CN112235387A - Multi-node cooperative computing unloading method based on energy consumption minimization - Google Patents

Abstract

The invention discloses a multi-node cooperative computing unloading method based on energy consumption minimization, which takes network energy consumption minimization as an objective function, comprehensively considers network delay and QoS requirements, converts an optimization process into an integer linear programming problem, and adopts a branch-and-bound method to realize an optimization target. Simulation analysis shows that compared with the traditional calculation unloading method, the multi-node cooperative calculation unloading method provided by the invention effectively reduces the network energy consumption and simultaneously ensures the execution of more data volume. The method can be applied to an intelligent home scene, green communication of the intelligent home is realized, and a local user side supporting Internet of things (IoT) divides the computing task and then parallelly unloads the computing task to a plurality of MEC nodes or cloud ends.

Description

Multi-node cooperative computing unloading method based on energy consumption minimization

Technical Field

The invention belongs to the field of cloud computing in a communication network, and particularly relates to a multi-node cooperative computing unloading method based on energy consumption minimization.

Background

With the development of internet of things (IoT) technology in recent years, devices in the IoT network have sensing and communication capabilities, and network clients can extend to any article in life to exchange and communicate information. Meanwhile, the IoT technology is also applied to various aspects in industrial production and daily life, and application scenarios include smart home, smart industry, smart city, and the like, aiming at transmission and network performance optimization. The application scene of the smart home is mainly considered, and since the IoT local user side in the smart home scene can be any article, user data in the IoT has the characteristic of diversity, and meanwhile, the smart electrical appliance is required to process task data more quickly and efficiently. Therefore, for some users with large data volume or sensitive time delay, a faster, efficient and safe task processing mode is needed to meet the user requirements. The traditional single cloud model cannot meet the user requirements, and for this reason, the concept of edge computing (MEC) is proposed on the basis of cloud computing. The edge calculation is a novel calculation model for providing intelligent services at the edge of a network close to an object or a data source, edge nodes are widely distributed and are closer to a user side, and the edge nodes can be installed on an edge server, such as a vehicle and the like, so that the connection requirements of different users are met. By combining the MEC, the data task realizes multi-node cooperation between the IoT local node and the MEC node through data transmission, so that the IoT local user data can be unloaded to an adjacent MEC server, the problem that the computation processing capacity of an IoT local user side in an intelligent home scene is limited is solved, and the user computation task pressure is shared. Due to the limited computing capability of the MEC nodes, when facing a computing task with a large data volume, a cooperation mode among multiple MEC nodes is needed.

The IoT local client can offload the computing task to the edge node, and the offloading method can be divided into full offloading and partial offloading. The all-offload scheme is to offload all the computation tasks to a certain edge node for execution, and document 1(j.liu, y.mao, j.zhang, and k.b.letaief, "Delay-optimal computation task scheduling for mobile-edge computing systems," in proc.ieee int.symp.inf.therapy (ISIT), Barcelona, Spain, jul.2016, pp.1451-1455) uses a one-dimensional search algorithm that reduces the execution Delay to the maximum extent, and comprehensively considers the application buffer queuing state and the available processing capacity. However, for the all-off method, the computing task is completely off-loaded to the edge node for processing, and there may be a problem that the computing power of the edge node cannot be satisfied, and a large transmission delay is caused. For this purpose, a partial unloading scheme is proposed, wherein the partial unloading scheme refers to that a computing task is partially executed locally, and the rest is unloaded to an edge node for executionSpecific contents of Partial offload are described in document 2(Z. Ning, P. Dong, X. Kong and F. Xia, "A Cooperative Partial Computation operable Scheme for Mobile Edge Computing Enabled Internet of Things," in IEEE Internet of Things Journal, vol.6, No.3, pp.4804-4814, June 2019). In a partial offloading scheme, it is necessary to determine the distribution location of the data task, and for this document 3(l.yang, j.cao, h.cheng, and y.ji, "Multi-user computing sharing for related sensitive mobile applications," IEEE trans.computer, vol.64, No.8, pp.2253-2266, aug.2015), the concept of task partition is proposed, and the purpose of task partition is to determine which module to offload and how to execute, i.e., offload to the edge and cloud nodes locally or remotely. Document 4(y.zhao, s.zhou, t.zhao, and z.niu, "Energy-efficiency task oriented for multi-user mobile closed computing," in proc.ieee/CIC Int.conf.command.china (ICCC), Shenzhen, China, nov.2015, pp.1-5) converts part of the unloading problem into a nonlinear constraint problem, and solves it by a linear programming method to achieve the purpose of optimization. Due to the diversity of network data, different sizes of data are generated, for which resource limitation becomes a key problem in a task offload process, document 5 (Zhao 31441; Uyu. resource-limited Mobile edge computing systems [ D)]Beijing university of post and telecommunications, 2019), document 6(O.

The problem was intensively studied by A.Pascal-island, and J.Visal, "Optimization of radio and comparative resources for energy efficiency in latency-constrained application of flooding," IEEE Trans.Veh.Technola, vol.64, No.10, pp.4738-4755, Oct.2015), document 7(C.You and K.Huang, "Multi user resource allocation for Mobile compliance," in Proc.IEEE Global.Commin.Conf. (GLOBOM), Washingg, DC, USA, Dec.2016, pp.1-6). In the document 5, for the problem of resource limitation, analysis is performed from the perspective of network capacity and data allocation, and a suitable processing position is selected for data, so that smooth execution of more data tasks is ensured. Reference 6 adoptsAnd analyzing and processing by using a partial unloading method, segmenting the data task, and sequentially transmitting segmented data to the edge node and the cloud node for execution, thereby solving the problem of resource limitation. Document 7 proposes an optimal resource allocation strategy for handling the waiting and sequencing situation of tasks in the case that the number of nodes and the processing capacity are limited in the tdma system, so as to ensure the processing efficiency of resources.

