A computing power integration method combining GPU virtualization and AI

Abstract

The present invention relates to the technical field of GPU computing power integration, and discloses a computing power integration method combining GPU virtualization and AI, including: receiving a computing power integration instruction initiated from an Internet of Things device, starting a collaborative computing model, using GPU virtualization technology to divide GPU hardware resources into n mutually independent virtual GPU instances, transmitting user data to an edge computing node, and intelligently allocating virtual GPU instances to the edge computing node to perform preprocessing on user data to obtain clean data, then transmitting the clean data to the backend of a cloud server, deploying the application programs and dependencies of the Internet of Things device on the edge computing node, when the deployment is successful, performing parameter configuration on the key device components of the virtual GPU instance, and when the configuration is successful, performing testing on the key device components to complete the computing power integration combining GPU virtualization and AI. The main purpose of the present invention is to consider the adaptation of virtual GPU instance resources and edge nodes, thereby improving the combination efficiency of GPU virtualization and AI technologies.

Description

Technical Field

[0001] The present invention relates to a computing power fusion method combining GPU virtualization and AI, belonging to the technical field of GPU computing power fusion. Background Art

[0002] With the rapid development of Internet of Things (IoT) technology, a large amount of data generated by devices needs to be processed quickly and efficiently. GPU virtualization technology improves resource utilization and flexibility by dividing physical GPU resources into multiple virtual GPU instances. Combined with artificial intelligence (AI), it can further enhance data processing capabilities and achieve intelligent data analysis and decision support. This computing power fusion method can improve computing efficiency. Especially in the fields of intelligent transportation, smart city, and intelligent manufacturing, the application prospect of this technology is broad, which is of great significance for promoting the digital transformation of society.

[0003] Traditional methods of combining GPU virtualization and AI do not consider the adaptation problem between virtual GPU instance resources and edge nodes, resulting in low efficiency in combining GPU virtualization and AI technologies. Summary of the Invention

[0004] The present invention provides a computing power fusion method, device, and computer-readable storage medium for combining GPU virtualization and AI. Its main purpose is to consider the adaptation between virtual GPU instance resources and edge nodes, thereby improving the combination efficiency of GPU virtualization and AI technologies.

[0005] To achieve the above object, a computing power fusion method for combining GPU virtualization and AI provided by the present invention includes:

[0006] Receiving a computing power fusion instruction initiated from an IoT device, receiving user data collected by the IoT device according to the computing power fusion instruction, and starting a collaborative computing model while receiving the user data. The collaborative computing model consists of an edge computing node combined with AI, a standard conversion unit, and a cloud server backend;

[0007] Using GPU virtualization technology, dividing GPU hardware resources into n mutually independent virtual GPU instances, where the GPU hardware resources include GPU computing power resources and GPU video memory resources;

[0008] Transmitting the user data to the edge computing node, intelligently allocating the virtual GPU instances to the edge computing node, using the edge computing node to perform preprocessing on the user data to obtain clean data, and then transmitting the clean data to the cloud server backend;

[0009] Deploy the application programs and dependencies of the Internet of Things devices to the edge computing nodes using the standard conversion unit. After successful deployment, perform parameter configuration on the key device components of the virtual GPU instance, where the key device components connect the edge computing nodes and the cloud server backend;

[0010] After successful configuration, perform tests on the key device components. The test content includes: whether the key device components are connected to the cloud server backend and whether they can direct process the clean data transmitted to the cloud server backend. After the test results show success, complete the computing power integration of GPU virtualization and AI.

[0011] Optionally, using the GPU virtualization technology, divide the GPU hardware resources into n independent virtual GPU instances, where the GPU hardware resources include GPU computing power resources and GPU video memory resources, including:

[0012] Through the GPU monitoring tool, real-time track and evaluate the performance index parameters of the GPU hardware resources, where the performance index parameters consist of the computing unit utilization rate and the memory bandwidth occupancy rate;

[0013] Monitor the performance index parameters, calculate the preset number of instances based on the computing unit utilization rate and the preset number of instances based on the memory bandwidth occupancy rate according to the performance index parameters, and perform a floor operation on the minimum value of the two to obtain the determined number of instances as n;

[0014] Based on the determined number of instances, use the GPU virtualization technology to divide the GPU hardware resources into n independent virtual GPU instances.

[0015] Optionally, the calculating the preset number of instances based on the computing unit utilization rate and the preset number of instances based on the memory bandwidth occupancy rate according to the performance index parameters includes:

[0016] Calculate the preset number of instances based on the computing unit utilization rate and the preset number of instances based on the memory bandwidth occupancy rate according to the following formula:

[0017] ;

[0018] where, represents the preset number of instances based on the computing unit utilization rate, represents the maximum number of virtual GPU instances that can run simultaneously on the GPU hardware resources, represents the computing unit utilization rate, represents the preset number of instances based on the memory bandwidth occupancy rate, represents the memory bandwidth occupancy rate.

