Resource state monitoring system and method of heterogeneous artificial intelligence all-in-one machine

Abstract

The invention discloses a resource state monitoring system and method of a heterogeneous artificial intelligence all-in-one machine. The system comprises a resource monitoring module and a data analysis module, the resource monitoring module and the data analysis module are connected with the artificial intelligence all-in-one machine, and the artificial intelligence all-in-one machine comprises a chip module, a communication module, a storage module, an input and output module and a power supply module. The resource monitoring module is used for carrying out resource state data state monitoring on the chip module, the communication module and the storage module of the artificial intelligence all-in-one machine and is also used for generating and comparing a large model of resource state data; and the data analysis module is used for integrating and analyzing the resource state data monitored by the data monitoring module and is also used for subsequent resource state optimization of the artificial intelligence integrated machine. The resource state is adjusted and optimized in the one-time calculation process of the artificial intelligence all-in-one machine, the stability of the calculation resource state of the artificial intelligence all-in-one machine is improved, and the calculation performance is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of computing system resource status monitoring, and in particular to a resource status monitoring system and method for a heterogeneous artificial intelligence all-in-one machine. Background Art

[0002] With the rapid development of artificial intelligence technology, heterogeneous artificial intelligence all-in-one machines have been widely used in various fields. These all-in-one machines integrate multiple key modules, such as chip modules for complex computing processing, communication modules for fast data transmission, storage modules for data storage, input and output modules for user interaction and data display, and power supply modules for stable power support for the entire device.

[0003] However, in actual application, how to effectively monitor and manage the resource status of these all-in-one machines has become a key issue. The existing monitoring technology has many shortcomings, such as the inability to comprehensively and accurately monitor a variety of key resource status information such as chip utilization, chip load status, running time, disk read and write status, disk read and write speed, available memory capacity, number of communication connections, communication speed, number of tasks, device computing amount and computing speed.

[0004] Traditional monitoring systems can often only monitor a single module or some key indicators, making it difficult to form a comprehensive resource status monitoring system. This makes it impossible to accurately understand the overall operating status of the all-in-one machine, and it is difficult to make timely and effective adjustments and optimizations when resources are not used reasonably. For example, when the chip load is too high and other module resources are idle, it is impossible to quickly and effectively allocate resources, which may cause the performance of the all-in-one machine to decline, affecting the efficiency and quality of task execution.

[0005] At the same time, the existing analysis methods are also relatively limited. It is impossible to conduct in-depth meta-analysis and calculation of the monitored data, and it is difficult to accurately compare the resource status data with the hardware and software usage parameters, and it is impossible to provide strong data support and decision-making basis for the resource status optimization of the all-in-one machine. Moreover, many monitoring systems will occupy the equipment hardware resources of the all-in-one machine during operation, affecting the performance and computing power of the all-in-one machine itself, and further reducing the work efficiency of the all-in-one machine.

[0006] Therefore, there is an urgent need for a resource status monitoring system and method that can be comprehensive, efficient, and does not affect the performance of the all-in-one machine itself, so as to achieve refined management and optimization of heterogeneous artificial intelligence all-in-one machines. Summary of the invention

[0007] Purpose of the invention: In view of the shortcomings of the prior art, the present invention provides a resource status monitoring system and method for a heterogeneous artificial intelligence all-in-one machine, which monitors and analyzes the resource status of the all-in-one machine and improves the computing performance and operating stability of the heterogeneous artificial intelligence all-in-one machine.

[0008] Technical solution: In the first aspect, a resource status monitoring system for a heterogeneous artificial intelligence all-in-one machine is provided, wherein the artificial intelligence all-in-one machine includes a chip module, a communication module, a storage module, an input / output module, and a power supply module, and the monitoring system includes a resource monitoring module, a data analysis module, and a calling module. The resource monitoring module is used to perform resource status data status monitoring on the chip module, communication module, and storage module of the artificial intelligence all-in-one machine, and is also used to generate a large data model based on the resource status data, and provide a data intuitive display channel; the data analysis module is used to integrate and analyze the resource status data monitored by the data monitoring module, and is also used for subsequent resource status optimization of the artificial intelligence all-in-one machine; the calling module is used to adjust and call the resource status of the chip module, communication module, and storage module of the artificial intelligence all-in-one machine in real time according to the analysis results of the data analysis module.

[0009] Furthermore, the resource monitoring module includes a data acquisition port, a data collection module, a data calling module, a data model generation module and a data model processing module; the data acquisition port is used to collect and transmit the physical state and operating resource status data of the artificial intelligence all-in-one machine; the data collection module is used to collect the data transmitted by the data acquisition port and store it temporarily; the data calling module is used to call the data collected and stored by the data collection module, and call the key data monitored in the data; the data model generation module is used to establish a large data model of the resource status with the key resource status data input by the data calling module; the data model processing module is used to intuitively digitize the established large data model, so that users can view the resource status more intuitively.

[0010] Furthermore, the physical state of the artificial intelligence all-in-one machine includes: one or more of device temperature, ambient humidity, and power supply voltage; the operating resource status data of the artificial intelligence all-in-one machine includes: one or more of chip utilization rate, chip load status, running time, disk read and write status, disk read and write speed, available memory capacity, number of communication connections, communication speed, number of tasks, device computing amount and computing speed.

[0011] Furthermore, the data acquisition port monitors the physical state of the artificial intelligence all-in-one machine by reading the signal sent by the temperature sensor, humidity sensor or voltage sensor set at the key position of the artificial intelligence all-in-one machine.

[0012] Furthermore, the data acquisition port monitors the operating resource status data of the artificial intelligence integrated machine including:

[0013] For chip utilization rate collection, the data collection port communicates with the chip's built-in performance monitoring registers or related system management interfaces, regularly reads the data in these registers, and calculates the ratio of the chip's actual working time to the total time within a certain period of time according to a preset algorithm, thereby obtaining the chip utilization rate;

[0014] For monitoring the chip load status, the load monitoring interface provided by the chip is used or the power supply current and voltage changes of the chip are monitored to indirectly determine the load status. The data acquisition port is connected to the power management circuit part of the chip. The current and voltage sensors are used to monitor the current and voltage fluctuations of the chip in real time during operation. The load status of the chip is determined according to the preset current and voltage threshold ranges.

[0015] For the collection of running time, the data collection port is connected to the system clock module of the chip to read the system time information of the chip. When the AI ​​all-in-one machine is started, the data collection port records the startup time, and in each subsequent collection cycle, the running time of the chip is calculated by the difference with the current time;

[0016] For monitoring the disk read / write status, the data acquisition port communicates with the disk controller and learns the disk read / write status by reading the disk controller's status register.

