Dynamic task scheduling method and system based on hardware state awareness
By monitoring hardware status in real time and dynamically adjusting task allocation and migration, the problems of uneven hardware resource utilization and high task failure rate in existing technologies are solved, thereby improving the system's resource utilization and the stability and efficiency of task execution.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HEFEI SUMICROELECTRONICS TECH CO LTD
- Filing Date
- 2025-12-03
- Publication Date
- 2026-05-12
AI Technical Summary
In existing technologies, task scheduling mechanisms lack the ability to perceive the real-time status of hardware, resulting in uneven utilization of hardware resources, overload, and high task failure rates. In particular, the stability and efficiency of the system are affected under high concurrency and high load scenarios.
By collecting hardware status data in real time, such as memory usage, core temperature and power consumption, the task allocation strategy is dynamically adjusted, and a task migration mechanism is activated when the hardware status exceeds the limit. Combined with feedback to optimize the scheduling strategy, the task allocation and seamless migration are achieved.
It achieves efficient utilization of hardware resources, reduces task failure rate, improves system robustness and task execution continuity, and enhances overall performance through adaptive optimization.
Smart Images
Figure CN122019070A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of computer system resource scheduling technology, and in particular to a task dynamic scheduling method and system based on hardware status awareness. Background Technology
[0002] In high-performance computing scenarios such as deep learning, scientific computing, real-time rendering, and edge computing, graphics processing units (GPUs) and neural processing units (NPUs) have become critical computing resources. Their efficient parallel processing capabilities enable the accelerated execution of complex tasks. However, in runtime environments where multiple tasks are executed concurrently, how to efficiently and stably schedule tasks to appropriate hardware resources has become a key factor affecting the overall performance and stability of the system.
[0003] Existing task scheduling mechanisms are mostly based on static configuration or simple load balancing strategies, lacking the ability to perceive the real-time status of hardware. Specifically, existing technologies have the following shortcomings: 1. Insufficient hardware status monitoring capabilities: Traditional scheduling systems typically rely on preset scheduling strategies and cannot acquire and respond to dynamic hardware status information in real time, such as key indicators like memory usage, core temperature, and power consumption. This makes it difficult for the system to make timely adjustments in situations of hardware resource scarcity or abnormalities.
[0004] 2. Disconnect between task allocation and hardware status: Existing scheduling strategies often ignore the actual hardware load, which can easily lead to overload of some hardware resources while other resources are idle, causing problems such as resource contention, increased task queuing delays, and even hardware overheating, affecting task execution efficiency and system reliability.
[0005] 3. High task failure rate under high load: When resources are scarce or hardware conditions are abnormal (such as memory overflow or excessive core temperature), the lack of an effective task migration or dynamic scheduling mechanism leads to an increased task execution failure rate, which challenges the overall stability of the system.
[0006] Therefore, there is an urgent need in this field for an intelligent scheduling mechanism that can sense hardware status in real time, dynamically adjust task scheduling strategies, and has the ability to migrate task overflows, so as to improve system resource utilization, ensure the continuity and stability of task execution, and adapt to a variety of high-concurrency and high-load application scenarios. Summary of the Invention
[0007] To address this, embodiments of the present invention provide a hardware-state-aware dynamic task scheduling method and system, which solves the problems of uneven resource utilization, system overload, and high task failure rate caused by the inability of the static scheduling strategy to adapt to dynamic hardware states in the prior art.
[0008] To address the aforementioned technical problems, embodiments of the present invention provide a hardware-state-aware dynamic task scheduling method, which includes the following steps: Real-time acquisition of hardware status data of at least one hardware computing unit, including memory usage, core temperature and power consumption; Based on the hardware status data, the tasks to be executed are dynamically allocated to hardware computing units whose load meets preset conditions. When the hardware status data of the target hardware computing unit is detected to exceed a preset threshold, the task overflow migration mechanism is activated to migrate at least one task on the target hardware computing unit to other available hardware computing units. Based on feedback data after task execution and the history of hardware status changes, the task allocation strategy is iteratively optimized.