[0019] Optionally, the intelligent allocation of the virtual GPU instances to the edge computing nodes includes:

[0020] Identifying the user data and analyzing the data task type of the user data, where the data task type includes the format type, byte count, complexity, and update frequency of the user data;

[0021] Determining the edge computing nodes and analyzing the computing power load conditions of the edge computing nodes;

[0022] Constructing a neural network model based on the edge computing nodes according to the data task type and the computing power load conditions;

[0023] Intelligently allocating the virtual GPU instances to the edge computing nodes by using the neural network model.

[0024] Optionally, the determining of the edge computing nodes and the analyzing of the computing power load conditions of the edge computing nodes include:

[0025] Using sensors pre-installed on the edge computing nodes to collect the key node metrics of the edge computing nodes in real time, where the key node metrics include the CPU utilization ratio and the memory usage ratio;

[0026] Reading the working log file of the edge computing node and performing parsing on the working log file. When the parsing is successful, key node information is obtained, where the key node information includes the request processing time, the task queue length, the error log record, and the description of abnormal events;

[0027] Statistically analyzing the key node metrics and the key node information to obtain the computing power load conditions of the edge computing nodes.

[0028] Optionally, the intelligently allocating of the virtual GPU instances to the edge computing nodes by using the neural network model includes:

[0029] Primarily allocating the virtual GPU instances to the edge computing nodes through the neural network model, where each edge computing node is allocated at least one virtual GPU instance;

[0030] When the allocation is successful, using the virtual GPU instances to perform iterative training on the neural network model and updating the parameters of the neural network model in a timely manner during the training process;

[0031] After completing the iterative training operation, performing an adjusted allocation on the virtual GPU instances again to complete the intelligent allocation of the virtual GPU instances to the edge computing nodes.

[0032] Optionally, after successful allocation, perform iterative training on the neural network model using the virtual GPU instances, and update the parameters of the neural network model in a timely manner during the training process, including:

[0033] Copy the neural network model to each virtual GPU instance to generate n copies of the neural network model loaded on the virtual GPU instances;

[0034] Determine the user data, evenly split the combined user data into sub-datasets, and sequentially allocate the sub-datasets to each virtual GPU instance;

[0035] Use each virtual GPU instance to independently perform iterative training on the neural network model copy. While performing iterative training, mirror the iterative training of the neural network model, and update the parameters of the neural network model in a timely manner during the training process.

[0036] Optionally, deploy the application program and dependencies of the Internet of Things device to the edge computing node using the standard conversion unit. After successful deployment, perform parameter configuration on the key device components of the virtual GPU instance, where the key device components connect the edge computing node and the cloud server backend, including:

[0037] Identify the application program and dependencies of the Internet of Things device, and use the standard conversion unit to perform standard packaging on the application program and dependencies;

[0038] After successful packaging, obtain a program image and deploy the program image to the edge computing node;

[0039] Confirm the distributed file system of the program image, and register the key device components of the edge computing node to the distributed file system, where the key device components connect the edge computing node and the cloud server backend;

[0040] After successful registration, perform performance parameter settings and system compatibility settings on the key device components to complete the parameter configuration of the key device components of the virtual GPU instance.

[0041] Optionally, registering the key device components of the edge computing node to the distributed file system includes:

[0042] Initialize the distributed file system, and send the configuration information and status data of the key device components to the distributed file system;

[0043] Use the distributed file system to compile the configuration information and status data of the key device components. After successful compilation, use the distributed file system to automatically identify the key device components of the edge computing node;

[0044] After successful recognition, a unique identifier is generated to complete the registration of the key device components of the edge computing node to the distributed file system.

[0045] To solve the above problems, the present invention also provides an electronic device, which includes:

[0046] At least one processor; and,

[0047] A memory communicatively connected to the at least one processor; wherein,

[0048] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to implement the above-mentioned computing power fusion method combining GPU virtualization and AI.

[0049] To solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one instruction is stored, and the at least one instruction is executed by a processor in an electronic device to implement the above-mentioned computing power fusion method combining GPU virtualization and AI.

