Heterogeneous resource scheduling method and system
Through the combination of deep learning network and classifier, accurate analysis and real-time scheduling of heterogeneous resources are achieved, the problem of inaccurate resource allocation in the MEC system is solved, and resource utilization and application performance are improved.
Patent Information
- Application Number
- CN201911237780.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2019-12-06
- Publication Date
- 2024-12-31
- Estimated Expiration
- 2039-12-06
AI Technical Summary
The existing MEC system cannot accurately allocate heterogeneous resources, resulting in the inability to meet the needs of edge computing applications for multi-class heterogeneous computing resources, and requires manual resource management and allocation.
Key features of edge applications are extracted through deep learning networks, classifiers are used to match resources, and combined with regression prediction of resource utilization, to achieve accurate analysis, allocation and real-time scheduling of heterogeneous resources.
The precise allocation of heterogeneous resources is achieved, resource utilization is improved, multi-class heterogeneous computing resource requirements for edge computing applications, and manual intervention is reduced.
Smart Images
Figure CN112925634B_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of edge computing, and in particular to a heterogeneous resource scheduling method and system. Background Art
[0002] OpenStack (a cloud computing platform jointly developed by Rackspace and NASA) manages heterogeneous resources through Cyborg. For MEC (Mobile Edge Computing) systems, the key point of the existing ETSI (European Telecommunications Standards Institute) solution is that heterogeneous computing resource management platforms represented by Cyborg only perform simple life cycle management on various heterogeneous resources themselves. At present, manual methods are still needed to understand the needs of edge computing applications for heterogeneous resources and manually estimate the allocation of heterogeneous resources to various applications.
[0003] Since the current MEC system does not support the precise allocation of various types of heterogeneous resources, it cannot meet the needs of edge computing applications for multiple types of heterogeneous computing resources. Summary of the invention
[0004] The present disclosure provides a solution for accurate analysis, allocation and real-time scheduling of heterogeneous resources.
[0005] According to a first aspect of an embodiment of the present disclosure, a heterogeneous resource scheduling method is provided, comprising: in response to a resource request sent by each edge application on a mobile edge computing platform, parsing an acceleration requirement file of the edge application so as to pre-allocate the required heterogeneous resources for the edge application; extracting key features of each edge application using a deep learning network; matching the key features of each edge application with the required heterogeneous resources using a classifier so as to obtain the ratio of various types of heterogeneous resources required by each edge application; determining a real-time application resource allocation strategy according to the ratio of various types of heterogeneous resources required by each edge application; performing regression prediction on the resource utilization of each edge application so as to obtain an overall heterogeneous resource allocation strategy within a predetermined time range; and uniformly allocating and scheduling heterogeneous resources according to the real-time application resource allocation strategy and the overall heterogeneous resource allocation strategy.
[0006] In some embodiments, parsing the acceleration requirement file of the edge application includes: calling the acceleration requirement file of the edge application through a requirement file library; parsing the acceleration requirement file using a requirement parser to obtain a pre-allocation strategy for heterogeneous resources required for the edge application; and pre-allocating the required heterogeneous resources for the edge application according to the pre-allocation strategy.
[0007] In some embodiments, performing regression prediction on the resource utilization of each edge application includes: using a gradient boosting decision tree (GBDT) regression predictor to perform regression prediction on the resource utilization of each edge application to obtain the distribution of resource requirements of each edge application within a predetermined time range; and determining the overall allocation strategy for heterogeneous resources within the predetermined time range based on the obtained distribution.
[0008] In some embodiments, the deep learning network is a convolutional neural network (CNN) feature extraction network; and the classifier is a Softmax classifier.
[0009] According to a second aspect of an embodiment of the present disclosure, a heterogeneous resource scheduling system is provided, including: a pre-allocation module, configured to respond to a resource request sent by each edge application on a mobile edge computing platform, parse the acceleration requirement file of the edge application, so as to pre-allocate the required heterogeneous resources for the edge application; a real-time allocation module, configured to use a deep learning network to extract key features of each edge application, use a classifier to match the key features of each edge application with the required heterogeneous resources, so as to obtain the ratio of various types of heterogeneous resources required by each edge application, and determine the real-time application resource allocation strategy according to the ratio of various types of heterogeneous resources required by each edge application; an overall allocation module, configured to perform regression prediction on the resource utilization of each edge application to obtain the overall heterogeneous resource allocation strategy within a predetermined time range; a heterogeneous resource scheduling module, configured to uniformly allocate and schedule heterogeneous resources according to the real-time application resource allocation strategy and the overall heterogeneous resource allocation strategy.
