Task queue-aware real-time channel allocation and task offloading method for edge computing

Abstract

A task queue-aware edge computing real-time channel allocation and task offloading method is proposed. According to the Lyapunov optimization framework, the task queue-aware base station channel allocation and user task offloading problems are transformed into a single time slot optimization model. According to game theory, users are divided and combined and cooperative games are carried out to form a convergent and stable combination set, and finally the channel allocation and task offloading strategy for each time slot is obtained.

Description

Technical Field

[0001] The present invention relates to a technology in the field of distributed computing, in particular to a real-time channel allocation and task unloading method based on Lyapunov optimization and game theory for task queue backlog perception. Background Art

[0002] Mobile Edge Computing (MEC) is widely regarded as an important technology to realize the vision of the next generation Internet. Traditional cloud computing systems rely on remote public clouds, which will cause long delays due to data exchange. MEC deploys a cloud computing platform at the edge of the wireless access network to provide computing, storage, network and communication resources for application services, which can ease the tension between computing-intensive applications and resource-limited mobile devices. Therefore, by offloading computing tasks on mobile devices to MEC servers, the distance to application services can be shortened, thereby reducing energy consumption and execution delays, and significantly improving the user's quality of experience.

[0003] Most of the existing computing task offloading technologies perform one-time channel allocation in the wireless transmission part, that is, the task does not change the channel during the transmission process. However, since the channel is time-varying, and the task needs to be transmitted in multiple time slots, it may cause a large transmission delay when the channel quality is poor. In addition, the current technology does not consider the task queue status of different application services on the edge server, which may cause a certain service to be overloaded while other services are idle, resulting in a large computing delay. Summary of the invention

[0004] In view of the problems that the prior art causes large wireless transmission delay and task calculation delay during channel allocation, and cannot perceive the status of different types of task queues of MEC servers, resulting in too many tasks of the same type arriving at the same time, causing a backlog in the task queue and a large calculation delay, and in user-dense places, the backlog of task queues of individual application services in the edge server leads to an increase in the average delay of all application tasks, the present invention proposes a task queue-aware edge computing real-time channel allocation and task offloading method, performs single-slot target conversion based on the Lyapunov optimization framework, analyzes the backlog degree of the task queue in the edge server and the remaining time of the task, and uses a game algorithm to perform real-time channel allocation and task offloading for user devices to minimize the average delay of the task.

[0005] The present invention is achieved through the following technical solutions:

[0006] The present invention relates to a task queue-aware edge computing real-time channel allocation and task unloading method, comprising:

[0007] Step A, based on the distance between the user and the base station, generates the channel gain matrix of each time slot and the transmission rate that the user can achieve in each sub-channel, and calculates the corresponding waiting delay, transmission delay and edge computing delay of each task according to the task volume of the edge server application service task queue, specifically including:

[0008] Step A-1: ​​Channel gain of the user on the subchannel Where: d m is the distance from user m to the base station, L l (d m ) is the distance between subchannel l and base station d m The path loss at is a small-scale zero-mean Gaussian distribution.

[0009] Step A-2: The received signal-to-noise ratio (SINR) between the user and the base station is: p l,m is the transmission power of user m in subchannel l, and the transmission power of the user in different channels does not change over time; σ is the variance of thermal noise power. Without loss of generality, the order of channel gain estimation for the user on the lth subchannel is

[0010] Step A-3, calculate the transmission rate of user m on subchannel l in the tth time slot Where: L is the total number of subchannels, W is the total channel bandwidth of the base station, which is evenly allocated to each subchannel; Assign decision variables to channels.

[0011] Step A-4, calculate the remaining size of each task in each time slot after the transmission is completed:

[0012] in: The task offloading decision variable for each user is ΔT, which is the duration of each time slot.

[0013] Step A-5, calculate the waiting delay of each task, that is, the delay from task generation to task start transmission and the transmission delay: in: is the generation time of the task, and The wireless transmission starts and ends at each task.

[0014] Step A-6: The task amount of the task queue of application service k in the edge server in the tth time slot is Q k (t), calculate the task queue length at the time when each task transmission is completed: in: is the transmission completion time of the nth task, (m,k)→n is the task index of the task queue mapped when the transmission of each task of the user is completed. When n=0, Task queue task volume Q k (0)=0.