[0050] Compared with the problems described in the background art, the present invention first receives a computing power fusion instruction initiated from an Internet of Things device, and receives user data collected by the Internet of Things device according to the computing power fusion instruction. While receiving the user data, a collaborative computing model is started. The collaborative computing model consists of an edge computing node combined with AI, a standard conversion unit, and a cloud server backend. Using GPU virtualization technology, the GPU hardware resources are divided into n independent virtual GPU instances. The GPU hardware resources include GPU computing power resources and GPU video memory resources. It can be seen that by dividing the GPU hardware resources into multiple independent virtual GPU instances, the present invention can dynamically allocate resources according to the actual needs of the edge computing node. This flexibility allows for intelligent adjustment of resource allocation according to the real-time computing load and task type, thereby improving resource utilization and computing efficiency. Then, the user data is transmitted to the edge computing node, and the virtual GPU instances are intelligently allocated to the edge computing node. The edge computing node is used to preprocess the user data to obtain clean data, and then the clean data is transmitted to the cloud server backend. The application programs and dependencies of the Internet of Things device are deployed on the edge computing node using the standard conversion unit. After the deployment is successful, parameter configuration is performed on the key device components of the virtual GPU instances. The key device components connect the edge computing node and the cloud server backend. Since the edge computing node of the present invention is close to the user data source, it can quickly preprocess and analyze the user data source, reducing data transmission latency. Importantly, by intelligently allocating virtual GPU instances to the edge computing node, it can be ensured that the edge node has sufficient computing power to execute these tasks, thereby accelerating the data processing speed. Finally, after the configuration is successful, testing is performed on the key device components. The test content includes whether the key device components are connected to the cloud server backend and whether they can direct the processing of the clean data transmitted to the cloud server backend. When the test result shows success, the computing power fusion combining GPU virtualization and AI is completed. Generally speaking, the independence and configurability of the virtual GPU instances of the present invention make the system easier to expand and maintain. And as the computing demand grows, only more virtual GPU instances need to be simply added, without the need to replace or upgrade physical hardware. Therefore, the computing power fusion method, electronic device, and computer-readable storage medium combining GPU virtualization and AI proposed by the present invention mainly aim to consider the adaptation of virtual GPU instance resources and edge nodes, thereby improving the combination efficiency of GPU virtualization and AI technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic flowchart of a computing power fusion method combining GPU virtualization and AI provided by an embodiment of the present invention;

[0052] Figure 2Schematic diagram of the structure of an electronic device for implementing the computing power fusion method combining GPU virtualization and AI provided by an embodiment of the present invention.

[0053] The implementation, functional features, and advantages of the present invention will be further described with reference to the embodiments and the accompanying drawings. Specific embodiments

[0054] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0055] The embodiments of the present application provide a computing power fusion method combining GPU virtualization and AI. The execution subject of the computing power fusion method combining GPU virtualization and AI includes, but is not limited to, at least one of electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided by the embodiments of the present application. In other words, the computing power fusion method combining GPU virtualization and AI can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to: a single server, a server cluster, a cloud server, or a cloud server cluster, etc.

[0056] Embodiment 1:

[0057] Refer to Figure 1 As shown, it is a flowchart of the computing power fusion method combining GPU virtualization and AI provided by an embodiment of the present invention. In this embodiment, the computing power fusion method combining GPU virtualization and AI includes:

[0058] S1. Receive a computing power fusion instruction initiated from an Internet of Things device, receive the user data collected by the Internet of Things device according to the computing power fusion instruction, and start a collaborative computing model while receiving the user data. The collaborative computing model consists of an edge computing node combined with AI, a standard conversion unit, and a cloud server backend.

[0059] It should be explained that the computing power fusion instruction is generally initiated by the background cloud computing engineers of the Internet of Things device. Exemplarily, Xiao Zhang is a cloud computing engineer of a certain Internet of Things device, and the corresponding artificial intelligence program of this Internet of Things device is an automotive autonomous driving APP. This APP needs to collect real-time road condition data, thus generating a large amount of data. Therefore, in order to improve the data processing efficiency of computing power resources, Xiao Zhang initiates a computing power fusion instruction for the Internet of Things device in the background. Its purpose is to accelerate data calculation and avoid program lag caused by large data calculation delays.

[0060] Furthermore, the collaborative computing model is a system that uses GPU virtualization and AI technology architecture to integrate computing power in real time, aiming to improve the data processing efficiency of computing power resources.

[0061] It is understandable that in the embodiments of the present invention, the collaborative computing model consists of an edge computing node combined with AI, a standard conversion unit, and a cloud server backend. Among them, the edge computing node combined with AI is a computing device located between the Internet of Things devices and the cloud server backend, distributed around the interfaces of the Internet of Things devices. Due to the proximity between the Internet of Things devices and the edge computing node, it can effectively reduce the data transmission delay between the Internet of Things devices and the edge computing node; the standard conversion unit is embedded with Docker engine technology and is mainly responsible for packaging the APP and dependencies of the artificial intelligence program into standard software units, enabling the APP to be adaptively transplanted from one computing environment to another for operation; finally, the main role of the cloud server backend is to perform accelerated computing based on GPU virtualization technology and can execute computing tasks with the computing power for processing massive data and complex data calculations.

[0062] S2. Using GPU virtualization technology, divide the GPU hardware resources into n mutually independent virtual GPU instances, where the GPU hardware resources include GPU computing power resources and GPU video memory resources.

[0063] It should be explained that through GPU virtualization technology, a single physical GPU hardware resource can be divided into n mutually independent virtual GPU instances, and in subsequent computing power work, they are intelligently allocated on demand according to different workloads, thereby improving the utilization efficiency of GPU hardware resources. Further, each virtual GPU instance is mutually independent, so it has a high degree of isolation, ensuring that different workloads will not affect each other and can improve the stability and security of the system computing power.

