value is Then the energy consumption expression is generated in represents CPU power, κ MEC is the effective switching capacitance of the MEC server CPU, then the generated time slot t is offloaded to the MEC server computing time and the expression Generate energy consumption and expression

[0022] Step 9: Generate the time and expression for offloading the time slot t to the MCC calculation Energy consumption and expressions where d MCC is the latency of sending a task from MEC to MCC, e MCC Calculate the energy cost of the task for the MCC server;

[0023] Step 10: Combine steps 1-9 and represent the offloading strategy of the kth user equipment as vector w k (t) = [x k (t), y k(t)] T , the uninstallation policies of all user devices are represented as a vector W(t)=[w1(t),...,w N (t)], the CPU frequency of the MEC server is represented as a vector The transmission power when the user equipment transmits tasks to the MEC is represented as a vector The sum of the delays of all user devices in time slot t and the sum of the unfinished task penalties is expressed as Where γ is the penalty coefficient of unfinished tasks in time, generating the total energy cost expression E(t) = E MEC (t)+E MCC (t), the following mathematical model for task offloading is established:

[0024]

[0025]

[0026]

[0027]

[0028]

[0029]

[0030]

[0031] question The problem described is to minimize the sum of delays and unfinished tasks of multiple user devices under the long-term energy consumption constraint. The decision variables of the problem include task offloading decision, allocation of MEC server CPU frequency and allocation of transmission channel bandwidth. Constraint (1) expresses the duality of offloading decision. Constraints (2)-(4) are the maximum value constraints of transmission power, transmission time and transmission energy consumption when the user device is offloading. Constraint (5) is the minimum value (computation time constraint) and maximum value constraint of MEC server frequency when offloading. Constraint (6) is the constraint of long-term energy consumption cost.

[0032] Step 11: According to the Lyapunov optimization algorithm, a virtual queue with an initial value of 0 is constructed, which is expressed as

[0033] Q(t+1)=max{Q(t)+E(t)-E max ,0}

[0034] Step 12: Based on steps 10 and 11, reshape the original long-term optimization problem into a deterministic optimization problem for each time slot:

[0035]

[0036] st(1)-(5)

[0037] Where V is a non-negative control parameter;

[0038] Step 13: Introduce the maximum time constraint τ for transmission tx , that is, the transmission needs to be tx to complete within the time frame to decouple and To simplify the solution, constraints (3) and (5) are modified as follows:

[0039]

[0040]

[0041] Step 14: Solve the optimal CPU frequency for each user device during MEC calculation, that is, solve the following problem:

[0042]

[0043] st(8)

[0044] The optimal solution can be obtained by analyzing the monotonicity of the optimization objective function:

[0045]

[0046] Step 15: Similarly, the optimal transmission power of each user equipment during unloading is solved, that is, the following problem is solved:

[0047]

[0048] st(2),(4),(7)

[0049] The optimal solution can be obtained by analyzing the monotonicity of the optimization objective function:

[0050]

[0051] in It is the equation The solution, It is the equation The solution;

[0052] Step 16: Express the local computation cost as The unloading cost is expressed as The MEC computation cost is expressed as The MCC calculation cost is expressed as For user device k, since it only needs to select MEC and MCC with the smaller computation cost when offloading, the optimal solution is y k (t) does not need to be decided again, so the unloading decision is expressed as x(t) = [x1(t), x2(t), ..., x N (t)], the unloading cost is expressed as According to the optimal MEC server CPU frequency and transmission power obtained in steps 14 and 15, the task offloading decision problem is constructed as an integer linear programming mathematical model:

[0053]

[0054] st(1)

[0055]

[0056] The constant M0=0, 1, ..., N represents the number of unloading, that is, it is necessary to solve N+1 such integer linear programming problems according to the different numbers of unloading, and then compare the optimal solutions of these N+1 problems, which is the optimal unloading strategy;