[0015] Step A-7, calculate the edge computing latency of each task: Where: k is the average task processing frequency of application service k in the edge server, specifically ν k =μ k / F K , where: μ k is the amount of computation required per bit for k-type applications, F K The computing resources allocated to the edge server for k types of application services.

[0016] Step B: Model the task queue-aware base station channel allocation and user task offloading problem as an optimization model with the goal of minimizing the average task delay, and transform it into a single-slot optimization target based on the Lyapunov optimization framework, which specifically includes:

[0017] Step B-1, establish an optimization model with the goal of minimizing the average task delay.

[0018] In step B-2, the optimization objective is converted to maximizing the cumulative sum of the maximum transmission and edge processing task sizes in all time slots.

[0019] Step B-3, based on the Lyapunov optimization framework, the optimization target in the time domain is converted into a single time slot model by establishing the delay remaining queue and the task queue task backlog queue: Among them: U k (a t ,c t ) is the utility function of each application service.

[0020] Step C, constructing a combination set based on cooperative game, that is, users who request the same type of application services are combined, and the users play games with the goal of maximizing the system utility. Finally, the combination set reaches convergence stability, which specifically includes:

[0021] Step C-1, initializing the combination set: the base station randomly allocates channels to users requesting services, each user randomly selects a task to be transmitted, and an initial combination set Π is constructed based on users with the same task transmission type.

[0022] Step C-2, calculate the utility u of user m under the combination m (φ), the utility U of each combination k (φ) and total system utility.

[0023] Step C-3, user m selects a combination from the combination set. If the user is not associated with a subchannel, the subchannel associated with the combination is first allocated to user m, and the user's own utility, combination utility, and total system utility after joining are calculated to determine whether user m meets the following transfer conditions:

[0024] Step C-3-1: User m selects from the current combination φ i Transfer to combination φ j When , its own utility is not less than the utility before joining;

[0025] Step C-3-2: User m selects from the current combination φ i Transfer to combination φ j , the system utility is greater than the system utility under the original combination set before joining.

[0026] Step C-4: When the transfer condition is met, the combination φ i Add to the candidate set; if the transfer condition is not met, select a new combination to join.

[0027] Step C-5: When the candidate set is not empty, select the combination φ that maximizes the system utility from the candidate combinations of user m opt , update the new and old combinations after the user joins.

[0028] Step C-6: When the combinations of all users no longer change, the game ends and a stable combination set is obtained.

[0029] Step C-7, generating channel allocation and task offloading strategies according to the final stable combination set.

[0030] In step D, tasks are transmitted in each time slot according to the obtained channel allocation and task offloading strategy, and the delay remaining queue and task queue task backlog queue are updated according to the queue update rule, and step C is continued until all tasks are transmitted.

[0031] The present invention relates to a system for implementing the above method, comprising: a data processing unit, a channel allocation and task unloading decision unit and a data updating unit, wherein: the data processing unit calculates the transmission rate that the user can achieve in each sub-channel according to the channel gain information and the user task information, and calculates the corresponding waiting delay, transmission delay and edge computing delay of each task according to the task amount of the task queue of each application service of the edge server; the channel allocation and task unloading decision unit performs a combination game between users according to the delay information of the user task and the task queue status to obtain a stable channel allocation and task unloading strategy; the data updating unit updates the delay remaining queue and the task backlog queue according to the current channel allocation and task unloading strategy.

[0032] Technical Effects

[0033] Compared with the prior art, the present invention considers the impact of data backlog in the task queue in the MEC server on channel allocation and task offloading decisions, and performs channel allocation and task offloading by jointly considering the data volume of the task queue and the remaining delay of the user task in a user-intensive scenario with multiple tasks, so that application tasks with less data volume in the task queue obtain higher priority, reduce server computing delay, and thus achieve lower average delay.