[0010] In some embodiments, the pre-allocation module is configured to call the acceleration requirement file of the edge application through the requirement file library, parse the acceleration requirement file using the requirement parser to obtain the pre-allocation strategy of the heterogeneous resources required by the edge application, and pre-allocate the required heterogeneous resources of the edge application according to the pre-allocation strategy.
[0011] In some embodiments, the overall allocation module is configured to use a gradient boosting decision tree (GBDT) regression predictor to perform regression prediction on the resource utilization of each edge application to obtain the distribution of resource requirements of each edge application within a predetermined time range, and determine the overall allocation strategy of heterogeneous resources within the predetermined time range based on the obtained distribution.
[0012] In some embodiments, the deep learning network is a convolutional neural network (CNN) feature extraction network; and the classifier is a Softmax classifier.
[0013] According to a third aspect of an embodiment of the present disclosure, a heterogeneous resource scheduling system is provided, comprising: a memory configured to store instructions; a processor coupled to the memory, the processor being configured to execute a method as described in any of the above embodiments based on the instructions stored in the memory.
[0014] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and when the instructions are executed by a processor, the method involved in any of the above embodiments is implemented.
[0015] Other features and advantages of the present disclosure will become apparent from the following detailed description of exemplary embodiments of the present disclosure with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] The accompanying drawings, which constitute a part of the specification, illustrate embodiments of the present disclosure and, together with the description, serve to explain the principles of the present disclosure.
[0017] The present disclosure may be more clearly understood from the following detailed description with reference to the accompanying drawings, in which:
[0018] Figure 1 is a flow chart of a heterogeneous resource scheduling method according to an embodiment of the present disclosure;
[0019] Figure 2 is a structural diagram of a heterogeneous resource scheduling system according to an embodiment of the present disclosure;
[0020] Figure 3 is a structural schematic diagram of a heterogeneous resource scheduling system according to another embodiment of the present disclosure;
[0021] Figure 4 It is a schematic diagram of the MEC system workflow according to an embodiment of the present disclosure.
[0022] It should be understood that the size of each part shown in the accompanying drawings is not drawn according to the actual proportional relationship. In addition, the same or similar reference numerals represent the same or similar components. DETAILED DESCRIPTION
[0023] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. The description of the exemplary embodiments is merely illustrative and is in no way intended to limit the present disclosure and its application or use. The present disclosure can be implemented in many different forms and is not limited to the embodiments described herein. These embodiments are provided to make the present disclosure thorough and complete and to fully express the scope of the present disclosure to those skilled in the art. It should be noted that unless otherwise specifically stated, the relative arrangement of the components and steps, the composition of the materials, and the numerical values set forth in these embodiments should be interpreted as being merely exemplary and not as limiting.
[0024] The words “include” or “comprising” and the like used in the present disclosure mean that the elements preceding the words include the elements listed after the words, and do not exclude the possibility of also including other elements.
[0025] All terms (including technical terms or scientific terms) used in the present disclosure have the same meanings as those understood by ordinary technicians in the field to which the present disclosure belongs, unless otherwise specifically defined. It should also be understood that terms defined in general dictionaries, such as general dictionaries, should be interpreted as having meanings consistent with their meanings in the context of the relevant technology, and should not be interpreted in an idealized or extremely formal sense, unless explicitly defined herein.
[0026] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered as part of the specification.
[0027] Figure 1 The following is a flow chart of a heterogeneous resource scheduling method according to an embodiment of the present disclosure. In some embodiments, the following steps of the heterogeneous resource scheduling method are executed by a heterogeneous resource scheduling system.
[0028] In step 101, in response to a resource request sent by each edge application on a mobile edge computing platform, an acceleration requirement file of the edge application is parsed to pre-allocate required heterogeneous resources of the edge application.
