Computing resource evaluation method and system for mobile edge application scenarios

Abstract

The present invention provides a computing resource evaluation method and system for mobile edge application scenarios. By constructing a task scenario description model based on task timing relationships, the complex relationship between task requirements, subtasks and resources can be effectively processed, so that task resources can be allocated based on a deep reinforcement learning algorithm, thereby achieving optimal allocation of task resources under resource constraints, improving task execution efficiency and resource utilization. Compared with traditional methods, it is more adaptable to dynamically changing task scenarios, effectively improves resource utilization and task execution efficiency, and reduces system latency and energy consumption.

Description

Technical Field

[0001] The present invention belongs to the field of computing, and in particular, relates to a computing resource evaluation method and system for mobile edge application scenarios. Background Art

[0002] In the traditional computing field, when the running resources required for a task are insufficient, the task may not be executed, or the system will dynamically adjust according to the existing resources, but such adjustments are often difficult to accurately meet the task requirements and may even be too rough. In addition, high-priority tasks may preempt the resources of low-priority tasks. To address this problem, engineering projects usually pre-configure larger computing resources to ensure task operation. However, in scenarios for mobile edge computing, the situation is different because the mobility itself is limited and too many devices cannot be carried.

[0003] In mobile edge scenarios, applications are usually fixed, business software is known, all tasks have preset priorities, and high-priority tasks must be met first. Before executing a task, it is necessary to determine whether the computing power equipment carried is sufficient to meet the task requirements. Summary of the invention

[0004] The main problem that the present invention solves is how to provide an accurate computing resource configuration solution for complex tasks in a mobile edge environment, and provides a computing resource evaluation method and system for mobile edge application scenarios.

[0005] In order to solve the above technical problems, the technical solution adopted is:

[0006] A computing resource evaluation method for mobile edge application scenarios includes the following steps:

[0007] Step 1: Obtain device resource quotas and task requirement sets based on task timing relationships;

[0008] Step 2: Construct a task scenario description model based on the task temporal relationship according to the task requirement set;

[0009] Step 3: According to the equipment resource quota, the task scenario description model is solved based on the deep reinforcement learning method to obtain an evaluation plan for computing power resources.

[0010] Furthermore, the method of constructing a task scenario description model based on task temporal relationship according to the task requirement set is:

[0011] Step 2.1: Traverse the task requirement set , select the task requirement set from the pre-set historical experience sub-task set The sth task requirement Related subtask sets , , The number of task requirements for task demand concentration;

[0012] Step 2.2: Subtask Set Each subtask in the Query the resource usage corresponding to each subtask in the The corresponding resource usage set ;

[0013] Step 2.3: Let the sth task requirement The task scenario description model is expressed as:

[0014] ;

[0015] Step 2.4: Set the task requirements After traversal, merge the requirements of each task Task scenario description model , get the task requirement set Task scenario description model:

[0016]

[0017] in, A set of subtasks required by each task The union of For each subtask set The corresponding resource usage set The union of .

[0018] Furthermore, a task set that matches the task requirement set is selected from the pre-set historical experience sub-task set. The sth task requirement Related subtask sets The method is:

[0019] According to the functions, task attributes and constraints of the task requirements, subtasks matching the functions, task attributes and constraints of the task requirements are selected from the historical subtask set to form the subtask set of the task requirements.

[0020] Furthermore, a deep deterministic policy gradient algorithm is adopted based on the deep reinforcement learning method.

[0021] Furthermore, the deep deterministic policy gradient algorithm takes the device resource quota and the task scenario description model as input, and outputs the minimum requirements for the device resources to be carried and the deployment location, CPU and memory resources of each subtask.

