Multi-functional dynamically configurable multi-computing-power extension method and multi-functional dynamically configurable multi-computing-power extension system

Abstract

The invention relates to the technical field of computing power extension, in particular to a multifunctional dynamically-configurable multi-computing-power extension method and system. The method comprises the following steps that an expansion Hub board is inserted into a host system, a connection and transmission assembly in a certain concentration mode is configured on the board, two or four computing power expansion modules can be further inserted into the board, and the modules can be arranged into two groups and are respectively connected and integrated with corresponding module interfaces in a cascade mode through a certain high-speed bus; a control assembly is realized on the board based on an MCU, and integration is performed by combining a multi-module architecture and a docking host interface; carrying out calculation task real-time monitoring on each calculation power extension module, and carrying out task priority determination and calculation power resource utilization evaluation analysis to obtain a calculation task priority and calculation power resource utilization efficiency; and computing power task scheduling analysis is carried out between the host system and each computing power extension module, and a multi-computing power task utilization scheduling strategy is generated. According to the invention, functional safety requirements under different operations can be met.

Description

Technical Field

[0001] The present invention relates to the technical field of computing power expansion, and in particular, to a multi-computing power expansion method and system that are multifunctional and dynamically configurable. Background Art

[0002] The application with publication number CN116714537A describes that a computing power expansion module is connected to the sensor system, in-vehicle network system, and power supply system of a vehicle, and is used to expand the computing power of a central domain controller. The application with publication number CN117742941A uniformly schedules and manages computing power resources through a virtualized resource pool, so that users do not need to care whether the target computing power resources called come from the main computing power module or the computing power expansion module. In addition, the application with publication number CN117742944A virtualizes in-vehicle computing power through the main computing power module and the computing power expansion module, uniformly schedules and manages computing power resources, and can support elastic expansion of computing power to flexibly meet the computing power requirements of different levels of intelligent driving and operation scenarios.

[0003] In the existing technologies, although computing power can be provided to the main application platform by expanding the computing power module, the basic architecture and implementation strategy mainly face the AI technology of traditional deep learning. When large language models (LLMs) are gradually deployed in end-side applications such as industrial control and vehicles, due to the huge number of parameters of the large language models, a single computing power expansion module may not be able to independently support an AI application, and multiple computing power modules need to be unified and coordinated and integrated to work together. In addition, the expanded computing power module should have two acceleration modes, one is the AI acceleration mode for accelerating AI processing, and the other is the graphics acceleration mode for accelerating graphics computing processing. The entire system should support different hardware acceleration requirements. Moreover, the functional safety requirements of AI hardware acceleration in the application of computing power expansion are not considered in the above existing technologies. Summary of the Invention

[0004] Based on this, it is necessary for the present invention to provide a multi-computing power expansion method and system that are multifunctional and dynamically configurable to solve at least one of the above technical problems.

[0005] To achieve the above object, a multifunctional and dynamically configurable multi-computing power expansion method includes the following steps:

[0006] Step S1: Configure and insert an expansion Hub board in the host system through a hot backup and dynamically configurable mechanism. A connection and transmission component H_Switch in a certain centralized manner is configured on the Hub board. The Hub board can further insert two or four computing power expansion modules. These computing power expansion modules can be set into two groups and are cascade-connected and integrated through a certain high-speed bus to the corresponding module interfaces of the H_Switch to generate a multifunctional and dynamically expandable module architecture;

[0007] Step S2: Implement a control component by introducing an H_MCU and connecting for configuration, and integrate and build it in combination with the multi-functional and dynamically extensible module architecture, H_Switch, and the docking host interface; monitor the computing tasks of each computing power expansion module in the multi-functional and dynamically extensible module architecture in real time through the corresponding H_MCU in the extended Hub board to obtain the computing power operation power consumption, task computing power calculation amount, and task computing power loading duration corresponding to the computing tasks on each expansion module.

[0008] Step S3: Obtain the real-time demand for computing tasks corresponding to the host system, and determine the task priority based on the real-time demand for computing tasks for the task computing power calculation amount and task computing power loading duration corresponding to the computing tasks on the expansion module, so as to obtain the corresponding computing task priority on the expansion module.

[0009] Step S4: Evaluate and analyze the utilization of computing power resources of each computing power expansion module in the multi-functional and dynamically extensible module architecture based on the computing power operation power consumption corresponding to the computing tasks on the expansion module to obtain the corresponding computing power resource utilization efficiency on the expansion module; perform computing power task scheduling analysis between the host system and each computing power expansion module based on the corresponding computing task priority and computing power resource utilization efficiency on the expansion module, and generate a multi-computing power task utilization scheduling strategy.

[0010] Further, the hot backup and dynamically configurable mechanism described in Step S1 is specifically that two groups of computing power expansion module groups are hot backups of each other. When one module in any group fails, it is dynamically adjusted so that each group only includes one valid module and the two groups are hot backups of each other; when all modules in any group fail and all modules in the other group are valid, it is dynamically adjusted to re-divide the two valid modules into two groups, with each group only including one module and the two groups being hot backups of each other.

[0011] Further, the computing power expansion module group includes two expansion modules or one expansion module, and the two expansion modules or one expansion module are specifically an AI acceleration working module or a graphics acceleration working module.

[0012] Further, the cascaded connection can process parallel processing tasks in the corresponding working mode, specifically including corresponding tensor parallel, data parallel, and pipeline parallel tasks.