[0009] Preferably, the method for real-time acquisition of hardware status data of at least one hardware computing unit specifically includes: The integrated hardware monitoring module periodically acquires runtime status metrics of the GPU and / or NPU in a polling or event-driven manner. The system performs anomaly diagnosis on the collected status data and generates an early warning signal when an anomaly is detected.
[0010] Preferably, the method for dynamically allocating tasks to be executed to hardware computing units whose load meets preset conditions specifically includes: Assess the current load level of each hardware computing unit based on real-time hardware status data; Based on the load assessment results, a load priority queue is constructed to prioritize the allocation of new tasks to hardware computing units with lower load levels; When the resource utilization of any hardware computing unit approaches its capacity threshold, limit or suspend the allocation of new tasks to that unit.
[0011] Preferably, the method for initiating the task overflow migration mechanism specifically includes: The preset thresholds include the core temperature limit, the memory usage limit, or the power consumption limit. The task migration process ensures the continuity of the execution state of the migrated task and resumes execution once the target hardware resources become available.
[0012] Preferably, the method for iteratively optimizing the task allocation strategy specifically includes: Collect performance data during task execution, including task execution latency and hardware resource utilization; By combining historical hardware status data and performance data, scheduling parameters can be adjusted through machine learning or rule engines to improve the adaptability and efficiency of subsequent task scheduling.
[0013] This invention also provides a hardware-state-aware dynamic task scheduling system, which implements the aforementioned hardware-state-aware dynamic task scheduling method and includes the following modules: The hardware monitoring module is used to collect hardware status data of at least one hardware computing unit in real time, including memory usage, core temperature and power consumption. The task allocation module is used to dynamically allocate tasks to be executed to hardware computing units whose load meets preset conditions based on the hardware status data. The task migration module is used to initiate a task overflow migration mechanism when the hardware status data of the target hardware computing unit exceeds a preset threshold, and to migrate at least one task on the target hardware computing unit to other available hardware computing units. The feedback optimization module is used to iteratively optimize the task allocation strategy based on feedback data after task execution and the history of hardware status changes.
[0014] Preferably, the hardware monitoring module is further used for: Perform anomaly diagnosis on the collected memory usage, core temperature and power consumption data; When an anomaly is detected, an early warning signal is sent to the task migration module and the task allocation module to trigger an adjustment of the scheduling strategy.
[0015] Preferably, the task allocation module is further configured to: Assess the current load level of each hardware computing unit based on real-time hardware status data, and construct a load priority queue. Based on the load priority queue, a task allocation instruction is generated.
[0016] Preferably, the task migration module is further configured to: Maintain a pool of available hardware resources; When the task overflow migration mechanism is initiated, a suitable spare hardware computing unit is selected from the resource pool, and the task migration operation is performed.
[0017] Preferably, the feedback optimization module is further configured to: By analyzing historical correlation data of task execution latency, resource utilization, and hardware status, the decision parameters in the scheduling strategy are dynamically adjusted to achieve continuous optimization of scheduling efficiency.
[0018] As can be seen from the above technical solutions, this invention application has the following beneficial effects: (1) Unlike traditional static scheduling strategies, this invention uses a hardware monitoring module to collect key indicators such as memory usage, core temperature, and power consumption in real time. The task allocation module dynamically evaluates the load of each computing unit based on these indicators and prioritizes assigning tasks to units with lower loads. This approach effectively avoids the waste of resources caused by some hardware being overloaded while others are idle, allowing expensive computing resources such as GPUs / NPUs to be fully utilized, thereby improving the overall throughput and computing efficiency of the system.