[0034] In the process of task transmission, the present invention divides the time domain into multiple time slots, comprehensively considers the task queue backlog and the remaining task delay of the application service in each time slot, and conducts a game between user combinations to obtain a stable channel allocation and task unloading strategy. The present invention combines game theory with channel allocation and task unloading, conducts a cooperative game between user combinations formed according to task types, forms a convergent and stable combination set, and then obtains a better channel allocation and task unloading strategy. The final results show that the present invention has better stability than traditional algorithms and can achieve lower average delay. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic diagram of an embodiment scenario;

[0036] Figure 2 is the relationship between the average task size and the average delay;

[0037] Figure 3 is a graph of the relationship between the average task size and the percentage of task timeout;

[0038] Figure 4 is a graph of the relationship between the average task size and the variance of the task queue backlog;

[0039] Figure 5 It is the relationship diagram between the distance between the user and the base station and the average delay;

[0040] Figure 6 is the relationship diagram between the distance between the user and the base station and the variance of the task queue backlog;

[0041] Figure 7 It is a flow chart of the specific steps of the present invention. DETAILED DESCRIPTION

[0042] The experimental environment of this embodiment is Windows 10 64bit operating system, the CPU is Intel i7-7600U, the memory is 16GB, and the experimental development language is Python. The MEC network scenario includes an edge server and a base station. The base station has L sub-channels, that is, l∈L={1,2,...,L}; there are M service request users within its coverage area, that is, m∈M={1,2,...,M}; each user has K types of application requests, that is, k∈K={1,2,...,K}; the average task generation size of k types of applications is B k , the same application task size b m,k,j The mean is B k The uniform distribution of k types of application tasks requires a computing amount per bit of μ k .

[0043] The time domain is divided into T time slots, that is, t∈T={1,2,...,T}, the duration of each time slot is ΔT, and the k-type applications of user m have a total of tasks, whose indexes are sorted by arrival time:

[0044] like Figure 7 As shown, this embodiment relates to a task queue-aware edge computing real-time channel allocation and task offloading method, and the specific steps include:

[0045] Step 1: Generate the channel gain matrix H of each time slot and the transmission rate that the user can achieve in each subchannel according to the distance between the user and the base station, and calculate the corresponding waiting delay, transmission delay and edge computing delay of each task according to the task queue task volume of each application service of the edge server;

[0046] The communication adopts Non-Orthogonal Multiple Access (NOMA) technology, which allows multiple users to use the same resource block at the same time, and further applies Successive Interference Cancellation (SIC) technology to alleviate the user's co-channel interference to effectively improve resource utilization. According to the rules of the NOMA protocol, the base station uses SIC for multi-user detection. Specifically, the base station sequentially decodes the signal from the device with higher channel gain and regards all other signals as interference.

[0047] The channel gain of the user on the subchannel is: Where: d m is the distance from user m to the base station, L l (d m) is the distance from the base station d from the subchannel l m The path loss at is a small-scale zero-mean Gaussian distribution. Without loss of generality, the order of channel gain estimation for users on the lth subchannel is Then the received signal-to-noise ratio SINR between the user and the base station is: p l,m is the transmission power of user m in subchannel l. The transmission power of the user in different channels does not change with time. σ is the thermal noise power variance. The signal-to-noise ratio of the user in all channels in each time slot is calculated. For data transmission, its transmission rate and transmission delay are calculated.

[0048] The transmission rate of user m on the tth time slot subchannel l is Where: W is the total channel bandwidth of the base station, which is evenly distributed to each sub-channel; Assign decision variables to channels when In the tth time slot, the base station allocates subchannel l to user m. To reduce inter-channel interference, each user can only be associated with one subchannel at most, that is,

[0049] According to the above wireless transmission rate, the remaining task size of each time slot user after the transmission associated with each sub-channel is completed is obtained as follows: in: For each user's task offloading decision variables, specifically: In the tth time slot, user m selects the task of type k application for transmission. That is, the remaining task size will change only when the base station allocates a subchannel to the user and the user selects the corresponding application. For the initial time slot, the remaining task size is the task generation size

[0050] By changing the remaining tasks, the wireless transmission start time slot of all tasks of the user can be obtained. and end slot Specifically For the same application of the same user, the transmission order of its tasks follows the FIFO order. The j+1th task of each application can only start transmission after the jth transmission is completed, that is,

[0051] The waiting delay and transmission delay of each task of each user are obtained through the above information: in: is the task generation time, and the task arrival rate of each application follows λ kThe Poisson distribution of the spatial domain is used to model the non-uniform mobile traffic using the log-normal distribution, i.e.