[0013] Further, the working mode includes an AI acceleration working mode and a graphics acceleration working mode.

[0014] Further, Step S3 includes the following steps:

[0015] Step S31: Obtain the real-time demand for computing tasks corresponding to the host system.

[0016] Step S32: Based on the real-time demand of the computing task, use the computing task progress calculation formula to perform task progress evaluation and calculation on the task computing power calculation amount and the task computing power loading duration corresponding to the computing task on the expansion module, so as to obtain the corresponding computing task progress on the expansion module;

[0017] Step S33: Determine the task priority of the corresponding computing task on the expansion module based on the corresponding computing task progress on the expansion module, so as to obtain the corresponding computing task priority on the expansion module.

[0018] Further, the computing task progress calculation formula described in Step S32 is specifically:

[0019]

[0020] In the formula, P is the computing task progress, Q is the real-time demand of the computing task, T c is the task computing power loading duration corresponding to the computing task on the expansion module, t is the time variable parameter, C(t) is the task computing power calculation amount of the computing task on the expansion module at time t, C max is the maximum computing amount required for the computing task, α1 is the computing amount influence weight coefficient, and η is the correction coefficient of the computing task progress.

[0021] Further, Step S4 includes the following steps:

[0022] Step S41: Obtain the computing task type and the computing task running status corresponding to the current computing task on each computing power expansion module;

[0023] Step S42: Based on the computing task type and the computing task running status corresponding to the current computing task, perform computing power resource allocation analysis on each computing power expansion module within the multi-functional dynamically expandable module architecture to obtain the computing power resource allocation amount corresponding to the computing task on the expansion module;

[0024] Step S43: Based on the computing power operation power consumption and the computing power resource allocation amount corresponding to the computing task on the expansion module, perform computing power resource utilization evaluation and analysis on the corresponding computing power expansion module to obtain the corresponding computing power resource utilization efficiency on the expansion module;

[0025] Step S44: Based on the corresponding computing task priority and the computing power resource utilization efficiency on the expansion module, perform computing power task scheduling analysis between the host system and each computing power expansion module to generate a multi-computing power task utilization scheduling strategy.

[0026] Further, the multi-computing power task utilization scheduling strategy described in step S44 is specifically that for computing tasks with short running time and high computing power requirements, a high priority is assigned to them so that they can be preferentially allocated to the computing power expansion module with fast computing speed and timely response of computing power resources; while for computing tasks with long running time and relatively stable computing power, a low priority is assigned to them so that they can be allocated to the computing power expansion module with relatively low resource utilization rate.

[0027] Further, the present invention also provides a multi-functional and dynamically configurable multi-computing power expansion system for executing the above-mentioned multi-functional and dynamically configurable multi-computing power expansion method. The multi-functional and dynamically configurable multi-computing power expansion system includes:

[0028] An expansion module cascade design function component for configuring and inserting two sets of computing power expansion modules in the host system through a hot backup and dynamically configurable mechanism and integrally designing them by cascading connection with the corresponding module interfaces through a high-speed bus to generate a multi-functional and dynamically expandable module architecture;

[0029] An expansion Hub board computing task monitoring function component for implementing a control component by introducing an H_MCU and connecting and configuring it, and integrating and constructing it in combination with the multi-functional and dynamically expandable module architecture, the H_Switch and the docking host interface; real-time monitoring of the computing tasks of each computing power expansion module in the multi-functional and dynamically expandable module architecture through the corresponding H_MCU in the expansion Hub board to obtain the computing power operation power consumption, task computing power calculation amount, and task computing power loading duration corresponding to the computing tasks on each expansion module;

[0030] A computing task priority determination function component for obtaining the real-time demand of the computing tasks corresponding to the host system and determining the task priority of the task computing power calculation amount and the task computing power loading duration corresponding to the computing tasks on the expansion module based on the real-time demand of the computing tasks to obtain the computing task priority corresponding to the expansion module;

[0031] A multi-computing power task scheduling function component for evaluating and analyzing the utilization of computing power resources of each computing power expansion module in the multi-functional and dynamically expandable module architecture based on the computing power operation power consumption corresponding to the computing tasks on the expansion module to obtain the computing power resource utilization efficiency corresponding to the expansion module; performing computing power task scheduling analysis between the host system and each computing power expansion module based on the corresponding computing task priority and computing power resource utilization efficiency on the expansion module, thereby generating a multi-computing power task utilization scheduling strategy.

[0032] The beneficial effects of the present invention:

[0033] 1. Compared with the prior art, the beneficial effect of the multi-computing power expansion method with multi-function and dynamic configuration proposed by the present invention is that the dynamic insertion configuration of the computing power expansion module is realized through the hot backup and dynamic configuration mechanism, and a multi-function and dynamically expandable module architecture is formed by cascading connection through the high-speed bus and the module interface. The biggest advantage of this architecture is that it can expand the computing power resources in real time according to actual needs. In practical applications, the host system often faces the situation of fluctuating computing loads, and traditional computing architectures are often difficult to quickly adapt to different load requirements. Through this mechanism, computing power modules can be quickly added when the load is high, and unnecessary expansion modules can be automatically reduced when the computing demand is reduced, thereby improving the utilization efficiency of system resources. Moreover, the hot backup mechanism ensures the high availability of the system. Even if a certain computing power expansion module fails, it can automatically switch to other standby modules to avoid the risk of computing task interruption or system crash, thus enhancing the maintainability and scalability of the system. Secondly, by introducing the H_MCU and connecting the control components, a centralized expansion Hub board control architecture is constructed, which plays a key scheduling and monitoring role in the multi-function expansion module. The H_MCU can not only monitor the execution of computing tasks of each expansion module in real time, but also evaluate the operating power consumption, computing power calculation amount and task loading duration of each module, so as to grasp the computing load situation of each module in real time, and thus make more accurate performance analysis and optimization decisions. The integration of the control components also enables the entire system to better coordinate the work between modules, avoid performance degradation caused by resource contention between modules, and improve the overall stability and reliability of the system. Then, by obtaining the computing task demand in real time and determining the task priorities on each expansion module based on the computing amount and loading duration of the tasks, it can perform accurate task scheduling according to the actual needs of the tasks. When multiple tasks in the system need to be executed simultaneously, the reasonable division of priorities can ensure that the most critical and urgent tasks are processed first, thus ensuring the efficient operation of the system. For example, some computationally intensive tasks with strong real-time requirements (such as data analysis or image processing) have strict time requirements, while some batch processing tasks can be performed when the load is low. In this case, by dynamically adjusting the task priorities, the advantages of computing power resources can be fully utilized, avoiding conflicts and resource waste between tasks, and thus greatly improving the operation efficiency and response ability of the system. Finally, by evaluating the utilization efficiency of the computing power resources of the expansion modules, more accurate resource scheduling and task allocation can be achieved. The computing tasks can be reasonably scheduled based on the utilization of the computing power resources of each expansion module to ensure that the computing resources are utilized to the greatest extent. Traditional computing systems are often difficult to accurately adjust the resource allocation according to actual needs, resulting in over-utilization of the computing power of some modules and under-utilization of the computing power of other modules.By monitoring and evaluating the computing power operation power consumption, modules with low resource utilization can be identified in a timely manner and rescheduled to other computing tasks, avoiding waste of computing power resources. In this way, the system can not only improve the computing efficiency, but also be configured as a hot standby for each other to meet the functional safety requirements, thereby improving the overall performance.

[0034] 2. The multi-computing power expansion system with multiple functions and dynamic configuration proposed by the present invention is generally composed of an expansion module cascade design module, an expansion Hub board computing task monitoring module, a computing task priority determination functional component, and a multi-computing power task scheduling functional component. It can implement any multi-computing power expansion method with multiple functions and dynamic configuration described in the present invention, and is used to realize the multi-computing power expansion method with multiple functions and dynamic configuration by coordinating the operations between computer programs running on each module. The internal structure of the system cooperates with each other, which can greatly reduce repetitive work and manpower investment, and can quickly and effectively provide a more accurate and efficient multi-computing power expansion process with multiple functions and dynamic configuration, thereby simplifying the operation process of the multi-computing power expansion system with multiple functions and dynamic configuration. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Other features, objects, and advantages of the present invention will become more apparent by reading the detailed description of the non-limiting embodiments with reference to the following drawings:

[0036] Figure 1 It is a schematic flow chart of the steps of the multi-computing power expansion method with multiple functions and dynamic configuration of the present invention;

[0037] Figure 2 is Figure 1 a detailed schematic flow chart of step S3 in

[0038] Figure 3 It is a schematic structural diagram of the expansion Hub board of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] The technical method of the present invention will be clearly and completely described below with reference to the drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0040] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities may be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0041] It should be understood that although terms such as "first" and "second" may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0042] To achieve the above object, please refer to Figures 1 to 3 , the present invention provides a multi-functional and dynamically configurable multi-computing power expansion method, and the method includes the following steps:

[0043] Step S1: Configure and insert an expansion Hub board in the host system through a hot backup and dynamically configurable mechanism. A connection and transmission component H_Switch in a certain centralized manner is configured on the Hub board. The Hub board can further insert two or four computing power expansion modules. These computing power expansion modules can be set into two groups and are respectively cascaded and connected to the module interfaces corresponding to the H_Switch through a certain high-speed bus for integrated design to generate a multi-functional and dynamically expandable module architecture;

[0044] Step S2: Implement a control component by introducing an H_MCU and connecting and configuring it, and integrate and construct it in combination with the multi-functional and dynamically expandable module architecture, the H_Switch, and the docking host interface; the H_MCU corresponding in the expansion Hub board monitors the computing tasks of each computing power expansion module in the multi-functional and dynamically expandable module architecture in real time to obtain the computing power operation power consumption, task computing power amount, and task computing power loading duration corresponding to the computing tasks on each expansion module;

[0045] Step S3: Obtain the real-time demand of the computing tasks corresponding to the host system, and determine the task priority based on the real-time demand of the computing tasks for the task computing power amount and the task computing power loading duration corresponding to the computing tasks on the expansion module to obtain the corresponding computing task priority on the expansion module;

[0046] Step S4: Based on the computing power operation power consumption corresponding to the computing tasks on the expansion module, perform an evaluation and analysis of the computing power resource utilization of each computing power expansion module within the multi-functional dynamically expandable module architecture to obtain the corresponding computing power resource utilization efficiency on the expansion module; based on the corresponding computing task priorities and computing power resource utilization efficiency on the expansion module, perform a computing power task scheduling analysis between the host system and each computing power expansion module, and generate a multi-computing power task utilization scheduling strategy.