[0019] (2) When the system detects that the core temperature, memory usage, or power consumption of a certain hardware computing unit exceeds the safety threshold, it will immediately start the task migration module to seamlessly migrate some tasks to other available hardware resource pools for execution, and ensure that the task status is not lost and the execution is not interrupted. This mechanism effectively prevents task failure or system crash due to abnormal conditions such as hardware overheating and memory overflow, and ensures the stable completion of tasks in long-running scenarios such as deep learning inference and scientific computing.
[0020] (3) This invention introduces a feedback optimization module, which continuously collects performance data such as task execution latency and resource utilization, and combines it with hardware status history to iteratively optimize scheduling parameters and strategies through machine learning or rule engines. This enables the system to learn from actual operating experience, continuously adapt to different workloads and application scenario characteristics, and achieve refined and intelligent scheduling strategies, thereby improving the accuracy of task scheduling and overall system performance in the long term and dynamically. Attached Figure Description
[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Referring to the drawings will make the features and advantages of the present invention clearer. The drawings are illustrative and should not be construed as limiting the present invention in any way. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart of a hardware-state-aware dynamic task scheduling method provided by the present invention; Figure 2 This is a block diagram of a hardware-state-aware dynamic task scheduling system provided by the present invention. Detailed Implementation
[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0023] Example 1: To address the problems of uneven resource utilization, system overload, and high task failure rates caused by the inability of existing static scheduling strategies to adapt to dynamic hardware states, such as... Figure 1 As shown, this invention proposes a hardware-state-aware dynamic task scheduling method, which includes the following steps: S1: Real-time acquisition of hardware status data of at least one hardware computing unit, including memory usage, core temperature and power consumption. S2: Based on hardware status data, dynamically allocate tasks to be executed to hardware computing units whose load meets preset conditions; S3: When the hardware status data of the target hardware computing unit is detected to exceed the preset threshold, the task overflow migration mechanism is activated to migrate at least one task on the target hardware computing unit to other available hardware computing units. S4: Iteratively optimize the task allocation strategy based on feedback data after task execution and the history of hardware status changes.
[0024] As can be seen from the above technical solution, this invention proposes a task dynamic scheduling method based on hardware status awareness. First, by collecting hardware status data in real time, it provides the system with a precise data foundation for dynamic scheduling, thereby realizing early detection and warning of hardware anomalies. Next, tasks are dynamically allocated according to the status data, achieving precise load balancing, effectively avoiding resource contention and overload, and significantly improving system resource utilization. Then, when the hardware status exceeds the limit, a task overflow migration mechanism is initiated to ensure that the execution status of critical tasks is not interrupted and can be seamlessly transferred, greatly enhancing the robustness of the system under high load and the continuity of task execution. Finally, based on feedback and historical data iterative optimization strategies, the system has self-learning and adaptive capabilities, enabling continuous optimization of scheduling decisions, thereby improving scheduling efficiency and overall system performance in the long term.
[0025] In step S1, hardware status data of at least one hardware computing unit is collected in real time. Specifically, through an integrated hardware monitoring module, runtime status indicators of the GPU and / or NPU, including memory usage, core temperature, and power consumption, are periodically acquired in a polling or event-driven manner. Anomaly diagnosis is performed on the collected status data (such as detecting memory overflow or temperature exceeding limits), and a warning signal is generated when an anomaly is diagnosed.
[0026] This invention enables the system to promptly detect changes in hardware status through real-time monitoring and anomaly diagnosis, providing a data foundation for dynamic scheduling and preventing task failures or system crashes due to hardware malfunctions. For example, in deep learning inference scenarios, real-time monitoring of GPU memory usage can prevent memory overflow and ensure the continuity of model inference.
[0027] In step S2, tasks to be executed are dynamically allocated to hardware computing units whose load meets preset conditions based on hardware status data. Specifically, the current load level of each hardware computing unit is evaluated based on real-time hardware status data, and a load priority queue is constructed. New tasks are preferentially allocated to hardware computing units with lower load levels. When the resource utilization of any hardware computing unit approaches its capacity threshold, the allocation of new tasks to that unit is restricted or suspended.