[0017] For the measurement of disk read / write speed, the data acquisition port calculates the disk read / write speed by recording the amount of data read / written by the disk within a period of time and dividing it by the time interval;

[0018] For monitoring the available memory capacity, the data acquisition port is connected to the memory management unit, and the data of memory usage is obtained by reading the relevant registers of the memory management unit, and the available memory capacity is calculated;

[0019] For statistics on the number of communication connections, the data acquisition port interacts with the communication module and obtains data on the number of communication connections by reading the connection status register of the communication module or the relevant statistical data interface;

[0020] For monitoring of communication speed, the data acquisition port calculates the communication speed by analyzing the amount of data transmitted by the communication module within a period of time;

[0021] For the statistics of the number of tasks, the data collection port communicates with the task management module of the operating system and periodically queries the task management module to obtain the data of the number of tasks;

[0022] For monitoring the computing power of the device, the data acquisition port estimates the computing power of the device by analyzing the amount of data processed by the chip and the number of computing instructions executed within a period of time;

[0023] For the measurement of computing speed, the data acquisition port calculates the computing speed by recording the time required to complete a certain computing task.

[0024] Furthermore, the process of establishing the big data model by the data model module is as follows:

[0025] Aggregate the data, obtain the status data collected through the data collection port, and collect the calculation data according to different calculation task types;

[0026] Classify data by type, and organize status data and calculation data separately;

[0027] Using data modeling technology, according to the characteristics of classified data and their interrelationships, the basic framework of the data model is constructed to store different types of data using different data structures;

[0028] Through attribute analysis, attribute labels and indexes are added to the data model so that the distribution of the same type of data on different attributes can be quickly viewed, which facilitates the statistics of the total amount of data with different characteristics.

[0029] Furthermore, the data analysis module includes a meta-analysis module, a data comparison module and a data storage module; the meta-analysis module is used to perform in-depth meta-analysis and calculation on the resource status data of the artificial intelligence all-in-one machine; the data comparison module is used to compare the resource status data calculated by the in-depth analysis with the existing hardware and software usage parameters; the data storage module is used to store the results of the in-depth analysis and calculation, so as to facilitate the subsequent rapid call of the resource status allocation data by the artificial intelligence all-in-one machine during the same calculation.

[0030] Furthermore, the meta-analysis module performs in-depth meta-analysis and calculation on the resource status data of the artificial intelligence all-in-one machine, including:

[0031] The data input into the meta-analysis module mainly includes various key data of the AI ​​all-in-one machine during operation, including running calculation amount, calculation progress, calculation content, calculation speed and calculation type;

[0032] The meta-analysis module is based on the running computing volume of the AI ​​all-in-one machine, and statistics the actual utilization rate of the AI ​​all-in-one machine chip using the computing content. It evaluates the core utilization rate required for allocating computing content according to multiple groups of computing progress, computing volume, computing speed and computing type. Based on different task characteristics and the current computing status, it comprehensively analyzes and calculates the optimal resource ratio that the chip should allocate to each task while ensuring the smooth progress of each computing task, that is, the core utilization rate. Finally, it outputs the optimal calling parameters for the hardware and software required for the current computing volume in the AI ​​all-in-one machine computing process.

[0033] Furthermore, the resource monitoring module and the data analysis module both adopt hardware hybrid management technology and utilize distributed computing to achieve resource monitoring and data analysis.

[0034] In a second aspect, a method for monitoring the resource status of a heterogeneous artificial intelligence all-in-one machine is provided, wherein the artificial intelligence all-in-one machine includes a chip module, a communication module, a storage module, an input and output module, and a power supply module, and the artificial intelligence all-in-one machine is connected to a resource monitoring module, a data analysis module, and a calling module, and the artificial intelligence all-in-one machine performs task calculations on input data; the resource monitoring module includes a data acquisition port, a data collection module, a data calling module, a data model generation module, and a data model processing module; the data analysis module includes a meta-analysis module, a data comparison module, and a data storage module; the method includes:

[0035] The artificial intelligence all-in-one machine inputs the required software and data for calculation and operation into the artificial intelligence all-in-one machine through the input port of the input and output module; the chip module, communication module and storage module of the artificial intelligence all-in-one machine calculate the data or run the software, and different hardware resource utilization is generated during the operation. The resource monitoring module monitors the resource status of the artificial intelligence all-in-one machine, including chip utilization rate, chip load status, running time, disk read and write status, disk read and write speed, available memory capacity, number of communication connections, communication speed, number of tasks, equipment calculation amount and calculation speed. The monitoring data is collected into the resource monitoring module through the data acquisition port, and then the resource status data is transmitted to the data collection module. The data collection module stores the collected resource status data for a short time. At this time, the resource status data is called through the data. Module, calls and transmits the key data in the collected resource status data to the data model generation module, the data model generation module generates a large data model from the resource status data, and generates intuitive image data for user viewing, and then transmits the generated data to the data model processing module, and then the data model processing module generates processed data from the model data, and then transmits it to the data analysis module; the meta-analysis module in the data analysis module obtains abnormal values ​​by analyzing and calculating the data, and then uses the data comparison module to compare the abnormal values ​​of the resource status of this calculation and operation, and after comparison, generates a resource allocation state that is more suitable for the operation of this artificial intelligence all-in-one machine, and then feeds this data back to the artificial intelligence all-in-one machine, and the artificial intelligence all-in-one machine calls different chips to be more suitable for the chip for this calculation or software operation.

[0036] Beneficial effects: Compared with the prior art, the present invention has the following beneficial effects:

[0037] 1. Comprehensive monitoring

[0038] Multi-module detailed monitoring:

[0039] The present invention can monitor the resource status data status of the chip module, communication module and storage module of the artificial intelligence integrated machine in an all-round way. For example, in terms of chip modules, not only can the chip usage rate and chip load status be monitored, but also subtle but key indicators such as running time can be paid attention to, which helps to accurately grasp the working conditions of the chip and timely discover potential performance bottlenecks.

[0040] For the communication module, the number of communication connections and the communication speed can be monitored in detail to ensure the stability and efficiency of data transmission. For example, in the scenario of large-scale data transmission, such as the real-time transmission of high-definition video streams or the download of large-scale data files, the load of the communication module can be accurately evaluated so that the data transmission strategy can be adjusted in time to avoid data transmission delays or losses caused by communication congestion.

[0041] In terms of storage modules, it not only pays attention to the disk read and write status, but also accurately monitors the disk read and write speed and available memory capacity, which is crucial for applications that need to read and write data frequently (such as database management systems or video editing software). Based on these detailed monitoring data, data storage and reading strategies can be optimized to improve overall system performance.