When the computing task of the user is large, even if a partial unloading method is adopted, a single MEC node still cannot meet the processing requirement of unloading the task from the user side, so that a plurality of nodes need to be selected for cooperative processing. Therefore, a multi-node cooperation method is proposed in document 8(w.fan, y.liu, b.tang, f.wu and z.wang, "Computing off-Computing Based on operation of Mobile Edge Computing-Enabled Base states," in IEEE Access, vol.6, pp.22622-22633,2018), for the case that a user end has a large Computing task and one MEC node cannot bear all data tasks, an adjacent MEC node is selected to share the Computing pressure of a target node, and an algorithm for solving an optimization problem is designed by using an interior node method and a logarithmic barrier function to optimize the energy consumption problem of a multi-node cooperation system. The multi-node cooperation method is mainly used for solving the problem that the computing capacity of a single node is insufficient, and for this purpose, a dynamic self-configuration multi-device mobile cloud system is used in document 9(k.habak, m.amar, k.a.harras, and e.zegura, "Femto clusters: Leveraging mobile device to product cluster service at the edge," in proc.ieee 8th int.conf.cloud company., jun.2015, pp.9-16), and the peripheral idle mobile devices are used as edge servers to execute corresponding computing tasks, so that the cloud system range is expanded to solve the problem that the network load is greater than the computing capacity of the node. With the multi-node cooperation method, the computing task should be distributed to multiple nodes to execute the task, which relates to the problem of node distribution and deployment, and document 10(y.c. hu, m.patel, d.sabella, n.sphere, and v.young, "Mobile edge computing-a key technology towards 5G," eur.telecom.standards institute, Sophia Antipolis, France, ETSI White Paper 11,2015, pp.1-16) mainly introduces the deployment location of MEC nodes, which are widely distributed, such as lte macro base station, multi-Radio Access Technology (RAT) cell aggregation site, etc. With the increasing popularization of MEC technology, multi-node cooperation technology is also increasingly applied to actual work and life.

In the above documents, a user unloads all or part of a computing task to one or more MEC nodes, optimizes a network structure and increases the capacity of the network to process the task, and explores resource optimization under a multi-node combined network structure. However, as the MEC nodes are widely distributed around the local user node, in the wireless network, a plurality of MEC nodes are selected to participate in the calculation, the more nodes are involved in the network, the greater the overall energy consumption of the network is, and with the proposal of the green MEC concept, the network energy consumption becomes one of the key concerns of people.

In the multi-node task allocation model, in order to reduce network energy consumption, edge nodes for executing computing tasks need to be reasonably selected. Document 11(w.zhang, y.wen, k.guan, d.kilper, h.luo, and d.o.wu, "Energy-optimal mobile computing under stored wireless channels," IEEE trans.wireless communications, vol.12, No.9, pp.4569-4581, sep.2013) proposes a single-user edge computing offload (MECO) method that determines an appropriate offload policy by comprehensively considering variable CPU cycles and variable transmission rates in the network with network Energy consumption as an optimization target. In addition to the above document, document 12(x.chen, l.jiao, w.li, and x.fu, "Efficient multi-user computing and offloading for mobile-edge computing," IEEE trans.net, vol.24, No.5, pp.2795-2808, oct.2016) considers the energy consumption and time delay of the end user comprehensively, and implements the optimal allocation of resources in the process of computing offloading by using the game theory. In document 13(d.han, s.li, y.pen and z.chen, "Energy Sharing-Based Energy and User Joint Allocation Method in Heterogeneous networks," in IEEE Access, vol.8, pp.37077-37086,2020), in order to cope with Energy shortage in Heterogeneous networks, a shared link is established between a plurality of Base Stations (BSs), and is expanded to the macro and micro domains for analysis. Document 14(d.w.k.ng, e.s.lo and r.schober, "Energy-Efficient Resource Allocation in Multi-Cell OFDMA Systems with Limited Backhaul Capacity," in IEEE Transactions on Wireless Communications, vol.11, No.10, pp.3618-3631, October2012) aims at the offloading priority problem, determines an offloading priority function by comprehensively considering the quantization fairness, the transmission channel and the local computation condition, realizes the optimal network Resource Allocation by analyzing the offloading priority function, and takes the overall network Energy consumption as a measurement index.

In summary, although many studies have been made in terms of multi-node cooperation and data offloading in the above documents 1 to 14, a user transmits data to a multi-node execution, and generally adopts a step-by-step transmission method, when a data amount of a user side is large, step-by-step transmission generates large delay accumulation, thereby destroying a delay constraint condition of the user and causing large network energy consumption.

Object of the Invention

The invention aims to overcome the defects in the aspects of multi-node cooperation and data unloading in the prior art, comprehensively considers the conditions of the distance between an MEC node and a user end, the channel characteristic, the CPU energy consumption and the like, and provides a multi-node cooperation calculation unloading method based on energy consumption minimization.

Disclosure of Invention

The invention provides a multi-node cooperative computing unloading method based on energy consumption minimization, which comprises the following steps:

s1: constructing a system model, specifically a local-edge-cloud edge computing network, wherein the system is provided with K IoT local clients which are served by N wireless base stations, and each base station is provided with an edge server, namely N edge nodes; the energy consumption in the local-edge-cloud edge computing network is as follows: the total energy consumption is calculation energy consumption + transmission energy consumption, wherein the calculation energy consumption comprises calculation energy consumption of an IoT local user, edge node cooperation and a cloud server; the transmission energy consumption comprises wireless transmission energy consumption between an IoT user and the edge node and transmission energy consumption between an IoT local user and the cloud server; the network delay of the local-edge-cloud edge computing network comprises computing delay and transmission delay, and network energy consumption needs to be minimized on the premise of meeting the requirement of network delay;

s2: constructing an objective function of a multi-node cooperative computing unloading model to realize the minimum overall energy consumption of the network under the condition that the time delay meets the time constraint;

s3: the objective function described in step S2 is optimized based on the branch definition algorithm.

Further, in the step S1, in the constructed local-edge-cloud edge computing network model, the computing model of the network is defined as a_k(R_k，s_k),R_kDenotes the task value of the local user terminal K, K denotes the kth user terminal, where K ═ 1,2_kThe task execution time of the user k is represented, and the computational energy consumption of the user k can be represented as shown in formula (1):

wherein C is_kM represents the number of CPU revolutions required for 1-bit data to perform a calculation task_kRepresenting the energy consumed by the CPU per revolution;

when the computing task cannot be performed locally in the IoT, the computing task needs to be offloaded to a suitable node to perform the computing task, and the offloading may bring a certain transmission energy consumption, where the transmission energy consumption is related to the transmission time and the transmission power of the task, and the transmission power is represented by formula (2):

wherein t is_kRepresenting the transmission time, p, of a computational task for user k_kRepresents the transmission power between user k and the offload node; the overall energy consumption of the user k for executing the task is the sum of the transmission energy consumption and the calculation energy consumption, and is represented by formula (3):

further, let parameters

For the CPU revolution number required by the 1bit task of the user k when the local, edge and cloud nodes execute,

respectively representing the energy consumed by each turn of the CPU when the computing task of the user k is executed at the local, edge and cloud ends; setting a data unit

Expressing IoT local user data in the form of data units, and dividing the data of user k into M_kA data unit represented as

Setting a parameter rho for each node, wherein rho_k→0The number of data units calculated locally by user k is represented, and N is {1,2.. N }, ρ is N MEC nodes in the network_k→nRepresenting offload from IoT local user k to MEC_nNumber of unit data for executing task, ρ_k→N+1The data unit number unloaded from an IoT local user side to a cloud end node to execute tasks is shown, wherein N +1 represents the cloud end node; setting a parameter beta for selection between IoT local user side and MEC node_m,nThe computing task of the mth block is unloaded to the node n for execution;

the IoT local user side unloads the computing task to a plurality of MECs and cloud nodes in a blocking manner, and for one data unit, one data unit can only be unloaded to one node, which is represented as

While an edge node receives multiple data units, denoted as

When n is 0, it means to execute locally on the IoTWhen N is equal to N +1, the cloud node executes the calculation task;

data for the kth user is as shown in equation 4):

let F₀,F_n,F_N+1Respectively representing the computing capacities of local nodes, edge nodes and cloud nodes, namely the number of CPU revolutions required by executing a computing task; in the formula (4)

Which indicates the size of the task to be performed,

the number of CPU revolutions required for executing a 1-bit task of a user k is represented, the task is executed on an IoT local user side when N is 0, the task is executed on an MEC node when N is {1,2.. N }, and the task is executed on a cloud node when N is N + 1.

