A slicing edge computing power management method

Abstract

The present invention discloses a slicing-based edge computing power management method, which can simplify the complexity of computing resource management by slicing edge node computing resources, and meet the computing power demands of computing service objects in various scenarios. The present invention reduces the complexity of computing power resource management by assembling distributed computing power resources into multiple computing power resource pools based on edge networks, and slicing and managing them according to computing power specifications. Different business scenarios apply for computing power use according to their own needs. Since edge computing power adopts slicing-based computing power resource management, it can quickly match needs and locate computing power resources in slices that meet the needs.

Description

Technical Field

[0001] The present invention belongs to the technical field related to edge computing, edge computing power management and scheduling between edge computing powers, and specifically relates to a slicing-type edge computing power management method. Background Art

[0002] Edge computing refers to the provision of edge intelligent services at the edge of the network close to objects or data sources. The computing resources that provide intelligent services are scattered in edge computing nodes. Due to various reasons (such as historical reasons, environmental factors, financial factors, etc.), the computing models deployed on edge computing nodes vary greatly, resulting in a wide variety of computing power types for edge computing nodes: for example, there are CPU, GPU and other computing power types, and there are X86, ARM and other CPU computing power types; even if they are of the same X86 type, they will be divided into different manufacturers, and the same manufacturer will have different periods and different series of CPU models.

[0003] In computing power management, it is necessary to distinguish and process different types of computing power. Various types of computing power increase the complexity of computing power management, which in turn increases the difficulty of computing power scheduling for edge computing. How to reasonably distinguish computing power types and accurately identify the capacity of resources of different computing power types is a difficult problem that must be faced in computing power management, and it is also the key to simplifying computing power scheduling.

[0004] In edge computing application scenarios, the service objects of intelligent services are diverse. The demand for computing power by service objects in different scenarios is also complex and changeable. For example, service objects in some scenarios have high requirements for GPU performance, but low requirements for CPU performance; service objects in some scenarios have high requirements for CPU performance, but no requirements for GPU; the changing computing power demand will generate demands for detailed management of computing power resources, and computing power resources need to be divided according to different scenarios to meet the computing power demand in various scenarios.

[0005] With the feature of supporting diversified service objects, the management of computing power of the whole system will become increasingly difficult. Often, the addition of a service object requires the increase of corresponding computing power division, and the impact on the original already divided computing power resources must be considered. Exploring suitable computing resources that can adaptively meet the service objects' demands for diverse computing resources can reduce the pressure of computing power management and conveniently connect the relationship between various computing power types, thereby improving the service efficiency of edge computing intelligent services.

[0006] The Chinese patent with publication number CN110995614A proposes a method and device for allocating computing power resources, which measures the processing time occupied by different business units as an indicator of computing power consumption, and dispatches the computing power resources of different business units according to the consumption of computing power resources of different business units, so as to realize flexible allocation of computing power resources of different business units and avoid the problem of low computing power resource utilization caused by physical division of computing power resources. The Chinese patent with publication number CN111240836A proposes a computing power resource management method, device, electronic device and storage medium, which platformizes computing power resources in a multi-cluster manner, allocates limited computing power resources to users in a smaller granularity, more flexible manner and more reasonably, and decouples users and computing power resources through strategies. On the one hand, computing power resources can be easily expanded and reduced by adding and reducing servers in the cluster and adding and reducing clusters, so as to realize reasonable allocation of computing power resources. On the other hand, computing power resources can be dispatched by adjusting strategies when the total amount of computing power resources remains unchanged, and computing power resources used by users can be reasonably allocated, thereby improving the flexibility of computing power resource allocation and resource utilization.

[0007] However, existing computing resources are relatively scattered, and computing management is extremely complex. At the same time, the demand for computing resources in different scenarios varies greatly. The above computing resource allocation and management methods still face difficulties in scheduling computing resources that meet scenario requirements. Summary of the invention

[0008] In view of the above, the present invention provides a sliced ​​edge computing power management method, which can simplify the complexity of computing resource management by slicing the edge node computing resources and meet the computing power demands of computing power service objects in various scenarios.

[0009] A slicing edge computing power management method comprises the following steps:

[0010] (1) Slicing edge computing power at multiple levels;

[0011] (2) Scheduling and managing edge computing resources based on slices;

[0012] (3) Multiple slices corresponding to the same edge computing power share the computing power resources and perform priority management;

[0013] (4) Provide external interfaces for slices and maintain slices in a unified manner, so that external parties can apply for computing resources based on demand and slice information.