[0064] Specifically, the step of using GPU virtualization technology to divide the GPU hardware resources into n mutually independent virtual GPU instances, where the GPU hardware resources include GPU computing power resources and GPU video memory resources, includes:

[0065] Through a GPU monitoring tool, real-time track and evaluate the performance index parameters of the GPU hardware resources, where the performance index parameters consist of the usage rate of the computing unit and the occupancy rate of the memory bandwidth;

[0066] Monitor the performance index parameters, calculate the preset number of instances based on the usage rate of the computing unit and the preset number of instances based on the occupancy rate of the memory bandwidth according to the performance index parameters, and perform a floor operation on the minimum value of the two to obtain the determined number of instances as n;

[0067] Based on the determined number of instances, use GPU virtualization technology to divide the GPU hardware resources to obtain n mutually independent virtual GPU instances.

[0068] It is understandable that GPU hardware resources refer to the component features that can support the GPU to execute efficient computing tasks, including GPU computing power resources and GPU video memory resources. Among them, GPU computing power resources refer to the parallel processing ability of the GPU for computing tasks, and the computing unit utilization rate represents the proportion of time occupied by the GPU in executing computing tasks or processing computing power data within a period of time, which can measure the performance index of the GPU. Therefore, the working efficiency of GPU computing power resources can be measured by the computing unit utilization rate; GPU video memory resources refer to the ability of the GPU to transmit and store core data, and the memory bandwidth occupancy rate represents the proportion of the bandwidth used within a specific time. Generally speaking, the larger the value of the memory bandwidth occupancy rate, the faster the system is transmitting data. Therefore, the working efficiency of GPU video memory resources can be measured by the memory bandwidth occupancy rate.

[0069] Further, calculating the preset number of computing examples based on the computing unit utilization rate and the preset number of computing examples based on the memory bandwidth occupancy rate according to the performance index parameters includes:

[0070] Calculating the preset number of computing examples based on the computing unit utilization rate and the preset number of computing examples based on the memory bandwidth occupancy rate according to the following formula:

[0071] ;

[0072] Wherein, represents the preset number of computing examples based on the computing unit utilization rate, represents the maximum number of virtual GPU computing examples that can run simultaneously on the GPU hardware resources, represents the computing unit utilization rate, represents the preset number of computing examples based on the memory bandwidth occupancy rate, represents the memory bandwidth occupancy rate.

[0073] It should be emphasized that after calculating the preset number of computing examples based on the computing unit utilization rate and the preset number of computing examples based on the memory bandwidth occupancy rate, a floor modulo operation needs to be performed on the smaller value of the two to ensure that the number of divided virtual GPU computing examples will not exceed the carrying capacity of the physical GPU.

[0074] S3. Transmit the user data to the edge computing node, and intelligently allocate the virtual GPU computing examples to the edge computing node, use the edge computing node to perform preprocessing on the user data to obtain clean data, and then transmit the clean data to the back end of the cloud server.

[0075] It should be noted that in the traditional cloud computing model, user data is often first transmitted to the data center through the network, and then the cloud servers in the data center are used for computing and processing. This model that overly concentrates computing power resources in the data center easily causes network congestion during data transmission, high data processing pressure, and low computing efficiency. Therefore, in the embodiments of the present invention, edge computing is considered to first allocate primary user data processing tasks to edge computing nodes close to Internet of Things devices. Since the edge computing nodes are closer to the location of user data, data transmission latency can be reduced, and real-time requirements can be responded to more quickly.

[0076] Specifically, the intelligent allocation of the virtual GPU instances to the edge computing nodes includes:

[0077] Identify the user data and analyze the data task type of the user data, where the data task type includes the format type, byte count, complexity, and update frequency of the user data;

[0078] Determine the edge computing nodes and analyze the computing power load situation of the edge computing nodes;

[0079] Construct a neural network model based on the edge computing nodes according to the data task type and the computing power load situation;

[0080] Intelligently allocate the virtual GPU instances to the edge computing nodes by using the neural network model.

[0081] It can be explained that the computing power load situation of the edge computing nodes describes the quantity and complexity of the computing tasks undertaken in the edge computing environment. Further, select an appropriate type of neural network model according to the data task type of the user data and the computing power load situation of the edge computing nodes, such as a convolutional neural network, a recurrent neural network, or a long short-term memory network, etc.

[0082] Further, the determination of the edge computing nodes and the analysis of the computing power load situation of the edge computing nodes include:

[0083] Use sensors pre-installed in the edge computing nodes to collect the key node metrics of the edge computing nodes in real time, where the key node metrics include the CPU utilization ratio and the memory usage ratio;

[0084] Read the work log file of the edge computing node and perform parsing on the work log file. When the parsing is successful, obtain the key node information, where the key node information includes the request processing time, the task queue length, the error log record, and the description of abnormal events;

[0085] Statistically analyze the key node metrics and the key node information to obtain the computing power load situation of the edge computing nodes.

[0086] Specifically, although edge computing nodes are closer to the location of user data and can reduce data transmission latency, due to the limitations of the volume and power consumption of edge node devices themselves, the computing power of node devices at the edge is relatively weak. Therefore, in the embodiments of the present invention, the GPU virtualization technology is used to divide the GPU hardware resources into n independent virtual GPU instances and then allocate them to edge computing nodes, additionally providing GPU computing power resources for edge computing nodes and effectively improving the processing ability of edge computing nodes.