Further, the computation latency of the local-edge-cloud-edge computing network is determined by the computation amount of the nodes, the number of CPU revolutions and the node capacity, and when the computation task is executed on the IoT local client, the computation latency of the data of the kth client is represented by equation (5):

calculating the maximum value of the time delay for each node, wherein the calculated time delay of a single node is shown as a formula (6):

the computation delay executed in the cloud is shown in equation (7):

setting a data unit

Is expressed by t_kWherein t is_k>0, represented by formula (8):

wherein r is_kRepresenting the data transfer rate from an IoT user k to a selected node by determining the number of bit cells ρ that the user offloads to the node_k→nTo calculate the total transmission time in the unloading process;

the transmission time unloaded from the IoT local user side to the node is the transmission time corresponding to the node with the largest unloading task, and the node satisfies the formula (9):

wherein T represents the delay to meet the QoS requirement of the user;

in the transmission model from the IoT local user terminal k to the selected node, the transmission rate from the IoT user k to the node n is represented by equation (10):

where W is the channel bandwidth, p_k,nIs the transmission power, h, between user k and node n_kIs the channel characteristic between the two; PL_kThe value of (A) satisfies a large-scale fading characteristic, and is expressed as a transmission distance

Wherein d represents the transmission distance, d₀Denotes a reference distance, n denotes a path loss exponent, X_σRepresents a mean value of0, standard deviation of σ²(ii) a gaussian random variable; w_E,W_CDenotes the bandwidth between the IoT local user and the edge node and the user and the cloud node, respectively, when N ═ 1,2.. N ·_k,nAnd PL_k,nRepresenting users k and MEC_nWhen N is equal to N +1, the transmission power and the transmission loss between the IoT local ue k and the cloud node are represented;

the transmission power from the IoT local user terminal k to the node n is represented by equation (11):

further, in step S2, the energy consumption when the task is executed on the IoT local user terminal is represented by equation (12):

wherein, a selection parameter rho is set_k→nN ═ 1,2.. N }, which represents the selection condition of the edge nodes, the overall energy consumption of the edge nodes includes the calculation energy consumption and the transmission energy consumption, and the overall energy consumption is represented by formula (13):

when the computing task is partially offloaded to the cloud server, the overall energy consumption of the cloud node is represented by equation (14):

the overall energy consumption in the local-edge-cloud edge computing network is the sum of the energy consumption of the IoT local user side, the edge and the cloud, and is represented by formula (15):

E_total＝E_L+E_E+E_C (15)；

the network delay of the local-edge-cloud edge computing network includes a computation delay and a transmission delay, where the computation delay of the IoT local user of the kth user is represented by equation (16):

the calculation delay is the maximum value of the calculation delay of each MEC node and each cloud node, and is shown in formula (17):

the overall computation delay of the local-edge-cloud edge computing network is represented by formula (18):

the overall time delay of the local-edge-cloud edge computing network is the sum of the computing time delay and the transmission time delay, and is represented by formula (20):

D_k＝s_k+t_k (20)；

the constraint conditions of the multi-node cooperative computing unloading model are shown in the formulas (21, 21-1-21-7):

min E_total (21)

D_k≤T (21-1)

t_k,n>0 (21-2)

in the formula, the limiting conditions (21-1) and (21-2) are transmission time limiting conditions, wherein (21-1) indicates that the overall time delay of the network should be smaller than the time delay limit of the user side, (21-3) and (21-4) indicate the distribution situation after the user side performs data segmentation, wherein (21-3) indicates that one unit data can only be unloaded to one edge node, (21-4) indicates the number of unit tasks processed by a certain edge node, and (21-5) to (21-7) respectively indicate the calculation capacity limits of the MEC node, the IoT local node, the MECs node and the cloud node.

Still further, in step S3, the formula (21) is used as an objective function, and ρ in the formula is calculated_k→0,ρ_k→1,…ρ_k→N,ρ_k→N+1Viewed as an argument, the objective function is viewed as a linear programming problem represented by the argument, where the argument ρ satisfies the condition as shown in equation (22):

ρ_k→0+ρ_k→1+…+ρ_k→N+ρ_k→N+1＝M_k (22)

for the above objective function (21), it is expressed as a condition shown in equation (23):

E_total＝v₀ρ_k→0+v₁ρ_k→1+…v_Nρ_k→N+v_N+1ρ_k→N+1 (23)

wherein v is₀,v₁,…v_N,v_N+1Coefficient representing the argument ρ front, let f ═ v₀ v₁…v_N v_N+1]^TF represents a coefficient vector of an argument in the objective function;

according to equation (21-1), equation (21) is converted to that shown as equation (24):

wherein

Represents the coefficient before the argument ρ in the objective function formula (21-1). The formula (21) is converted into the formulas (25) to (27) according to the constraints (21-6) to (21-8) of the formula (21):

a₁₀ρ_k→0+a₁₁ρ_k→1+…a_1Nρ_k→N+a_1N+1ρ_k→N+1≤F_n (25)

a₂₀ρ_k→0+a₂₁ρ_k→1+…a_2Nρ_k→N+a_2N+1ρ_k→N+1≤F₀ (26)

a₃₀ρ_k→0+a₃₁ρ_k→1+…+a_3Nρ_k→N+a_3N+1ρ_k→N+1≤F_N+1 (27)

the above (22), (24), (25), (26) and (27) convert the constraint of the objective function (21) into a standard form with respect to the argument ρ, as shown in the formula (28)

Where a represents a constraint matrix formed by the constraint equation set, and b is [ T F ]_n F₀ F_N+1 M_k]^TAnd b represents the right vector of the system of constraint equations.

Drawings

FIG. 1 illustrates a Benzender-edge-cloud edge computing network model constructed in accordance with the present invention.

FIG. 2 is a graph comparing energy consumption for three unloading models.

FIG. 3 is a task allocation of each node of a multi-node collaborative computing offload model.