[0028] This invention avoids resource contention and overload issues through dynamic load balancing, thereby improving resource utilization and system throughput. For example, in scientific computing scenarios, tasks are preferentially allocated to GPUs with lower temperatures, reducing performance degradation caused by overheating.
[0029] In step S3, when the hardware status data of the target hardware computing unit exceeds a preset threshold, a task overflow migration mechanism is initiated. Specifically, the preset thresholds include a core temperature limit, a memory usage limit, or a power consumption limit. The task migration process ensures the continuity of the execution state of the migrated task and resumes execution once the target hardware resources become available. The system maintains a pool of available hardware resources and selects a suitable backup hardware computing unit during migration.
[0030] This invention ensures continuous task execution under high load or abnormal conditions through a task overflow migration mechanism, reducing task failure rate and improving system reliability. For example, in a real-time rendering scenario, when a GPU temperature is too high, low-priority tasks are migrated to a backup GPU, avoiding rendering interruption.
[0031] In step S4, the task allocation strategy is iteratively optimized based on feedback data after task execution and historical hardware state changes. Specifically, performance data during task execution is collected, including task execution latency and hardware resource utilization. Combining historical hardware state data and performance data, scheduling parameters (such as load thresholds or migration strategies) are adjusted through machine learning or a rule engine to improve the adaptability and efficiency of subsequent task scheduling.
[0032] This invention utilizes feedback optimization, enabling the system to adapt to different application scenarios and continuously improve scheduling efficiency and system performance. For example, in edge computing scenarios, the optimized scheduling strategy reduces task latency and improves the utilization rate of resource-constrained devices.
[0033] Example 2: Figure 2 As shown, the present invention provides a hardware-state-aware dynamic task scheduling system. This system is used to implement the hardware-state-aware dynamic task scheduling method of Embodiment 1 above, and includes the following modules: The hardware monitoring module is used to collect hardware status data of at least one hardware computing unit in real time. The hardware status data includes memory usage, core temperature and power consumption. The task allocation module is used to dynamically allocate tasks to be executed to hardware computing units whose load meets preset conditions based on hardware status data. The task migration module is used to initiate a task overflow migration mechanism when the hardware status data of the target hardware computing unit exceeds a preset threshold, and to migrate at least one task on the target hardware computing unit to other available hardware computing units. The feedback optimization module is used to iteratively optimize the task allocation strategy based on feedback data after task execution and the history of hardware status changes.
[0034] Furthermore, the hardware monitoring module collects real-time hardware status data (including memory usage, core temperature, and power consumption) from at least one hardware computing unit and performs anomaly diagnosis on the collected data. When an anomaly is detected, it sends an alert signal to the task migration module and the task allocation module to trigger scheduling strategy adjustments.
[0035] This module enables the system to respond proactively to hardware anomalies through real-time monitoring and early warning, preventing task execution failures. For example, in deep learning inference scenarios, when the hardware monitoring module detects abnormal memory usage, it immediately triggers task migration to ensure uninterrupted inference tasks.
[0036] Furthermore, the task allocation module assesses the current load level of each hardware computing unit based on real-time hardware status data and constructs a load priority queue. Based on the queue, task allocation instructions are generated, prioritizing the allocation of new tasks to hardware computing units with lower load levels, and restricting new task allocation when resource utilization approaches a threshold.
[0037] This module optimizes resource utilization through dynamic task allocation, reduces task queuing latency, and improves system efficiency. For example, in scientific computing scenarios, the task allocation module ensures that computing tasks are evenly distributed, avoiding overloading of some GPUs.
[0038] Furthermore, the task migration module maintains a pool of available hardware resources. When the hardware status data of the target hardware computing unit exceeds a preset threshold, the task overflow migration mechanism is activated to select a suitable backup hardware computing unit from the resource pool and perform the task migration operation to ensure the continuity of task status.