[0042] Input and output module monitoring includes:

[0043] The present invention also includes input and output modules in the monitoring scope, including data input modules, image output modules, text output modules and audio output modules. This enables a comprehensive understanding of the interaction between the all-in-one machine and the user during the monitoring process. For example, in some application scenarios that require real-time response, such as virtual reality interactive systems or online voice recognition systems, the system resource allocation can be adjusted according to the resource status data of the input and output modules to ensure the smoothness of the user experience.

[0044] For example, when a large amount of data input or high-resolution image output demand is detected, resource allocation can be dynamically adjusted to prioritize the performance of input and output modules and avoid freezes or delays caused by insufficient resources.

[0045] 2. Data processing and analysis capabilities

[0046] Advanced data model generation and processing:

[0047] The resource monitoring module of the present invention has a unique data model generation and processing capability. Through the data model generation module, key resource status data can be established as a data model of resource status. This large model can integrate the complex data relationships of multiple modules, such as performing correlation analysis on data such as chip utilization, communication speed, and disk read and write speed, so as to more comprehensively understand the overall operating status of the integrated machine.

[0048] The data model processing module can process the established data model intuitively, so that users can view the resource status more intuitively. For example, it can display resource usage trends and resource allocation ratios between modules in the form of charts, so that administrators can quickly and accurately understand and evaluate the resource status of the all-in-one machine, and make more scientific decisions.

[0049] In-depth meta-analysis and comparison:

[0050] The meta-analysis module in the data analysis module can perform in-depth meta-analysis and calculation on resource status data. This in-depth analysis can dig out hidden patterns and potential problems behind the data. For example, by performing time series analysis on resource status data over a period of time, certain periodic resource usage peaks or valleys can be discovered, so that resource allocation can be prepared in advance.

[0051] The data comparison module can compare the resource status data calculated by deep analysis with the existing hardware and software usage parameters, which provides an accurate basis for resource optimization. For example, when it is found that the actual resource usage is different from the optimal resource configuration designed by the software, the software parameters or hardware resource allocation can be adjusted in a targeted manner to achieve the best performance.

[0052] 3. Resource optimization capabilities

[0053] Real-time resource calling and optimization:

[0054] The artificial intelligence integrated machine of the present invention is equipped with a calling module, which can adjust and call the resource status of the chip module, communication module and storage module in real time. For example, when it is detected that a task requires a large amount of computing resources, more chip resources can be dynamically allocated to the task; when the network communication load is high, the parameters of the communication module can be adjusted to improve communication efficiency.

[0055] The optimization suggestions provided by the data analysis module can optimize the resource status of the AI ​​all-in-one machine. For example, according to the data analysis results, the task scheduling strategy is automatically adjusted to prioritize resources for key tasks or tasks with higher user priority, thereby improving the overall system operation efficiency and response speed.

[0056] Does not occupy the hardware resources of the all-in-one device:

[0057] The resource monitoring module and data analysis module of the present invention both adopt hardware hybrid management technology, use distributed computing to achieve resource monitoring and data analysis, and do not occupy the device hardware of the artificial intelligence all-in-one machine at all. Compared with the traditional monitoring system, this greatly reduces the impact on the performance of the all-in-one machine itself.

[0058] For example, in some application scenarios with limited resources, such as embedded artificial intelligence devices or small edge computing all-in-one machines, traditional monitoring systems may cause the performance of the all-in-one machine to decline due to the occupation of device hardware resources, affecting its normal operation. However, the design of the present invention can ensure that the all-in-one machine can maintain efficient computing power and response speed while performing resource monitoring and analysis.

[0059] 4. Efficient computing power supported by cloud computing

[0060] Independent data analysis of computing power:

[0061] The meta-analysis module uses cloud computing, so that the data analysis module does not occupy the computing power of the AI ​​all-in-one machine itself when monitoring and analyzing the resource status of the AI ​​all-in-one machine. This means that the all-in-one machine can concentrate all its computing power on executing user tasks, ensuring the performance stability of the all-in-one machine when performing complex computing tasks.

[0062] For example, in application scenarios that require a lot of computing power, such as large-scale data training or real-time intelligent reasoning, traditional monitoring systems may cause computing task delays or performance degradation due to sharing computing power with the all-in-one machine. The data analysis module supported by cloud computing in the present invention can quickly complete the analysis and processing of resource status data without affecting the computing power of the all-in-one machine, providing timely support for the resource optimization of the all-in-one machine.

[0063] Fast resource call response:

[0064] Cloud computing also enables the data analysis module to quickly process a large amount of resource status data and provide timely resource optimization suggestions based on the analysis results. For example, when dealing with sudden high-load tasks, cloud computing can quickly analyze the resource status of the all-in-one machine and guide the calling module to quickly allocate resources, ensuring that the all-in-one machine can adapt to changes in task requirements in the shortest possible time, thereby improving the reliability and flexibility of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a system schematic diagram of the artificial intelligence all-in-one machine of the present invention;

[0066] Figure 2 It is a system schematic diagram of the chip module of the present invention;

[0067] Figure 3 A system schematic diagram of a resource monitoring module of the present invention;

[0068] Figure 4 is a system schematic diagram of a data analysis module of the present invention;

[0069] Figure 5 The present invention is a monitoring and use flow chart of the artificial intelligence all-in-one system. DETAILED DESCRIPTION

[0070] The technical solutions in the embodiments of the present invention will be described clearly and completely below with reference to the accompanying drawings.

[0071] See also Figure 1 This embodiment provides a resource status monitoring system for a heterogeneous artificial intelligence all-in-one machine, including a chip module, a communication module, a storage module, an input-output module, and a power supply module. The chip module, the communication module, the storage module, the input-output module, and the power supply module are applied to the artificial intelligence all-in-one machine, and the artificial intelligence all-in-one machine is connected to a resource monitoring module and a data analysis module.

[0072] Chip module: As the core of the AI ​​all-in-one machine, the chip module usually includes high-performance processors such as CPU, GPU, FPGA or ASIC. These processors are responsible for executing complex algorithms and computing tasks and are the basis for AI computing.

[0073] Communication module: The communication module is responsible for data exchange between the AI ​​machine and other devices or networks. It ensures high-speed and stable communication between devices through wired or wireless methods such as Ethernet, Wi-Fi, 5G, etc.

[0074] Storage module: The storage module is used to store the operating system, application programs, algorithm models, and data required for the operation of the AI ​​all-in-one machine. It usually includes high-speed storage devices such as hard disks and SSDs to meet the needs of fast reading and writing of large amounts of data.

[0075] Input and output module: The input and output module includes various interfaces and devices, such as USB, HDMI, DisplayPort, etc., which are used to connect external devices to realize data input and output and device control. Specifically, the input and output module includes a data input module, an image output module, a text output module, and an audio output module, which are used for the input of computing task data of the AI ​​all-in-one machine, as well as the output of data, images, and audio after the computing task is completed.

