Cloud edge collaborative computing power dynamic distribution system

Abstract

The invention relates to the field of resource scheduling, in particular to a cloud-side collaborative computing power dynamic distribution system, which comprises a task management module, a cloud-side collaborative computing power dynamic distribution module, a cloud-side collaborative computing power dynamic distribution module and a cloud-side collaborative computing power dynamic distribution module, the edge information management module is used for collecting information of edge equipment at an edge end and collecting calculation basic information and resource utilization rate of the edge equipment by issuing a calculation power test standard task; the computing power collaboration module is used for acquiring data of the edge information management module and distributing the tasks to the corresponding edge ends according to the computing power requirements of the tasks and the residual available computing power of the edge devices; the calculation feedback module is used for acquiring the isomerism and task calculation type of the edge equipment through the cloud edge communication module; the parameter dynamic optimization module sets a weight coefficient according to the heterogeneity of the edge end and the task calculation type, and the cloud edge collaborative computing power dynamic distribution system flexibly, efficiently and reliably carries out data distribution and optimizes a cloud edge collaborative scheduling distribution strategy.

Description

Technical Field

[0001] The present invention relates to the field of resource scheduling, and in particular to a cloud-edge collaborative computing power dynamic allocation system. Background Art

[0002] With the development of distributed computing and Internet of Things technology, cloud and edge are a common architecture. In the existing common cloud-edge collaborative computing power usage, there are generally a large number of edge devices, but the computing power of a single edge device is limited. In large-scale cloud-edge resource scheduling, there are problems such as unbalanced resource allocation and waste of algorithm resources.

[0003] Although there are many methods for adjusting resource allocation at present, the effects are not ideal. For example, in the patent document CN202410407649.9, a data allocation method and system based on cloud-edge collaboration, it is proposed to calculate the data delay in the cloud based on pre-stored data to predict the edge execution efficiency. It does not take into account that the computing power of the edge device may change in real time due to various factors, such as device status, task load, etc., and lacks a dynamic change processing flow of computing power; for example, in the patent document CN202310986681.2, an edge-cloud collaborative scheduling method and system, it is proposed to evaluate the processing efficiency of data based on the characteristic parameters, urgency, planned processing time, total computing power of the edge server, and maximum number of processes of the data to be scheduled, so as to ensure the reasonable allocation of computing power. However, the allocation is based on the task volume and planned processing time, without considering the heterogeneity of edge devices and the diversity of tasks, and the load balancing effect is poor. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a cloud-edge collaborative computing power dynamic allocation system, which flexibly, efficiently and reliably allocates data and optimizes the cloud-edge collaborative scheduling and allocation strategy.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A cloud-edge collaborative computing power dynamic allocation system includes a task management module, an edge information management module, a computing power collaboration module, a computing feedback module and a parameter dynamic optimization module.

[0007] The task management module is used for collecting tasks on the cloud and edge;

[0008] The edge information management module is used to collect information of the edge device at the edge end. The edge information management module is provided with a computing power detection submodule. The computing power detection submodule collects the basic computing information and resource utilization rate of the edge device by issuing a computing power test standard task;

[0009] The computing power coordination module is used for data acquisition of the edge information management module, and the computing power coordination module allocates the task to the corresponding edge end through the cloud-edge communication module according to the computing power requirement of the task and the remaining available computing power of the edge device;

[0010] The computing feedback module is used to obtain the heterogeneity and task computing type of the edge device through the cloud-edge communication module;

[0011] The parameter dynamic optimization module is used for data acquisition of the calculation feedback module. The parameter dynamic optimization module sets the weight coefficient according to the heterogeneity of the edge end and the task calculation type, and the computing power coordination module adjusts the task allocation of the edge device through the weight coefficient.

[0012] Furthermore, the task management module queues the tasks according to the urgency of the tasks.

[0013] Furthermore, the computing basic information includes device status, CPU size, GPU size, NPU size, and storage capacity; the resource utilization rate includes CPU utilization rate, GPU utilization rate, and NPU utilization rate; and the edge information management module calculates the CPU based on the computing basic information and resource utilization rate. _Score , GPU _Score , NPU_Sore benchmark scores.