[0087] Further, the intelligent allocation of virtual GPU instances to edge computing nodes by using the neural network model includes:

[0088] Initial allocation of virtual GPU instances to edge computing nodes through the neural network model, where each edge computing node is at least allocated one virtual GPU instance;

[0089] After successful allocation, perform iterative training on the neural network model using the virtual GPU instance and update the parameters of the neural network model in a timely manner during the training process;

[0090] After completing the iterative training operation, perform adjusted allocation on the virtual GPU instance again to complete the intelligent allocation of the virtual GPU instance to the edge computing node.

[0091] It should be emphasized that in the process of intelligently allocating virtual GPU instances to edge computing nodes in the embodiments of the present invention, first, the virtual GPU instances are initially allocated to edge computing nodes through the neural network model, where each edge computing node is at least allocated one virtual GPU instance to ensure that each edge computing node has a computing task to execute and avoid waste of computing power resources; perform iterative training on the neural network model through the virtual GPU instances allocated to edge computing nodes so that during the training process, the parameters of the neural network model are updated in a timely manner to gradually approach the optimal allocation solution; since the computing tasks on different nodes may vary in terms of task execution time, resource utilization, etc. due to complexity, data volume, or other factors, after completing the iterative training operation, it is necessary to use the neural network model again to re-allocate the virtual GPU instances to achieve more efficient task execution and resource utilization.

[0092] Further, the step of performing iterative training on the neural network model through the virtual GPU instance and updating the parameters of the neural network model in a timely manner after successful allocation includes:

[0093] Copy the neural network model to each virtual GPU instance to generate copies of the neural network model loaded on the virtual GPU instances;

[0094] Determine the user data, combine the user data, evenly split it to obtain sub-datasets, and sequentially allocate the sub-datasets to each virtual GPU instance;

[0095] Use each of the virtual GPU instances to independently perform iterative training on copies of the neural network model. While performing iterative training, mirror and perform iterative training of the neural network model, and update the parameters of the neural network model in a timely manner during the training process.

[0096] Therefore, after the edge computing node is additionally provided with GPU computing power resources, it performs computationally intensive data preprocessing tasks such as data cleaning and data integration on the user data, thereby reducing the data processing pressure on the back end of the cloud server. The back end of the cloud server can focus more on performing complex analysis, decision-making, and other tasks, improving the performance of the overall computing system.

[0097] S4. Use the standard conversion unit to deploy the application programs and dependencies of the IoT device to the edge computing node. After successful deployment, perform parameter configuration on the key device components of the virtual GPU instance, where the key device components connect the edge computing node and the back end of the cloud server.

[0098] It should be explained that the standard conversion unit is embedded with Docker engine technology. Therefore, it can package the APP and dependencies of the artificial intelligence program in the IoT device into a standard software unit, enabling the APP to be adaptively transplanted from one computing environment to another for operation, realizing a smooth transplant between the IoT device environment and the edge computing node.

[0099] Specifically, the step of using the standard conversion unit to deploy the application programs and dependencies of the IoT device to the edge computing node, and after successful deployment, performing parameter configuration on the key device components of the virtual GPU instance, where the key device components connect the edge computing node and the back end of the cloud server, includes:

[0100] Identify the application programs and dependencies of the IoT device, and use the standard conversion unit to perform standard packaging on the application programs and dependencies;

[0101] After successful packaging, obtain a programmatic image, and deploy the programmatic image to the edge computing node;

[0102] Confirm the distributed file system of the programmatic image, and register the key device components of the edge computing node to the distributed file system, where the key device components connect the edge computing node and the back end of the cloud server;

[0103] After successful registration, perform performance parameter settings and system compatibility settings on the key device components, and complete the parameter configuration for the key device components of the virtual GPU example.

[0104] It can be understood that registering the key device components of the edge computing node to the distributed file system means integrating the relevant driver programs and configuration information of the virtual GPU example resources into the operating system of the programmatic image. Therefore, when running program tasks, the programmatic images generated by the application programs and dependencies of the IoT devices can recognize and use the virtual GPU example resources.

[0105] Furthermore, registering the key device components of the edge computing node to the distributed file system includes:

[0106] Initialize the distributed file system and send the configuration information and status data of the key device components to the distributed file system;

[0107] Use the distributed file system to compile the configuration information and status data of the key device components. After successful compilation, use the distributed file system to automatically identify the key device components of the edge computing node;

[0108] After successful identification, generate a unique identifier to complete the registration of the key device components of the edge computing node to the distributed file system.

[0109] In addition, the key device components connect the edge computing node and the cloud server backend, and perform parameter configuration on the key device components of the virtual GPU example to ensure correct interaction with the programmatic image. Performing parameter configuration on the key device components includes settings in terms of performance parameters and system compatibility.

[0110] S6. After successful configuration, perform tests on the key device components. The test content is: whether the key device components are connected to the cloud server backend, and whether they can directedly process the clean data transmitted to the cloud server backend. When the test results show success, complete the computing power integration of GPU virtualization and AI combination.