[0022] Furthermore, the reward function of the deep deterministic policy gradient algorithm is ;

[0023]

[0024] in, Describe the model for the task scenario, For the The device resource quota within a time slot, wherein the time slot refers to a time period divided into equal parts for the entire duration of solving the task, and the time slots are arranged in a time sequence, Indicates that the task scenario description model is , in The device resource quota in a time slot is The reward function of the algorithm is , , Respectively represent the resource quota of the input device After that, the time slot The number of tasks in the solution, the number of discarded tasks, and the number of timed tasks. Weight is the weight of the weighted sum of latency and energy consumption in the reward function. is the reward weighting function based on the task scenario description model, For the time slot Internal, computing task scenario description model The collection The model parameters corresponding to the task requirements, For the time slot Internal, computing task scenario description model Middle The discount factor preset by the task requirement, the discard refers to the given device resource quota and time slot The number of tasks that the system cannot handle within the limit will be discarded without running. The timeout refers to a task that has started but cannot be successfully run due to insufficient resources or long execution time.

[0025] Furthermore, when matching subtasks according to task requirements, the principles to be followed are:

[0026] When the equipment resource quota is sufficient, different subtasks are selected based on the premise of maximum subtask processing efficiency and least constraints; when the equipment resource quota is insufficient, different subtasks are selected based on the premise of minimizing subtask resource occupation and giving priority to ensuring operation.

[0027] The present invention also provides a computing power resource evaluation system for mobile edge application scenarios, which is implemented using the steps of the computing power resource evaluation method for mobile edge application scenarios described above.

[0028] By adopting the above technical solution, the present invention has the following beneficial effects:

[0029] The present invention provides a computing resource evaluation method and system for mobile edge application scenarios. By constructing a task scenario description model based on task timing relationships, the historical subtask set and the historical resource usage set are fully utilized to effectively handle the complex relationship between task requirements, subtasks and resources, so that task resources can be allocated based on a deep reinforcement learning algorithm, thereby achieving optimal allocation of task resources under resource constraints, improving task execution efficiency and resource utilization. Compared with traditional methods, it is more adaptable to dynamically changing task scenarios, effectively improves resource utilization and task execution efficiency, and reduces system latency and energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a schematic diagram of the framework of the present invention. DETAILED DESCRIPTION

[0031] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] Figure 1 A specific embodiment of a computing resource evaluation method for a mobile edge application scenario of the present invention is shown, comprising the following steps:

[0033] Step 1: Obtain device resource quotas and task requirement sets based on task timing relationships.

[0034] Step 2: Construct a task scenario description model based on the task temporal relationship according to the task requirement set.

[0035] In this embodiment, considering that there are many elements in the task scenario and there are complex multi-dimensional relationships between most elements, this application introduces an ontology to construct a task scenario description model, abstractly defines the scene elements into different ontologies, and establishes related relationships between ontologies, mainly including influence, use, mutual exclusion, and collaboration. The following mainly constructs domain knowledge from three aspects: mission requirements, subtasks, and resources required for tasks, and describes them with a triple.

[0036] The method of constructing a task scenario description model based on the task temporal relationship for the task scenario is:

[0037] Step 2.1: Traverse the task requirement set , select the task requirement set from the pre-set historical experience sub-task set The sth task requirement Related subtask sets , , The number of task requirements for task demand concentration;

[0038] Step 2.2: Subtask Set Each subtask in the Query the resource usage corresponding to each subtask in the The corresponding resource usage set . In this embodiment, the historical resource usage set R and the historical subtask set are subtasks pre-set in historical data and resource usage sets corresponding to the subtasks one by one. Therefore, through the functions, task attributes and constraints of each task requirement, subtasks that match the functions, task attributes and constraints of the task requirements are selected from the historical subtask set to form the subtask set of the task requirement. The function of the task requirement generally refers to the goal of the task requirement. The task attributes include the type, priority, timing requirements, reliability requirements, etc. of the task requirement. The constraints include resource constraints (such as CPU and memory limitations), time constraints (such as task completion time windows), etc. After obtaining the subtask set of the task requirement, the historical resource usage corresponding to each subtask in the subtask set can be found from the historical resource usage set, thereby forming a resource usage set corresponding to the subtask set of the current task requirement.

[0039] Step 2.3: Let the sth task requirement The task scenario description model is expressed as:

[0040] .

[0041] In this embodiment, through the ternary model of task scenario description, the task requirement is matched with the subtask set matching the task requirement and the resource usage required to complete each subtask in the subtask set, so that complex problems can be described.