[0014] Furthermore, the task allocation step of the computing power coordination module includes:

[0015] S1. Receive the task through the task management module to obtain the computing power requirements of the task;

[0016] S2. Obtain the remaining available computing power set of the edge device through the edge information management module;

[0017] S3. Obtain the data of step S1 and step S2 to calculate the expected computing power usage ratio based on the computing power demand and the remaining available computing power set;

[0018] S4. Allocate the task to the edge device that can support its operation according to the estimated computing power usage ratio, and select the edge device corresponding to the smallest estimated computing power usage ratio as the object of the task allocation;

[0019] S5. After the task is assigned to the edge device, the remaining available computing power set is updated.

[0020] Furthermore, in step S1, the computing power requirement of the task is: 1 , D 2 , D 3, …D m}; where m is the number of tasks to be executed;

[0021] In step S2, the remaining available computing power set is {F 1 , F 2 , F 3 , …, F n}; where n is the number of edge devices;

[0022] The estimated computing power usage ratio of step S3 is Among them, M j is the estimated usage ratio of computing power of task j; D j is the computing power required for task j; Fi is the remaining available computing power of edge device i.

[0023] Furthermore, in step S4, the optimal estimated computing power usage ratio of task j is:

[0024]

[0025] Among them, select the minimum value M j The corresponding edge device is used as the assignment object of the task; when When the remaining computing power of the edge end cannot bear the running of the task to be executed.

[0026] Further, in step S5, after the task j is assigned to the edge device i, the remaining computing power of the edge device i is Fi 1 =Fi-D j ; The remaining available computing power set obtained by updating is {F 1 , F 2 , F 3 ,···,Fi 1 , …, F n}.

[0027] Furthermore, the heterogeneity of the edge device in the computing feedback module includes processor type, network status, and storage capacity; the task computing type includes computing-intensive and IO-intensive.

[0028] Furthermore, the weight coefficient calculation method of the edge device i in the parameter dynamic optimization module is:

[0029]

[0030] Among them, Score i is the weight coefficient of edge device i, and the weight coefficient of edge device i is dynamically adjusted according to the real-time status and task characteristics of the edge device; ω CPU is the CPU weight of edge device i; CPU ScoreScore the CPU benchmark of edge device i; ω GPU is the GPU weight of edge device i; GPU Score Score GPU benchmarks for edge devices i; ω NPU is the NPU weight score of edge device i; NPU Score Scoring the NPU benchmark for edge devices i; ω Network Network weight points; Network Score Score network benchmarks; Storage Storage weight points; Score Scoring storage benchmarks;

[0031] The estimated execution efficiency E of task j on edge device i is calculated based on the weight coefficient of edge device i. ij , then

[0032]

[0033] Among them, TaskDemand j The computing power coordination module selects the largest E through a greedy algorithm to meet the computing requirements of task j. ij , until all tasks are assigned or there is no suitable edge device i.

[0034] Furthermore, the computing power coordination module is also used to reserve preset resources for status monitoring and status reporting of the edge device.

[0035] The beneficial effects of the present invention are:

[0036] The edge information management module can collect basic features and pre-analyze computing power, so as to accurately evaluate the total computing power and local computing power, achieve an in-depth understanding of the heterogeneity of different edge devices, and thus perform more refined resource management. Due to the difference in the actual cloud-edge resource scheduling scenario, there are more edge devices but the computing power of a single edge device is limited. The present invention uses the task management module to first prioritize high-priority tasks, and then uses the computing power coordination module to allocate tasks to the edge end with higher remaining computing power according to the computing power required for the task and the remaining computing power of the edge end, thereby achieving highly adaptive task allocation. The parameter dynamic optimization module introduces weights, considers the execution efficiency of algorithm tasks by different processors, and integrates the influence of network and storage, so that task allocation is more in line with actual hardware performance and improves the efficiency of task execution; at the same time, the parameter dynamic optimization module uses the dynamic evaluation and secondary evaluation mechanism of the computing power allocation process, so that the present invention can adapt to real-time changing resource status and continuous task requirements, and has better flexibility and adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a structural block diagram of a cloud-edge collaborative computing power dynamic allocation system according to a preferred embodiment of the present invention.