[0039] This module enhances the system's resilience under high load conditions through a task migration mechanism, ensuring the execution of critical tasks. For example, in real-time rendering scenarios, the migration module transfers overloaded tasks to idle GPUs, maintaining the stability of the rendering frame rate.
[0040] Furthermore, the feedback optimization module collects historical correlation data on task execution latency, resource utilization, and hardware status, and dynamically adjusts the decision parameters in the scheduling strategy (such as load assessment weights or migration trigger conditions) through machine learning or rule engines to achieve continuous optimization of scheduling efficiency.
[0041] This module enables the system to adapt to dynamic workloads through iterative optimization, resulting in long-term performance improvements. For example, in edge computing scenarios, the optimization module adjusts task allocation strategies based on historical data, reducing overall energy consumption.
[0042] This embodiment provides a hardware-state-aware task dynamic scheduling system for implementing the aforementioned hardware-state-aware task dynamic scheduling method. Therefore, the specific implementation of the hardware-state-aware task dynamic scheduling system can be found in the previous embodiment section of the hardware-state-aware task dynamic scheduling method. For example, the hardware monitoring module, task allocation module, task migration module, and feedback optimization module are used to implement steps S1, S2, S3, and S4 in the aforementioned hardware-state-aware task dynamic scheduling method. Therefore, the specific implementation can be referred to the description of the corresponding embodiments. To avoid redundancy, it will not be repeated here.
[0043] The following describes the implementation of the present invention in specific scenarios: 1. Deep learning inference scenarios Implementation: The hardware monitoring module monitors the GPU's memory usage and core temperature in real time. The task allocation module prioritizes assigning new inference tasks to GPUs with lower temperatures and available memory. When the temperature of a GPU exceeds a threshold, the task migration module migrates some tasks to other GPUs to ensure continuous execution of inference tasks.
[0044] Advantages: It avoids task interruptions caused by hardware overload, and improves model inference efficiency and system stability.
[0045] 2. Scientific computing scenarios Implementation: In multi-task concurrent scientific computing, when the hardware monitoring module detects that the temperature of a GPU core exceeds a threshold, the task migration module transfers some computing tasks to other GPUs. After the temperature returns to normal, the task allocation module reassigns tasks to that GPU.
[0046] Advantages: It ensures the continuity and efficiency of computing tasks and reduces the risk of hardware overheating.
[0047] 3. Real-time scene rendering Implementation: The task allocation module dynamically allocates rendering tasks based on video memory usage. Under high load, the task migration module migrates low-priority tasks to a backup GPU for execution, while the hardware monitoring module provides real-time status visualization.
[0048] Advantages: Optimized rendering task allocation, improved real-time rendering performance and system reliability.
[0049] 4. Edge computing scenarios Implementation: In resource-constrained edge devices, the feedback optimization module adjusts the task allocation strategy based on historical data, and the hardware monitoring module optimizes resource utilization in real time.
[0050] Advantages: Improves resource utilization efficiency of edge devices, reduces task latency and energy consumption.
[0051] In summary, this invention achieves the following advantages through real-time hardware status awareness, dynamic task allocation, task overflow migration, and feedback optimization: (1) Improved resource utilization: Dynamic load balancing avoids resource contention and overload, and improves hardware resource utilization.
[0052] (2) Enhanced task execution stability: The task overflow migration mechanism ensures the continuous execution of tasks under high load or abnormal conditions.
[0053] (3) System performance optimization: By combining real-time monitoring and feedback optimization, the efficiency of task scheduling and system throughput have been improved.
[0054] (4) Operational transparency: Provides real-time visualization of hardware status and task scheduling, which facilitates operation and maintenance management and optimization.
[0055] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0056] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0057] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0058] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.