[0029] In some embodiments, the acceleration requirement file of the edge application is called through the requirement file library, and the acceleration requirement file is parsed by the requirement parser to obtain the pre-allocation strategy of the heterogeneous resources required by the edge application. The required heterogeneous resources of the edge application are pre-allocated according to the pre-allocation strategy.
[0030] In step 102, a deep learning network is used to extract key features of each edge application.
[0031] In some embodiments, the deep learning network is a CNN (Convolutional Neural Network) feature extraction network.
[0032] In step 103, a classifier is used to match the key features of each edge application with the required heterogeneous resources to obtain the ratio of various types of heterogeneous resources required by each edge application.
[0033] In some embodiments, the classifier is a Softmax classifier.
[0034] In step 104, a real-time application resource allocation strategy is determined according to the ratio of various types of heterogeneous resources required by each edge application.
[0035] In step 105, a regression prediction is performed on the resource utilization of each edge application to obtain an overall heterogeneous resource allocation strategy within a predetermined time range.
[0036] In some embodiments, the resource utilization of each edge application is stored in a resource monitoring file library, and a GBDT (Gradient Boosting Decision Tree) regression predictor is used to perform regression prediction on the resource utilization of each edge application to obtain the distribution of resource requirements of each edge application within a predetermined time range, and determine the overall allocation strategy of heterogeneous resources within the predetermined time range based on the obtained distribution.
[0037] In step 106, the heterogeneous resources are uniformly allocated and scheduled according to the real-time application resource allocation strategy and the overall heterogeneous resource allocation strategy.
[0038] In the heterogeneous resource scheduling method provided in the above embodiment of the present disclosure, by acquiring and analyzing application demand information to obtain a pre-allocation strategy for resources required for edge applications, a deep learning algorithm is used to extract key features of various applications, thereby achieving precise analysis, monitoring and scheduling of heterogeneous computing resources, thereby maximizing resource utilization and edge application capabilities.
[0039] Figure 2 FIG. 1 is a schematic diagram of a heterogeneous resource scheduling system according to an embodiment of the present disclosure. Figure 2 As shown, the heterogeneous resource scheduling system includes a pre-allocation module 21, a real-time allocation module 22, an overall allocation module 23 and a heterogeneous resource scheduling module 24.
[0040] The pre-allocation module 21 is configured to respond to resource requests sent by each edge application on the mobile edge computing platform and parse the acceleration requirement file of the edge application so as to pre-allocate the required heterogeneous resources of the edge application.
[0041] In some embodiments, the pre-allocation module 21 calls the acceleration requirement file of the edge application through the requirement file library, and uses the requirement parser to parse the acceleration requirement file to obtain the pre-allocation strategy of the heterogeneous resources required by the edge application. The required heterogeneous resources of the edge application are pre-allocated according to the pre-allocation strategy.
[0042] The real-time allocation module 22 is configured to use a deep learning network to extract the key features of each edge application, and use a classifier to match the key features of each edge application with the required heterogeneous resources to obtain the ratio of various heterogeneous resources required by each edge application, and determine the real-time application resource allocation strategy based on the ratio of various heterogeneous resources required by each edge application.
[0043] In some embodiments, the deep learning network is a CNN feature extraction network, and the classifier is a Softmax classifier.
[0044] The overall allocation module 23 is configured to perform regression prediction on the resource utilization of each edge application to obtain an overall heterogeneous resource allocation strategy within a predetermined time range.
[0045] In some embodiments, the overall allocation module 23 stores the resource utilization of each edge application into a resource monitoring file library, and uses a GBDT regression predictor to perform regression prediction on the resource utilization of each edge application to obtain the distribution of resource requirements of each edge application within a predetermined time range, and determine the overall allocation strategy of heterogeneous resources within the predetermined time range based on the obtained distribution.
[0046] The heterogeneous resource scheduling module 24 is configured to uniformly allocate and schedule heterogeneous resources according to the real-time application resource allocation strategy and the overall heterogeneous resource allocation strategy.
[0047] Figure 3 FIG. 1 is a schematic diagram of a heterogeneous resource scheduling system according to another embodiment of the present disclosure. Figure 3 As shown, the heterogeneous resource scheduling system includes a memory 31 and a processor 32 .