[0076] Power supply module: The power supply module provides a stable power supply for the AI ​​all-in-one machine to ensure the normal operation of each module. The power supply module usually includes a power adapter, battery, etc. to meet the needs of different usage scenarios.

[0077] Resource monitoring module: The resource monitoring module is used to monitor the operating status of the AI ​​all-in-one machine in real time, including the usage of CPU, memory, storage, network and other resources. By monitoring data, resource allocation can be optimized and equipment operation efficiency can be improved.

[0078] Data analysis module: The data analysis module is one of the core functions of the AI ​​all-in-one machine, responsible for processing and analyzing input data. By using machine learning, deep learning and other algorithms, it can mine, predict and analyze data and provide users with valuable information.

[0079] See also Figure 2 According to an embodiment of the present invention, the chip module includes a CPU, a GPU, an ASIC, an FPGA, a TPU and an NPU. The above chips are all integrated in the chip module of the artificial intelligence all-in-one machine. The chip module is provided with a corresponding number of interfaces according to the chip type, and is connected to the artificial intelligence all-in-one machine using different chip interfaces. The above chips are all connected to the artificial intelligence all-in-one machine and supplied to the artificial intelligence all-in-one machine for use, so as to meet different computing requirements for multiple scenarios, facilitate corresponding chip calls, and improve the efficient use of computing resources of the artificial intelligence all-in-one machine.

[0080] GPU (graphics processing unit) was originally used for image processing. It has powerful parallel computing capabilities and is suitable for processing high-dimensional data such as images and videos. The advantages of GPU are maturity, versatility, and easy programming, but it has the disadvantages of high power consumption, high cost, and performance limited by memory bandwidth.

[0081] FPGA (Field Programmable Gate Array) is a chip that can be reconfigured as needed to implement customized hardware logic and is suitable for processing different algorithms and models. The advantages of FPGA are flexibility, customizability, and low latency, but the disadvantages are high programming difficulty, high power consumption, and limited resources.

[0082] ASIC (Application Specific Integrated Circuit) is a chip designed for specific applications. It can achieve the most optimized hardware architecture and is suitable for processing fixed tasks and models. The advantages of ASIC are high performance, low power consumption and low cost, but the disadvantages are long development cycle, unchangeability and non-universality.

[0083] NPU (neural network processor) is a chip designed specifically for neural networks. It can achieve efficient matrix operations and data stream processing, and is suitable for processing artificial intelligence tasks such as deep learning. The advantages of NPU are high performance, low power consumption, and easy programming, but its disadvantages are novel technology, poor compatibility, and inconsistent standards.

[0084] TPU (Tensor Processing Unit) is an AI accelerator designed specifically for accelerating deep learning workloads. It is specially optimized for large-scale matrix multiplication operations and provides very high efficiency and speed.

[0085] The above are collectively referred to as AI chips, also known as AI accelerators or computing cards, which are modules specifically used to handle a large number of computing tasks in artificial intelligence applications. The non-computing tasks of the artificial intelligence all-in-one machine are handled by the CPU, which meets the different computing needs of multiple scenarios and improves the resource status and stability of the artificial intelligence all-in-one machine.

[0086] In the present invention, the resource monitoring module is used to monitor the resource status data status of the chip module, communication module and storage module of the artificial intelligence integrated machine, and is also used for large model generation and comparison of resource status data, and provides a data intuitive display channel. The data analysis module is used to integrate and analyze the resource status data monitored by the data monitoring module, and is also used for subsequent resource status optimization of the artificial intelligence integrated machine.

[0087] According to an embodiment of the present invention, both the resource monitoring module and the data analysis module adopt hardware hybrid management technology. The resource monitoring module can monitor hardware resources of different types and brands, including servers, storage devices, network devices, etc., to ensure that the operating status of all hardware resources is under control. The data analysis module can process data from different hardware architectures, such as x86, ARM, etc., which improves the flexibility and scope of data processing. Through the hardware hybrid management technology, the resource monitoring module can display the real-time status of all hardware in a unified interface, which is convenient for administrators to centrally manage. By managing high-performance computing resources, the data analysis module can quickly complete complex data analysis tasks, such as large-scale parallel processing, real-time data analysis, etc. This technology enables resource monitoring and data analysis modules to work seamlessly in multi-cloud and hybrid cloud environments. In addition, the resource monitoring module adopts a hybrid cloud management platform that can freely convert the working environment and information of the artificial intelligence all-in-one machine between reality and virtuality, and can quickly take over the original virtualization cluster without affecting the normal operation of any application of the artificial intelligence all-in-one machine. In actual use, the artificial intelligence all-in-one machine is connected to the Internet, and the Internet cloud computing method is used to split the large task calculated by the artificial intelligence all-in-one machine into multiple groups of small tasks, which are then shared by the resources of the Internet cloud server. After a large number of small tasks are calculated, the calculated data is transmitted from the Internet to the artificial intelligence all-in-one machine, and these small task data are integrated again to complete cloud computing. The process does not take up too many computing resources of the artificial intelligence all-in-one machine, and can better handle computing tasks of large amounts of data.

[0088] The artificial intelligence all-in-one machine resource status monitoring system also includes a calling module for real-time adjustment and calling of the resource status of the chip module, communication module and storage module of the artificial intelligence all-in-one machine. The artificial intelligence all-in-one machine can call chips with different functions and call computing chips that are more advantageous for the current computing type to perform artificial intelligence computing. For example, if the artificial intelligence all-in-one machine is in a computing project of computing compilation, the chip used is an FPGA chip, which is suitable for programming and compiling data. When the artificial intelligence all-in-one machine completes the calculation of the above-mentioned calculation content, the content of the next calculation tends to be image processing. At this time, the FPGA chip's ability to process images decreases, which reduces the speed of calculation. At this time, the chip module calls the module to set the GPU chip call as the main computing chip of the artificial intelligence all-in-one machine, and then processes the project in the image. In this way, the artificial intelligence all-in-one machine can better adapt to the calculation of different projects and improve the calculation accuracy and efficiency of the artificial intelligence all-in-one machine. And through the data analysis module, the resource status of the artificial intelligence all-in-one machine can be optimized. The calling module can adjust the chip of the artificial intelligence all-in-one machine in real time after the data analysis module identifies the feedback data, and use different artificial intelligence all-in-one machine chips to call, so that the chip can adapt to the artificial intelligence all-in-one machine calculation, thereby reducing the abnormal resource status burden of the artificial intelligence all-in-one machine and improving the stability of the use of the artificial intelligence all-in-one machine.