Claims
1. A hardware-state-aware dynamic task scheduling method, characterized in that, Includes the following steps: Real-time acquisition of hardware status data of at least one hardware computing unit, including memory usage, core temperature and power consumption; Based on the hardware status data, the tasks to be executed are dynamically allocated to hardware computing units whose load meets preset conditions. When the hardware status data of the target hardware computing unit is detected to exceed a preset threshold, the task overflow migration mechanism is activated to migrate at least one task on the target hardware computing unit to other available hardware computing units. Based on feedback data after task execution and the history of hardware status changes, the task allocation strategy is iteratively optimized.
2. The task dynamic scheduling method based on hardware state awareness according to claim 1, characterized in that, The method for real-time acquisition of hardware status data of at least one hardware computing unit specifically includes: The integrated hardware monitoring module periodically acquires runtime status metrics of the GPU and / or NPU in a polling or event-driven manner. The system performs anomaly diagnosis on the collected status data and generates an early warning signal when an anomaly is detected.
3. The task dynamic scheduling method based on hardware state awareness according to claim 1, characterized in that, The method for dynamically allocating tasks to be executed to hardware computing units whose load meets preset conditions specifically includes: Assess the current load level of each hardware computing unit based on real-time hardware status data; Based on the load assessment results, a load priority queue is constructed to prioritize the allocation of new tasks to hardware computing units with lower load levels; When the resource utilization of any hardware computing unit approaches its capacity threshold, limit or suspend the allocation of new tasks to that unit.
4. The task dynamic scheduling method based on hardware state awareness according to claim 1, characterized in that, The method for initiating the task overflow migration mechanism specifically includes: The preset thresholds include the core temperature limit, the memory usage limit, or the power consumption limit. The task migration process ensures the continuity of the execution state of the migrated task and resumes execution once the target hardware resources become available.
5. The task dynamic scheduling method based on hardware state awareness according to claim 1, characterized in that, The method for iteratively optimizing the task allocation strategy specifically includes: Collect performance data during task execution, including task execution latency and hardware resource utilization; By combining historical hardware status data and performance data, scheduling parameters can be adjusted through machine learning or rule engines to improve the adaptability and efficiency of subsequent task scheduling.
6. A task dynamic scheduling system based on hardware state awareness, characterized in that, The system is used to implement the hardware-state-aware dynamic task scheduling method according to any one of claims 1 to 5, and includes the following modules: The hardware monitoring module is used to collect hardware status data of at least one hardware computing unit in real time, including memory usage, core temperature and power consumption. The task allocation module is used to dynamically allocate tasks to be executed to hardware computing units whose load meets preset conditions based on the hardware status data. The task migration module is used to initiate a task overflow migration mechanism when the hardware status data of the target hardware computing unit exceeds a preset threshold, and to migrate at least one task on the target hardware computing unit to other available hardware computing units. The feedback optimization module is used to iteratively optimize the task allocation strategy based on feedback data after task execution and the history of hardware status changes.
7. The hardware-state-aware task dynamic scheduling system according to claim 6, characterized in that, The hardware monitoring module is also used for: Perform anomaly diagnosis on the collected memory usage, core temperature and power consumption data; When an anomaly is detected, an early warning signal is sent to the task migration module and the task allocation module to trigger an adjustment of the scheduling strategy.
8. The task dynamic scheduling system based on hardware state awareness according to claim 6, characterized in that, The task allocation module is also used for: Assess the current load level of each hardware computing unit based on real-time hardware status data, and construct a load priority queue. Based on the load priority queue, a task allocation instruction is generated.
9. The hardware-state-aware dynamic task scheduling system according to claim 6, characterized in that, The task migration module is also used for: Maintain a pool of available hardware resources; When the task overflow migration mechanism is initiated, a suitable spare hardware computing unit is selected from the resource pool, and the task migration operation is performed.
10. The hardware-state-aware task dynamic scheduling system according to claim 6, characterized in that, The feedback optimization module is also used for: By analyzing historical correlation data of task execution latency, resource utilization, and hardware status, the decision parameters in the scheduling strategy are dynamically adjusted to achieve continuous optimization of scheduling efficiency.