[0042] In this embodiment, the subtask set Ts is defined as:

[0043]

[0044] For subtask set The subtasks; The number of subtasks contained in the subtask set.

[0045] Step 2.4: Set the task requirements After traversal, merge the requirements of each task Task scenario description model , get the task requirement set Task scenario description model:

[0046]

[0047] in, A set of subtasks required by each task The union of is called the total subtask set of the task requirement set M. For each subtask set The corresponding resource usage set The union of is called the total resource usage set of the task requirement set M.

[0048] In this embodiment, the description of the task scenario description model D is mainly based on, for a specific complex task, a task requirement set with a time sequence relationship, the task requirements in the task requirement set are in chronological order, the total subtask set T is a subtask set Ts corresponding to the current task requirements matched according to the functions of each task requirement from the determined historical experience subtask set, R is a known resource usage set corresponding to each subtask, T and R are a relationship set corresponding to the subtask and resource usage pre-set according to historical experience, according to the task requirement set proposed by the user, and combined with the time sequence relationship between the task requirements, the corresponding subtask set is matched through each task requirement, when matching, functional screening is performed according to the function of each subtask, and a subtask set with specific functions is selected so that each subtask can meet the task requirements, and the resource usage set R is the resource requirement of the corresponding subtask obtained by querying the historical resource usage set for each subtask.

[0049] Step 3: According to the equipment resource quota, the task scenario description model is solved based on the deep reinforcement learning method to obtain an evaluation plan for computing power resources.

[0050] In this embodiment, the mapping problem between task requirements and subtasks is achieved by determining the amount of resources through a reinforcement learning method. A deep deterministic policy gradient algorithm DDPG (Deep Deterministic Policy Gradient algorithm) is used based on a deep reinforcement learning method.

[0051] In this embodiment, the deep deterministic policy gradient algorithm takes the quantitative device resource quota and the task scenario description model as input, and outputs the minimum requirements for the device resources to be carried and the deployment location, CPU and memory resources of each subtask.

[0052] In this embodiment, the deep deterministic policy gradient algorithm DDPG is used as one of the deep reinforcement learning DRL. The reward function of the deep deterministic policy gradient algorithm is ;

[0053]

[0054] in, Describe the model for the task scenario, For the The device resource quota within a time slot, wherein the time slot refers to a time period divided into equal parts for the entire duration of solving the task, and the time slots are arranged in a time sequence, Indicates that the task scenario description model is , in The device resource quota in a time slot is The reward function of the algorithm is , , Respectively represent the resource quota of the input device After that, the time slot The number of tasks in the solution, the number of discarded tasks, and the number of timed tasks. Weight is the weight of the weighted sum of latency and energy consumption in the reward function. is the reward weighting function based on the task scenario description model, For the time slot Internal, computing task scenario description model The collection The model parameters corresponding to the task requirements, For the time slot Internal, computing task scenario description model Middle The discount factor preset by the task requirement, the discard refers to the given device resource quota and time slot The number of tasks that the system cannot handle within the limit will be discarded without running. The timeout refers to a task that has started but cannot be successfully run due to insufficient resources or long execution time.

[0055] In this embodiment, by designing a reward function, when resources are limited, as many resources as possible are allocated to tasks per unit time, while minimizing the total system delay and energy consumption, and reducing the task loss and timeout rate. The resource allocation amount is gradually converged in the direction of a larger reward function value, and finally the optimal reward output (i.e., the device resource carrying plan) that meets the task operation requirements is output. According to the reward function, the subtasks in the task scenario description model Dt within the time slot t are obtained one by one, and then the reward weighting of each subtask is performed by the input device resource quota until the input device resource quota is maximized and effectively utilized.