[0052] For the edge server, the task queue task volume of application service k in the edge server at the t time slot is Q k (t), each task of the user is mapped to the task index of the task queue when the transmission is completed, that is, when the transmission of the jth task of user m and application k is completed, it is the nth task in the task queue of application service k, (m, k)→n.

[0053] According to the transmission delay, the transmission completion time of the nth task is Then the task queue task volume at the time when each task transmission is completed is For n = 0, Task queue task volume Q k (0)=0.

[0054] For the task queue of the application service, the task size cannot exceed its maximum length, that is, in: The maximum length of the task queue serving k types of applications in the edge server.

[0055] The edge computing delay of each task can be obtained by the task queue length: Where: k is the average task processing frequency of application service k in the edge server, specifically ν k =μ k / F K , where: μ k is the amount of computation required per bit for k-type applications, F K The computing resources allocated to the edge server for k types of application services.

[0056] Step 2: Model the task queue-aware base station channel allocation and user task offloading problems as an optimization model with the goal of minimizing the average task delay, and transform it into a single-slot optimization objective based on the Lyapunov optimization framework.

[0057] The task queue-aware base station channel allocation and user task offloading problem refers to: when the task queue of an application service in the edge server is crowded with tasks, the application service load of the edge server can be balanced by allocating channels to users with other types of application tasks and giving priority to transmitting tasks of idle applications, so as to reduce the waiting time of tasks in the task queue and reduce the computing delay, thereby reducing the average delay of the task and improving user satisfaction.

[0058] The optimization model with the goal of minimizing the average task delay is specifically:

[0059]

[0060]

[0061]

[0062]

[0063]

[0064] Where: D m,k,j is the total latency of the jth task of application k of user m, specifically:

[0065] The optimization model must satisfy the requirement that the total latency of all tasks is within the range required by the application, i.e.

[0066] The above optimization model of minimizing the average user delay is transformed into the optimization target of time domain integration, which is equivalent to maximizing the cumulative sum of the maximum transmission and edge processing task sizes in all time slots, specifically:

[0067]

[0068]

[0069]

[0070]

[0071]

[0072]

[0073]

[0074] Based on Lyapunov optimization theory, the task delay constraint and task queue constraint are transformed into the delay remaining queue and task queue task backlog queue, specifically: in: The remaining execution time of the jth task of type k application for user m is:

[0075] The single time slot optimization target is based on the Lyapunov optimization framework. Through the delay remaining queue and task queue task backlog queue established above, the optimization target in the above time domain is converted into a single time slot, that is, the strategy of the entire time domain is obtained by solving the channel allocation and task transmission strategy in each time slot. The single time slot optimization target is specifically:

[0076]

[0077]

[0078]

[0079] Among them: U k (a t ,c t ) is the utility function of each application service, specifically:

[0080] Step 3: Construct a combination set based on cooperative game, that is, users who request the same type of application services are combined, and users compete with each other to maximize the system utility, and finally the combination set reaches convergence stability.

[0081] The combination set is a user combination set Π={φ1, φ2, ..., φ K}, where: k≠k′, When there is no combination φ k Users in the group will join other groups by k′ To change the current partition, the combined set Π is a stable set.

[0082] The cooperative game mentioned above means that each user selects any type of task to transmit and builds a task queue of the application service in the edge server together with other users in the combination. Specifically, for user m, define > m is a complete transitive relation on all possible combinations that user m may form; when φ i > m φ j , which means that user m is better than combination φ j Prefer to join group φ i This preference relationship will affect the formation of the final combination set. Users compete with each other to form combinations, and consider whether to add new combinations based on this preference relationship, i.e., the combination rule, to finally achieve the stability of all combinations. In the combination formation game, the preference order can ensure the existence of combination stability.