Fig. 4 is a comparison of energy consumption at different network bandwidths.

Fig. 5 is a comparison of the offloading scheme for the case of large task volumes.

Detailed Description

The following detailed description of the invention refers to the accompanying drawings.

First, a system model is constructed.

The present invention constructs a local-edge-cloud edge computing network, as shown in fig. 1, a system has K IoT local clients, which are served by N wireless base stations, and each base station is configured with an edge server, that is, N edge nodes. The computing task at the IoT local user side can be selected to be executed locally, or can be selected to be partially offloaded to an edge node or transmitted to a cloud server through a router of a base station to be executed. Before task unloading, the IoT local user side divides a computing task according to a certain rule, and the divided task selects a proper task unloading edge node or a cloud server according to parameters such as time delay, energy consumption and edge node computing capacity. For example, without loss of generality, we assume that the local user UE_kThe calculation task at a certain time can be divided into N task blocks, and the lower subscript k represents the kth local user end. Task block 1 may perform computational tasks on IoT local clients and task block 2 may choose to offload to node MEC₁Where the computing task is executed, subscript 1 denotes the 1 st edge node, and task blocks 3, 4, 5 may be optionally offloaded to the MEC₂The nodes execute the calculation tasks, the lower corner 2 represents the 2 nd edge node, and the task blocks 6 and 7 can be selectively unloaded to the node MEC₃And executing a calculation task, wherein a subscript 3 represents a 3 rd edge node, and the remaining task blocks are selectively transmitted to the cloud to execute calculation due to the fact that the calculation capacity of the edge node cannot meet the requirements of the remaining task blocks. By adopting the method of unloading a plurality of task blocks in parallel, the network delay can be effectively reduced, and the overall energy consumption of the network can be reduced.

In the constructed local-edge-cloudIn the edge computing network model, since IoT local user data is offloaded to multiple nodes in parallel for execution, to minimize network energy consumption, computation and transmission of the network should be considered comprehensively, that is, computation energy consumption for the IoT local user side to transmit computation tasks to the MEC node and the cloud end node in parallel, and transmission energy consumption for the IoT local user side to transmit the computation tasks to each node. Wherein the computational model of the network is defined as A_k(R_k，s_k),R_kDenotes the task value of the local user terminal K, K denotes the kth user terminal, where K ═ 1,2_kRepresenting the task execution time of user k. The calculated energy consumption of user k can be expressed as shown in equation (1):

wherein C is_kM represents the number of CPU revolutions required for 1-bit data to perform a calculation task_kRepresenting the energy consumed by the CPU per revolution.

When the computing task cannot be performed locally in the IoT, the computing task needs to be offloaded to a suitable node to perform the computing task, and the offloading may bring a certain transmission energy consumption, where the transmission energy consumption is related to the transmission time and the transmission power of the task, and the transmission power is represented by formula (2):

wherein t is_kRepresenting the transmission time, p, of a computational task for user k_kRepresenting the transmission power between user k and the offload node. The overall energy consumption of the user k for executing the task is the sum of the transmission energy consumption and the calculation energy consumption, and is represented by formula (3):

setting parameters

For the CPU revolution number required by the 1bit task of the user k when the local, edge and cloud nodes execute,

representing the energy consumed by the CPU per revolution when the computing task for user k is performed locally, marginally, and in the cloud. The invention sets a multi-node cooperation mode, so that data is segmented at an IoT local user side, and the segmented data is transmitted to the MEC or MCC node for calculation. In order to conveniently observe the unloading condition of the divided task, a data unit is arranged

Expressing IoT local user data in the form of data units, and dividing the data of user k into M_kA data unit represented as

Setting a parameter rho for each node, wherein rho_k→0The number of data units calculated locally by user k is represented, and N is {1,2.. N }, ρ is N MEC nodes in the network_k→nRepresenting offload from IoT local user k to MEC_nNumber of unit data for executing task, ρ_k→N+1The number of data units offloaded from the IoT local client to the cloud node to perform the task is shown, where N +1 represents the cloud node. Setting a parameter beta for the selection problem between the IoT local user terminal and the MEC node_m,nIndicating that the computing task of the mth block is unloaded to the node n for execution. In the model set up herein, since IoT local clients offload computing tasks to multiple MECs and cloud nodes in blocks, for one data unit, one data unit can only be offloaded to one node, denoted as

But at the same time one edge node may receive multiple data units, denoted as

When N is 0, it means that the computation task is executed locally on the IoT, and when N is N +1, it means that the computation task is executed on the cloud node.

Since data is split at the IoT local user side and transmitted to multiple node computers simultaneously, the data allocated to each node should meet the computing capacity of the node. Analyzing the data of the kth user, wherein the formula (4) is shown in the specification:

let F₀,F_n,F_N+1Respectively representing the computing power of local, edge nodes and cloud nodes, namely the number of CPU revolutions required for executing computing tasks. In the above formula

Which indicates the size of the task to be performed,

the number of CPU revolutions required for executing a 1-bit task of a user k is represented, the task is executed on an IoT local user side when N is 0, the task is executed on an MEC node when N is {1,2.. N }, and the task is executed on a cloud node when N is N + 1.

The computation delay is determined by the computation amount of the node, the number of CPU turns, and the node capacity, and when the computation task is executed at the IoT local client, the computation delay of the data at the kth client is expressed as formula (5):

since data is split at an IoT local user side and transmitted to multiple MEC nodes to be executed, the calculation delay is the maximum value of the calculation delay for each node, where the calculation delay of a single node is shown in formula (6):

for a computation task that cannot be executed at the IoT local client and cannot be executed at the MEC node, the computation task needs to be transmitted to the cloud server to be executed, and the computation delay executed at the cloud is represented by equation (7):

the transmission link in the network is divided into a wireless communication link between the edge server and the UE, a VLAN for transmission between the edge servers, and a transmission link between the edge server and the cloud server. In the network transmission process, the relationship between the computation capacity of the network and the network transmission capacity should be considered, and if the computation capacity is too large, channel resources in the network cannot be orderly allocated to IoT users, which may cause channel congestion and increase network delay. Let IoT local user terminal k need to process data size R_k(bit) wherein

Representing the size of the computing task performed locally on the IoT,

indicating the size of the task performed at the MEC node,

representing the size of the task performed at the cloud node. Tasks executed locally on the IoT do not need to be transmitted, there is no transmission energy consumption, and offloading of computing tasks to the MEC and MCC nodes for execution would result in transmission energy consumption.