[0056] In this embodiment, when matching subtasks according to task requirements, the principles followed are:

[0057] When the equipment resource quota is sufficient, different subtasks are selected based on the premise of the highest subtask processing efficiency and the least constraints; when the equipment resource quota is not sufficient, different subtasks are selected based on the premise of the least subtask resource occupation and priority guarantee of operation. Since the subtasks in the task scenario description model Dt are obtained one by one, the reward weighting of each subtask is performed by the input equipment resource quota. Therefore, under a certain amount of equipment resource quota, it is guaranteed that at least one subtask of the same functional type is running, meeting the most basic task requirements and the corresponding task timing relationship. On this basis, the task operation efficiency is maximized and the constraints are reduced, and the scalability is improved. Finally, the optimal equipment resource carrying plan that meets the task requirements is output, including the minimum requirements for the equipment resources to be carried, and the specific requirements of each subtask, such as the specific deployment location of each subtask, the requirements for CPU and memory resources, etc.

[0058] Taking the flood disaster rescue scene as an example, the specific example process is as follows:

[0059] 1. Task scenario modeling, including a set of task requirements with temporal relationships :

[0060] : Preliminary reconnaissance; Timing relationship: Executed first, providing basic information for subsequent tasks;

[0061] : Detailed reconnaissance and assessment; Sequential relationship: Following the initial reconnaissance, further determine the extent of the disaster and rescue needs;

[0062] : Emergency rescue; Timing relationship: Carried out simultaneously with or immediately after detailed reconnaissance and assessment, giving priority to rescuing trapped people;

[0063] : Temporary resettlement and material distribution; Timing relationship: Carry out simultaneously with emergency rescue to ensure the basic living needs of the affected people;

[0064] : Infrastructure assessment and emergency repair; Timing relationship: It will be carried out after emergency rescue and temporary resettlement work has achieved certain results, in order to restore the infrastructure in the disaster area.

[0065] The timing relationship is expressed as: .

[0066] Subtask set T and task required resource set R:

[0067] : Preliminary reconnaissance A; : CPU resources: 2 cores, memory resources: 4GB;

[0068] : Preliminary reconnaissance B; : CPU resources: 2 cores, memory resources: 8GB;

[0069] : Detailed reconnaissance and assessment; : CPU resources: 2 cores, memory resources: 4GB;

[0070] : Emergency rescue; : CPU resources: 4 cores, memory resources: 8GB;

[0071] : Temporary resettlement and material distribution; : CPU resources: 3 cores, memory resources: 6GB;

[0072] : Infrastructure assessment and repair; : CPU resources: 5 cores, memory resources: 4GB.

[0073] 2. The resource optimization allocation process based on the DDPG algorithm is as follows:

[0074] Suppose the rescue operation needs to be deployed in a temporary command center, which has certain computing resources that need to be reasonably allocated according to mission requirements.

[0075] (1) Input: Dt: scenario description model, including task requirements such as reconnaissance, rescue, temporary resettlement, and infrastructure repair. at: a certain amount of equipment resource quota.

[0076] (2) Resource allocation: Based on the timing relationship, first ensure and The resource allocation of the corresponding subtasks is used to obtain accurate information. The matching subtasks are and ,according to and The task attributes and resource usage are marked, and different subtasks are selected when resources are insufficient or sufficient in the future. Matching subtask selection Then according to the reward function ,distribute resources, matching the corresponding subtasks and ensure the timeliness of rescue operations. resources, matching subtasks Ensure the basic living needs of the affected people. If resources are insufficient, the principle of matching subtasks according to task requirements is followed. In the case of insufficient resources, different subtasks are selected based on the premise of minimizing the subtasks’ resource consumption and ensuring operation first. The corresponding subtask selects the one that takes up the least resources Finally, allocate according to the actual situation resources, select the corresponding subtask , and then gradually restore the infrastructure.

[0077] (3) Output result: optimal resource carrying plan. The specific subtasks are: , , , , , the corresponding total amount of device resources carried is: CPU has 16 cores and memory resources are 26GB.

[0078] Through the above steps, the rescue team can ensure that the required resources are effectively allocated according to the task sequence relationship under limited resources to achieve maximum efficiency and effectiveness of the rescue operation.