[0111] It should be explained that after successful configuration, in order to ensure that the cloud server backend can receive the computing task requests from the virtual GPU example, it is necessary to test whether the key device components have been successfully connected to the cloud server backend, and verify that the cloud server backend can receive and correctly process the clean data, ensuring that when performing computing tasks, the virtual GPU example resources can be called to perform acceleration processing, and complete the computing power integration of GPU virtualization and AI combination.

[0112] Compared with the problems described in the background art, the present invention first receives a computing power fusion instruction initiated from an Internet of Things device, and receives user data collected by the Internet of Things device according to the computing power fusion instruction. While receiving the user data, a collaborative computing model is started. The collaborative computing model consists of an edge computing node combined with AI, a standard conversion unit, and a cloud server backend. Using GPU virtualization technology, the GPU hardware resources are divided into n independent virtual GPU instances. Among them, the GPU hardware resources include GPU computing power resources and GPU video memory resources. It can be seen that by dividing the GPU hardware resources into multiple independent virtual GPU instances, the present invention can dynamically allocate resources according to the actual needs of the edge computing node. This flexibility allows for intelligent adjustment of resource allocation according to the real-time computing load and task type, thereby improving resource utilization and computing efficiency. Then, the user data is transmitted to the edge computing node, and the virtual GPU instances are intelligently allocated to the edge computing node. The edge computing node is used to preprocess the user data to obtain clean data, and then the clean data is transmitted to the cloud server backend. The application programs and dependencies of the Internet of Things device are deployed on the edge computing node using the standard conversion unit. After the deployment is successful, parameter configuration is performed on the key device components of the virtual GPU instances. The key device components connect the edge computing node and the cloud server backend. Since the edge computing node of the present invention is close to the user data source, it can quickly preprocess and analyze the user data source, reducing data transmission latency. Importantly, by intelligently allocating virtual GPU instances to the edge computing node, it can be ensured that the edge node has sufficient computing power to execute these tasks, thereby accelerating the data processing speed. Finally, after the configuration is successful, tests are performed on the key device components. The test content includes: whether the key device components are connected to the cloud server backend, and whether they can directionally process the clean data transmitted to the cloud server backend. When the test results show success, the computing power fusion combining GPU virtualization and AI is completed. Generally speaking, the independence and configurability of the virtual GPU instances of the present invention make the system easier to expand and maintain. And as the computing demand grows, only more virtual GPU instances need to be simply added, without the need to replace or upgrade physical hardware. Therefore, the computing power fusion method, electronic device, and computer-readable storage medium combining GPU virtualization and AI proposed by the present invention mainly aim to consider the adaptation of virtual GPU instance resources and edge nodes, thereby improving the combination efficiency of GPU virtualization and AI technology.

[0113] Embodiment 2:

[0114] As Figure 2 shown, it is a schematic structural diagram of an electronic device for implementing the computing power fusion method combining GPU virtualization and AI according to an embodiment of the present invention.

[0115] The electronic device 1 may include a processor 10, a memory 11, a bus 12, and a communication interface 13. It may also include a computer program stored in the memory 11 and executable on the processor 10, such as a computing power integration program combining GPU virtualization and AI.

[0116] Among them, the memory 11 includes at least one type of readable storage medium, which includes flash memory, mobile hard disk, multimedia card, card-type memory (such as SD or DX memory, etc.), magnetic memory, magnetic disk, optical disc, etc. In some embodiments, the memory 11 may be an internal storage unit of the electronic device 1, such as the mobile hard disk of the electronic device 1. In some other embodiments, the memory 11 may also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. equipped on the electronic device 1. Further, the memory 11 may also include both the internal storage unit and the external storage device of the electronic device 1. The memory 11 can be used not only to store application software installed on the electronic device 1 and various types of data, such as the code of the computing power integration program combining GPU virtualization and AI, but also to temporarily store data that has been output or will be output.

[0117] In some embodiments, the processor 10 may be composed of integrated circuits. For example, it may be composed of a single packaged integrated circuit, or may be composed of multiple integrated circuits with the same or different functions, including a combination of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control core (Control Unit) of the electronic device, connecting various components of the entire electronic device through various interfaces and lines, and by running or executing programs or modules (such as the computing power integration program combining GPU virtualization and AI, etc.) stored in the memory 11, and calling data stored in the memory 11, to execute various functions of the electronic device 1 and process data.

[0118] The bus may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. The bus is configured to enable connection communication between the memory 11 and at least one processor 10, etc.

[0119] Figure 2 Only an electronic device with components is shown. Those skilled in the art can understand that Figure 2 the shown structure does not constitute a limitation on the electronic device 1, and it may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0120] For example, although not shown, the electronic device 1 may further include a power source (such as a battery) for powering each component. Preferably, the power source can be logically connected to the at least one processor 10 through a power management device, so as to implement functions such as charging management, discharging management, and power consumption management through the power management device. The power source may also include any components such as one or more DC or AC power sources, a recharge device, a power failure detection circuit, a power converter or inverter, a power status indicator, etc. The electronic device 1 may also include various sensors, a Bluetooth module, a Wi-Fi module, etc., which will not be elaborated here.