[0057] Step 17: Based on the pseudo-matrix theory, calculate the cost difference between the offload calculation and the local calculation of each user device, that is, The user equipment with the smallest cost difference is selected for uninstallation in a loop until the number of uninstallations reaches M0, and the remaining user equipment is calculated locally;

[0058] Step 18: All user devices are unloaded according to the optimal unloading decision obtained in step 17, the CPU frequency of the user device calculated locally is set according to step 6, and no bandwidth resources are allocated, transmission power and bandwidth are allocated to the unloaded user device according to steps 14 and 15, and the CPU frequency of the MEC server is allocated to the user device that is offloaded to the MEC server for calculation;

[0059] Step 19: Update the task buffer of the virtual queue and the user equipment before the end of the current time slot, and repeat steps 2 to 18 in the next time slot.

[0060] The present invention adopts a joint offloading decision and resource allocation optimization method in a green edge computing system, which takes advantage of MEC and enables users to obtain higher service quality in applications with low latency requirements. First, based on Lyapunov optimization and the long-term energy consumption cost budget of the system, the long-term cost constraint is converted into a deterministic problem for each time slot; then, for this problem, the independence and monotonicity of its variables are analyzed, and the optimal user equipment transmission power and MEC server CPU frequency are first obtained during the optimal offloading. After completing the optimal resource allocation, a minimum cost greedy algorithm based on pseudo-matrix theory is designed for the offloading problem to solve the optimal task offloading decision of the user equipment.

[0061] The beneficial effects of the present invention are as follows: by utilizing the low latency advantage of mobile edge computing, while meeting the constraints of long-term energy budget costs, the task offloading decisions and the allocation of computing and communication resources are jointly optimized, ultimately achieving the effect of reducing system latency and reducing the number of unfinished tasks of user devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a schematic diagram of the joint task offloading and resource allocation model of multiple energy capture user devices under the green MEC system of the present invention. DETAILED DESCRIPTION

[0063] The present invention will be further described below in conjunction with the accompanying drawings.

[0064] Combination Figure 1, a joint offloading decision and resource allocation optimization method in a green edge computing system, comprising the following steps:

[0065] Step 1: If Figure 1 There are N user devices supporting energy capture in the system. The devices are fully powered by green energy in the environment, including local computing energy consumption and transmission energy consumption during unloading. The time of the edge computing system is divided into several time slots of length τ. The number of user equipment in the system is represented by k, Each device has a task buffer to store newly arrived tasks and unfinished tasks in the current time slot. In each time slot, the device can choose to calculate the task locally, offload it to the MEC server, or offload it to the MCC server. If it is to be offloaded to the MCC server, it must first be transmitted to the MEC server;

[0066] Step 2: Ignore the time and energy consumption of the calculation results, and take the green energy E captured by the user device in time slot t as in (t) Task transfer for local computing or offloading;

[0067] Step 3: Generate the k-th user equipment buffer update formula in time slot t: S k (t+1)=min(A k (t)+S k (t)-C k (t), S max ), where S max The task buffer capacity of each user device, S k (t) is the task size of the buffer at the initial stage of the time slot, A k (t) is the size of the random arriving task at the end of the time slot, C k (t) is the amount of tasks completed in the current time slot, Put the task that arrives at the end of the time slot into the task buffer so that it can be calculated in the next time slot;

[0068] Step 4: When the task is offloaded, the task is first transmitted to the MEC server. According to the Shannon formula, the transmission rate expression of the task when the kth user equipment is unloaded in time slot t is generated:

[0069]

[0070] in is the transmission power of the kth user equipment in time slot t, and its maximum value is B is the channel bandwidth available to the user equipment, σ is the noise power, h k (t) is the channel power gain, and the offloading strategy of the kth user equipment in time slot t is represented as x k(t), y k (t):

[0071]

[0072]

[0073] The number of uninstall users is calculated

[0074] Step 5: According to step 4, calculate the transmission time Transmission energy consumption

[0075] Step 6: Calculate the CPU frequency value that should be taken by the local calculation of the k-th user equipment in time slot t:

[0076]

[0077] in is the maximum CPU frequency value of the user device, μ is the CPU cycle required for the user device and the MEC server CPU to process 1 bit of data, and κ L Effective switched capacitor for the CPU of the user device,

[0078] Step 7: Generate the power expression calculated locally by the kth user equipment in time slot t Energy consumption expression in For the calculation time, the time of all user devices performing local calculations in time slot t and the expression

[0079] Step 8: Generate the computation time expression for offloading the k-th user equipment to the MEC server in time slot t in is the CPU core frequency of the MEC server, the maximum value is Then the energy consumption expression is generated in represents CPU power, κ MEC is the effective switching capacitance of the MEC server CPU, then the generated time slot t is offloaded to the MEC server computing time and the expression Generate energy consumption and expression

[0080] Step 9: Generate the time and expression for offloading the time slot t to the MCC calculation Energy consumption and expressions where dMCC is the delay of sending the task from MEC to MCC, which is a constant. This is because the time it takes to send a task between MEC and MCC is mainly composed of the propagation delay, which is determined by the geographical distance, while the transmission delay difference caused by the amount of data sent can be ignored. MCC The energy cost of computing tasks for the MCC server is a constant, because the duration of renting the MCC server is a fixed value;

[0081] Step 10: Combine steps 1-9 and represent the offloading strategy of the kth user equipment as vector w k (t) = [x k (t), y k (t)] T , the uninstallation policies of all user devices are represented as a vector W(t)=[w1(t),...,w N (t)], the CPU frequency of the MEC server is represented as a vector The transmission power when the user equipment transmits tasks to the MEC is represented as a vector The sum of the delays of all user devices in time slot t and the sum of the unfinished task penalties is expressed as Where γ is the penalty coefficient of unfinished tasks in time, generating the total energy cost expression E(t) = E MEC (t)+E MCC (t), the following mathematical model for task offloading is established:

[0082]

[0083]

[0084]

[0085]

[0086]

[0087]

[0088]

[0089] question The problem described is to minimize the sum of delays and unfinished tasks of multiple user devices under the long-term energy consumption constraint. The decision variables of the problem include task offloading decision, allocation of MEC server CPU frequency and allocation of transmission channel bandwidth. Constraint (1) expresses the duality of offloading decision. Constraints (2)-(4) are the maximum value constraints of transmission power, transmission time and transmission energy consumption when the user device is offloading. Constraint (5) is the minimum value (computation time constraint) and maximum value constraint of MEC server frequency when offloading. Constraint (6) is the constraint of long-term energy consumption cost.

[0090] Step 11: According to the Lyapunov optimization algorithm, a virtual queue with an initial value of 0 is constructed, which is expressed as

[0091] Q(t+1)=max{Q(t)+E(t)-E max ,0}

[0092] Step 12: Based on steps 10 and 11, reshape the original long-term optimization problem into a deterministic optimization problem for each time slot:

[0093]

[0094] st(1)-(5)

[0095] Where V is a non-negative control parameter;

[0096] Step 13: Introduce the maximum time constraint τ for transmission tx , that is, the transmission needs to be tx to complete within the time frame to decouple and To simplify the solution, constraints (3) and (5) are modified as follows:

[0097]

[0098]

[0099] Step 14: Solve the optimal CPU frequency for each user device during MEC calculation, that is, solve the following problem:

[0100]

[0101] st (8)

[0102] The optimal solution can be obtained by analyzing the monotonicity of the optimization objective function:

[0103]

[0104] Step 15: Similarly, the optimal transmission power of each user equipment during unloading is solved, that is, the following problem is solved:

[0105]

[0106] st(2),(4),(7)

[0107] The optimal solution can be obtained by analyzing the monotonicity of the optimization objective function:

[0108]

[0109] in It is the equation The solution, It is the equation The solution;