[0048] The memory 31 is used to store instructions. The processor 32 is coupled to the memory 31. The processor 32 is configured to execute the instructions stored in the memory to implement the following. Figure 1 The method of any one of the embodiments.
[0049] like Figure 3 As shown, the heterogeneous resource scheduling system further includes a communication interface 33 for information exchange with other devices. At the same time, the system also includes a bus 34, through which the processor 32, the communication interface 33, and the memory 31 communicate with each other.
[0050] The memory 31 may include high-speed RAM (Random Access Memory) and may also include NVM (Non-Volatile Memory). For example, at least one disk storage. The memory 31 may also be a storage array. The memory 31 may also be divided into blocks, and the blocks may be combined into virtual volumes according to certain rules.
[0051] In addition, the processor 32 may be a central processing unit, or may be an ASIC (Application Specific Integrated Circuit), or may be configured to implement one or more integrated circuits of the embodiments of the present disclosure.
[0052] The present disclosure also provides a computer-readable storage medium. The computer-readable storage medium stores computer instructions, which are executed by a processor to implement the following Figure 1 The method of any one of the embodiments.
[0053] Figure 4 FIG. 1 is a schematic diagram of the working process of the MEC system according to an embodiment of the present disclosure. Figure 4 As shown, the heterogeneous resource scheduling system involved in the present disclosure is set in the VIM (Virtualized Infrastructure Manager) of the MEC system.
[0054] It should be noted here that Nova and Glance are the main modules in OpenStack. Nova is the controller of the cloud organization, which provides a tool to deploy the cloud, including running instances, managing networks, and controlling users. Glance is a storage, query and retrieval system for virtual machine images. In addition, NFVI (Network Functions Virtualization Infrastructure) is the network function virtualization infrastructure, MEP (ME Platform) is the mobile edge platform, MEAO (Mobile Edge Application Orchestrator) is the mobile edge application orchestrator, MEPM (MEP Manager)-V (NFV) is the network function virtualization of the mobile edge platform manager, and NFVM (NFV Manager) is the network function virtualization manager.
[0055] The corresponding workflow is as follows:
[0056] 1. Select a suitable computing node and use the image provided by ME-APP to run the virtual machine;
[0057] 2. Analyze the acceleration resources, performance and priority of the ME-APP acceleration requirement description file;
[0058] 3. According to the analysis results, call the required ME-APP acceleration software;
[0059] 4. Perform resource discovery and pool management on FPGA (Field-Programmable Gate Array) or GPU (Graphics Processing Unit); burn the acceleration software of ME-App into the target acceleration hardware instance according to the needs;
[0060] 5. Bind the burned acceleration hardware instance to ME-App;
[0061] 6.Nova sends the currently running application information;
[0062] 7. The CNN feature extraction network extracts application features to obtain various resources and resource allocation information required by the application;
[0063] 8. Send the application-resource status utilization information to the resource monitoring file library for storage;
[0064] 9. Through the application-resource status information, the GBDT regression predictor is used to predict the overall distribution of resource requirements of various applications over a period of time;
[0065] 10. Send the real-time application resource allocation strategy and the overall heterogeneous resource pre-allocation strategy to the heterogeneous resource pre-allocator for implementation.
[0066] The present disclosure provides an acceleration requirement description file and provides an initial allocation strategy for edge applications according to the requirement file. Through deep learning algorithms, real-time and refined resource allocation strategies are provided for edge applications. In addition, by monitoring the use of heterogeneous resources, it is used to count and predict the overall heterogeneous resource demand rules and obtain the pre-allocation strategy of heterogeneous resources. The system automatically generates resource allocation strategies, provides resources on demand in a refined manner, and significantly improves the overall resource utilization rate.
[0067] In some embodiments, the above-mentioned functional modules can be implemented as a general-purpose processor, a programmable logic controller (PLC), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components or any appropriate combination thereof for performing the functions described in the present disclosure.
[0068] So far, the embodiments of the present disclosure have been described in detail. In order to avoid obscuring the concept of the present disclosure, some details known in the art are not described. Based on the above description, those skilled in the art can fully understand how to implement the technical solutions disclosed herein.