[0083] The combination rule means that when the combination utility of user m after joining the combination is higher than the combination utility before joining the combination and the user's own utility is improved, the user will join the new combination, that is, when the user chooses to join the combination φ i When , its own utility increases, and it will increase the total utility of the system. The specific combination rule is: in: For the user's utility, Add group φ for user m i The new combination φ i The original combination j The combined effect of is user m joining group φ i The previous combination i The original combination j Since the addition of user m will only affect the new and old combinations and will not affect other combinations, it is feasible to consider the combined utility of the new and old combinations and then affect the total system utility.

[0084] The combination set reaches convergence stability means that all users continuously play games according to the combination rules and finally converge to form a stable combination set, which specifically includes:

[0085] 3.1) Initialize the combination set: The base station randomly allocates channels to users who request services, and each user randomly selects a task to be transmitted. The initial combination set Π is constructed based on users with the same task transmission type.

[0086] 3.2) Calculate the utility u of user m under the combination m (φ), the utility U of each combination k (φ) and total system utility.

[0087] 3.3) User m selects a combination from the combination set. When the user is not associated with a subchannel, the subchannel associated with the combination is first assigned to user m, and its own utility, combination utility and total system utility after joining are calculated.

[0088] Determine whether user m meets the following transfer conditions:

[0089] a) User m from the current combination φ i Transfer to combination φ j When , its own utility is not less than the utility before joining;

[0090] b) User m from the current combination φ i Transfer to combination φ j , the system utility is greater than the system utility under the original combination set before joining.

[0091] When the transfer condition is met, the combination φ i Add to the candidate set; if the transfer condition is not met, select a new combination to join.

[0092] 3.4) When the candidate set is not empty, select the combination φ that maximizes the system utility from the candidate combinations of user m opt , update the new and old combinations after the user joins.

[0093] 3.5) When the combinations of all users no longer change, the game ends and a stable combination set is obtained.

[0094] Channel allocation and task offloading strategies are generated based on the final stable combination set.

[0095] Step 4: In each time slot, tasks are transmitted according to the obtained channel allocation and task offloading strategy, and the delay remaining queue and task queue task backlog queue are updated according to the queue update rule, and step 3 is continued until all tasks are transmitted.

[0096] like Figure 1 As shown in the figure, the specific application scenarios involved in this embodiment include: a base station, a MEC server connected to the base station, 5 sub-channels, 8 users and 3 different types of applications. The users are randomly distributed in a simulation area with a radius of 200m, the total bandwidth W = 5MHz, and the average task generation size is B k =1MB, the simulation results obtained according to the above method are as follows Figure 2-Figure 6 shown.

[0097] like Figure 2 As shown, the average delay of the task is compared with the task queue-aware real-time channel allocation and task offloading method proposed in the present invention and the greedy algorithm and random algorithm under different average task sizes. It can be seen from the figure that the present invention can achieve the lowest delay.

[0098] like Figure 3 As shown, the percentage of tasks that exceed the maximum delay in the total tasks is compared with the task queue-aware real-time channel allocation and task offloading method proposed in the present invention and the greedy algorithm and random algorithm under different average task sizes. It can be seen from the figure that the present invention can meet the experimental requirements of more tasks.

[0099] like Figure 4 As shown, the variance of task backlogs of task queues of different types of application services in edge servers is compared with the task queue-aware real-time channel allocation and task offloading method proposed in the present invention and the greedy algorithm and random algorithm under different average task sizes. As can be seen from the figure, the present invention can achieve better load balancing.

[0100] like Figure 5 As shown, the average delay of the task changes with the distance between the user and the base station, and the comparison between the task queue-aware real-time channel allocation and task unloading method proposed in the present invention and the greedy algorithm and the random algorithm. As can be seen from the figure, the present invention can achieve the lowest delay.

[0101] like Figure 6 As shown, the variance of the task backlog of different types of application service task queues in the edge server changes with the distance between the user and the base station. The task queue-aware real-time channel allocation and task offloading method proposed in the present invention is compared with the greedy algorithm and the random algorithm. As can be seen from the figure, the present invention can achieve better load balancing.