Setting a data unit

Is expressed by t_kWherein t is_k>0, which can be expressed by the formula (8):

wherein r is_kRepresenting the data transfer rate from an IoT user k to a selected node, determining the number of bit cells ρ that the user offloads to the node_k→nTo calculate the total transfer time during the offloading process. Assuming that n edge servers receive data from the ue, the IoT local ue splits the data and transmits the data simultaneously. But p is the time of task transmission in each part_k→nOne computing task will experience ρ_k→nt_kBecause the basic parameters of different nodes are different and the data volume to be offloaded is also different, the transmission time to be offloaded from the IoT local ue to a node should be the transmission time corresponding to the node with the largest offloading task, and the node should satisfy the formula (9):

where T denotes the delay to meet the QoS requirements of the user.

In the transmission model from the IoT local user terminal k to the selected node, the transmission rate from the IoT user k to the node n is represented by equation (10):

where W is the channel bandwidth, p_k,nIs the transmission power, h, between user k and node n_kThe channel characteristics are different between the nodes due to the different distances between the nodes and the users. PL_kThe value of (A) satisfies a large-scale fading characteristic, and is expressed as a transmission distance

Wherein d represents the transmission distance, d₀Denotes a reference distance, n denotes a path loss exponent, X_σDenotes a mean value of 0 and a standard deviation of σ²Gaussian random variable of (2). W_E,W_CDenotes the bandwidth between the IoT local user and the edge node and the user and the cloud node, respectively, when N ═ 1,2.. N ·_k,nAnd PL_k,nRepresenting users k and MEC_nWhen N is equal to N +1, the transmission power and the transmission loss between the IoT local ue k and the cloud node are represented.

According to the formulas (1) and (3), the data transmission rate r_kThe transmission power from the IoT local user terminal k to the node n is expressed by equation (11):

in summary, through the analysis of the transmission and the computation conditions in the local-edge-cloud edge computing network, the energy consumption in the network is as follows: the total energy consumption is calculation energy consumption + transmission energy consumption, and the calculation energy consumption comprises calculation energy consumption of an IoT local user, edge node cooperation and a cloud server. The transmission energy consumption comprises wireless transmission energy consumption between the IoT user and the edge node, and transmission energy consumption between the IoT local user and the cloud server. Meanwhile, on the basis of considering network energy consumption, network delay is also comprehensively considered, wherein the network delay comprises calculation delay and transmission delay, and the network energy consumption is minimized on the premise of meeting the network delay requirement.

Next, the multi-node cooperative computing offloading method of the present invention is described

For the local-edge-cloud edge computing network model constructed by the invention, a part of computing tasks are executed at an IoT local user side, the rest computing tasks are unloaded to proper nodes, IoT user data is segmented according to a certain rule, and the IoT user data is unloaded to a plurality of nodes at the same time. Since the edge node is close to the IoT user and the data transmission time is short, but the computing power of the edge node is limited, the offloading position needs to be reasonably selected according to the user requirement.

The invention discloses a multi-node cooperative computing unloading method, which comprises the following steps:

step one, constructing an objective function.

According to the formula (3), the overall network energy consumption is composed of calculation and transmission energy consumption, and as the model is a local-edge-cloud edge calculation network model, each level can be assigned with a certain task to process, namely, the network energy consumption is composed of calculation and transmission energy consumption of an IoT local user side, an edge network and a cloud network.

When the task is executed on the IoT local user side, only the computing energy consumption is involved, and the energy consumption of the task when executed locally is represented by equation (12):

when the computing task is offloaded to the edge node for execution, since there are multiple edge nodes around the IoT local user, the edge node is selected, and the selection parameter ρ is set_k→nN ═ 1,2.. N }, which represents the selection condition of the edge nodes, the overall energy consumption of the edge nodes includes the calculation energy consumption and the transmission energy consumption, and the overall energy consumption is represented by formula (13):

when the computing task is partially offloaded to the cloud server, the overall energy consumption of the cloud node is represented by equation (14):

the overall energy consumption in the network is the sum of the energy consumption of the IoT local user side, the edge and the cloud, and is represented by formula (15):

E_total＝E_L+E_E+E_C (15)

for network latency, the latency includes computation latency and transmission latency because computation and transmission of IoT local user data are considered in the network model. As for the computation delay, due to the local-edge-cloud edge computation network model, computation delays of IoT local users, edges, and cloud ends need to be considered, where the computation delay of the IoT local user of the kth user is shown in equation (16):

different from the node-to-node computation offload model, a task is split at an IoT local user side and is transmitted to multiple nodes in parallel for simultaneous processing, so that the computation delay should be the maximum value of the computation delay of each MEC node and a cloud node, as shown in formula (17):

the calculated delay of the entire network is expressed by equation (18):

the network delay also comprises transmission delay, and the transmission delay mainly comprises a wireless transmission link from the IoT local user end to the edge node and a VLAN transmission network from the IoT local user end to the cloud end. Since data is split at the IoT local user side and the split data is transmitted by selecting an appropriate node, the overall transmission delay is shown in equation (19) because of the parallel transmission of data:

t_k＝max{ρ_k→nt_k,n,n＝0,1…N,N+1} (19)

the overall network delay is the sum of the calculated delay and the transmission delay, and is represented by formula (20):

D_k＝s_k+t_k (20)

the goal of constructing the multi-node cooperative computing unloading model is to realize optimization of network energy and realize minimum overall energy consumption of the network under the condition that time delay meets time constraint. The multi-node cooperative computing unloading optimization system is shown in the following formulas (21, 21-1-21-7):

min E_total (21)

D_k≤T (21-1)

t_k,n>0 (21-2)

in the formula, the limiting conditions (21-1) and (21-2) are transmission time limiting conditions, where (21-1) indicates that the overall time delay of the network should be smaller than the time delay limit of the user side, (21-3) (21-4) indicates the distribution situation after the user side performs data segmentation, where (21-3) indicates that one unit data can only be offloaded to one edge node, (21-4) indicates the number of unit tasks processed by one edge node, and (21-5) - (21-7) respectively indicate the computation capacity limits of the MEC node, the IoT local node, the MECs node, and the cloud node.

For the above problem, in the formula (21), the calculation task is offloaded to the parameter ρ for the size of each node_k→nBy analyzing ρ_k→nThe node task allocation condition under the condition of realizing optimal energy consumption is judged according to the value taking condition. Due to the parameter ρ_k→nThe number of data units is determined so that the value can only take integer values, and the optimization problem becomes an integer programming problem.

And step two, optimizing the objective function in the step one based on a branch definition algorithm.

A branch definition algorithm (BB) is adopted to decide the resource allocation scheme of each MEC node, and the basic idea of the algorithm is to search all feasible solution spaces of the optimization problem with constraint conditions. The algorithm, when executed in detail, partitions the overall feasible solution space into smaller and smaller subsets and computes a lower or upper bound for the values of the solution within each subset. The branch-and-bound method solves the general linear programming problem by adopting the simplex on the integer programming problem, divides the decision variable of the non-integer value into two nearest integers, divides the two integers into a column condition, adds the two integers into the original problem, and solves the updated constraint vector. From which an upper or lower bound of the value is sought.