[0121] Furthermore, the electronic device 1 may further include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device 1 and other electronic devices.

[0122] Optionally, the electronic device 1 may further include a user interface. The user interface may be a display, an input unit (such as a keyboard), and optionally, the user interface may also be a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch liquid crystal display, and an OLED (Organic Light-Emitting Diode) toucher, etc. Among them, the display may also be appropriately referred to as a display screen or a display unit, which is used to display the information processed in the electronic device 1 and to display a visual user interface.

[0123] It should be understood that the embodiments are only for illustration purposes and are not limited by this structure in the scope of the patent application.

[0124] The computing power fusion program combining GPU virtualization and AI stored in the memory 11 of the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can implement:

[0125] Receive a computing power integration instruction initiated from an Internet of Things device, receive user data collected by the Internet of Things device according to the computing power integration instruction, and start a collaborative computing model while receiving the user data. The collaborative computing model consists of an edge computing node combined with AI, a standard conversion unit, and a cloud server backend;

[0126] Using GPU virtualization technology, divide the GPU hardware resources into independent virtual GPU instances. Among them, the GPU hardware resources include GPU computing power resources and GPU video memory resources;

[0127] Transmit the user data to the edge computing node, and intelligently allocate the virtual GPU instances to the edge computing node. Use the edge computing node to perform preprocessing on the user data to obtain clean data, and then transmit the clean data to the cloud server backend;

[0128] Use the standard conversion unit to deploy the application programs and dependencies of the Internet of Things device on the edge computing node. After successful deployment, perform parameter configuration on the key device components of the virtual GPU instances. The key device components connect the edge computing node and the cloud server backend;

[0129] After successful configuration, perform tests on the key device components. The test content includes: whether the key device components are connected to the cloud server backend, and whether they can directedly process the clean data transmitted to the cloud server backend. When the test results show success, complete the computing power integration of GPU virtualization and AI combination.

[0130] Specifically, the specific implementation method of the above instructions by the processor 10 can refer to Figures 1 to 2 the description of the relevant steps in the corresponding embodiments, which will not be elaborated here.

[0131] Furthermore, if the modules / units integrated in the electronic device 1 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM, Read-Only Memory).

[0132] The present invention also provides a computer-readable storage medium. The readable storage medium stores a computer program, and when the computer program is executed by the processor of the electronic device, it can implement:

[0133] Receive the computing power integration instruction initiated by the Internet of Things device, receive the user data collected by the Internet of Things device according to the computing power integration instruction, and start the collaborative computing model while receiving the user data. The collaborative computing model consists of an edge computing node combined with AI, a standard conversion unit, and a cloud server backend;

[0134] Using GPU virtualization technology, divide the GPU hardware resources into n independent virtual GPU instances. Among them, the GPU hardware resources include GPU computing power resources and GPU video memory resources;

[0135] Transmit the user data to the edge computing node, and intelligently allocate the virtual GPU instances to the edge computing node. Use the edge computing node to perform preprocessing on the user data to obtain clean data, and then transmit the clean data to the cloud server backend;

[0136] Use the standard conversion unit to deploy the application programs and dependencies of the Internet of Things device on the edge computing node. After successful deployment, perform parameter configuration on the key device components of the virtual GPU instances. The key device components connect the edge computing node and the cloud server backend;

[0137] After successful configuration, perform tests on the key device components. The test content is: whether the key device components are connected to the cloud server backend, and whether they can directedly process the clean data transmitted to the cloud server backend. When the test results show success, complete the computing power integration of GPU virtualization and AI combination.

[0138] The modules described as separate components may or may not be physically separated. The components shown as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed to multiple network units. You can select some or all of the modules according to actual needs to achieve the purpose of the solution of this embodiment.

[0139] In addition, in each embodiment of the present invention, the functional modules can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware, or in the form of hardware plus software functional modules.

[0140] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A computing power fusion method combining GPU virtualization and AI, characterized in that, The method includes: Receiving a computing power integration instruction initiated from an Internet of Things device, receiving user data collected by the Internet of Things device according to the computing power integration instruction, and starting a collaborative computing model while receiving the user data, where the collaborative computing model consists of an edge computing node combined with AI, a standard conversion unit, and a cloud server backend; Using GPU virtualization technology, dividing the GPU hardware resources into n independent virtual GPU instances, where the GPU hardware resources include GPU computing power resources and GPU video memory resources. The using GPU virtualization technology to divide the GPU hardware resources into n independent virtual GPU instances, where the GPU hardware resources include GPU computing power resources and GPU video memory resources, includes: Real-time tracking and evaluating the performance index parameters of the GPU hardware resources through a GPU monitoring tool, where the performance index parameters consist of the usage rate of the computing unit and the occupancy rate of the memory bandwidth; Monitoring the performance index parameters, calculating the preset number of instances based on the usage rate of the computing unit and the preset number of instances based on the occupancy rate of the memory bandwidth according to the performance index parameters, and performing a floor operation on the minimum value of the two to obtain the determined number of instances as n; Based on the determined number of instances, using GPU virtualization technology to divide the GPU hardware resources into n independent virtual GPU instances; Transmitting the user data to the edge computing node, and intelligently allocating the virtual GPU instances to the edge computing node, using the edge computing node to perform preprocessing on the user data to obtain clean data, and then transmitting the clean data to the cloud server backend. The intelligently allocating the virtual GPU instances to the edge computing node includes: Identifying the user data and analyzing the data task type of the user data, where the data task type includes the format type, byte count, complexity, and update frequency of the user data; Determining the edge computing node and analyzing the computing power load situation of the edge computing node; Constructing a neural network model based on the edge computing node according to the data task type and the computing power load situation; Intelligently allocating the virtual GPU instances to the edge computing node using the neural network model; Deploying the application program and dependencies of the Internet of Things device to the edge computing node using the standard conversion unit, and performing parameter configuration on the key device components of the virtual GPU instance when the deployment is successful, where the key device components connect the edge computing node and the cloud server backend; When the configuration is successful, performing a test on the key device components, where the test content is: whether the key device components are connected to the cloud server backend and whether they can directedly process the clean data transmitted to the cloud server backend. When the test result shows success, complete the computing power integration combining GPU virtualization and AI.

