CN110099384A - Resource regulating method is unloaded based on side-end collaboration more MEC tasks of multi-user - Google Patents

Abstract

Resource regulating method is unloaded based on side-end collaboration more MEC tasks of multi-user the invention discloses a kind of, include the following steps: step 1, task analysis: analysis task completes time delay, determines that task completes benefit；Step 2, problem is formed: completing total benefit as target formation optimization problem to maximize task；Step 3, guarantee stable state: guaranteeing the stability of task backlog queue to simplify problem；Step 4, channel distribution: Given task unloading allocation strategy determines optimal channel distribution；Step 5, task schedule: given channel resource allocation policy determines optimal task schedule；Step 6, combined optimization: joint step 4 and 5 obtains optimal channel distribution and task schedule.The present invention fully takes into account business diversity, carries out priority division to task, to improve user side income to greatest extent, realizes that the more MEC tasks of multi-user unload scheduling of resource.

Description

Resource regulating method is unloaded based on side-end collaboration more MEC tasks of multi-user

Technical field

Resource regulating method is unloaded based on side-end collaboration more MEC tasks of multi-user the present invention relates to a kind of, belongs to network Resource allocation techniques field.

Background technique

Mobile edge calculations (Mobile Edge Computing, MEC) technology belongs to a kind of distributed computing, by data Processing, application program the operation even realization of some function services be put on the node of network edge.Mobile edge cloud by One or more Edge Server compositions are it equipped with the server for calculating store function that is, on traditional base station, will be traditional Base station is updated to mobile edge calculations base station.Since the base station MEC is close to user terminal, processing task on mobile terminal can be helped, Reduction task processing delay, and reduce the energy consumption of mobile terminal.From different, the side cloud computing (Cloud Computing) Edge calculating is transferred to the functions such as data processing on network edge node by network center, handles data nearby, without inciting somebody to action Mass data uploads to the hard core control platform of distal end, it is possible to reduce the time in the round-trip cloud of data and network bandwidth cost.

Complete edge calculations uninstall process is divided into following three parts: 1) mobile terminal sends task unloading request to MEC Service area (some necessary informations including task computation amount)；2) task is offloaded to MEC server process；3) MEC server is sent out Send unloading response (including task processing result etc.) to mobile terminal.Time is divided into frame, wherein frame be divided into control subframe and Calculate unloading subframe.In control subframe, exchange of control information is between MEC server and mobile terminal to determine discharge time Table.In calculating unloading subframe, workload is sent to MEC server first, and MEC server will after completing processing workload As a result mobile terminal is returned to.For the edge cloud being made of multiple MEC servers, if unloading task unloads task It is loaded onto which MEC server is handled, how to ensure that all mobile terminals and edge Cloud Server can distribute and efficiently The adaptation energy and the variation of user demand be still outstanding question.

Comprehensively consider the task amount of user, the computing capability of MEC server, the computing capability of mobile terminal and channel resource Occupancy situation completes maximizing the benefits as target using task, establishes the more MEC task unloading resource dispatching models of multi-user, for There is very big Practical significance based on the more MEC task unloading scheduling of resource of side-end collaboration multi-user.

Summary of the invention

Resource regulating method is unloaded based on side-end collaboration more MEC tasks of multi-user the object of the present invention is to provide a kind of, To realize the more MEC task unloading scheduling of resource of optimal multi-user.

To achieve the above object, the technical solution adopted by the present invention are as follows:

It is a kind of to unload resource regulating method based on side-end collaboration more MEC tasks of multi-user, include the following steps:

Step 1, task analysis: analysis task completes time delay, determines that task completes benefit；

Step 2, problem is formed: completing total benefit as target formation optimization problem to maximize task；

Step 3, guarantee stable state: guaranteeing the stability of task backlog queue to simplify problem；

Step 4, channel distribution: Given task unloading allocation strategy determines optimal channel distribution；

Step 5, task schedule: given channel resource allocation policy determines optimal task schedule；

Step 6, combined optimization: joint step 4 and 5 obtains optimal channel distribution and task schedule.