[0014] Furthermore, the specific implementation method of step (1) is: first identify the type of each edge computing power, and classify and design the computing power in layers according to performance, type, and manufacturer model; then slice the edge computing power according to the stratification results, and place each computing power resource into the corresponding slices, so that the computing power resources in a single slice are distributed in each edge computing power pool, and the computing power resources in an edge computing power pool belong to multiple slices.

[0015] Furthermore, the specific method of hierarchical classification and slicing of computing power is as follows: first, all computing power is divided into three categories according to performance: high-performance computing power, medium-performance computing power, and low-performance computing power, which exist as the top level of computing power type and identify the computing power category; then the bottom level is divided according to computing power type and manufacturer model (according to CPU type, such as X86, ARM; according to GPU type, such as PowerVR, Adreno, Mali, Tegra; according to different manufacturers, such as Intel, AMD, Huawei, MediaTek, Qualcomm, NVIDIA, etc.), which exist as the bottom level of computing power type and identify the computing power category; after the distributed computing power resources are online, they are merged into the edge computing power pool, and at the same time, they are divided into corresponding slices according to the previous hierarchical classification. Therefore, hierarchical classification and slicing of computing power resources can better divide computing power into flexible management particles.

[0016] Slicing is to cut a large-scale module or function into thin slices. Because the original module is large or the function is complex, the sliced ​​module or function will be very simple and flexible, and easy to manage. For example, 5G slicing solves the network processing under complex scene functions. Slicing greatly simplifies the network functions under 5G and meets the needs of 5G applications more flexibly.

[0017] For example, if the slicing is divided into two types, (high performance + PowerVR) and (high performance + Intel), then the computing power of the high-performance GPU type is PowerVR and the manufacturer is Intel will be divided into these two slices. If the slicing is divided into one type, (high performance + PowerVR + Intel), then the computing power of the high-performance GPU type is PowerVR and the manufacturer is Intel will be divided into this one slice. In actual applications, it is often enough to divide the computing power into one slice, but there is also a situation where it is divided into two slices. When it is divided into two slices, it is necessary to solve the problem of logical management and coordination between slices.

[0018] The edge computing power pool is a form of pooling distributed computing power resources. The so-called resource pooling is to put the computing power of all devices into one pool and then distribute them uniformly.

[0019] Furthermore, the specific implementation method of step (2) is as follows: when a certain edge computing power is idle, the logical resources corresponding to the computing power in all slices containing the computing power are in an idle state, and these resources are marked as allocated for use; for a certain edge computing power, if the logical resources corresponding to the computing power in one of the slices containing the computing power are used, the logical resources corresponding to the computing power in other slices containing the computing power will also be marked as used and will no longer be allocated for use in other scenarios; for the newly-added edge computing power, the computing power resources are divided into corresponding slices according to the computing power type after it goes online, and can be divided into multiple slices as needed, and the newly-added computing power logical resources will be added to the corresponding slice; for the edge computing power that is deleted from the line, the logical resources corresponding to the computing power in all slices containing the computing power are deleted.

[0020] Furthermore, the specific implementation method of step (3) is as follows: when an edge computing power is added to multiple slices, the computing power resources will be shared within these slices by default. When the computing power resources are shared, different slices allocate computing power resources according to priority. If the slices are not prioritized or have equal priorities, the computing power resources are allocated among these slices according to the FIFO (First Input First Output) principle.

[0021] Furthermore, when computing resources are shared, applications for computing resources in different business scenarios need to be allocated within the slice. Priority is a strategy formulated for the computing resources themselves, that is, a certain scenario has a high priority right to use a certain slice. If no corresponding priority is specified, computing resources within the slice are allocated on a first-come, first-served basis.

[0022] Furthermore, the specific implementation method of step (4) is as follows: first, a unified usage interface is provided for the slices that contain computing power resources. When computing power resources are applied for, the computing power of the slices is allocated according to the computing power resource application requirements and priorities. For computing power resource applications, it is necessary to use the information of the computing power resource slices; then, the slices are maintained uniformly, and slice management is used instead of global computing power resource management. The computing power resources in the slices are maintained and managed on the edge computing cloud center side and the edge node side.

[0023] The present invention assembles distributed computing resources into multiple computing resource pools based on the edge network, and manages them by slicing them according to computing power specifications, thereby reducing the complexity of computing power resource management. Different business scenarios apply for computing power usage according to their own needs. Since edge computing power adopts slice-based computing power resource management, it can quickly match needs and locate computing power resources in slices that meet the needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a hierarchical schematic diagram of the slice-type edge computing power management system of the present invention.