[0047] In the embodiment of the present invention, please refer to Figure 1 As shown in the figure, it is a schematic diagram of the step flow of the multi-computing power expansion method with multi-functional dynamic configuration of the present invention. In this example, the multi-functional dynamic configuration multi-computing power expansion method includes the following steps:

[0048] Step S1: Configure and insert an expansion Hub board into the host system through a hot standby dynamic configuration mechanism. A connection and transmission component H_Switch in a certain centralized manner is configured on the Hub board. The Hub board can further insert two or four computing power expansion modules. These computing power expansion modules can be set into two groups and cascaded and connected to the corresponding module interfaces of the H_Switch through a certain high-speed bus for integrated design to generate a multi-functional dynamically expandable module architecture;

[0049] In the embodiment of the present invention, by configuring two groups of computing power expansion module groups inside the host system, and respectively cascading and connecting these two groups of computing power expansion module groups to the H_Switch and the corresponding module interfaces through a high-speed bus to ensure the scalability and efficient parallel processing ability of the system. The two groups of computing power expansion module groups are backed up by each other through a hot standby mechanism to ensure the high availability of the system. That is, when a certain module in a certain group of modules fails, the system will automatically perform dynamic adjustment to ensure that only one valid module is retained in each group of modules, and the hot standby function is still maintained between the two groups. In this way, when a group of modules fails completely, the valid module in the other group can temporarily take over the task to ensure the continuous operation of the system; and if all the modules in any one group fail and all the modules in the other group are valid, dynamic adjustment will be performed based on the remaining valid modules, reallocating the valid modules so that each group maintains one valid module and ensuring that the two groups of modules are hot standby to each other. Each computing power expansion module may specifically include an AI acceleration working module and a graphics acceleration working module, or two AI acceleration working modules, or two graphics acceleration working modules. Each expansion module can efficiently process parallel computing tasks in multiple working modes through cascaded connection, specifically including tensor parallelism, data parallelism, and pipeline parallelism tasks, to meet the different requirements in the AI acceleration working mode and the graphics acceleration working mode, and finally generate a multi-functional dynamically expandable module architecture through integrated design.

[0050] Step S2: Implement a control component by introducing an H_MCU and connecting for configuration, and integrate and build it in combination with the multi-functional dynamically extensible module architecture, the H_Switch, and the docking host interface; monitor the computing tasks of each computing power extension module in the multi-functional dynamically extensible module architecture in real time through the corresponding H_MCU in the extended Hub board, so as to obtain the computing power operation power consumption, task computing power calculation amount, and task computing power loading duration corresponding to the computing tasks on each extension module.

[0051] In the embodiment of the present invention, by introducing an H_MCU (i.e., the MCU main control and coordination unit) and connecting and configuring it as a control component, this control component is responsible for coordinating the work of all computing power extension modules in the multi-functional dynamically extensible module architecture. Through this integration solution, an extended Hub board that docks with the H_Switch and the host system interface is generated (as Figure 3 shown). This extended Hub board has the communication ability with the computing power extension modules in the multi-functional dynamically extensible module architecture and can monitor the computing tasks on each extension module in real time. The H_MCU is responsible for collecting data related to the computing tasks on the extension modules, such as the computing power operation power consumption, task computing power calculation amount, and task computing power loading duration of the computing tasks. The acquisition of these data requires real-time monitoring of the power consumption and load conditions of the module, and then the scheduling and optimization of the computing tasks are carried out to ensure that the host system can accurately control the allocation and usage efficiency of the computing power resources, and finally the computing power operation power consumption, task computing power calculation amount, and task computing power loading duration corresponding to the computing tasks on each extension module are obtained through real-time monitoring.

[0052] Step S3: Obtain the real-time demand of the computing tasks corresponding to the host system, and determine the task priority based on the real-time demand of the computing tasks for the task computing power calculation amount and the task computing power loading duration of the computing tasks on the extension module, so as to obtain the corresponding computing task priority on the extension module.

[0053] In an embodiment of the present invention, by obtaining the computing task demand required by the host system in real time, the acquisition of this demand depends on the real-time state of the tasks currently executed by the system. Based on this demand data, the H_MCU processes the computing tasks on the expansion module, and performs priority sorting according to the computing power consumption of each task and the task computing power loading duration. Specifically, the H_MCU will, through a preset scheduling strategy, perform priority sorting on different computing tasks according to the urgency of the tasks, the computing power requirements, and the complexity of the tasks. When the computing demand of a task is high and it needs to be completed as soon as possible, its priority will be automatically increased. When the task requires less computing power or is relatively simple, its priority is relatively low. This priority determination process will ensure that the host system can preferentially process important tasks or tasks that require a large amount of computing power under limited resources, thereby maximizing the utilization efficiency of computing power resources, and finally obtaining the corresponding computing task priorities on the expansion module.

[0054] Step S4: Based on the computing power operation power consumption corresponding to the computing tasks on the expansion module, perform an evaluation and analysis of the computing power resource utilization of each computing power expansion module within the multi-functional dynamically expandable module architecture to obtain the corresponding computing power resource utilization efficiency on the expansion module; based on the corresponding computing task priorities and computing power resource utilization efficiency on the expansion module, perform a computing power task scheduling analysis between the host system and each computing power expansion module, and generate a multi-computing power task utilization scheduling strategy.