[0069] Although some specific embodiments of the present disclosure have been described in detail by way of examples, it should be understood by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present disclosure. It should be understood by those skilled in the art that the above embodiments may be modified or some technical features may be replaced by equivalents without departing from the scope and spirit of the present disclosure. The scope of the present disclosure is defined by the appended claims.
Claims
1. A heterogeneous resource scheduling method, comprising: In response to a resource request sent by each edge application on the mobile edge computing platform, parsing an acceleration requirement file of the edge application to pre-allocate required heterogeneous resources for the edge application; Use deep learning networks to extract key features of each edge application; Use classifiers to match the key features of each edge application with the required heterogeneous resources to obtain the ratio of various heterogeneous resources required by each edge application; Determine the real-time application resource allocation strategy based on the ratio of various heterogeneous resources required by each edge application; By performing regression prediction on the resource utilization of each edge application, the overall heterogeneous resource allocation strategy within a predetermined time range is obtained; The heterogeneous resources are uniformly allocated and scheduled according to the real-time application resource allocation strategy and the overall heterogeneous resource allocation strategy.
2. The method according to claim 1, wherein: Parsing the acceleration requirement file of the edge application includes: Calling the acceleration requirement file of the edge application through the requirement file library; Parsing the acceleration requirement file using a requirement parser to obtain a pre-allocation strategy for heterogeneous resources required by the edge application; The required heterogeneous resources of the edge application are pre-allocated according to the pre-allocation strategy.
3. The method according to claim 1, wherein: The resource utilization of each edge application is regressed and predicted, including: The resource utilization of each edge application is regressed and predicted using the gradient boosted decision tree (GBDT) regression predictor to obtain the distribution of resource requirements of each edge application within a predetermined time range. The overall allocation strategy of heterogeneous resources within a predetermined time range is determined according to the obtained distribution.
4. The method according to claim 1, wherein: The deep learning network is a convolutional neural network (CNN) feature extraction network; The classifier is a Softmax classifier.
5. A heterogeneous resource scheduling system, comprising: A pre-allocation module is configured to parse the acceleration requirement file of each edge application in response to a resource request sent by each edge application on the mobile edge computing platform, so as to pre-allocate the required heterogeneous resources of the edge application; The real-time allocation module is configured to use a deep learning network to extract key features of each edge application, use a classifier to match the key features of each edge application with the required heterogeneous resources, so as to obtain the ratio of various heterogeneous resources required by each edge application, and determine the real-time application resource allocation strategy according to the ratio of various heterogeneous resources required by each edge application; The overall allocation module is configured to obtain an overall heterogeneous resource allocation strategy within a predetermined time range by performing regression prediction on the resource utilization of each edge application; The heterogeneous resource scheduling module is configured to uniformly allocate and schedule heterogeneous resources according to the real-time application resource allocation strategy and the overall heterogeneous resource allocation strategy.
6. The system according to claim 5, wherein: The pre-allocation module is configured to call the acceleration requirement file of the edge application through the requirement file library, parse the acceleration requirement file using the requirement parser to obtain the pre-allocation strategy of the heterogeneous resources required by the edge application, and pre-allocate the required heterogeneous resources of the edge application according to the pre-allocation strategy.
7. The system according to claim 5, wherein: The overall allocation module is configured to use the gradient boosting decision tree (GBDT) regression predictor to perform regression prediction on the resource utilization of each edge application to obtain the distribution of resource requirements of each edge application within a predetermined time range, and determine the overall allocation strategy of heterogeneous resources within the predetermined time range based on the obtained distribution.
8. The system according to claim 5, wherein: The deep learning network is a convolutional neural network (CNN) feature extraction network; The classifier is a Softmax classifier.
9. A heterogeneous resource scheduling system, comprising: a memory configured to store instructions; A processor is coupled to the memory, and the processor is configured to execute the method according to any one of claims 1 to 4 based on instructions stored in the memory.
10. A computer-readable storage medium, wherein: The computer-readable storage medium stores computer instructions, and when the instructions are executed by a processor, the method according to any one of claims 1 to 4 is implemented.