[0038] Figure 2 It is an allocation flow chart of a cloud-edge collaborative computing power dynamic allocation system according to a preferred embodiment of the present invention.

[0039] In the figure, 1-task management module, 2-edge information management module, 21-computing power detection sub-module, 3-computing power coordination module, 4-computing feedback module, 5-parameter dynamic optimization module, 6-cloud-edge communication module. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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.

[0041] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0042] Please also see Figure 1 and Figure 2 A cloud-edge collaborative computing power dynamic allocation system of a preferred embodiment of the present invention includes a task management module 1, an edge information management module 2, a computing power collaboration module 3, a computing feedback module 4 and a parameter dynamic optimization module 5.

[0043] The task management module 1 is used for collecting tasks on the cloud and edge; the task management module 1 queues the tasks according to their urgency.

[0044] In the cloud and edge, there are a large number of uncompleted tasks due to insufficient computing power. The task management module 1 of this embodiment is responsible for collecting tasks to be calculated in the cloud and edge, queuing them according to their urgency, so as to complete the calculation of the tasks as soon as possible.

[0045] The edge information management module 2 is used to collect information of edge devices at the edge end. The edge information management module 2 is provided with a computing power detection submodule 21. The computing power detection submodule 21 collects the basic computing information and resource utilization rate of edge devices by issuing computing power test standard tasks. At the same time, the edge information management module 2 can receive the computing operation progress and results of edge devices.

[0046] The computing basic information of this embodiment includes device status, CPU size, GPU size, NPU size, and storage capacity; resource utilization includes CPU utilization, GPU utilization, and NPU utilization; and the computing power detection submodule 21 obtains CPU based on the computing basic information and resource utilization. _Score , GPU _Score , NPU_Sore benchmark scores.

[0047] In the computing power detection submodule 21, the idle resources of a single device = device hardware resources × usage rate. The available disk storage size, CPU size, GPU size, and NPU size are calculated through the idle resources, which provides data basis for the subsequent cloud-edge collaboration and the allocation of computing tasks in the computing power collaboration module 3.

[0048] The edge information management module 2 can collect basic features and pre-analyze computing power, so as to accurately evaluate the total computing power and local computing power, achieve a deep understanding of the heterogeneity of different edge devices, and thus perform more refined resource management.

[0049] The computing power coordination module 3 is used for data acquisition of the edge information management module 2, and the computing power coordination module 3 allocates the task to the corresponding edge end through the cloud-edge communication module 6 according to the computing power requirements of the task and the remaining available computing power of the edge device.

[0050] In this embodiment, the task allocation step of the computing power coordination module 3 includes:

[0051] S1. Receive the task through the task management module 1 to obtain the computing power requirement of the task; in step S1, the computing power requirement of the task is: {D 1 , D 2 , D 3 , …D m}; where m is the number of tasks to be executed. In this embodiment, the computing power required for each task j to be executed is D j Based on historical experience.

[0052] S2. Obtain the remaining available computing power set of the edge device through the edge information management module 2; in step S2, the remaining available computing power set is {F 1 , F 2 , F 3 , …, F n}; where n is the number of edge devices.

[0053] S3. Obtain the data of step S1 and step S2 to calculate the expected computing power usage ratio based on the computing power demand and the remaining available computing power set; the expected computing power usage ratio of step S3 is Among them, M jis the estimated usage ratio of computing power of task j; D j is the computing power required for task j; Fi is the remaining available computing power of edge device i.

[0054] S4. Allocate the task to the edge device that can support its operation based on the expected computing power usage ratio, and select the corresponding edge device with the smallest expected computing power usage ratio as the task allocation object.