Table 1 lists the basic flow of the branch definition algorithm (BB).

TABLE 1 basic flow of Branch-and-bound Algorithm

Solving the energy consumption optimization problem by using a branch definition algorithm, taking a formula (21) as an objective function, and taking rho in the formula_k→0,ρ_k→1,…ρ_k→N,ρ_k→N+1Viewed as an argument, the objective function can be viewed as a linear programming problem represented by the argument. Wherein the independent variable ρ satisfies a condition shown by the formula (22):

ρ_k→0+ρ_k→1+…+ρ_k→N+ρ_k→N+1＝M_k (22)

for the above objective function (21), it can be expressed as a condition as shown in equation (23):

E_total＝v₀ρ_k→0+v₁ρ_k→1+…v_Nρ_k→N+v_N+1ρ_k→N+1 (23)

wherein v is₀,v₁,…v_N,v_N+1Coefficient before the argument ρ is expressed by the equation f ═ v₀ v₁…v_N v_N+1]^TAnd f denotes a coefficient vector of an argument in the objective function. For the constraint (1) on latency in equation (21), it is converted as shown in equation (24):

wherein

A coefficient before an argument ρ in a constraint condition (1) in an objective function (21) is expressed. The constraints (6) to (8) of the equation (21) can be converted into equations (25) to (27):

a₁₀ρ_k→0+a₁₁ρ_k→1+…a_1Nρ_k→N+a_1N+1ρ_k→N+1≤F_n (25)

a₂₀ρ_k→0+a₂₁ρ_k→1+…a_2Nρ_k→N+a_2N+1ρ_k→N+1≤F₀ (26)

a₃₀ρ_k→0+a₃₁ρ_k→1+…+a_3Nρ_k→N+a_3N+1ρ_k→N+1≤F_N+1 (27)

the above (22), (24), (25), (26) and (27) convert the constraint of the objective function (21) into a standard form with respect to the argument ρ, as shown in the formula (28)

Where a represents a constraint matrix formed by the constraint equation set, and b is [ T F ]_n F₀ F_N+1 M_k]^TAnd b represents the right vector of the system of constraint equations. The range of the argument ρ can be expressed by the constraints (6) to (8) of the objective function, and is expressed by lb, ub in the algorithm.

Through simulation, the multi-node computing unloading method is compared with the traditional unloading method, the traditional unloading method comprises a single cloud unloading method and a node mutual transmission computing unloading method, and the network energy consumption of the three unloading methods under the condition of processing different data volumes is respectively judged by taking the overall network energy consumption as a measurement standard. For the unloading model of multi-node cooperation, 3 MEC nodes are set, and rho is set₀,ρ₁,ρ₂,ρ₃,ρ₄The number of the user data units allocated to the IoT local user terminal, the three MEC nodes, and the cloud node is respectively represented. The three MEC nodes correspond to different CPU parameters and have distances d from IoT local user ends respectively₁,d₂,d₃Analyzing the network energy consumption problem of the data task on one IoT local user terminal.

Changing the network bandwidth from the user side to the MEC node, wherein the network bandwidth influences the transmission rate of the data according to a formula

L_E,iRepresents the transmission task amount, r, of the ith MEC node_E,iIndicating the data transmission rate of the ith MEC node. Three different situations are set for the network bandwidth, and the corresponding transmission rates are shown in table 2:

table 2 transmission rate table for three situations

The basic parameters used for the simulation are shown in table 3 below:

table 3 sets forth simulation parameters

The following are relevant comparison results:

energy consumption comparison of three calculation unloading models

Selecting the calculated data quantity M_k(1000,2500), analyzing the network energy consumption of the three computation offloading models, as shown in fig. 2, the network energy consumption of the multi-node collaborative computation offloading model is smaller than that of the single-cloud computation offloading model and the node mutual transmission computation offloading model. When the data volume unloaded by the IoT local user side is less than 1500kbit, the network energy consumption of the data parallel transmission multi-node cooperative computing unloading model and the node mutual transmission computing unloading model is basically the same, and when the unloading task volume is greater than the value, the network energy consumption of the multi-node cooperative computing unloading model is less than the energy consumption of the node mutual transmission model. The analysis can be carried out, when the unloading data volume is small, the new unloading model and the traditional node mutual transmission calculation unloading model have similar network optimization effect, but the calculation tasks of the multi-node cooperation calculation unloading model are transmitted in parallel, so that the method has certain advantages in the aspect of network time delay. When the unloading data volume of the network is large, the multi-node cooperative computing unloading model has great advantages in terms of network energy consumption and network time delay.

The task allocation of each node of the multi-node cooperative computing unloading model is shown in fig. 3, and the resource allocation condition of the node with the unloading data volume in the range of 1000-2500 kbit is shown. Analysis can be carried out, as the amount of calculation unloading data increases, the number of nodes for resource allocation gradually increases, and when the amount of IoT local user data is not large, data is only processed between the local server and the MEC server. And the task is executed without being unloaded to the cloud node. The larger the amount of tasks to be processed, the greater the requirement on the number of nodes and the greater the number of nodes at the same time. The larger the number of IoT local user tasks is, the greater the advantage is, and the network overall optimization is facilitated.

When the network bandwidth is changed, the influence of the network information transmission rate on the network energy consumption is considered. For example, as shown in fig. 4, the energy consumption ratio under different network bandwidths is that the network energy consumption in case 3 is the smallest, the network transmission rate in case 3 is the largest, and the transmission rate in case 1 is the smallest, and the larger the network transmission rate is, the smaller the overall energy consumption of the network is. In the multi-node cooperative computing offloading model, all computing tasks are transmitted simultaneously, and the larger the transmission rate of the network is, the larger the data volume that can be transmitted simultaneously is, and the larger the total offloading amount that the IoT local user side can perform offloading operations is. The simulation result shows that the total amount of data processed in case 1 is 5000kbit, the total amount of data processed in case 2 is 7500kbit, and the total amount of data processed in case 3 is 17500 kbit. In summary, when the network transmission rate is higher, the overall data amount that the network can process is increased, and meanwhile, under the condition of calculating the same data amount, the network energy consumption is reduced.

When the data volume processed by the IoT local user side is very large, the overall data transmission efficiency of the network needs to be improved, and when the data volume of the processing task in the network calculation is 10000-50000 kbit, the network data transmission rate is increased to 2Gbit/s, and the task allocation situation of the node is as shown in fig. 5.