[0055] In an embodiment of the present invention, the H_MCU evaluates and analyzes the computing power resource utilization efficiency of each computing power expansion module based on the computing power operation power consumption of the computing tasks on the expansion module. This process requires monitoring the working state of each computing power expansion module, obtaining parameters such as the real-time power consumption, computing volume, and task processing time of the module, and evaluating the computing power resource utilization of the module based on this. Specifically, the H_MCU will analyze the load of each module, judge its current working efficiency, and generate a computing power resource utilization efficiency report. Based on these evaluation results, the H_MCU will further optimize the task scheduling between the host system and each computing power expansion module. This scheduling process needs to combine the priorities of each computing task and the utilization efficiency of the module to generate an optimized computing power task scheduling strategy. This multi-computing power task utilization scheduling strategy will ensure that the host system can efficiently utilize the resources of each computing power expansion module, reasonably allocate tasks, avoid resource waste, and improve the overall computing performance, and finally generate a multi-computing power task utilization scheduling strategy.

[0056] Further, the hot backup dynamic configuration mechanism described in step S1 is specifically that two computing power expansion module groups are in hot backup with each other. When one module in any one group fails, it is dynamically adjusted so that each group only includes one valid module and the two groups are in hot backup with each other; when all modules in any one group fail and all modules in the other group are valid, it is dynamically adjusted to re-divide the two valid modules into two groups, with each group only including one module and the two groups being in hot backup with each other.

[0057] Further, the computing power expansion module group includes two expansion modules or one expansion module, and the two expansion modules or one expansion module are specifically AI acceleration working modules or graphics acceleration working modules.

[0058] Further, the cascaded connection can cascade and process parallel processing tasks in the corresponding working mode, specifically including corresponding tensor parallel, data parallel, and pipeline parallel tasks.

[0059] Further, the working mode includes an AI acceleration working mode and a graphics acceleration working mode.

[0060] Further, step S3 includes the following steps:

[0061] Step S31: Obtain the real-time demand for computing tasks corresponding to the host system;

[0062] Step S32: Based on the real-time demand for computing tasks, use the computing task progress calculation formula to perform task progress evaluation calculations on the task computing power calculation amount and the task computing power loading duration corresponding to the computing tasks on the expansion module, so as to obtain the corresponding computing task progress on the expansion module;

[0063] Step S33: Determine the task priority of the computing tasks corresponding to the expansion module based on the corresponding computing task progress on the expansion module, so as to obtain the corresponding computing task priority on the expansion module.

[0064] As an embodiment of the present invention, refer to Figure 2 shown, for Figure 1 the detailed step flow schematic diagram of step S3 in

[0065] Step S31: Obtain the real-time demand for computing tasks corresponding to the host system;

[0066] In an embodiment of the present invention, by monitoring the operating state of the host system and the task scheduling system, the real-time demand of each computing task in the current host system is obtained. The real-time demand of the computing task is dynamically obtained by the task scheduling system based on factors such as the computing requirements of each task (such as CPU and memory resource requirements) and the priority of task submission. Specifically in implementation, the monitoring system first collects task queue information, identifies the resource amount required for each task, and uses hardware monitoring tools to collect the load situation of the host in real time, obtains the resource consumption status of each task, and then generates real-time demand data. For high-priority tasks, the real-time demand will be dynamically adjusted to meet the urgent computing needs. Through the calculation of the task demand, the computing task demand in the host system is quantified, and finally the real-time demand of the computing task is obtained.

[0067] Step S32: Based on the real-time demand of the computing task, use the computing task progress calculation formula to perform a task progress evaluation calculation on the task computing power calculation amount and the task computing power loading duration corresponding to the computing task on the expansion module, so as to obtain the computing task progress corresponding to the expansion module.

[0068] In an embodiment of the present invention, by combining the real-time demand of the computing task, the task computing power loading duration, the task computing power calculation amount, the maximum computing amount required by the computing task, the computing amount influence weight coefficient, and related parameters, a suitable computing task progress calculation formula is formed for evaluation calculation, so as to quantitatively obtain the execution state progress of each computing task on the expansion module, and finally obtain the computing task progress corresponding to the expansion module.

[0069] Step S33: Based on the computing task progress corresponding to the expansion module, determine the task priority of the computing task corresponding to the expansion module, so as to obtain the task priority corresponding to the expansion module.

[0070] In the embodiments of the present invention, the priority of each task is dynamically adjusted according to the previously calculated progress of the computing task. Specifically, when implementing, a priority evaluation criterion is set. For example, a comprehensive evaluation is carried out according to factors such as the progress percentage of the task, the remaining computing power requirement, and the execution duration. For tasks with slow progress and large demand, their priorities will be automatically increased to ensure that the tasks can be completed on time; conversely, the priorities of tasks with faster progress or lower resource requirements will be correspondingly reduced. In addition, the determination of the priority also takes into account factors such as the task type, the urgency of the task, and the overall utilization efficiency of the resources. For example, if a task is performing important real-time calculations and is approaching its deadline, the priority of this task will be automatically increased and more computing resources will be obtained. After the task priority is determined, the scheduling system is used to re-adjust the execution order and resource allocation of the tasks in the expansion module to ensure that high-priority tasks are executed first, thereby improving the execution efficiency and completion quality of the overall computing task, and finally obtaining the corresponding computing task priorities on the expansion module.