[0055] In step S4, the optimal estimated computing power usage ratio of task j is:

[0056]

[0057] Among them, select the minimum value M j The corresponding edge device is used as the task assignment object; when When the remaining computing power at the edge is insufficient to handle the execution of tasks to be executed.

[0058] S5. Update the remaining available computing power set after the task is assigned to the edge device. In step S5, after task j is assigned to edge device i, the remaining computing power of edge device i is Fi 1 =Fi-D j ; The remaining available computing power set obtained by updating is {F 1 , F 2 , F 3 ,···,Fi 1 , …, F n}.

[0059] Due to the difference in actual cloud-edge resource scheduling scenarios, there are more edge devices but the computing power of a single edge device is limited. In this paper, the task management module 1 is used to first prioritize high-priority tasks, and then the computing power coordination module 3 is used to allocate tasks to the edge end with higher remaining computing power according to the computing power required for the tasks and the remaining computing power of the edge end, thereby achieving highly adaptable task allocation.

[0060] The computing power coordination module 3 is responsible for allocating computing tasks based on the basic information of edge devices and the remaining computing power, and allocating appropriate tasks to the edge. The goal is to be able to process more valuable tasks on the edge devices at the same time, fully mobilize the computing power of the edge, and improve the efficiency of task execution without wasting computing power. The computing power allocation mechanism matches the computing power of a single edge device, so that the edge can process more tasks.

[0061] In this embodiment, the computing power coordination module 3 is also used to reserve preset resources for status monitoring and status reporting of edge devices. In actual application scenarios, various device problems may occur in edge devices, such as network problems that affect task delivery and edge status synchronization, unknown anomalies that cause task interruption during operation, and parallel tasks on the edge affecting each other. Therefore, when the edge device is running, a relatively independent operating environment should be considered, and certain resources should be reserved for device status monitoring and reporting to ensure stable operation of the edge device.

[0062] The computing feedback module 4 is used to obtain the heterogeneity of edge devices and the task computing type through the cloud-edge communication module 6. The heterogeneity of edge devices in the computing feedback module 4 includes processor type, network status, and storage capacity; the task computing type includes computing-intensive and IO-intensive.

[0063] A timed active reporting mechanism is set up on the edge to report the latest device messages to the computing feedback module 4 through the cloud-edge communication module 6. The main device information includes: device status, processor (CPU, GPU, NPU) size, storage, etc. The computing power is quantified based on the device's processor. The quantification method is based on the processor information and usage, and the CPU utilization rate, GPU memory occupancy, and NPU utilization rate are calculated.

[0064] The parameter dynamic optimization module 5 is used to obtain data from the calculation feedback module 4. The parameter dynamic optimization module 5 sets the weight coefficient according to the heterogeneity of the edge end and the task calculation type, and the computing power coordination module 3 adjusts the task allocation of the edge device through the weight coefficient.

[0065] The weight coefficient calculation method of edge device i in parameter dynamic optimization module 5 is:

[0066]

[0067] Among them, Score i is the weight coefficient of edge device i, and the weight coefficient of edge device i is dynamically adjusted according to the real-time status and task characteristics of the edge device; ω CPU is the CPU weight of edge device i; CPU Score Score the CPU benchmark of edge device i; ω GPU is the GPU weight of edge device i; GPU Score Score GPU benchmarks for edge devices i; ω NPU is the NPU weight score of edge device i; NPU Score Scoring the NPU benchmark for edge devices i; ω Network Network weight points; Network Score Score network benchmarks; StorageStorage weight points; Score Scoring storage benchmarks;

[0068] The estimated execution efficiency E of task j on edge device i is calculated based on the weight coefficient of edge device i. ij , then

[0069]

[0070] Among them, TaskDemand j To meet the computing requirements of task j, computing power coordination module 3 selects the largest E through a greedy algorithm. ij , until all tasks are assigned or there is no suitable edge device i.