[0102] In summary, this method models the base station channel allocation and user task offloading problems as an optimization model with the goal of minimizing the average task delay. It comprehensively considers the task backlog and remaining task delay of the task queue of the application service, and converts the time domain model into a single time slot model based on the Lyapunov optimization framework. In each time slot, users are divided into different combinations according to game theory for cooperative games. Channel allocation and task transmission are performed according to the stable combination set formed by convergence, and the remaining delay queue and task backlog queue are updated until all tasks are transmitted. The final result significantly reduces the average task delay.

[0103] The above-mentioned specific implementation can be partially adjusted in different ways by those skilled in the art without departing from the principle and purpose of the present invention. The protection scope of the present invention shall be based on the claims and shall not be limited by the above-mentioned specific implementation. Each implementation scheme within its scope shall be subject to the constraints of the present invention.

Claims

1. A task queue-aware edge computing real-time channel allocation and task offloading method, characterized in that: Based on the distance between the user and the base station, the channel gain matrix of each time slot and the transmission rate that the user can achieve in each sub-channel are generated, and the corresponding waiting delay, transmission delay and edge computing delay of each task are calculated according to the task volume of the edge server application service task queue; the base station channel allocation and user task offloading problem perceived by the task queue is modeled as an optimization model with the goal of minimizing the average task delay, and is converted into a single time slot optimization target according to the Lyapunov optimization framework; in each time slot, tasks are transmitted according to the obtained channel allocation and task offloading strategies, and the delay remaining queue and the task queue task backlog queue are updated according to the queue update rule, and the game of the combination set is carried out until all tasks are transmitted; The game of the combination set refers to: constructing a combination set based on cooperative game and users playing games with the goal of maximizing system utility until the combination set reaches convergence stability; The waiting delay, transmission delay and edge computing delay corresponding to each task are calculated in the following way: Step A-1: ​​Channel gain of the user on the subchannel Where: d m is the distance from user m to the base station, L l (d m ) is the distance between subchannel 1 and base station d m The path loss at is a small-scale zero-mean Gaussian distribution; Step A-2: The received signal-to-noise ratio (SINR) between the user and the base station is: p l,m is the transmission power of user m in subchannel 1, and the transmission power of the user in different channels does not change over time; σ is the variance of thermal noise power; without loss of generality, the order of channel gain estimation for the user on the first subchannel is Step A-3, calculate the transmission rate of user m on subchannel 1 in the tth time slot Where: L is the total number of subchannels, W is the total channel bandwidth of the base station, which is evenly allocated to each subchannel; Assign decision variables to channels; Step A-4, calculate the remaining size of each task in each time slot after the transmission is completed: in: The task offloading decision variable for each user, ΔT is the duration of each time slot; Step A-5, calculate the waiting delay of each task, that is, the delay from task generation to task start transmission and the transmission delay: in: is the generation time of the task, and The wireless transmission start time slot and end time slot for each task; Step A-6: The task amount of the task queue of application service k in the edge server in the tth time slot is Q k (t), calculate the task queue length at the time when each task transmission is completed: in: is the transmission completion time of the nth task, (m, k)→n is the task index of the task queue mapped when the transmission of each task of the user is completed; when n=0, Task queue task volume Q k (0) = 0; Step A-7, calculate the edge computing latency of each task: Where: k is the average task processing frequency of application service k in the edge server, specifically ν k =μ k / F K , where: μ k is the amount of computation required per bit for k types of applications, F K The computing resources allocated to the edge server for k types of application services; The single time slot optimization target is obtained by: Step B-1, establishing an optimization model with the goal of minimizing the average task delay; Step B-2, the optimization objective is converted to maximize the cumulative sum of the maximum transmission and edge processing task sizes in all time slots; Step B-3, based on the Lyapunov optimization framework, the optimization target in the time domain is converted into a single time slot model by establishing the delay remaining queue and the task queue task backlog queue: Among them: U k (a t ,c t ) is the utility function for each application service; The game with the goal of maximizing system utility specifically refers to: Step C-1, initializing the combination set: the base station randomly allocates channels to users requesting services, each user randomly selects a task to be transmitted, and an initial combination set Π is constructed based on users with the same task transmission type; Step C-2, calculate the utility u of user m under the combination m (φ), the utility U of each combination k (φ) and total system utility; Step C-3, user m selects a combination from the combination set. If the user is not associated with a subchannel, the subchannel