When the data volume of the IoT local user is large, the data transmission rate of the network is increased, and it can be seen from analyzing fig. 5 that the data volume of the IoT local user is increased and then is unloaded to ρ ″₃The data volume of the node is obviously increased due to rho₃The CPU power consumption of the node is minimal. When the computing task is large, the energy consumption of the CPU becomes a main influence factor of the network energy consumption, and meanwhile, the total amount of data unloaded to the cloud nodes is continuously increased along with the increase of the unloaded data volume, so that the superiority of the edge cloud cooperation mechanism under the condition of large data volume is reflected.

The invention has the following advantages:

1. the data tasks are transmitted to the edge by the IoT local user side and the cloud server executes the computing tasks, and the computing energy consumption formed by distributing the computing tasks to each node by the IoT local user side and executing the computing tasks and the transmission energy consumption unloaded from the IoT local user side to each node and the like are comprehensively considered. In contrast to past studies, network computations and transmissions are considered herein comprehensively. The energy consumed by the computation of the local user side of the internet of things, the MEC node and the cloud is covered in the former, the energy consumed by the computation of the local user side of the cloud mainly comprises transmission energy consumption between the IoT user side and the MEC node and between the IoT user side and the cloud node, the overall energy consumption of the network is taken as an optimization target, and the network delay is comprehensively considered.

2. By means of a multi-MEC node cooperation method and in combination with a partial unloading theory, computing tasks of an IoT local user side are transmitted to a plurality of nodes in parallel to be executed, and energy loss caused by data transmission among the nodes is solved through task parallel transmission. Meanwhile, in the aspect of time delay, due to the fact that computing tasks of the user side are unloaded in parallel, the distribution position of data is determined by comprehensively considering network bandwidth and node parameters, due to the fact that data are transmitted in parallel, time delay of each part cannot be accumulated directly, and time for each node to receive and process the data is analyzed to determine the total time delay of the network. And forming an integer linear programming problem based on a network energy consumption optimization target, and analyzing the task unloading condition of a single user by using a branch-and-bound algorithm to minimize the overall network energy consumption.

3. The simulation result shows that with the increase of the unloading data volume of the IoT user side, the demand of the quantity of the MEC nodes is larger, and meanwhile, compared with the traditional calculation unloading mode, the multi-node cooperation model has obvious advantages in the aspects of energy consumption and time delay, and the advantages are more obvious with the increase of the unloading data volume. For the parameter setting of the nodes, the overall energy consumption of the network is determined by parameters such as network bandwidth and CPU energy consumption, and the optimal network energy consumption is realized by selecting the unloading nodes of the IoT client data.

To sum up, in order to realize green communication in an intelligent home scene, the invention firstly creates a multi-node cooperative computing unloading model. In the model, an IoT local user side divides a computing task according to a certain rule, reasonably distributes the divided data, and transmits the data to a plurality of nodes in parallel to execute operation. On the basis, two traditional unloading methods, namely a single-cloud computing unloading model and a node mutual-transmission computing unloading model, are comprehensively analyzed, the overall energy consumption of a network is taken as an optimization target, time delay is taken as an optimization condition, the resource distribution condition between the MEC nodes is determined through a branch-and-bound algorithm, and the resource distribution condition is compared with the two traditional models. When the unloading task amount is large, the network energy consumption of the data segmentation transmission multi-node cooperative computing unloading model is small, the time delay characteristic is good, meanwhile, the analysis of parameters among the MEC nodes is carried out, and the influence of CPU parameters of the nodes on the network energy consumption is large. For the condition of large data volume, the multi-MEC node and the edge-cloud cooperation model have better network characteristics, and the network bandwidth and the information transmission rate have certain influence on the data unloading condition of the network. In the multi-node cooperative computing unloading model, the divided data tasks are transmitted in parallel, and when the data transmission rate is higher, the amount of the computing tasks which can be processed is larger, and the overall energy consumption of the network is smaller.

It will be understood by those skilled in the art that the foregoing detailed description is merely exemplary of the spirit and concepts of the invention, and should not be construed as limiting the scope of the invention, which is defined by the appended claims as broadly as possible, since various modifications and substitutions can be made therein without departing from the spirit and concepts of the invention.

Claims (6)

1. A multi-node cooperative computing unloading method based on energy consumption minimization is characterized by comprising the following steps:

s1: constructing a system model, specifically a local-edge-cloud edge computing network, wherein the system is provided with K IoT local clients which are served by N wireless base stations, and each base station is provided with an edge server, namely N edge nodes; the energy consumption in the local-edge-cloud edge computing network is as follows: the total energy consumption is calculation energy consumption + transmission energy consumption, wherein the calculation energy consumption comprises calculation energy consumption of an IoT local user, edge node cooperation and a cloud server; the transmission energy consumption comprises wireless transmission energy consumption between an IoT user and the edge node and transmission energy consumption between an IoT local user and the cloud server; the network delay of the local-edge-cloud edge computing network comprises computing delay and transmission delay, and network energy consumption needs to be minimized on the premise of meeting the requirement of network delay;

s2: constructing an objective function of a multi-node cooperative computing unloading model to realize the minimum overall energy consumption of the network under the condition that the time delay meets the time constraint;

s3: the objective function described in step S2 is optimized based on the branch definition algorithm.

2. The multi-node cooperative computing offloading method of claim 1, wherein in the step S1, in the constructed local-edge-cloud edge computing network model, the computing model of the network is defined as a_k(R_k，s_k),R_kDenotes the task value of the local user terminal K, K denotes the kth user terminal, where K ═ 1,2_kThe task execution time of the user k is represented, and the computational energy consumption of the user k can be represented as shown in formula (1):

wherein C is_kM represents the number of CPU revolutions required for 1-bit data to perform a calculation task_kRepresenting the energy consumed by the CPU per revolution;

when the computing task cannot be performed locally in the IoT, the computing task needs to be offloaded to a suitable node to perform the computing task, and the offloading may bring a certain transmission energy consumption, where the transmission energy consumption is related to the transmission time and the transmission power of the task, and the transmission power is represented by formula (2):

wherein t is_kRepresenting the transmission time, p, of a computational task for user k_kRepresents the transmission power between user k and the offload node; the overall energy consumption of the user k for executing the task is the sum of the transmission energy consumption and the calculation energy consumption, and is expressed asFormula (3):

3. the multi-node cooperative computing offload method of claim 2,

setting parameters

For the CPU revolution number required by the 1bit task of the user k when the local, edge and cloud nodes execute,

respectively representing the energy consumed by each turn of the CPU when the computing task of the user k is executed at the local, edge and cloud ends; setting a data unit

Expressing IoT local user data in the form of data units, and dividing the data of user k into M_kA data unit represented as

Setting a parameter rho for each node, wherein rho_k→0The number of data units calculated locally by user k is represented, and N is {1,2.. N }, ρ is N MEC nodes in the network_k→nRepresenting offload from IoT local user k to MEC_nNumber of unit data for executing task, ρ_k→N+1The data unit number unloaded from an IoT local user side to a cloud end node to execute tasks is shown, wherein N +1 represents the cloud end node; setting a parameter beta for selection between IoT local user side and MEC node_m,nThe computing task of the mth block is unloaded to the node n for execution;

the IoT local user side unloads the computing task to a plurality of MECs and cloud end nodes in a blocking way, and for one data unit, one data unitA unit can only be offloaded to one node, denoted as

While an edge node receives multiple data units, denoted as

When N is 0, the computing task is executed locally on the IoT, and when N is N +1, the computing task is executed on the cloud end node;

data for the kth user is as shown in equation 4):

let F₀,F_n,F_N+1Respectively representing the computing capacities of local nodes, edge nodes and cloud nodes, namely the number of CPU revolutions required by executing a computing task; in the formula (4)

Which indicates the size of the task to be performed,

the number of CPU revolutions required for executing a 1-bit task of a user k is represented, the task is executed on an IoT local user side when N is 0, the task is executed on an MEC node when N is {1,2.. N }, and the task is executed on a cloud node when N is N + 1.