In the step 1, benefit is determined by two aspect factors: the expected revenus value that task self attributes generate, at task The time delay loss generated during reason；The specific steps of the step 1 are as follows:

Step 11, consider within K moment, all users there are I task needs to locally execute or be offloaded in moment t MEC processing, wherein the set J={ 1,2,3 ..., j..., J } of MEC, the collection of task is combined intoIndicate i-th of task, i=1,2,3 ..., I；Using task model To describe task m_iSize: D_i(t) (bit), i.e. task m_iData package size；Task m_iThe workload that need to be handled is D_i(t)X_i (CPU cycles), wherein X_iCpu cycle needed for indicating processing 1bit data volume；Spectral bandwidth that MEC j possesses is set as B_j (t), there is N_j(t) a subcarrier, each subcarrier bandwidth areAccording to shannon formula, task m on subcarrier n_iData pass Defeated rate is

Task m_iOverall data transmission rate be

Wherein, N_j(t) sub-carrier number, π are indicated_{I, n}(t) it is channel allocation indicator, works as π_{I, n}(t)=1 when, indicate that son carries Wave n distributes to task m_iIt is unloaded；Work as π_{I, n}(t)=0 when, indicate that subcarrier n is not yet assigned to task m_iIt is unloaded；p_iIt indicates Task m_iThe transmission power of place terminal；h_{N, j}Indicate the channel gain of user's subcarrier n, user during setting task unloading Mobility is not high, so h_{N, j}=127+logd_{I, j}, d_{I, j}Expression task m_iPlace user terminal is at a distance from MEC j；σ²It is channel Noise power；Task m_iThe transmission delay for being offloaded to MEC j is

Step 12, when task is stayed in processing locality by user, time delay only includes the task processing time；Consider that mobile subscriber is whole Hold u_iCPU processing capacity beThen locally executing time delay is

Step 13, task is offloaded to MEC server execution time delay to be made of following three parts: when a, task discharge conveyor Between, b, when have magnanimity task need to be offloaded to edge cloud processing when, more than MEC server load, task may be needed in each MEC Server queues wait, i.e. queue waiting time, and c, task handle the time；

Wherein, the discharge conveyor time isQueue waiting time isAt the CPU for setting MEC j Reason ability isTask m_iThe processing time beTherefore, task m_iIt is offloaded to execution at MEC server j Overall delay is

Step 14, for any one MEC server, the arrival process of task is modeled as Bernoulli process, if The task arrival rate for determining MEC server j is λ_j；The task quantity waited in the queue is assumed to be quene state: Q_j(t)=0,1, 2,3 ... }, the queue Q of MEC j_j(t) more new formula is

Q_j(t+1)=Q_j(t)-V_j(t)+A_j(t)

Wherein, V_j(t) processing speed for indicating task at MEC j, i.e., handle within the time that moment t/length is 1 and complete V_j(t) a task；A_j(t) indicate whether reach in moment t task, A_j(t) { 0,1 } ∈；Therefore, there is Pr { A_j(t)=1 }=λ_j And Pr { A_j(t)=0 }=1- λ_j；Based on Little's Law, consider in K moment, including waits in line time delay and processing delay exists Interior execution delay is directly proportional to the average queue length of task buffer area, and average queue length such as following formula indicates:

Step 15, if u_iExpression task m_iIn the expected revenus formulated according to its priority, L (T_i) indicate task m_iWhen Between T_iIt is interior to complete paid time delay loss；

Wherein, C is proportionality coefficient, determines that C is bigger to the susceptibility of time delay according to system, is indicated as caused by time delay Time delay loss is bigger；ρ_iFor the tolerance to time delay is lost, when time delay is less than tolerance, time delay will not make user satisfaction At influence, i.e., the income of user will not be lost, when time delay is greater than ρ_i, time delay causes influence to user satisfaction, corresponding to produce Time delay loss is given birth to.

In the step 2, user's benefit value is introduced as the index for measuring system performance, to maximize in a period of time It is that target establishes optimization problem that user side task, which completes total benefit,；Specific steps are as follows:

Step 21, task m_iBeing offloaded to the income that MEC processing generates is Wherein, u_iExpression task m_iIn the expected revenus formulated according to its priority, L (T_{I, j} It (t)) is task m_iIt is offloaded to the time delay loss that MEC processing generates；

Step 22, task m_iIt is performed locally the income of generation

Step 23, pass through combined optimization subcarrier allocator π_{I, n}(t), task allocator s_{I, j}(t), it obtains to maximize one User side task in the section time completes the optimization problem that total revenue is target:

P1:

S.t.C1:

C2:

C3:

C4:

C5:

C6:

C7:

Wherein, C1 ensures that a task can only select in processing locality or be offloaded to a MEC server execution；C2 ensures s_{I, j}It (t) is binary variable；C3 ensures π_{I, n}It (t) is binary variable；C4, which ensures a subcarrier at most, can only distribute to a use Family；C5 ensures that base station is the maximum transmission power that the transmission power of user's distribution is no more than base station, p^maxIt is the emission maximum of base station Power；C6 ensures that discharge conveyor energy is no more than task m_iPlace mobile terminal device dump energyC7 ensures that task is held Row time delay meets maximum delay requirementDue to the expected utility u of each task in the objective function of optimization problem P1_iIt is solid It is fixed, not at any time t and change, time delay loss function L () is linear function, therefore the optimization problem P2 being simplified:

P2:

S.t.C1:

C2:

C3:

C4:

C5:

C6:

C7:

In the step 3, guarantee the stability of each MEC server task backlog queue, being based on Lyapunov's theory will Problem reduction is the OPTIMAL TASK unloading resource dispatching strategy solved under limit, specific steps are as follows:

Step 31, if it is Bernoulli process that the task of each queue, which reaches state, Θ (t)=(Q is enabled₁(t), Q₂(t) ..., Q_j(t) ..., Q_J(t)) quene state is indicated, Θ (t) is according to task arrival rate λ_jDevelop on time slot t ∈ { 0,1,2 ... }； Define secondary liapunov function:

ω_jIndicate that weight set, different weights will lead to different queue status locating in task scheduling strategy not Together, if all ω_jIt is all 1；Obviously, which is non-negative, and if only if all Θ_j(t) be 0 when, L (Θ (t)) it is equal to 0；

Step 32, in order to predict the variation of each quene state, the secondary liapunov function between a moment is defined Difference mean value be Liapunov drift function Δ (Θ (t)):

Wherein,Indicate the mean value of the difference of secondary liapunov function；

This drift is the performance of expected change that liapunov function is engraved at one；

Step 33, each moment t observes current Θ (t) value and takes control action, greedy according to consistent Θ (t) Minimum formula drift plus penalty expectation:

Step 34, delay sensitive parameter v is determined₀If v₀=1, optimization problem P2 abbreviation are as follows:

P3:

S.t.C1:

C2:

C3:

C4:

C5:

C6:

In the step 4, if converting channel resource for optimization problem P3 in the case that task unloading allocation strategy is given Assignment problem, and optimal channel distribution is solved using KKT condition；Specific steps are as follows:

Step 41, if Given task unloads allocation strategy S '_{I, j}(t), optimization problem P3 is one about R_{I, j}(t) convex is asked Topic, it is assumed that there is l task to be offloaded to MEC processing, i.e. S_{I, j}(t)=1 number is l, and optimization object function is converted into such as following formula institute Show:

f(R_ij(t), S '_ijIt (t)) is about R_ij(t) function；

Step 42, due to f (R_{I, j}(t), S '_{I, j}It (t)) is convex function, and institute's Prescribed Properties are linear function, so Optimization problem is convex optimization problem, according to KKT condition, be can get about R_{I, j}(t) optimal solution

Step 43, the Lagrangian of constitution optimization problem, as follows:

Wherein, μ_{I, j}It is the undetermined coefficient of each constraint condition；

If R_{I, j}(t) and μ_{I, j}All meet KKT condition in arbitrary point, obtain:

By solving above formula, optimal R is obtained_{I, j}(t):

It can thus be concluded that pinned task unloads allocation strategy S '_{I, j}(t) optimal solution:

In the step 5, if given channel resource allocation policy, optimization problem P3 is converted into Zero-one integer programming problem；Tool Body step are as follows:

Step 51, if given channel resource allocation policy, optimization problem P3 is converted into Zero-one integer programming problem, following institute Show:

P4:

S.t.C1:

C2:

Step 52, each moment t is gone out on missions allocation strategy S using minimizing all tasks processing overall delays as object solving (t), that is, the corresponding best MEC server of each task is solved, and obtains each task and is offloaded to best MEC j^*Server The time delay of processing

Step 53, calculating task stays in the time delay T of processing locality_i(t), the time delay of MEC processing will be offloaded toWith (T_i(t)+δ) compare, δ is time delay tolerance, ifTask is then in MEC j^*Processing, otherwise in local Processing updates Task Assigned Policy S (t).

The specific steps of the step 6 are as follows:

Step 61, the preferred channels resource allocation under pinned task unloading distribution is obtained according to step 4.

Step 62, the OPTIMAL TASK unloading allocation strategy under fixed channel is obtained according to step 5.