2. The computing power fusion method combining GPU virtualization and AI according to claim 1, characterized in that The calculating the preset number of instances based on the usage rate of the computing unit and the preset number of instances based on the occupancy rate of the memory bandwidth according to the performance index parameters includes: Calculating the preset number of instances based on the usage rate of the computing unit and the preset number of instances based on the occupancy rate of the memory bandwidth according to the following formula: ； Among them, represents the preset number of calculation cases based on the utilization rate of the computing unit, represents the maximum number of virtual GPU calculation cases that can run simultaneously on the GPU hardware resources, represents the utilization rate of the computing unit, represents the preset number of calculation cases based on the memory bandwidth occupancy rate, represents the memory bandwidth occupancy rate.

3. The computing power fusion method combining GPU virtualization and AI according to claim 2, wherein, Determining the edge computing node and analyzing the computing power load situation of the edge computing node includes: Using sensors pre-installed on the edge computing node to collect the node key metrics of the edge computing node in real time, where the node key metrics include the CPU utilization ratio and the memory usage ratio; Reading the working log file of the edge computing node and performing parsing on the working log file. When the parsing is successful, node key information is obtained, where the node key information includes the request processing time, the task queue length, the error log record, and the exception event description; Statistically analyzing the node key metrics and the node key information to obtain the computing power load situation of the edge computing node.

4. The computing power fusion method combining GPU virtualization and AI according to claim 3, wherein The intelligent allocation of virtual GPU instances to edge computing nodes using the neural network model includes: Initializing the allocation of virtual GPU instances to edge computing nodes through the neural network model, where each edge computing node is allocated at least one virtual GPU instance; After the allocation is successful, performing iterative training on the neural network model through the virtual GPU instances, and updating the parameters of the neural network model in a timely manner during the training process; After completing the iterative training operation, performing an adjusted allocation on the virtual GPU instances again to complete the intelligent allocation of the virtual GPU instances to edge computing nodes.

5. The computing power fusion method combining GPU virtualization and AI according to claim 4, wherein The performing iterative training on the neural network model through the virtual GPU instances and updating the parameters of the neural network model in a timely manner during the training process after the allocation is successful includes: Copying the neural network model to each virtual GPU instance to generate n copies of the neural network model loaded on the virtual GPU instances; Determining the user data, uniformly splitting the combined user data into sub-datasets, and sequentially allocating the sub-datasets to each virtual GPU instance; Using each virtual GPU instance to independently perform iterative training on the neural network model copies. During the iterative training, mirroring the iterative training of the neural network model and updating the parameters of the neural network model in a timely manner during the training process.

6. The computing power fusion method combining GPU virtualization and AI according to claim 5, characterized in that, Deploying the application program and dependencies of the Internet of Things device to the edge computing node using the standard conversion unit. After the deployment is successful, performing parameter configuration on the key device components of the virtual GPU instances, where the key device components connect the edge computing node and the cloud server backend, includes: Identifying the application program and dependencies of the Internet of Things device, and using the standard conversion unit to perform standardized packaging on the application program and dependencies; After the packaging is successful, obtaining a program image and deploying the program image to the edge computing node; Confirming the distributed file system of the program image and registering the key device components of the edge computing node to the distributed file system, where the key device components connect the edge computing node and the cloud server backend; After the registration is successful, performing performance parameter settings and system compatibility settings on the key device components to complete the parameter configuration of the key device components of the virtual GPU instances.

7. The computing power fusion method combining GPU virtualization and AI according to claim 6, characterized in that The registering the key device components of the edge computing node to the distributed file system includes: Initialize the distributed file system and send the configuration information and status data of the key device components to the distributed file system; Use the distributed file system to compile the configuration information and status data of the key device components. After successful compilation, use the distributed file system to automatically identify the key device components of the edge computing nodes; After successful identification, generate a unique identifier to complete the registration of the key device components of the edge computing nodes to the distributed file system.

Images