[0110] Step 16: Express the local computation cost as The unloading cost is expressed as The MEC computation cost is expressed as The MCC calculation cost is expressed as For user device k, since it only needs to select MEC and MCC with the smaller computation cost when offloading, the optimal solution is y k (t) does not need to be decided again, so the unloading decision is expressed as x(T) = [x1(t), x2(t), ..., x N (t)], the unloading cost is expressed as According to the optimal MEC server CPU frequency and transmission power obtained in steps 14 and 15, the task offloading decision problem is constructed as an integer linear programming mathematical model:

[0111]

[0112] st(1)

[0113]

[0114] The constant M0=0, 1, ..., N represents the number of unloading, that is, it is necessary to solve N+1 such integer linear programming problems according to the different numbers of unloading, and then compare the optimal solutions of these N+1 problems, which is the optimal unloading strategy;

[0115] Step 17: Based on the pseudo-matrix theory, calculate the cost difference between the offload calculation and the local calculation of each user device, that is, The user equipment with the smallest cost difference is selected for uninstallation in a loop until the number of uninstallations reaches M0, and the remaining user equipment is calculated locally;

[0116] Step 18: All user devices are unloaded according to the optimal unloading decision obtained in step 17, the CPU frequency of the user device calculated locally is set according to step 6, and no bandwidth resources are allocated, transmission power and bandwidth are allocated to the unloaded user device according to steps 14 and 15, and the CPU frequency of the MEC server is allocated to the user device that is offloaded to the MEC server for calculation;

[0117] Step 19: Update the task buffer of the virtual queue and the user equipment before the end of the current time slot, and repeat steps 2 to 18 in the next time slot.