[0025] Figure 2 This is a schematic diagram of the computing power slicing principle of the present invention.

[0026] Figure 3 This is a schematic diagram of the slice management process of the present invention.

[0027] Figure 4 Schematic diagram of the computing resource scheduling and allocation process of the present invention. DETAILED DESCRIPTION

[0028] In order to describe the present invention more specifically, the technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0029] like Figure 1 As shown, the present invention divides the processing framework of computing power resources into three layers: computing power layer, slicing layer, and application layer. The computing power layer and application layer are the management of computing power and scenario applications. The slicing layer is in the middle layer. Constructing the slicing layer is a model implementation of computing power resource slice management.

[0030] Computing resources are distributed at each edge computing end. The computing resources at the edge computing end form a huge computing pool through network connections. The computing pool is sliced ​​and the computing resources are divided into corresponding slices. The computing resource slices are managed to meet the needs of scenario applications for computing resources. When applications in computing demand scenarios apply for computing resources, they only pay attention to the information of computing resource slices. The computing resource slices satisfy the allocation of computing resources.

[0031] (1) Slice and layer.

[0032] Slicing management reduces the complexity of computing power management and simplifies computing power grouping / computing power sharing / computing power multi-scenario applications; computing power slicing is performed based on the computing power pool and relies on computing power type information for computing power slicing; computing power slices are provided to scene applications for use, and applications do not directly interact with the raw computing power, but only interact with computing power slices and only pay attention to computing power slices; computing power slices shield the computing power information inside the computing power pool to avoid cross-calling of the computing power pool by the application layer; the specific implementation process is as follows:

[0033] Step 1: Distributed computing power varies greatly in performance. It is divided into high-performance computing power, medium-performance computing power, and low-performance computing power according to computing power performance. It is classified according to the existing mainstream computing power. It exists as the top level of computing power type and identifies the computing power category.

[0034] Step 2: Divide the computing power type and manufacturer into the bottom layer, divide it into CPU type (such as X86, ARM, etc.), divide it into GPU type (such as PowerVR, Adreno, Mali, Tegra, etc.), and divide it into different manufacturers (such as Intel, AMD, Huawei, MediaTek, Qualcomm, NVIDIA, etc.). As the bottom layer of computing power type, it identifies the computing power category.

[0035] Step 3: After the distributed computing resources are online, they are merged into the edge computing resource pool and are divided into corresponding computing slices according to the previous layered slicing conditions.

[0036] Computing power resources are sliced ​​and classified in layers to better divide computing power into granularities that can be flexibly managed.

[0037] (2) Slicing principle and implementation.

[0038] like Figure 2 As shown in the figure, after slicing, the computing power is dispersed in different edge computing power pools. The slice contains information about all computing power resources. The specific implementation method is as follows:

[0039] Step 1: Slice the computing power type according to the above method. In this example, it is sliced ​​into three slices: slice 1, slice 2, and slice 3.

[0040] Step 2: After slicing, the resources in a single slice are distributed in each edge computing pool. The computing power in the edge computing pool belongs to multiple slices. Slice 1 contains the computing power resources in edge computing pools 1 / 2 / 3.

[0041] Step 3: The computing resources within the computing power slice are managed in a unified manner. The slices are isolated from each other, and the computing power resources belong to only one slice. For example, slice 1 and slice 2 will not have the same computing power resources of edge computing pool 1 at the same time. The slices can be re-allocated (can be reset through management).

[0042] The computing resources within the computing power slice are managed uniformly, providing a logical layer for computing power resources. The same slice is also adjusted according to the changes in the computing power pool.

[0043] (3) Slice management.

[0044] like Figure 3 As shown in the figure, slice management includes the full life cycle management of computing resources in the edge computing pool. The specific implementation process is as follows:

[0045] Step 1: Add computing resources. First, determine whether the computing resources belong to the existing computing resource slice type. If so, directly assign them to the computing resource slice and change the corresponding slice information. If not, create a new computing resource slice and add the new computing resources to the new computing slice.

[0046] Step 2: Delete computing resources. For newly added computing resources that are included in the computing resource slice, if the computing resource is deleted, the computing resource slice needs to be updated. If the deleted computing resource slice only contains computing resources, the computing resource slice will also be deleted.

[0047] Step 3: Computing resource abnormality. After the computing resources are added, the status of the computing resources needs to be continuously monitored. Once a computing resource abnormality occurs, it needs to be marked in the computing resource slice. The abnormal computing resources will not be allocated when the computing resources are applied for subsequently.

[0048] (4) Scheduling and allocation of computing resources.