[0071] Further, the specific calculation formula for the progress of the computing task described in step S32 is as follows:

[0072]

[0073] In the formula, P is the progress of the computing task, Q is the real-time demand of the computing task, T c is the task computing power loading duration corresponding to the computing task on the expansion module, t is the time variable parameter, C(t) is the task computing power calculation amount of the computing task on the expansion module at time t, C max is the maximum calculation amount required for the computing task, α1 is the calculation amount influence weight coefficient, and η is the correction coefficient of the computing task progress.

[0074] The present invention obtains a calculation task progress calculation formula through the use of a specific mathematical model and verification, which is used to evaluate the task progress by calculating the corresponding task computing power and the task computing power loading duration. The integral part in the calculation task progress calculation formula comprehensively considers the changes in time and the task computing power, which helps to dynamically reflect the progress of the calculation task. By continuously updating the time variable, the progress of the task can be calculated in real time, making the progress of the task more in line with the actual running state. By considering the ratio between the task computing power and the maximum required computing power and multiplying it by the corresponding weight coefficient, the demand and consumption of computing power resources by different tasks at different stages can be reflected, thereby affecting the progress of the task. This design helps to more accurately evaluate the task progress. In addition, by introducing a correction coefficient, a flexible parameter is provided, which can adjust the result of the task progress evaluation according to the actual situation. This correction coefficient can be used to compensate for the progress evaluation errors caused by external environment changes (such as hardware failures, resource shortages, etc.) or task nature differences, making the evaluation result more accurate and reliable. By accurately evaluating the progress of the calculation task, the actual progress state of the task can be better understood, which helps to make more reasonable task priority rankings and resource allocations, ensuring that important tasks obtain more resources and avoiding task delays caused by improper resource allocation as much as possible. In summary, the formula fully considers the calculation task progress P, the real-time demand Q of the calculation task, the task computing power loading duration T of the calculation task corresponding to the expansion module c , the time variable parameter t, the task computing power C(t) of the calculation task on the expansion module at time t, and the maximum required computing power C of the calculation task max , the computing power influence weight coefficient α1, the correction coefficient η of the calculation task progress, and a functional relationship is formed according to the mutual correlation between the calculation task progress P and the above parameters At the same time, the introduction of the correction coefficient η of the calculation task progress can be adjusted according to the error situation in the calculation process, thereby improving the accuracy and applicability of the calculation task progress calculation formula.

[0075] Further, step S4 includes the following steps:

[0076] Step S41: Obtain the calculation task type and the calculation task running state corresponding to the current calculation task on each computing power expansion module;

[0077] In the embodiments of the present invention, by monitoring each computing power expansion module, detailed information about the computing tasks being executed on each module is obtained. Specifically, through the task management module built into the system, the types of computing tasks currently running on each computing power expansion module (e.g., matrix operations, image processing, data analysis, etc.) and the running status of the computing tasks (such as running, pending execution, completed, error, etc.) are obtained. Metadata of task execution, such as the priority of the task, remaining execution time, executed time, etc., can be obtained by accessing the status register of the module or through the network interface. This process can be carried out in a timed scheduling manner to ensure that the system updates the task status data of each module in real time, and finally obtains the computing task type corresponding to the current computing task and the running status of the computing task.

[0078] Step S42: Based on the computing task type corresponding to the current computing task and the running status of the computing task, perform an analysis of the computing power resource allocation for each computing power expansion module within the multi-functional dynamically expandable module architecture to obtain the computing power resource allocation amount corresponding to the computing task on the expansion module;

[0079] In the embodiments of the present invention, the task management system analyzes the current computing task type and running status of each computing power expansion module, and combines the computing requirements of each task, such as the amount of computation, task complexity, time requirements, etc., to analyze the required computing power resources. In this process, a resource allocation algorithm is used. According to the type of task, tasks with high computing requirements and high priorities are given priority and higher computing power resources are allocated to them. Through the resource management module within the system, the resource allocation of tasks on each computing power expansion module is dynamically adjusted based on the actual execution status of the tasks (such as running time, resource consumption). The calculation method of the resource allocation amount can be based on the characteristics of the computing tasks (such as data scale, parallelism) and historical data, combined with the load balancing strategy, to ensure that the tasks can obtain the most appropriate amount of computing power resource data on different expansion modules, and finally obtain the computing power resource allocation amount corresponding to the computing task on the expansion module.

[0080] Step S43: Based on the computing power operation power consumption and the computing power resource allocation amount corresponding to the computing task on the expansion module, perform an evaluation and analysis of the computing power resource utilization of the corresponding computing power expansion module to obtain the corresponding computing power resource utilization efficiency on the expansion module;

[0081] In an embodiment of the present invention, by evaluating the utilization efficiency of computing power resources of each computing power expansion module, on each expansion module, the system will monitor and compare the power consumption of the module with the allocated computing power resources. Specifically, the system obtains the power consumption information of each computing power expansion module through the hardware monitoring interface, and compares it with the previously allocated amount of computing power resources to evaluate the energy efficiency and the utilization of computing power resources of the module when executing the current task. Using calculation formulas, such as the ratio of power consumption to the amount of computing resource allocation, to evaluate the resource utilization efficiency of each expansion module. Based on the evaluation results, the resource allocation strategy can be further adjusted to improve the overall utilization rate of computing power resources, reduce unnecessary resource waste, and finally obtain the corresponding computing power resource utilization efficiency on the expansion module.