4. The multi-node cooperative computing offloading method of claim 3, wherein the computation latency of the local-edge-cloud-edge computing network is determined by the computation amount of the node, the number of CPU revolutions and the node capacity, and when the computation task is executed on the IoT local client, the computation latency of the data of the kth client is expressed as formula (5):

calculating the maximum value of the time delay for each node, wherein the calculated time delay of a single node is shown as a formula (6):

the computation delay executed in the cloud is shown in equation (7):

setting a data unit

Is expressed by t_kWherein t is_k>0, represented by formula (8):

wherein r is_kRepresenting the data transfer rate from an IoT user k to a selected node by determining the number of bit cells ρ that the user offloads to the node_k→nTo calculate the total transmission time in the unloading process;

the transmission time unloaded from the IoT local user side to the node is the transmission time corresponding to the node with the largest unloading task, and the node satisfies the formula (9):

wherein T represents the delay to meet the QoS requirement of the user;

in the transmission model from the IoT local user terminal k to the selected node, the transmission rate from the IoT user k to the node n is represented by equation (10):

where W is the channel bandwidth, p_k,nIs the transmission power, h, between user k and node n_kIs the channel characteristic between the two; PL_kThe value of (A) satisfies a large-scale fading characteristic, and is expressed as a transmission distance

Wherein d represents the transmission distance, d₀Denotes a reference distance, n denotes a path loss exponent, X_σDenotes a mean value of 0 and a standard deviation of σ²(ii) a gaussian random variable; w_E,W_CDenotes the bandwidth between the IoT local user and the edge node and the user and the cloud node, respectively, when N ═ 1,2.. N ·_k,nAnd PL_k,nRepresenting users k and MEC_nWhen N is equal to N +1, the transmission power and the transmission loss between the IoT local ue k and the cloud node are represented;

the transmission power from the IoT local user terminal k to the node n is represented by equation (11):

5. the multi-node cooperative computing offloading method of claim 4, wherein in step S2, the energy consumption when the task is executed at the IoT local user side is represented by equation (12):

wherein, a selection parameter rho is set_k→nN ═ 1,2.. N }, which represents the selection of edge nodes, and the overall energy consumption of the edge nodes includes the calculation energyAnd the total energy consumption of the energy consumption and the transmission energy consumption is expressed by the formula (13):

when the computing task is partially offloaded to the cloud server, the overall energy consumption of the cloud node is represented by equation (14):

the overall energy consumption in the local-edge-cloud edge computing network is the sum of the energy consumption of the IoT local user side, the edge and the cloud, and is represented by formula (15):

E_total＝E_L+E_E+E_C (15)；

the network delay of the local-edge-cloud edge computing network includes a computation delay and a transmission delay, where the computation delay of the IoT local user of the kth user is represented by equation (16):

the calculation delay is the maximum value of the calculation delay of each MEC node and each cloud node, and is shown in formula (17):

the overall computation delay of the local-edge-cloud edge computing network is represented by formula (18):

the overall time delay of the local-edge-cloud edge computing network is the sum of the computing time delay and the transmission time delay, and is represented by formula (20):

D_k＝s_k+t_k (20)；

the constraint conditions of the multi-node cooperative computing unloading model are shown in the formulas (21, 21-1-21-7):

minE_total (21)

D_k≤T (21-1)

t_k,n>0 (21-2)

in the formula, the limiting conditions (21-1) and (21-2) are transmission time limiting conditions, wherein (21-1) indicates that the overall time delay of the network should be smaller than the time delay limit of the user side, (21-3) and (21-4) indicate the distribution situation after the user side performs data segmentation, wherein (21-3) indicates that one unit data can only be unloaded to one edge node, (21-4) indicates the number of unit tasks processed by a certain edge node, and (21-5) to (21-7) respectively indicate the calculation capacity limits of the MEC node, the IoT local node, the MECs node and the cloud node.

6. The multi-node cooperative computing offloading method of claim 5, wherein in step S3, taking equation (21) as an objective function, p in the equation is expressed by_k→0,ρ_k→1,…ρ_k→N,ρ_k→N+1Viewed as an argument, the objective function is viewed as a linear programming problem represented by the argument, where the argument ρ satisfies the condition as shown in equation (22):

ρ_k→0+ρ_k→1+…+ρ_k→N+ρ_k→N+1＝M_k (22)

for the above objective function (21), it is expressed as a condition shown in equation (23):

E_total＝v₀ρ_k→0+v₁ρ_k→1+…v_Nρ_k→N+v_N+1ρ_k→N+1 (23)

wherein v is₀,v₁,…v_N,v_N+1Coefficient representing the argument ρ front, let f ═ v₀ v₁…v_N v_N+1]^TF represents a coefficient vector of an argument in the objective function;

according to equation (21-1), equation (21) is converted to that shown as equation (24):

wherein

A coefficient representing the argument ρ before in the objective function formula (21-1); the formula (21) is converted into the formulas (25) to (27) according to the constraints (21-6) to (21-8) of the formula (21):

a₁₀ρ_k→0+a₁₁ρ_k→1+…a_1Nρ_k→N+a_1N+1ρ_k→N+1≤F_n (25)

a₂₀ρ_k→0+a₂₁ρ_k→1+…a_2Nρ_k→N+a_2N+1ρ_k→N+1≤F₀ (26)

a₃₀ρ_k→0+a₃₁ρ_k→1+…+a_3Nρ_k→N+a_3N+1ρ_k→N+1≤F_N+1 (27)

the above (22), (24), (25), (26) and (27) convert the constraint of the objective function (21) into a standard form with respect to the argument ρ, as shown in the formula (28)

Where a represents a constraint matrix formed by the constraint equation set, and b is [ T F ]_n F₀ F_N+1 M_k]^TAnd b represents the right vector of the system of constraint equations.

Images