Step 63, step 61 and 62 is repeated until obtaining optimal channel distribution and task scheduling strategy.

The utility model has the advantages that the purpose of the present invention is comprehensively consider the task amount of user, the computing capability of MEC server, movement The computing capability and channel resource occupancy situation of terminal complete maximizing the benefits as target using task, carry out computing resource and letter The distribution of road resource establishes the calculating unloading frame of the more MEC of multi-user, guarantees that MEC service is had a high regard for using Lyapunov's theory Business squeezes the stability of queue and KKT condition is solved, and realizes the more MEC task unloading scheduling of resource of optimal multi-user.This Invention fully takes into account business diversity, carries out priority division to task, to improve user side income to greatest extent, realizes The more MEC tasks of multi-user unload scheduling of resource.

Detailed description of the invention

Fig. 1 is based on the more MEC task unloading scheduling of resource figures of side-end collaboration multi-user；

Fig. 2 is to calculate unloading schematic diagram of a scenario based on side-end collaboration more MEC of multi-user.

Specific embodiment

The invention will be further described below in conjunction with the accompanying drawings.

The present invention comprehensively considers the task amount of user, the computing capability of MEC server, the computing capability of mobile terminal and letter Road occupation condition completes maximizing the benefits as target using task, carries out the distribution of computing resource and channel resource, establish more The calculating of the more MEC of user unloads frame, using Lyapunov's theory guarantee MEC server task squeeze queue stability and KKT condition is solved, and realizes the more MEC task unloading scheduling of resource of optimal multi-user.

Of the invention is a kind of based on the more MEC tasks unloading resource regulating methods of side-end collaboration multi-user, including walks as follows It is rapid:

Step 1, task analysis: for mobile terminal, completing different task can produce different benefit, and benefit is main Determined by two aspect factors: the expected revenus value that task self attributes (such as priority) generate generates in task processes Time delay loss；Specific steps are as follows:

Step 11, as shown in Fig. 2, considering within K moment, all users have I task to need local hold in moment t Row is offloaded to MEC processing, wherein the set J={ 1,2,3 ..., j..., J } of MEC, the collection of task is combined intoIndicate i-th of task, i=1,2,3 ..., I；Using task model To describe task m_iSize: D_i(t) (bit), i.e. task m_iData package size；Task m_iThe workload that need to be handled is D_i(t)X_i (CPU cycles), wherein X_iCpu cycle needed for indicating processing 1bit data volume；Spectral bandwidth that MEC j possesses is set as B_j (t), there is N_j(t) a subcarrier, each subcarrier bandwidth areAccording to shannon formula, task m on subcarrier n_iData pass Defeated rate is

Task m_iOverall data transmission rate be

Wherein, N_j(t) sub-carrier number, π are indicated_{I, n}(t) it is channel allocation indicator, works as π_{I, n}(t)=1 when, indicate that son carries Wave n distributes to task m_iIt is unloaded；Work as π_{I, n}(t)=0 when, indicate that subcarrier n is not yet assigned to task m_iIt is unloaded；p_iIt indicates Task m_iThe transmission power of place terminal；h_{N, j}Indicate the channel gain of user's subcarrier n, user during setting task unloading Mobility is not high, so h_{N, j}=127+logd_{I, j}, d_{I, j}Expression task m_iPlace user terminal is at a distance from MEC j；σ²It is channel Noise power；Task m_iThe transmission delay for being offloaded to MEC j is

Step 12, when task is stayed in processing locality by user, time delay only includes the task processing time；Consider that mobile subscriber is whole Hold u_iCPU processing capacity beThen locally executing time delay is

Step 13, task is offloaded to MEC server execution time delay to be made of following three parts: when a, task discharge conveyor Between, b, when have magnanimity task need to be offloaded to edge cloud processing when, more than MEC server load, task may be needed in each MEC Server queues wait, i.e. queue waiting time, and c, task handle the time；

Wherein, the discharge conveyor time isQueue waiting time isAt the CPU for setting MEC j Reason ability isTask m_iThe processing time beTherefore, task m_iIt is offloaded to execution at MEC server j Overall delay is

Step 14, for any one MEC server, the arrival process of task is modeled as Bernoulli process, if The task arrival rate for determining MEC server j is λ_j；The task quantity waited in the queue is assumed to be quene state: Q_j(t)=0,1, 2,3 ... }, the queue Q of MEC j_j(t) more new formula is