[0082] Step S44: Based on the corresponding computing task priorities and computing power resource utilization efficiencies on the expansion modules, perform computing power task scheduling analysis between the host system and each computing power expansion module to generate a multi-computing power task utilization scheduling strategy.

[0083] In an embodiment of the present invention, by formulating a reasonable multi-computing power task scheduling strategy according to the priorities of tasks, computing requirements, and utilization efficiencies of computing power resources, perform dynamic scheduling analysis based on the priorities of each computing task and the computing power resource utilization efficiency of the expansion module. For tasks with high computing requirements and need to be completed in a short time, the system will preferentially allocate them to expansion modules with fast computing power response speed and high resource utilization rate to ensure that the tasks can be completed as early as possible; while for tasks with stable computing volume and long execution time, they will be allocated to computing power expansion modules with relatively low resource utilization rate to reasonably utilize system resources and reduce power consumption. Through task scheduling algorithms, such as priority queue scheduling, combined with the power consumption and resource utilization efficiency of the module, generate a suitable scheduling strategy to ensure that each task can be executed efficiently and will not cause resource waste. That is, for computing tasks with high computing requirements and short running time, assign them a high priority to be preferentially allocated to computing power expansion modules with fast computing speed and timely computing power resource response; while for computing tasks with long running time and relatively stable computing volume, assign them a low priority to be allocated to computing power expansion modules with relatively low resource utilization rate.

[0084] Furthermore, the present invention also provides a multi-computing power expansion system with multi-function and dynamic configuration for executing the multi-computing power expansion method with multi-function and dynamic configuration as described above. The multi-computing power expansion system with multi-function and dynamic configuration includes:

[0085] An expansion module cascade design functional component, used to configure and insert two groups of computing power expansion modules in the host system through a hot backup dynamic configuration mechanism and cascade connect and integrate them with the corresponding module interfaces through a high-speed bus respectively to generate a multi-functional dynamically expandable module architecture;

[0086] The extended Hub board computing task monitoring function component is used to implement a control component by introducing an H_MCU and connecting and configuring it, and is integrated and constructed in combination with a multi-functional dynamically extensible module architecture, an H_Switch, and a docking host interface; the corresponding H_MCU in the extended Hub board is used to monitor the computing tasks of each computing power extension module in the multi-functional dynamically extensible module architecture in real time, so as to obtain the computing power operation power consumption, task computing power calculation amount, and task computing power loading duration corresponding to the computing tasks on each extension module.

[0087] The computing task priority determination function component is used to obtain the real-time demand of the computing tasks corresponding to the host system, and determine the task priority based on the real-time demand of the computing tasks for the task computing power calculation amount and task computing power loading duration corresponding to the computing tasks on the extension module, so as to obtain the corresponding computing task priority on the extension module.

[0088] The multi-computing power task scheduling function component is used to evaluate and analyze the utilization of computing power resources of each computing power extension module in the multi-functional dynamically extensible module architecture based on the computing power operation power consumption corresponding to the computing tasks on the extension module, so as to obtain the corresponding computing power resource utilization efficiency on the extension module; perform computing power task scheduling analysis between the host system and each computing power extension module based on the corresponding computing task priority and computing power resource utilization efficiency on the extension module, so as to generate a multi-computing power task utilization scheduling strategy.

[0089] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, it is intended to cover all changes falling within the meaning and scope of the equivalent elements of the application documents within the present invention.

[0090] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A multifunctional and dynamically configurable multi-computing power expansion method, characterized in that: The following steps are involved: Step S1: A Hub expansion board is configured and inserted in the host system through a hot standby dynamically configurable mechanism. A centralized connection and transmission component H_Switch is configured on the Hub board. Two or four computing power expansion modules can be further inserted into the Hub board. These computing power expansion modules can be arranged into two groups and connected in cascade with the module interfaces corresponding to the H_Switch through a high-speed bus for an integrated design, so as to generate a multifunctional dynamically scalable module architecture; Step S2: A control component is implemented by introducing an H_MCU on the Hub expansion board and connecting and configuring it, and is integrated and constructed in combination with the multifunctional dynamically expandable module architecture, the H_Switch, and the docking host interface; the computing tasks of each computing power expansion module in the multifunctional dynamically expandable module architecture are monitored in real time by expanding the corresponding H_MCU in the Hub board, so as to obtain the computing power operation power consumption, task computing power calculation amount, and task computing power loading time corresponding to the computing tasks on each expansion module; Step S3: obtaining the real-time demand of the computing task corresponding to the host system, and determining the task priority of the task computing power calculation amount and the task computing power loading duration corresponding to the computing task on the extension module based on the real-time demand of the computing task, so as to obtain the corresponding computing task priority on the extension module; Step S4: performing a computing resource utilization evaluation and analysis on each computing expansion module in the multi-functional dynamically expandable module architecture based on the computing power operation power consumption corresponding to the computing task on the expansion module, so as to obtain the computing resource utilization efficiency corresponding to the expansion module; Based on the corresponding computing task priority and computing resource utilization efficiency on the extension module, the computing task scheduling analysis is performed between the host system and each computing expansion module to generate a multi-computing task utilization scheduling strategy.