[0089] See also Figure 3 The resource monitoring module includes a data acquisition port, a data collection module, a data call module, a data model generation module and a data model processing module. The data acquisition port is used to collect and transmit the resource status data of the artificial intelligence integrated machine; the data collection module is used to collect the data transmitted by the data acquisition port and store it temporarily; the data call module is used to call the data collected and stored by the data collection module, and call the key data monitored in the data; the data model generation module is used to establish a data model of the resource status with the key resource status data input by the data call module; the data model processing module is used to intuitively process the established data model into data, so as to facilitate a more intuitive viewing of the resource status.

[0090] The resource monitoring module can monitor the status and data of the AI ​​all-in-one machine's operation process by collecting various indicators during the operation of the AI ​​all-in-one machine, such as the device's temperature, computing usage, computing speed, and computing accuracy.

[0091] Regarding the physical state of the artificial intelligence all-in-one machine, it can be monitored by setting sensors at key locations of the artificial intelligence all-in-one machine, such as temperature sensors, humidity sensors, voltage sensors, etc.

[0092] For the temperature collection of the device, the data acquisition port in the present invention is connected to the temperature sensors distributed at the key positions of the artificial intelligence all-in-one machine. These temperature sensors can sense the temperature changes of different parts of the device in real time, and transmit the temperature data to the data acquisition port in the form of electrical signals. The data acquisition port continuously receives these electrical signals and converts them into digital signals for storage and subsequent processing. For example, a high-precision temperature sensor is installed near the chip heat sink of the all-in-one machine. When the chip is running and heating, the temperature sensor can quickly detect the temperature change and transmit the temperature data to the data acquisition port in real time.

[0093] For humidity monitoring, the humidity sensor is also connected to the data acquisition port. The humidity sensor usually adopts the capacitive or resistive principle and can accurately detect the humidity level of the environment around the device. The data acquisition port reads the humidity data from the humidity sensor at a certain time interval so as to timely grasp the humidity condition of the environment where the device is located and prevent damage to the device due to excessive or low humidity. For example, a humidity sensor is installed inside the chassis of the all-in-one machine, and the data acquisition port reads the humidity data at a certain interval (such as every 5 minutes) to ensure that the device is running in a suitable humidity environment.

[0094] The voltage sensor is connected to the data acquisition port to monitor the power supply voltage of the AI ​​​​machine. The voltage sensor can detect the voltage value of the power input in real time and transmit it to the data acquisition port. The data acquisition port can determine whether the power supply is stable and whether there is excessive voltage fluctuation by analyzing the voltage data. For example, when the voltage of the machine suddenly drops during operation, the data acquisition port can detect it in time and issue an alarm so that appropriate protective measures can be taken.

[0095] The data generated by the AI ​​all-in-one machine during operation and the system occupancy and usage can be collected and fed back through chip-specific data collection technology. The resource status information of the AI ​​all-in-one machine includes chip utilization, chip load status, running time, disk read and write status, disk read and write speed, available memory capacity, number of communication connections, communication speed, number of tasks, device computing capacity and computing speed.

[0096] For chip utilization rate collection, the data acquisition port communicates with the chip's built-in performance monitoring registers or related system management interfaces. These registers or interfaces usually record the chip's working status and resource usage in different time periods. The data acquisition port periodically reads the data in these registers and calculates the ratio of the chip's actual working time to the total time within a certain period of time based on a preset algorithm, thereby deriving the chip utilization rate. For example, at a fixed time point every second, the data acquisition port sends a query command to the chip, and the chip returns data containing the current working status information. The data acquisition port analyzes and calculates based on this data to obtain real-time data on chip utilization.

[0097] For monitoring the chip load status, the load monitoring interface provided by the chip is used or the power supply current and voltage changes of the chip are monitored to indirectly determine the load status. The data acquisition port is connected to the power management circuit part of the chip, and the current and voltage fluctuations of the chip when working are monitored in real time through high-precision current and voltage sensors. When the chip processes a large number of computing tasks, the current and voltage will increase accordingly. The data acquisition port determines whether the chip load status is light load, medium load or heavy load based on the preset current and voltage threshold range. For example, if the current exceeds a preset high load current threshold, the data acquisition port determines that the chip is in a high load state.

[0098] For the collection of running time, the data acquisition port is connected to the system clock module of the chip to read the system time information of the chip. When the AI ​​all-in-one machine starts, the data acquisition port records the startup time, and in each subsequent acquisition cycle, the chip's running time is calculated by the difference with the current time. For example, at regular intervals (such as every minute), the data acquisition port reads the system time once, compares it with the startup time, and updates the chip's running time data.

[0099] For monitoring the disk read and write status, the data acquisition port communicates with the disk controller. The disk controller can provide information about whether the disk is currently performing a read or write operation. The data acquisition port understands the disk's read and write status by reading the disk controller's status register. For example, when the disk is writing data, the corresponding status bit of the disk controller will be set, and the data acquisition port can detect this status bit to know that the disk is in the write state.

[0100] For the measurement of disk read / write speed, the data acquisition port calculates the disk read / write speed by recording the amount of data read / written by the disk within a period of time and dividing it by the time interval. For example, within a period of time (such as 10 seconds), the data acquisition port counts the number of data bytes read or written by the disk, and then divides it by the time interval to obtain the average value of the disk read / write speed.

[0101] For monitoring the available memory capacity, the data acquisition port is connected to the memory management unit. The memory management unit can provide information about the currently used memory and available memory. The data acquisition port obtains the memory usage data by reading the relevant registers of the memory management unit and calculates the available memory capacity. For example, at regular intervals, the data acquisition port queries the memory management unit to obtain the latest memory usage data so as to keep abreast of the changes in available memory.

[0102] For statistics of the number of communication connections, the data acquisition port interacts with the communication module. The communication module can provide information on the number of communication connections currently established. The data acquisition port obtains data on the number of communication connections by reading the connection status register of the communication module or the related statistical data interface. For example, in network communication, the data acquisition port can count the number of TCP / IP connections currently established by querying the connection status information of the network interface.

[0103] For monitoring the communication speed, the data acquisition port calculates the communication speed by analyzing the amount of data transmitted by the communication module over a period of time. For example, within a period of time (such as 1 minute), the data acquisition port counts the number of data bytes received and sent by the communication module, and then divides it by the time interval to obtain the average value of the communication speed.

[0104] For the statistics of the number of tasks, the data acquisition port communicates with the task management module of the operating system. The task management module can provide information about the number of tasks currently running. The data acquisition port periodically queries the task management module to obtain data on the number of tasks. For example, at regular intervals (such as every 10 seconds), the data acquisition port sends a query request to the task management module to obtain the number of tasks currently running.

[0105] For monitoring the computing power of the device, the data acquisition port estimates the computing power of the device by analyzing the amount of data processed by the chip and the number of computing instructions executed within a period of time. For example, the data acquisition port can count the number of addition, multiplication and other operation instructions executed by the chip within a certain period of time, as well as the number of data bytes processed, and estimate the computing power of the device based on the preset calculation model.