[0071] Repeat the project allocation action, update the weight coefficient and comprehensive score according to the task execution feedback and device status to complete the task allocation. The edge information management module 2 of this embodiment can obtain the computing power weight of the edge device, and the computing power coordination module 3 adjusts the computing power weight according to the weight coefficient.

[0072] In this embodiment, since most of the time there are many tasks and insufficient computing power at the edge, it is necessary to consider the weight coefficients of tasks on different devices based on the heterogeneity of edge devices and the computing type of the tasks to improve execution efficiency. Therefore, this embodiment introduces an evaluation standard based on the weight coefficient, which will be determined based on the device's processor type, network status, storage capacity, and computing type of the task.

[0073] By continuously adjusting the weights dynamically, the weight coefficients will be dynamically adjusted according to the real-time status of the device and the characteristics of the task. For example, if the current CPU usage of an edge device is high, the processor type weight ω CPU The allocation is then reduced proportionally to reduce the allocation of CPU tasks to the device.

[0074] The parameter dynamic optimization module 5 of this embodiment introduces weights, takes into account the execution efficiency of different processors on algorithm tasks, and integrates the impact of network and storage, so that task allocation is more in line with actual hardware performance and improves the efficiency of task execution. At the same time, the parameter dynamic optimization module 5 uses the dynamic evaluation and secondary evaluation mechanism of the computing power allocation process to enable this embodiment to adapt to real-time changing resource status and continuous task requirements, and has better flexibility and adaptability.

[0075] like Figure 2 As shown, the allocation method of the cloud-edge collaborative computing power dynamic allocation system of this embodiment includes the following steps:

[0076] A1. Collect cloud and edge tasks through task management module 1; collect edge device information through edge information management module 2;

[0077] A2. Detect the computing power of the edge device in the edge information management module 2;

[0078] A3. In the computing power coordination module 3, according to the computing power requirements of the task and the remaining available computing power of the edge device, the task is assigned to the corresponding edge end through the cloud-edge communication module 6;

[0079] A4. Feedback the heterogeneity of edge devices and task computing types to the parameter dynamic optimization module 5 through the computing feedback module 4 and using the cloud-edge communication module 6;

[0080] A5. The parameter dynamic optimization module 5 sets the weight coefficient according to the heterogeneity of the edge end and the task calculation type, and the computing power coordination module 3 adjusts the task allocation of the edge device through the weight coefficient.

[0081] This embodiment analyzes and evaluates the computing power of edge devices, and achieves a more efficient and reasonable computing power scheduling and allocation effect through a resource evaluation mechanism in multiple scenarios. It can give full play to the remaining computing power of edge devices, so that a large number of tasks to be calculated can be calculated synchronously on multiple edges, thereby improving computing efficiency. Compared with existing methods based on predicted computing power and static allocation strategies, this embodiment can provide a more flexible, efficient and reliable data allocation solution, thereby optimizing the cloud-edge collaborative scheduling and allocation strategy.

Claims

1. A cloud-edge collaborative computing power dynamic allocation system, characterized in that: It includes a task management module (1), an edge information management module (2), a computing power coordination module (3), a computing feedback module (4) and a parameter dynamic optimization module (5). The task management module (1) is used for collecting tasks on the cloud and edge terminals; The edge information management module (2) is used for collecting information of the edge device at the edge end, and the edge information management module (2) is provided with a computing power detection submodule (21). The computing power detection submodule (21) collects basic computing information and resource utilization rate of the edge device by issuing a computing power test standard task; The computing power coordination module (3) is used for data acquisition of the edge information management module (2), and the computing power coordination module (3) allocates the task to the corresponding edge end through the cloud-edge communication module (6) according to the computing power requirement of the task and the remaining available computing power of the edge device; The computing feedback module (4) is used to obtain the heterogeneity of the edge device and the task computing type through the cloud-edge communication module (6); The parameter dynamic optimization module (5) is used for data acquisition of the calculation feedback module (4), the parameter dynamic optimization module (5) sets a weight coefficient according to the heterogeneity of the edge end and the task calculation type, and the computing power coordination module (3) adjusts the task allocation of the edge device through the weight coefficient.