associated with the combination is first allocated to user m, and the user's own utility, combination utility, and total system utility after joining are calculated to determine whether user m meets the transfer condition. Step C-4: When the transfer condition is met, the combination φ i Add to the candidate set; if the transfer condition is not met, select a new combination to join; Step C-5: When the candidate set is not empty, select the combination φ that maximizes the system utility from the candidate combinations of user m opt , update the new and old combinations after the user joins; Step C-6: When all the user combinations no longer change, the game ends and a stable combination set is obtained; Step C-7, generating a channel allocation and task offloading strategy according to the final stable combination set; The transfer conditions include: Step C-3-1: User m selects from the current combination φ i Transfer to combination φ j When , its own utility is not less than the utility before joining; Step C-3-2: User m selects from the current combination φ i Transfer to combination φ j When , the system utility is greater than the system utility under the original combination set before joining; The optimization model with the goal of minimizing the average task delay is specifically: Where: D m,k,j is the total latency of the jth task of application k of user m, specifically: The single time slot optimization objectives are specifically: Among them: U k (a t ,c t ) is the utility function of each application service, specifically: The cooperative game mentioned above means that each user selects any type of task to transmit and builds a task queue of the application service in the edge server together with other users in the combination. Specifically, for user m, define > m is a complete transitive relation on all possible combinations that user m may form; when φ i > m φ j , which means that user m is better than combination φ j Prefer to join group φ i This preference relationship will affect the formation of the final combination set. Users compete with each other to form combinations. According to this preference relationship, that is, the combination rule, whether to add a new combination is considered, and finally all combinations are stable. In the combination formation game, the preference order ensures the existence of combination stability; The combination rule means that when the combination utility of user m after joining the combination is higher than the combination utility before joining the combination and the user's own utility is improved, the user will join the new combination, that is, when the user chooses to join the combination φ i When , its own utility increases, and it will increase the total utility of the system. The specific combination rule is: in: For the user's utility, Add group φ for user m i The new combination φ i And the original combination φ j The combined effect of is user m joining the group φ i The previous combination i And the original combination φ j Since the addition of user m will only affect the new and old combinations and will not affect other combinations, it is feasible to consider the combined utility of the new and old combinations and then affect the total system utility; The combination set reaches convergence stability means that all users continuously play games according to the combination rules and finally converge to form a stable combination set, which specifically includes: 3.1) Initialize the combination set: The base station randomly allocates channels to users who request services, and each user randomly selects a task to be transmitted. The initial combination set Π is constructed based on users with the same task transmission type; 3.2) Calculate the utility u of user m under the combination m (φ), the utility U of each combination k (φ) and total system utility; 3.3) User m selects a combination from the combination set. If the user is not associated with a subchannel, the subchannel associated with the combination is first assigned to user m, and the user's own utility, combination utility, and total system utility after joining are calculated; Determine whether user m meets the following transfer conditions: a) User m from the current combination φ i Transfer to combination φ j When , its own utility is not less than the utility before joining; b) User m from the current combination φ i Transfer to combination φ j When , the system utility is greater than the system utility under the original combination set before joining; When the transfer condition is met, the combination φ i Add to the candidate set; if the transfer condition is not met, select a new combination to join; 3.4) When the candidate set is not empty, select the combination φ that maximizes the system utility from the candidate combinations of user m opt , update the new and old combinations after the user joins; 3.5) When the combinations of all users no longer change, the game ends and a stable combination set is obtained.

2. A system for implementing the task queue-aware edge computing real-time channel allocation and task offloading method of claim 1, characterized in that: include: A data processing unit, a channel allocation and task unloading decision unit, and a data updating unit, wherein: the data processing unit calculates the transmission rate that the user can achieve in each sub-channel according to the channel gain information and the user task information, and calculates the corresponding waiting delay, transmission delay and edge computing delay of each task according to the task amount of the task queue of each application service of the edge server; the channel allocation and task unloading decision unit conducts a combination game between users according to the delay information of the user task and the task queue status to obtain a stable channel allocation and task unloading strategy; the data updating unit updates the delay remaining queue and the task backlog queue according to the current channel allocation and task unloading strategy.

Images