[0079] Through this embodiment, it is ensured that in the mobile edge environment, in the face of limited equipment resources, computing resources can be pre-evaluated and reasonably configured to meet the needs of task execution. Even if the equipment resources carried may not be sufficient to support the operation of the task, we can determine the equipment required to be carried based on the resource evaluation results before departure to ensure that the task can be smoothly executed in the mobile edge environment while achieving the goal of minimizing the total system delay and energy consumption. Using the DDPG algorithm, the resource allocation strategy can be adaptively adjusted according to the real-time environment and task requirements. Compared with traditional methods, it can better adapt to dynamically changing task scenarios, effectively improve resource utilization and task execution efficiency, and reduce system delay and energy consumption.

[0080] The present invention also provides a computing power resource evaluation system for mobile edge application scenarios, which is implemented using the steps of the computing power resource evaluation method for mobile edge application scenarios described above.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A computing resource evaluation method for mobile edge application scenarios, characterized in that: The following steps are involved: Step 1: Obtain device resource quotas and task requirement sets based on task timing relationships; Step 2: Construct a task scenario description model based on the task temporal relationship according to the task requirement set; Step 3: According to the equipment resource quota, the task scenario description model is solved based on the deep reinforcement learning method to obtain an evaluation plan for computing resources; The method of constructing a task scenario description model based on task timing relationship according to the task requirement set is: Step 2.1: Traverse the task requirement set , select the task requirement set from the pre-set historical experience sub-task set The sth task requirement Related subtask sets , , The number of task requirements for task demand concentration; Step 2.2: Subtask Set Each subtask in the Query the resource usage corresponding to each subtask in the The corresponding resource usage set ; Step 2.3: Let the sth task requirement The task scenario description model is expressed as: ; Step 2.4: Set the task requirements After traversal, merge the requirements of each task Task scenario description model , get the task requirement set Task scenario description model: in, A set of subtasks required by each task The union of For each subtask set The corresponding resource usage set The union of; Adopting deep deterministic policy gradient algorithm based on deep reinforcement learning method; The reward function of the deep deterministic policy gradient algorithm is ; in, Describe the model for the task scenario, For the The device resource quota within a time slot, wherein the time slot refers to a time period divided into equal parts for the entire duration of solving the task, and the time slots are arranged in a time sequence, Indicates that the task scenario description model is , in The device resource quota in a time slot is The reward function of the algorithm is , , Respectively represent the resource quota of the input device After that, the time slot The number of tasks in the solution, the number of discarded tasks, and the number of timed tasks. Weight is the weight of the weighted sum of latency and energy consumption in the reward function. is the reward weighting function based on the task scenario description model, For the time slot Internal, computing task scenario description model The collection The model parameters corresponding to the task requirements, For the time slot Internal, computing task scenario description model Middle The discount factor preset by the task requirement, the discard refers to the given device resource quota and time slot The number of tasks that the system cannot handle within the limit will be discarded without running. The timeout refers to a task that has started but cannot be successfully run due to insufficient resources or long execution time.

2. The computing resource evaluation method according to claim 1, characterized in that: Select the task requirement set from the pre-set historical experience sub-task set The sth task requirement Related subtask sets The method is: According to the functions, task attributes and constraints of the task requirements, subtasks matching the functions, task attributes and constraints of the task requirements are selected from the historical subtask set to form the subtask set of the task requirements.

3. The computing resource evaluation method according to claim 2, characterized in that: The deep deterministic policy gradient algorithm takes the device resource quota and the task scenario description model as input, and outputs the minimum requirements for the device resources to be carried and the deployment location, CPU and memory resources of each subtask.

4. The computing resource evaluation method according to claim 3, characterized in that: When matching subtasks according to task requirements, the principles to be followed are: When the equipment resource quota is sufficient, different subtasks are selected based on the premise of maximum subtask processing efficiency and minimum constraints; When the device resource quota is not sufficient, different subtasks are selected based on the premise of minimizing the resource usage and ensuring operation first.

5. A computing resource evaluation system for mobile edge application scenarios, characterized in that: The steps of a computing resource evaluation method for mobile edge application scenarios as described in any one of claims 1 to 4 are implemented.

Images