2. According to claim 1, a cloud-edge collaborative computing power dynamic allocation system is characterized by: The task management module (1) queues the tasks according to their urgency.

3. According to claim 1, a cloud-edge collaborative computing power dynamic allocation system is characterized by: The computing basic information includes device status, CPU size, GPU size, NPU size, and storage capacity; the resource utilization rate includes CPU utilization rate, GPU utilization rate, and NPU utilization rate; and the edge information management module (2) calculates the CPU based on the computing basic information and resource utilization rate. _Score , GPU _Score , NPU_Sore benchmark scores.

4. The cloud-edge collaborative computing power dynamic allocation system according to claim 3 is characterized by: The task allocation step of the computing power coordination module (3) includes: S1. Receiving the task through the task management module (1) to obtain the computing power requirement of the task; S2. Obtaining the remaining available computing power set of the edge device through the edge information management module (2); S3. Obtain the data of step S1 and step S2 to calculate the expected computing power usage ratio based on the computing power demand and the remaining available computing power set; S4. Allocate the task to the edge device that can support its operation according to the estimated computing power usage ratio, and select the edge device corresponding to the smallest estimated computing power usage ratio as the object of the task allocation; S5. After the task is assigned to the edge device, the remaining available computing power set is updated.

5. The cloud-edge collaborative computing power dynamic allocation system according to claim 4 is characterized by: In step S1, the computing power requirement of the task is: {D1, D2, D3, ... D m }; where m is the number of tasks to be executed; In step S2, the remaining available computing power set is {F1, F2, F3, ..., F n }; where n is the number of edge devices; The estimated computing power usage ratio of step S3 is Among them, M j is the estimated usage ratio of computing power of task j; D j is the computing power required for task j; Fi is the remaining available computing power of edge device i.

6. A cloud-edge collaborative computing power dynamic allocation system according to claim 5, characterized in that: In step S4, the optimal estimated computing power usage ratio of task j is: Among them, select the minimum value M j The corresponding edge device is used as the assignment object of the task; when When the remaining computing power of the edge end cannot bear the running of the task to be executed.

7. The cloud-edge collaborative computing power dynamic allocation system according to claim 5 is characterized by: In step S5, after the task j is assigned to the edge device i, the remaining computing power of the edge device i is Fi1 = Fi-D j ; The remaining available computing power set obtained by updating is {F1, F2, F3, ···, Fi1, …, F n }.

8. The cloud-edge collaborative computing power dynamic allocation system according to claim 3 is characterized by: The heterogeneity of the edge devices in the computing feedback module (4) includes processor type, network status, and storage capacity; the task computing type includes computing intensive and IO intensive.

9. The cloud-edge collaborative computing power dynamic allocation system according to claim 8, characterized in that: The weight coefficient calculation method of the edge device i in the parameter dynamic optimization module (5) is: Among them, Score i is the weight coefficient of edge device i, and the weight coefficient of edge device i is dynamically adjusted according to the real-time status and task characteristics of the edge device; ω CPU is the CPU weight of edge device i; CPU Score Score the CPU benchmark of edge device i; ω GPU is the GPU weight of edge device i; GPU Score Score GPU benchmarks for edge devices i; ω NPU is the NPU weight score of edge device i; NPU Score Scoring the NPU benchmark for edge device i;ω Network Network weight points; Network Score Score network benchmarks; Storage Storage weight points; Score Scoring storage benchmarks; The estimated execution efficiency E of task j on edge device i is calculated based on the weight coefficient of edge device i. ij , then Among them, TaskDemand j The computing power coordination module (3) selects the largest E through a greedy algorithm to meet the computing requirements of task j. ij , until all tasks are assigned or there is no suitable edge device i.

10. The cloud-edge collaborative computing power dynamic allocation system according to claim 1, characterized in that: The computing power coordination module (3) is also used to reserve preset resources for status monitoring and status reporting of the edge device.

Images