Q_j(t+1)=Q_j(t)-V_j(t)+A_j(t)

Wherein, V_j(t) processing speed for indicating task at MEC j, i.e., handle within the time that moment t/length is 1 and complete V_j(t) a task；A_j(t) indicate whether reach in moment t task, A_j(t) { 0,1 } ∈；Therefore, there is Pr { A_j(t)=1 }=λ_j And Pr { A_j(t)=0 }=1- λ_j；Based on Little's Law, consider in K moment, including waits in line time delay and processing delay exists Interior execution delay is directly proportional to the average queue length of task buffer area, and average queue length such as following formula indicates:

Step 15, if u_iExpression task m_iIn the expected revenus formulated according to its priority, L (T_i) indicate task m_iWhen Between T_iIt is interior to complete paid time delay loss；

Wherein, C is proportionality coefficient, determines that C is bigger to the susceptibility of time delay according to system, is indicated as caused by time delay Time delay loss is bigger, sets C=1 in the present invention；ρ_iFor the tolerance to time delay is lost, when time delay is less than tolerance, time delay pair User satisfaction will not impact, i.e., the income of user will not be lost, when time delay is greater than ρ_i, time delay causes user satisfaction It influences, accordingly produces time delay loss.

Step 2, problem is formed: introducing user's benefit value as the index for measuring system performance, to maximize a period of time It is that target establishes optimization problem that interior user side task, which completes total benefit,；Specific steps are as follows:

Step 21, task m_iBeing offloaded to the income that MEC processing generates is Wherein, u_iExpression task m_iIn the expected revenus formulated according to its priority, L (T_{I, j} It (t)) is task m_iIt is offloaded to the time delay loss that MEC processing generates；

Step 22, task m_iIt is performed locally the income of generation

Step 23, pass through combined optimization subcarrier allocator π_{I, n}(t), task allocator s_{I, j}(t), it obtains to maximize one User side task in the section time completes the optimization problem that total revenue is target:

P1:

S.t.C1:

C2:

C3:

C4:

C5:

C6:

C7:

Wherein, C1 ensures that a task can only select in processing locality or be offloaded to a MEC server execution；C2 ensures s_{I, j}It (t) is binary variable；C3 ensures π_{I, n}It (t) is binary variable；C4, which ensures a subcarrier at most, can only distribute to a use Family；C5 ensures that base station is the maximum transmission power that the transmission power of user's distribution is no more than base station, p^maxIt is the emission maximum of base station Power；C6 ensures that discharge conveyor energy is no more than task m_iPlace mobile terminal device dump energyC7 ensures task execution Time delay meets maximum delay requirementDue to the expected utility u of each task in the objective function of optimization problem P1_iIt is fixed , not at any time t and change, time delay loss function L () is linear function, therefore the optimization problem P2 being simplified:

P2:

S.t.C1:

C2:

C3:

C4:

C5:

C6:

C7:

Step 3, guarantee stable state: guaranteeing the stability of each MEC server task backlog queue, be based on Lyapunov's theory It is to solve the OPTIMAL TASK under limit to unload resource dispatching strategy by problem reduction；Specific steps are as follows:

Step 31, if it is Bernoulli process that the task of each queue, which reaches state, Θ (t)=(Q is enabled₁(t), Q₂(t) ..., Q_j(t) ..., Q_J(t)) quene state is indicated, Θ (t) is according to task arrival rate λ_jDevelop on time slot t ∈ { 0,1,2 ... }； Define secondary liapunov function:

ω_jIndicate that weight set, different weights will lead to different queue status locating in task scheduling strategy not Together, if all ω_jIt is all 1；Obviously, which is non-negative, and if only if all Θ_j(t) be 0 when, L (Θ (t)) it is equal to 0；

Step 32, in order to predict the variation of each quene state, the secondary liapunov function between a moment is defined Difference mean value be Liapunov drift function Δ (Θ (t)):

Wherein,Indicate the mean value of the difference of secondary liapunov function；

This drift is the performance of expected change that liapunov function is engraved at one；

Step 33, each moment t observes current Θ (t) value and takes control action, greedy according to consistent Θ (t) Minimum formula drift plus penalty expectation:

Step 34, delay sensitive parameter v is determined₀If v₀=1, optimization problem P2 abbreviation are as follows:

P3:

S.t.C1:

C2:

C3:

C4:

C5:

C6:

Step 4, channel distribution: set task unloading allocation strategy it is given in the case where, convert channel for optimization problem P3 Resource allocation problem, and optimal channel distribution is solved using KKT condition；Specific steps are as follows:

Step 41, if Given task unloads allocation strategy S '_{I, j}(t), optimization problem P3 is one about R_{I, j}(t) convex is asked Topic, it is assumed that there is l task to be offloaded to MEC processing, i.e. S_{I, j}(t)=1 number is l, and optimization object function is converted into such as following formula institute Show:

f(R_ij(t), S '_ijIt (t)) is about R_ij(t) function；

Step 42, due to f (R_{I, j}(t), S '_{I, j}It (t)) is convex function, and institute's Prescribed Properties are linear function, so Optimization problem is convex optimization problem, according to KKT condition, be can get about R_{I, j}(t) optimal solution

Step 43, the Lagrangian of constitution optimization problem, as follows:

Wherein, μ_{I, j}It is the undetermined coefficient of each constraint condition；

If R_{I, j}(t) and μ_{I, j}All meet KKT condition in arbitrary point, obtain:

By solving above formula, optimal R is obtained_{I, j}(t):

It can thus be concluded that pinned task unloads allocation strategy S '_{I, j}(t) optimal solution:

Step 5, task schedule: setting given channel resource allocation policy, and optimization problem P3 can be converted into Zero-one integer programming and ask Topic；Specific steps are as follows:

Step 51, if given channel resource allocation policy, optimization problem P3 is converted into Zero-one integer programming problem, following institute Show:

P4:

S.t.C1:

C2:

Step 52, each moment t is gone out on missions allocation strategy S using minimizing all tasks processing overall delays as object solving (t), that is, the corresponding best MEC server of each task is solved, and obtains each task and is offloaded to best MEC j^*Server The time delay of processing

Step 53, calculating task stays in the time delay T of processing locality_i(t), the time delay of MEC processing will be offloaded toWith (T_i(t)+δ) compare, δ is time delay tolerance, ifTask is then in MEC j^*Processing, otherwise in local Processing updates Task Assigned Policy S (t).

Step 6, combined optimization: alternating iteration step 4 and step 5 have until maximizing user's income in a period of time Body step are as follows:

Step 61, the preferred channels resource allocation under pinned task unloading distribution is obtained according to step 4；

Step 62, the OPTIMAL TASK unloading allocation strategy under fixed channel is obtained according to step 5；

Step 63, step 61 and 62 is repeated until obtaining optimal channel distribution and task scheduling strategy.

The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered It is considered as protection scope of the present invention.

Claims (7)

1. a kind of unload resource regulating method based on side-end collaboration more MEC tasks of multi-user, it is characterised in that: including as follows Step:

Step 1, task analysis: analysis task completes time delay, determines that task completes benefit；

Step 2, problem is formed: completing total benefit as target formation optimization problem to maximize task；

Step 3, guarantee stable state: guaranteeing the stability of task backlog queue to simplify problem；

Step 4, channel distribution: Given task unloading allocation strategy determines optimal channel distribution；

Step 5, task schedule: given channel resource allocation policy determines optimal task schedule；

Step 6, combined optimization: joint step 4 and 5 obtains optimal channel distribution and task schedule.

2. according to claim 1 unload resource regulating method based on side-end collaboration more MEC tasks of multi-user, special Sign is: in the step 1, benefit is determined by two aspect factors: the expected revenus value that task self attributes generate, task processing The time delay loss generated in the process；The specific steps of the step 1 are as follows:

Step 11, considering within K moment, all users have I task needs to locally execute or be offloaded to MEC processing in moment t, Wherein, the set J={ 1,2,3 ..., j..., J } of MEC, the collection of task is combined into Indicate i-th of task, i=1,2,3 ..., I；Task m is described using task model_iSize: D_i(t) (bit), i.e., Task m_iData package size；Task m_iThe workload that need to be handled is D_i(t)X_i(CPU cycles), wherein X_iIndicate processing 1bit Cpu cycle needed for data volume；Spectral bandwidth that MEC j possesses is set as B_j(t), there is N_j(t) a subcarrier, each subcarrier Bandwidth isAccording to shannon formula, task m on subcarrier n_iMessage transmission rate be