[0049] like Figure 4 As shown in the figure, the scenario application selects computing resources in the appropriate slice through scheduling and allocates computing resources. The specific implementation process is as follows:

[0050] Step 1: Computing resource applications are matched with computing resource slices. The top-level category matching of computing resources is high performance / medium performance / low performance. The bottom-level category matching of computing resources is CPU type, GPU type, and manufacturer. If the computing resource application does not focus on a certain category, all slices of this category are used. If no matching computing slice is found, the computing resource application fails. This is because the computing resource application is not performed in accordance with the specifications of computing resource slices.

[0051] Step 2: Computing resource slice resource margin management. After Step 1, one or more computing resource slices are matched. For multiple computing resource slices, one of the computing resource slices is selected for computing resource allocation according to the priority logic established in advance.

[0052] Step 3: Allocate computing resources within the computing resource slice, apply for computing resources, mark this computing resource, and update the slice information.

[0053] The above description of the embodiments is to facilitate the understanding and application of the present invention by those skilled in the art. It is obvious that those skilled in the art can easily make various modifications to the above embodiments and apply the general principles described herein to other embodiments without creative work. Therefore, the present invention is not limited to the above embodiments, and improvements and modifications made to the present invention by those skilled in the art based on the disclosure of the present invention should be within the scope of protection of the present invention.

Claims

1. A slicing edge computing power management method, comprising the following steps: (1) Slicing edge computing power by dividing it into multiple layers. The specific implementation method is as follows: first, identify the type of each edge computing power, and classify and design the computing power by performance, type, and manufacturer model; then slice the edge computing power according to the stratification results, and place each computing power resource into the corresponding slice, so that the computing power resources in a single slice are distributed in each edge computing power pool, and the computing power resources in an edge computing power pool belong to multiple slices; The specific method of layering, classifying and slicing computing power is as follows: first, all computing power is divided into three categories according to performance: high-performance computing power, medium-performance computing power, and low-performance computing power. This exists as the top layer of computing power type and identifies the computing power category; then, the bottom layer is divided according to computing power type and manufacturer model. This exists as the bottom layer of computing power type and identifies the computing power category; after the distributed computing power resources are online, they are merged into the edge computing power pool and divided into corresponding slices according to the previous layered classification. (2) Scheduling and managing edge computing resources based on slices; (3) Multiple slices corresponding to the same edge computing power share the computing power resources and perform priority management; (4) Provide external interfaces for slices and maintain slices in a unified manner, so that external parties can apply for computing resources based on demand and slice information.

2. The slicing edge computing power management method according to claim 1, characterized in that: The specific implementation method of step (2) is as follows: when a certain edge computing power is idle, the logical resources corresponding to the computing power in all slices containing the computing power are in an idle state, and these resources are marked as allocated for use; for a certain edge computing power, if the logical resources corresponding to the computing power in one of the slices containing the computing power are used, the logical resources corresponding to the computing power in other slices containing the computing power will also be marked as used and will no longer be allocated for use in other scenarios; for the newly launched edge computing power, the computing power resources are divided into corresponding slices according to the computing power type after it goes online, and can be divided into multiple slices as needed, and the newly launched computing power logical resources will be added to the corresponding slice; for the edge computing power that is deleted from the line, the logical resources corresponding to the computing power in all slices containing the computing power are deleted.

3. The slicing edge computing power management method according to claim 1, characterized in that: The specific implementation method of step (3) is as follows: when an edge computing power is added to multiple slices, the computing power resources will be shared within these slices by default. When the computing power resources are shared, different slices allocate computing power resources according to priority. If the slices are not prioritized or have equal priorities, the computing power resources are allocated among these slices according to the FIFO principle.

4. The slicing edge computing power management method according to claim 3 is characterized in that: When computing resources are shared, applications for computing resources in different business scenarios need to be allocated within the slice. Priority is a strategy formulated for the computing resources themselves, that is, a certain scenario has a high priority right to use a certain slice. If no corresponding priority is specified, computing resources within the slice are allocated on a first-come, first-served basis.

5. The slicing edge computing power management method according to claim 1, characterized in that: The specific implementation method of step (4) is as follows: first, a unified usage interface is provided for the slices that contain computing power resources. When computing power resources are applied for, the computing power of the slices is allocated according to the computing power resource application requirements and priorities. For computing power resource applications, it is necessary to use the information of the computing power resource slices; then, the slices are maintained uniformly, and slice management is used instead of global computing power resource management. The computing power resources in the slices are maintained and managed on the edge computing cloud center side and the edge node side.

Images