2. The multi-functional dynamically configurable multi-computing power expansion method according to claim 1, characterized in that: The hot backup dynamically configurable mechanism described in step S1 is specifically that two groups of computing power expansion module groups serve as hot backups for each other. When a module in any group fails, it is dynamically adjusted so that each group includes only one valid module and the two groups serve as hot backups for each other; when all modules in any group fail and all modules in the other group are valid, it is dynamically adjusted so that the two valid modules are redivided into two groups, each group includes only one module and the two groups serve as hot backups for each other.

3. The multi-functional dynamically configurable multi-computing power expansion method according to claim 2 is characterized in that: The computing power expansion module group includes two expansion modules or one expansion module, wherein the two expansion modules or one expansion module can specifically be an AI acceleration working module or a graphics acceleration working module.

4. The multi-functional dynamically configurable multi-computing power expansion method according to claim 1, characterized in that: The cascade connection can cascade process parallel processing tasks in the corresponding working mode, specifically including corresponding tensor parallelism, data parallelism and pipeline parallelism tasks.

5. The multi-functional dynamically configurable multi-computing power expansion method according to claim 4 is characterized in that: The working modes include an AI acceleration working mode and a graphics acceleration working mode.

6. The multi-functional dynamically configurable multi-computing capacity expansion method according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: obtaining the real-time demand of the computing task corresponding to the host system; Step S32: Based on the real-time demand of the computing task, the computing task progress calculation formula is used to perform task progress evaluation calculation on the task computing power calculation amount and the task computing power loading time corresponding to the computing task on the extension module to obtain the corresponding computing task progress on the extension module; Step S33: Determine the task priority of the corresponding computing task on the extension module based on the progress of the corresponding computing task on the extension module to obtain the corresponding computing task priority on the extension module.

7. The multi-functional dynamically configurable multi-computing power expansion method according to claim 6, characterized in that: The calculation formula for the calculation task progress described in step S32 is specifically: In the formula, P is the progress of the computing task, Q is the real-time demand of the computing task, and T c is the task computing power loading duration corresponding to the computing task on the extension module, t is the time variable parameter, C(t) is the task computing power calculation amount of the computing task on the extension module at time t, C max is the maximum amount of computation required for the task, α1 is the weight coefficient of the computational effort, and η is the correction coefficient of the task progress.

8. The multi-functional dynamically configurable multi-computing power expansion method according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: Obtain the computing task type and computing task running status corresponding to the current computing task on each computing power expansion module; Step S42: performing a computing resource allocation analysis on each computing power expansion module in the multifunctional dynamically expandable module architecture based on the computing task type and computing task running status corresponding to the current computing task, and obtaining the computing resource allocation amount corresponding to the computing task on the expansion module; Step S43: Based on the computing power running power consumption and computing power resource allocation amount corresponding to the computing task on the expansion module, the computing power resource utilization evaluation and analysis is performed on the corresponding computing power expansion module to obtain the computing power resource utilization efficiency corresponding to the expansion module; Step S44: Based on the corresponding computing task priority and computing resource utilization efficiency on the expansion module, a computing task scheduling analysis is performed between the host system and each computing power expansion module to generate a multi-computing task utilization scheduling strategy.

9. The multi-functional dynamically configurable multi-computing capacity expansion method according to claim 8, characterized in that: The multi-computing task scheduling strategy described in step S44 is to give high priority to computing tasks that run for a short time and have high computing requirements, so that they are preferentially allocated to computing expansion modules with fast computing speed and timely computing resource response; and for computing tasks that run for a long time and have relatively stable computing requirements, they are given low priority so that they are allocated to computing expansion modules with relatively low resource utilization.

10. A multifunctional dynamically configurable multi-computing power expansion system, characterized in that: Used to execute the multi-functional dynamically configurable multi-computing power expansion method as claimed in claim 1, the multi-functional dynamically configurable multi-computing power expansion system comprises: The expansion module cascade design component is used to configure the insertion of expansion Hub boards in the host system through a hot backup dynamic configuration mechanism. The Hub board includes two groups of computing power expansion modules and is cascaded and integrated with the H_Switch and the corresponding module interface through a high-speed bus to generate a multi-functional dynamically expandable module architecture; The computing task monitoring component of the extended Hub board is used to realize a control component by introducing an H_MCU and connecting and configuring it, and integrate it with the multi-functional dynamically expandable module architecture, H_Switch and the docking host interface for construction; the computing tasks of each computing power expansion module in the multi-functional dynamically expandable module architecture are monitored in real time by extending the corresponding H_MCU in the Hub board to obtain the computing power operation power consumption, task computing power calculation amount and task computing power loading time corresponding to the computing tasks on each expansion module; The computing task priority determination function component is used to obtain the real-time demand of the computing task corresponding to the host system, and determine the task priority of the task computing power calculation amount and the task computing power loading time corresponding to the computing task on the extension module based on the real-time demand of the computing task, so as to obtain the corresponding computing task priority on the extension module; The multi-computing task scheduling functional component is used to evaluate and analyze the computing resource utilization of each computing expansion module within the multi-functional dynamically expandable module architecture based on the computing power consumption corresponding to the computing tasks on the expansion module, so as to obtain the corresponding computing resource utilization efficiency on the expansion module; based on the corresponding computing task priority and computing resource utilization efficiency on the expansion module, the computing task scheduling analysis is performed between the host system and each computing expansion module, so as to generate a multi-computing task utilization scheduling strategy.

Images