[0106] For the measurement of computing speed, the data acquisition port calculates the computing speed by recording the time required to complete a certain computing task. For example, the data acquisition port starts a specific computing task (such as performing a set of large-scale matrix operations), records the start time and end time, and then calculates the computing speed based on the computing amount of the task and the time difference.

[0107] During the operation of the heterogeneous AI all-in-one machine, the results are continuously generated by calculating the data. During the calculation process, the AI ​​all-in-one machine will generate status data and a large amount of calculation data. Based on the various physical data and operation data collected by the above data collection port, the chip module, communication module and storage module of the AI ​​all-in-one machine can be fully monitored for resource status data.

[0108] In addition, the present invention also includes input and output modules in the monitoring scope, including data input modules, image output modules, text output modules and audio output modules. This enables a comprehensive understanding of the interaction between the all-in-one machine and the user during the monitoring process. For example, in some application scenarios that require real-time response, such as virtual reality interaction systems or online speech recognition systems, the system resource allocation can be adjusted according to the resource status data of the input and output modules to ensure the smoothness of the user experience. For example, when a large amount of data input or high-resolution image output requirements are detected, resource allocation can be dynamically adjusted to prioritize the performance of the input and output modules and avoid freezes or delays caused by insufficient resources.

[0109] The resource monitoring module of the present invention has unique data model generation and processing capabilities. Through the data model generation module, key resource status data can be established as a data big model of resource status. The so-called big data model is a comprehensive data model that integrates various types of data generated by the artificial intelligence all-in-one machine during operation and is constructed according to a certain data structure and logical relationship. This model not only includes status data such as chip utilization, chip load status, and running time, but also covers specific calculation data generated by various computing tasks. For example, image feature data generated in image recognition tasks, text semantic data generated in natural language processing tasks, etc., are all included in this big data model.

[0110] The process of building a big data model is as follows: First, aggregate the data. For status data, collect relevant information from various key components such as chips, communication modules, and storage modules through data acquisition ports, such as obtaining chip usage, load status, and other data from the chip, collecting communication connection numbers, communication speed, and other data from the communication module, and collecting disk read and write status, read and write speed, and other data from the storage module. For computing data, classify and collect according to different computing task types, such as image computing task data, text processing task data, etc.

[0111] Then, classify the data according to its type. Classify the status data and calculation data separately. At the same time, the calculation data can be further subdivided according to the task type, data format, etc. For example, image calculation data can be classified into one category and text calculation data into another category.

[0112] Next, we use data modeling technology to build the basic framework of the data model based on the characteristics of the classified data and the relationships between them. For example, we can use the table structure of a relational database to store and manage some structured state data, while for unstructured computing data such as images and audio, we can use specialized big data storage technologies such as distributed file systems or object storage systems.

[0113] Finally, through attribute analysis, attribute labels and indexes are added to the data model so that the distribution of the same type of data on different attributes can be quickly viewed, which is convenient for counting the total amount of data with different characteristics. In this process, the measurement method is reasonably selected. Commonly used measurement methods include: data type, data volume, number of duplicates, sum, maximum value, minimum value and mean, etc. For example, for a set of numerical calculation data, statistical indicators such as its mean and standard deviation can be calculated and added to the big data model as attribute characteristics of the data. Through such processing, the establishment of the big data model is completed.

[0114] The AI ​​integrated machine uses cloud computing monitoring with hardware hybrid management to collect data on the status of each resource of the AI ​​integrated machine and generate intuitive images or digital models for users to quickly view. Subsequent users can use the model data to achieve timely regulation and use, thereby improving the efficiency of AI integrated machine software operation and calculation.

[0115] See also Figure 4 The data analysis module includes a meta-analysis module, a data comparison module and a data storage module. The meta-analysis module is used to perform in-depth meta-analysis and calculation on the resource status data of the artificial intelligence all-in-one machine; the data comparison module is used to compare the resource status data calculated by the in-depth analysis with the existing hardware and software usage parameters; the data storage module is used to store the results of the in-depth analysis and calculation, so that the subsequent artificial intelligence all-in-one machine can quickly call the allocation data of the resource status during the same calculation.

[0116] The method for obtaining the hardware usage parameters of the existing implementation is as follows: obtain relevant data through the underlying driver or hardware monitoring interface directly connected to the hardware device. For example, the clock frequency, operating voltage and other parameters of the chip can be read through the specific interface provided by the chip manufacturer or the underlying hardware monitoring tool of the system. These tools can monitor the operating status of the hardware in real time and feed back the relevant parameter information to the data comparison module. For the hardware parameters of the storage device, such as the rotation speed of the disk, the cache size, etc., detailed information can be obtained through the disk's SMART (Self-Monitoring, Analysis and Reporting Technology) technology.

[0117] The ways to obtain software usage parameters are relatively diverse. For software at the operating system level, the data comparison module can obtain relevant parameters through the system management interface or performance monitoring tool provided by the operating system. For example, the task manager of the operating system can provide parameters such as running process information, memory usage, CPU usage time, etc. For software at the application level, some software itself will provide interfaces or log files to record key parameters during its operation, such as query execution time and cache hit rate of database software. The data comparison module can obtain software usage parameters by reading these interface data or parsing log files. In addition, special monitoring tools can be deployed in the software running environment. These tools can monitor the software's network communication traffic, file read and write operations and other behaviors in real time, and feedback relevant parameters to the data comparison module.

[0118] The data input to the meta-analysis module mainly includes various key data of the AI ​​all-in-one machine during operation, such as the amount of running calculations, calculation progress, calculation content, calculation speed and calculation type. These data come from the real-time monitoring of each module of the all-in-one machine by the data acquisition port and the tracking and recording of the calculation tasks. For example, the data acquisition port will continuously collect the working status data of the chip when processing different calculation tasks, including the amount of data processed in each calculation cycle, the calculation steps completed and the time spent. At the same time, for the ongoing calculation tasks, detailed information such as the task type, the required computing resources and the current progress will also be recorded.

[0119] The meta-analysis module is based on the computing capacity of the AI ​​machine, which is a key starting indicator that represents the total amount of data or the scale of computing tasks that the machine needs to process within a certain period of time. For example, if the machine is performing a large-scale data analysis task, the computing capacity may include the number of bytes of massive data that need to be processed and the corresponding complex computing operations.

[0120] Next, the actual utilization rate of the AI ​​all-in-one chip is counted. By interacting with the chip's monitoring interface or performance management system, the actual working status of the chip when processing computing tasks can be obtained in real time. For example, over a period of time, the proportion of the time the chip is in working state to the total time is counted to determine the actual utilization rate. This step can help understand the current usage of the chip, whether it is running under high load or there are idle resources.