[0118] The contents described in the embodiments of this specification are merely an enumeration of the implementation forms of the inventive concept. The protection scope of the present invention should not be regarded as limited to the specific forms described in the embodiments. The protection scope of the present invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A method for joint offloading decision and resource allocation optimization in a green edge computing system, comprising the following steps: Step 1: Divide the time of the edge computing system into several time slots of length τ. The number of N user devices supporting energy capture in the system is represented by k, Step 2: Ignore the time and energy consumption of the calculation results, and take the green energy E captured by the user device in time slot t as in (t) Task transfer for local computing or offloading; Step 3: Generate the k-th user equipment buffer update formula in time slot t: S k (t+1)=min(A k (t)+S k (t)-C k (t), S max ), where S max The task buffer capacity of each user device, S k (t) is the task size of the buffer at the initial stage of the time slot, A k (t) is the size of the random arriving task at the end of the time slot, C k (t) is the amount of tasks completed in the current time slot, Put the task that arrives at the end of the time slot into the task buffer so that it can be calculated in the next time slot; Step 4: When the task is offloaded, the task is first transmitted to the MEC server. According to the Shannon formula, the transmission rate expression of the task when the kth user equipment is unloaded in time slot t is generated: in is the transmission power of the kth user equipment in time slot t, and its maximum value is B is the channel bandwidth available to the user equipment, σ is the noise power, h k (t) is the channel power gain, and the offloading strategy of the kth user equipment in time slot t is represented as x k (t), y k (t): The number of uninstall users is calculated Step 5: Based on step 4, generate the transmission time expression of the kth user equipment in time slot t: Transmission energy consumption expression Step 6: Calculate the CPU frequency value that should be taken by the local calculation of the k-th user equipment in time slot t: in is the maximum CPU frequency value of the user device, μ is the CPU cycle required for the user device and the MEC server CPU to process 1 bit of data, and κ L Effective switched capacitor for the CPU of the user device, Step 7: Generate the power expression calculated locally by the kth user equipment in time slot t Energy consumption expression in For the calculation time, the time of all user devices performing local calculations in time slot t and the expression Step 8: Generate the computation time expression for offloading the k-th user equipment to the MEC server in time slot t in is the CPU core frequency of the MEC server, the maximum value is Then the energy consumption expression is generated in represents CPU power, κ MEC is the effective switching capacitance of the MEC server CPU, then the generated time slot t is offloaded to the MEC server computing time and the expression Generate energy consumption and expression Step 9: Generate the time and expression for offloading the time slot t to the MCC calculation Energy consumption and expressions where d MCC is the latency of sending a task from MEC to MCC, e MCC Calculate the energy cost of the task for the MCC server; Step 10: Combine steps 1-9 and represent the offloading strategy of the kth user equipment as vector w k (t) = [x k (t), y k (t)] T , the uninstallation policies of all user devices are represented as a vector W(t)=[w1(t),...,w N (t)], the CPU frequency of the MEC server is represented as a vector The transmission power when the user equipment transmits tasks to the MEC is represented as a vector The sum of the delays of all user devices in time slot t and the sum of the unfinished task penalties is expressed as Where γ is the penalty coefficient of unfinished tasks in time, generating the total energy cost expression E(t) = E MEC (t)+E MCC (t), the following mathematical model for task offloading is established: question The problem described is to minimize the sum of delays and unfinished tasks of multiple user devices under the long-term energy consumption constraint. The decision variables of the problem include task offloading decision, allocation of MEC server CPU frequency and allocation of transmission channel bandwidth. Constraint (1) expresses the duality of offloading decision. Constraints (2)-(4) are the maximum value constraints of transmission power, transmission time and transmission energy consumption when the user device is offloading. Constraint (5) is the minimum value (computation time constraint) and maximum value constraint of MEC server frequency when offloading. Constraint (6) is the constraint of long-term energy consumption cost. Step 11: According to the Lyapunov optimization algorithm, a virtual queue with an initial value of 0 is constructed, which is expressed as Q(t+1)=max{Q(t)+E(t)-E max ,0} Step 12: Based on steps 10 and 11, reshape the original long-term optimization problem into a deterministic optimization problem for each time slot: st(1)-(5) Where V is a non-negative control parameter; Step 13: Introduce the maximum time constraint τ for transmission tx , that is, the transmission needs to be tx to complete within the time frame to decouple and To simplify the solution, constraints (3) and (5) are modified as follows: Step 14: Solve the optimal CPU frequency for each user device during MEC calculation, that is, solve the following problem: st(8) The optimal solution can be obtained by analyzing the monotonicity of the optimization objective function: Step 15: Similarly, the optimal transmission power of each user equipment during unloading is solved, that is, the following problem is solved: st(2),(4),(7) The optimal solution can be obtained by analyzing the monotonicity of the optimization objective function: in It is the equation The solution, It is the equation The solution; Step 16: Express the local computation cost as The unloading cost is expressed as The MEC computation cost is expressed as The MCC calculation cost is expressed as For user device k, since it only needs to select MEC and MCC with the smaller computation cost when offloading, the optimal solution is y k (t) does not need to be decided again, so the unloading decision is expressed as x(t) = [x1(t), x2(t), ..., x N (t)], the unloading cost is expressed as According to the optimal MEC server CPU frequency and transmission power obtained in steps 14 and 15, the task offloading decision problem is constructed as an integer linear programming mathematical model: st(1) The constant M0=0, 1, ..., N represents the number of unloading, that is, it is necessary to solve N+1 such integer linear programming problems according to the different numbers of unloading, and then compare the optimal solutions of these N+1 problems, which is the optimal unloading strategy; Step 17: Based on the pseudo-matrix theory, calculate the cost difference between the offload calculation and the local calculation of each user device, that is, The user equipment with the smallest cost difference is selected for uninstallation in a loop until the number of uninstallations reaches M0, and the remaining user equipment is calculated locally; Step 18: All user devices are unloaded according to the optimal unloading decision obtained in step 17, the CPU frequency of the user device calculated locally is set according to step 6, and no bandwidth resources are allocated, transmission power and bandwidth are allocated to the unloaded user device according to steps 14 and 15, and the CPU frequency of the MEC server is allocated to the user device that is offloaded to the MEC server for calculation; Step 19: Update the task buffer of the virtual queue and the user equipment before the end of the current time slot, and repeat steps 2 to 18 in the next time slot.

Images