Task m_iOverall data transmission rate be

Wherein, N_j(t) sub-carrier number, π are indicated_{I, n}(t) it is channel allocation indicator, works as π_{I, n}(t)=1 when, subcarrier n points are indicated Dispensing task m_iIt is unloaded；Work as π_{I, n}(t)=0 when, indicate that subcarrier n is not yet assigned to task m_iIt is unloaded；p_iExpression task m_iThe transmission power of place terminal；h_{N, j}Indicate the channel gain of user's subcarrier n, the movement of user during setting task unloading Property is not high, so h_{N, j}=127+logd_{I, j}, d_{I, j}Expression task m_iPlace user terminal is at a distance from MEC j；σ²It is interchannel noise Power；Task m_iThe transmission delay for being offloaded to MEC j is

Step 12, when task is stayed in processing locality by user, time delay only includes the task processing time；Consider mobile subscriber terminal u_i CPU processing capacity beThen locally executing time delay is

Step 13, task MEC server execution time delay is offloaded to be made of following three parts: a, task discharge conveyor time, B, when there is magnanimity task to need to be offloaded to the processing of edge cloud, more than MEC server load, task may need to take in each MEC Business device is waited in line, i.e. queue waiting time, and c, task handle the time；

Wherein, the discharge conveyor time isQueue waiting time isThe CPU for setting MEC j handles energy Power isTask m_iThe processing time beTherefore, task m_iBe offloaded at MEC server j execute it is total when Prolonging is

Step 14, for any one MEC server, the arrival process of task is modeled as Bernoulli process, sets MEC The task arrival rate of server j is λ_j；The task quantity waited in the queue is assumed to be quene state: Q_j(t)=0,1,2, 3 ... }, the queue Q of MEC j_j(t) more new formula is

Q_j(t+1)=Q_j(t)-V_j(t)+A_j(t)

Wherein, V_j(t) processing speed of task at MEC j is indicated, i.e., processing completes V within the time that moment t/length is 1_j (t) a task；A_j(t) indicate whether reach in moment t task, A_j(t) { 0,1 } ∈；Therefore, there is Pr { A_j(t)=1 }=λ_jAnd Pr{A_j(t)=0 }=1- λ_j；Based on Little's Law, consider in K moment, including waiting in line time delay and processing delay Execution delay it is directly proportional to the average queue length of task buffer area, average queue length for example following formula expression:

Step 15, if u_iExpression task m_iIn the expected revenus formulated according to its priority, L (T_i) indicate task m_iIn time T_iIt is interior Complete paid time delay loss；

Wherein, C is proportionality coefficient, determines that C is bigger to the susceptibility of time delay according to system, indicates the time delay as caused by time delay It loses bigger；ρ_iFor the tolerance to time delay is lost, when time delay is less than tolerance, time delay not will cause shadow to user satisfaction It rings, i.e., the income of user will not be lost, when time delay is greater than ρ_i, time delay causes influence to user satisfaction, produces accordingly Time delay loss.

3. according to claim 1 unload resource regulating method based on side-end collaboration more MEC tasks of multi-user, special Sign is: in the step 2, introducing user's benefit value as the index for measuring system performance, to maximize in a period of time It is that target establishes optimization problem that user side task, which completes total benefit,；Specific steps are as follows:

Step 21, task m_iBeing offloaded to the income that MEC processing generates is Wherein, u_iExpression task m_iIn the expected revenus formulated according to its priority, L (T_{I, j} It (t)) is task m_iIt is offloaded to the time delay loss that MEC processing generates；

Step 22, task m_iIt is performed locally the income of generation

Step 23, pass through combined optimization subcarrier allocator π_{I, n}(t), task allocator s_{I, j}(t), when obtaining to maximize one section Interior user side task completes the optimization problem that total revenue is target:

P1:

S.t.C1:

C2:

C3:

C4:

C5:

C6:

C7:

Wherein, C1 ensures that a task can only select in processing locality or be offloaded to a MEC server execution；C2 ensures s_{I, j} It (t) is binary variable；C3 ensures π_{I, n}It (t) is binary variable；C4, which ensures a subcarrier at most, can only distribute to a user； C5 ensures that base station is the maximum transmission power that the transmission power of user's distribution is no more than base station, p^maxIt is the emission maximum function of base station Rate；C6 ensures that discharge conveyor energy is no more than task m_iPlace mobile terminal device dump energyWhen C7 ensures task execution Prolong and meets maximum delay requirementDue to the expected utility u of each task in the objective function of optimization problem P1_iBe it is fixed, Not at any time t and change, time delay loss function L () is linear function, therefore the optimization problem P2 being simplified:

P2:

S.t.C1:

C2:

C3:

C4:

C5:

C6:

C7:

4. according to claim 1 unload resource regulating method based on side-end collaboration more MEC tasks of multi-user, special Sign is: in the step 3, guaranteeing the stability of each MEC server task backlog queue, will be asked based on Lyapunov's theory Topic is reduced to solve the OPTIMAL TASK unloading resource dispatching strategy under limit, specific steps are as follows:

Step 31, if it is Bernoulli process that the task of each queue, which reaches state, Θ (t)=(Q is enabled₁(t), Q₂(t) ..., Q_j (t) ..., Q_J(t)) quene state is indicated, Θ (t) is according to task arrival rate λ_jDevelop on time slot t ∈ { 0,1,2 ... }；It is fixed The secondary liapunov function of justice:

ω_jIndicate that weight set, different weights will lead to different queue status difference locating in task scheduling strategy, if All ω_jIt is all 1；Obviously, which is non-negative, and if only if all Θ_j(t) be 0 when, L (Θ (t)) Equal to 0；

Step 32, in order to predict the variation of each quene state, the difference of the secondary liapunov function between a moment is defined The mean value of value is Liapunov drift function Δ (Θ (t)):

Wherein,Indicate the mean value of the difference of secondary liapunov function；

This drift is the performance of expected change that liapunov function is engraved at one；

Step 33, each moment t observes current Θ (t) value and takes control action, according to consistent Θ (t), it is greedy most The drift of smallization formula plus penalty expectation:

Step 34, delay sensitive parameter v is determined₀If v₀=1, optimization problem P2 abbreviation are as follows:

P3:

S.t.C1:

C2:

C3:

C4:

C5:

C6:

5. according to claim 1 unload resource regulating method based on side-end collaboration more MEC tasks of multi-user, special Sign is: in the step 4, if converting channel resource for optimization problem P3 in the case that task unloading allocation strategy is given Assignment problem, and optimal channel distribution is solved using KKT condition；Specific steps are as follows:

Step 41, if Given task unloads allocation strategy S '_ij(t), optimization problem P3 is one about R_{I, j}(t) convex problem, it is false MEC processing, i.e. S are offloaded to equipped with l task_{I, j}(t)=1 number is l, and optimization object function, which is converted into, to be shown below:

f(R_ij(t), S '_ij(t)) about R_ij(t) function；

Step 42, due to f (R_{I, j}(t), S '_{I, j}It (t)) is convex function, and institute's Prescribed Properties are linear function, so optimal Change problem is convex optimization problem, according to KKT condition, be can get about R_{I, j}(t) optimal solution

Step 43, the Lagrangian of constitution optimization problem, as follows:

Wherein, μ_{I, j}It is the undetermined coefficient of each constraint condition；

If R_{I, j}(t) and μ_{I, j}All meet KKT condition in arbitrary point, obtain:

By solving above formula, optimal R is obtained_{I, j}(t):

It can thus be concluded that pinned task unloads allocation strategy S '_{I, j}(t) optimal solution:

6. according to claim 1 unload resource regulating method based on side-end collaboration more MEC tasks of multi-user, special Sign is: in the step 5, if given channel resource allocation policy, optimization problem P3 is converted into Zero-one integer programming problem；Tool Body step are as follows:

Step 51, if given channel resource allocation policy, optimization problem P3 is converted into Zero-one integer programming problem, as follows:

P4:

S.t.C1:

C2:

Step 52, each moment t is gone out on missions allocation strategy S (t) using minimizing all tasks processing overall delays as object solving, i.e., The corresponding best MEC server of each task is solved, and obtains each task and is offloaded to best MEC j^*Server process Time delay

Step 53, calculating task stays in the time delay T of processing locality_i(t), the time delay of MEC processing will be offloaded toWith (T_i(t)+ δ) comparing, δ is time delay tolerance, ifTask is then in MEC j^*Processing, otherwise in processing locality, more New task allocation strategy S (t).

7. according to claim 1 unload resource regulating method based on side-end collaboration more MEC tasks of multi-user, special Sign is: the specific steps of the step 6 are as follows:

Step 61, the preferred channels resource allocation under pinned task unloading distribution is obtained according to step 4.

Step 62, the OPTIMAL TASK unloading allocation strategy under fixed channel is obtained according to step 5.

Step 63, step 61 and 62 is repeated until obtaining optimal channel distribution and task scheduling strategy.