[0121] Then, the core utilization required to allocate the computing content is evaluated based on the computing progress, computing content, computing speed and computing type of multiple groups. Different computing tasks have different requirements for chip resources. For example, image rendering tasks may rely more on the graphics processing capabilities of the chip, while large-scale data computing tasks focus more on the computing performance and data throughput of the chip. The meta-analysis module will comprehensively analyze and calculate the optimal resource ratio, i.e., core utilization, that the chip should allocate to each task under the premise of ensuring the smooth progress of each computing task based on these different task characteristics and the current computing status.

[0122] In the entire computing process, cloud computing technology plays an important supporting role. The cloud computing platform provides powerful computing power and data storage resources, which can quickly process large amounts of computing data and complex algorithm operations. The meta-analysis module uploads the collected data to the cloud computing platform, and uses the distributed computing architecture and advanced data analysis algorithms of the cloud computing platform to achieve efficient processing and analysis of the data. For example, the cloud computing platform can simultaneously analyze the data of multiple computing tasks in parallel, quickly calculate the best resource allocation plan for each task, and greatly improve computing efficiency and accuracy.

[0123] After the above complex calculation process, the meta-analysis module finally outputs the hardware and software better calling parameters required for the amount of calculation in the process of optimizing the calculation of the artificial intelligence all-in-one machine. These parameters include but are not limited to chip operating frequency adjustment suggestions, memory allocation strategies, communication module data transmission rate settings, and software-level task scheduling priorities. For example, if the meta-analysis module finds that a certain computing task is progressing slowly and some cores of the chip are idle, it may output a suggestion to assign the task to more chip cores for processing, and adjust the memory cache strategy accordingly to increase the data reading speed. At the same time, at the software level, the execution order of tasks may be adjusted to prioritize the execution of critical tasks in order to make full use of limited computing resources, improve the efficiency of the use of artificial intelligence all-in-one machines, and avoid competition or idleness of chip computing power.

[0124] The meta-analysis module uses cloud computing. Cloud computing accesses the Internet through heterogeneous AI all-in-one machines, accesses network-shared software and hardware resources and information through the Internet, distributes the huge amount of computing power into multiple small computing programs on demand, and performs calculations through network information resources, so that the data analysis module will not occupy the computing power of the AI ​​all-in-one machine itself when monitoring and analyzing the resource status of the AI ​​all-in-one machine, ensuring the stability of the computing power of the AI ​​all-in-one machine. The purpose of meta-analysis is to obtain more accurate conclusions by integrating data, and to be able to summarize and promote to a wider range. Using meta-analysis calculation statistics, we can better analyze the optimization points of the monitoring resource status data this time. By analyzing data, the AI ​​all-in-one machine can adjust the allocation of resources in real time and improve the computing stability and efficiency of the AI ​​all-in-one machine.

[0125] See also Figure 5 When in use, the artificial intelligence all-in-one machine inputs the required software and data for calculation and operation into the artificial intelligence all-in-one machine through the input port of the input and output module. At this time, the chip module, communication module and storage module of the artificial intelligence all-in-one machine calculate the data or run the software, and different hardware resources are utilized during the operation. At this time, the resource monitoring module monitors the resource status of the artificial intelligence all-in-one machine, including chip utilization rate, chip load status, running time, disk read and write status, disk read and write speed, available memory capacity, number of communication connections, communication speed, number of tasks, device calculation amount and calculation speed. The monitoring data is collected into the resource monitoring module through the data acquisition port, and then the resource status data is transmitted to the data collection module. The data collection module stores the collected resource status data for a short time. At this time, the collected resource status data is collected through the data call module. The key data in the resource status data of the set is called and transmitted to the data model generation module. At this time, the resource monitoring module generates intuitive image data from the resource status data to facilitate user viewing. The generated data is then transmitted to the data model processing module, and then the data model processing module generates processed data from the model data, which is then transmitted to the data analysis module. At this time, the meta-analysis module in the data analysis module obtains abnormal values ​​by analyzing and calculating the data, and then uses the data comparison module to compare the abnormal values ​​of the resource status of this calculation and run. After comparison, a resource allocation state that is more suitable for the operation of this artificial intelligence all-in-one machine is generated, and then this data is fed back to the artificial intelligence all-in-one machine, and then the artificial intelligence all-in-one machine calls different chips to be more suitable for the chip that is more suitable for this calculation or software operation, thereby making the operation of the artificial intelligence all-in-one machine more stable and efficient.

[0126] Among them, the heterogeneous AI all-in-one machine performs calculations through chips. When indicators such as the speed, accuracy and excessive chip utilization of the chip calculation appear during the calculation process, as well as the calculated data error or calculated data abnormality, the state of the AI ​​all-in-one machine and the calculated data abnormality are judged. The calculation data abnormality can be detected using the box plot anomaly detection method, also known as the box and whisker plot, which is a statistical chart that displays the distribution of a set of continuous data by calculating several key numerical points. When the data in the chart destroys continuity or the error value between the data and the overall data is large, the abnormal data in the AI ​​calculation can be detected, which is particularly suitable for detecting outliers in the data.

[0127] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A resource status monitoring system for a heterogeneous artificial intelligence integrated machine, characterized in that: The artificial intelligence all-in-one machine includes a chip module, a communication module, a storage module, an input and output module and a power supply module. The monitoring system includes a resource monitoring module, a data analysis module and a calling module. The resource monitoring module is used to monitor the resource status data status of the chip module, the communication module and the storage module of the artificial intelligence all-in-one machine, and is also used to generate a large data model based on the resource status data, and provide a data intuitive display channel; the data analysis module is used to integrate and analyze the resource status data monitored by the data monitoring module, and is also used for subsequent resource status optimization of the artificial intelligence all-in-one machine; the calling module is used to adjust and call the resource status of the chip module, the communication module and the storage module of the artificial intelligence all-in-one machine in real time according to the analysis results of the data analysis module.

2. The system according to claim 1, characterized in that The resource monitoring module includes a data acquisition port, a data collection module, a data calling module, a data model generation module and a data model processing module; the data acquisition port is used to collect and transmit the physical state and operating resource state data of the artificial intelligence all-in-one machine; the data collection module is used to collect the data transmitted by the data acquisition port and store it temporarily; The data calling module is used to call the data collected and stored by the data collecting module, and call the key data monitored in the data; The data model generation module is used to establish a data model of resource status using the key resource status data input by the data calling module; the data model processing module is used to perform intuitive data processing on the established data model to facilitate users to view resource status more intuitively.

3. The system according to claim 2, characterized in that The physical state of the artificial intelligence all-in-one machine includes: one or more of device temperature, ambient humidity, and power supply voltage; the operating resource status data of the artificial intelligence all-in-one machine includes: one or more of chip utilization rate, chip load status, running time, disk read and write status, disk read and write speed, available memory capacity, number of communication connections, communication speed, number of tasks, device computing amount and computing speed.

4. The system according to claim 3, characterized in that The data acquisition port monitors the physical state of the artificial intelligence integrated machine by reading the signals sent by the temperature sensor, humidity sensor or voltage sensor set at the key position of the artificial intelligence integrated machine.

5. The system according to claim 3, characterized in that The data acquisition port monitors the running resource status data of the artificial intelligence integrated machine including: For chip utilization rate collection, the data collection port communicates with the chip's built-in performance monitoring registers or related system management interfaces, regularly reads the data in these registers, and calculates the ratio of the chip's actual working time to the total time within a certain period of time according to a preset algorithm, thereby obtaining the chip utilization rate; For monitoring the chip load status, the load monitoring interface provided by the chip is used or the power supply current and voltage changes of the chip are monitored to indirectly determine the load status. The data acquisition port is connected to the power management circuit part of the chip. The current and voltage sensors are used to monitor the current and voltage fluctuations of the chip in real time during operation. The load status of the chip is determined according to the preset current and voltage threshold ranges. For the collection of running time, the data collection port is connected to the system clock module of the chip to read the system time information of the chip. When the AI ​​all-in-one machine is started, the data collection port records the startup time, and in each subsequent collection cycle, the running time of the chip is calculated by the difference with the current time; For monitoring the disk read / write status, the data acquisition port communicates with the disk controller and learns the disk read / write status by reading the disk controller's status register. For the measurement of disk read / write speed, the data acquisition port calculates the disk read / write speed by recording the amount of data read / written by the disk within a period of time and dividing it by the time interval; For monitoring the available memory capacity, the data acquisition port is connected to the memory management unit, and the data of memory usage is obtained by reading the relevant registers of the memory management unit, and the available memory capacity is calculated; For statistics on the number of communication connections, the data acquisition port interacts with the communication module and obtains data on the number of communication connections by reading the connection status register of the communication module or the relevant statistical data interface; For monitoring of communication speed, the data acquisition port calculates the communication speed by analyzing the amount of data transmitted by the communication module within a period of time; For the statistics of the number of tasks, the data collection port communicates with the task management module of the operating system and periodically queries the task management module to obtain the data of the number of tasks; For monitoring the computing power of the device, the data acquisition port estimates the computing power of the device by analyzing the amount of data processed by the chip and the number of computing instructions executed within a period of time; For the measurement of computing speed, the data acquisition port calculates the computing speed by recording the time required to complete a certain computing task.

6. The system according to claim 2, characterized in that The process of establishing a big data model by the data model module is as follows: Aggregate the data, obtain the status data collected through the data collection port, and collect the calculation data according to different calculation task types; Classify data by type, and organize status data and calculation data separately; Using data modeling technology, according to the characteristics of classified data and their interrelationships, the basic framework of the data model is constructed to store different types of data using different data structures; Through attribute analysis, attribute labels and indexes are added to the data model so that the distribution of the same type of data on different attributes can be quickly viewed, which facilitates the statistics of the total amount of data with different characteristics.

7. The system according to claim 1, characterized in that The data analysis module includes a meta-analysis module, a data comparison module and a data storage module; the meta-analysis module is used to perform in-depth meta-analysis and calculation on the resource status data of the artificial intelligence all-in-one machine; the data comparison module is used to compare the resource status data calculated by the in-depth analysis with the existing hardware and software usage parameters; the data storage module is used to store the results of the in-depth analysis and calculation, so as to facilitate the subsequent rapid call of the resource status allocation data by the artificial intelligence all-in-one machine during the same calculation.

8. The system according to claim 7, characterized in that The meta-analysis module performs in-depth meta-analysis and calculation on the resource status data of the artificial intelligence all-in-one machine, including: The data input into the meta-analysis module mainly includes various key data of the AI ​​all-in-one machine during operation, including running calculation amount, calculation progress, calculation content, calculation speed and calculation type; The meta-analysis module is based on the running computing volume of the AI ​​all-in-one machine, and statistics the actual utilization rate of the AI ​​all-in-one machine chip using the computing content. It evaluates the core utilization rate required for allocating computing content according to multiple groups of computing progress, computing volume, computing speed and computing type. Based on different task characteristics and the current computing status, it comprehensively analyzes and calculates the optimal resource ratio that the chip should allocate to each task while ensuring the smooth progress of each computing task, that is, the core utilization rate. Finally, it outputs the optimal calling parameters for the hardware and software required for the current computing volume in the AI ​​all-in-one machine computing process.

9. The system according to claim 1, characterized in that The resource monitoring module and the data analysis module both adopt hardware hybrid management technology and utilize distributed computing to achieve resource monitoring and data analysis.

10. A method for monitoring resource status of a heterogeneous artificial intelligence integrated machine, characterized in that: The artificial intelligence all-in-one machine includes a chip module, a communication module, a storage module, an input and output module and a power supply module. The artificial intelligence all-in-one machine is connected to a resource monitoring module, a data analysis module and a call module. The artificial intelligence all-in-one machine performs task calculations on the input data; the resource monitoring module includes a data acquisition port, a data collection module, a data call module, a data model generation module and a data model processing module; The data analysis module includes a meta-analysis module, a data comparison module and a data storage module; the method includes: The AI ​​all-in-one machine inputs the software and data required for computing and running into the AI ​​all-in-one machine through the input port of the input-output module; The chip module, communication module and storage module of the artificial intelligence all-in-one machine calculate data or run software, and different hardware resource utilization is generated during the operation. The resource monitoring module monitors the resource status of the artificial intelligence all-in-one machine, including chip utilization rate, chip load status, running time, disk read and write status, disk read and write speed, available memory capacity, number of communication connections, communication speed, number of tasks, device computing amount and computing speed. The monitoring data is collected into the resource monitoring module through the data acquisition port, and then the resource status data is transmitted to the data collection module. The data collection module stores the collected resource status data for a short time. At this time, the key data in the collected resource status data is called and transmitted through the data calling module. The data is input into the data model generation module, which generates a large data model from the resource status data and generates intuitive image data for user viewing. The generated data is then transmitted to the data model processing module, and then the data model processing module generates processed data from the model data, which is then transmitted to the data analysis module; the meta-analysis module in the data analysis module obtains abnormal values ​​by analyzing and calculating the data, and then uses the data comparison module to compare the abnormal values ​​of the resource status of this calculation run. After comparison, a resource allocation state that is more suitable for the operation of this artificial intelligence all-in-one machine is generated, and then this data is fed back to the artificial intelligence all-in-one machine. The artificial intelligence all-in-one machine calls different chips to be more suitable for the chip that is more suitable for this calculation or software operation.

Images