A hierarchical execution method and system for loss-aware scheduling decisions

By adaptively switching hardware management paths through a four-level execution channel, the problem of hardware wear scheduling and adaptation for devices with multiple permissions is solved, achieving smooth scheduling of hardware wear and long-term reduction of aging. It is suitable for servers, PCs, mobile terminals, edge gateways and smart home appliances.

CN123068079APending Publication Date: 2026-11-10HEFEI MOXING COMPUTING TECHNOLOGY CO LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202611199061.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-18
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing hardware wear-aware scheduling strategies cannot adapt to multi-privilege and multi-type host computing devices. Low-privilege and firmware-restricted environments make it difficult to perform hardware management, resulting in rapid hardware aging and shortened lifespan.

Method used

The design incorporates a four-level adaptive execution channel, switching hardware control paths based on program execution permissions. High-permission users directly manipulate registers, medium-permission users call system interfaces, and low-permission users output data packets for firmware execution. This achieves bidirectional isolation between decision logic and firmware execution, making it compatible with devices such as servers, PCs, mobile terminals, and edge gateways.

Benefits of technology

It enables the adaptive execution of the same hardware wear scheduling strategy under different permission environments, expands the applicability of wear-aware scheduling, reduces hardware aging wear, and is compatible with multiple operating systems and home appliance main control devices.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123068079A_ABST
    Figure CN123068079A_ABST
Patent Text Reader

Abstract

This invention discloses a hierarchical execution method and system for loss-aware scheduling decisions, belonging to the field of hardware loss scheduling technology. It is a divisional application of the applicant's original heterogeneous computing power operating system application submitted on July 18, 2026. This solution constructs a scheduling strategy based on real-time hardware loss factors and resource health gain / loss factors. It divides the execution channels into four levels according to the host device's running permissions: high kernel privileges, medium operating system privileges, low user privileges, and firmware collaboration. The high-privilege channel directly operates on hardware registers, the medium-privilege channel calls the system's standard power interface, and the low-privilege channel outputs standardized suggestion data packets for firmware parsing and execution. The firmware collaboration channel isolates the decision-making process from the firmware code. After execution, hardware loss data is collected for closed-loop iterative optimization of scheduling weights. This invention has a universal architecture, adaptable to various host computing devices such as servers, PCs, mobile terminals, and edge gateways. It can also be extended to various smart home appliance main control hardware, solving the problem of difficulty in implementing loss scheduling strategies under different permission environments. It requires no modification to the kernel or firmware and can balance hardware load and delay device aging over the long term.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] Explanation of Case Division Relationship

[0002] This application is a divisional application, originating from the application filed on July 18, 2026, application number 202611070355.7 entitled "A Heterogeneous Computing Power Operating System Based on Hardware Loss Awareness and Its Computing Power Assetization Method and Storage Medium", inheriting the filing date of the original application; the original application disclosed a full-stack heterogeneous computing power operating system covering hardware awareness, computing power normalization and conversion, zone partitioning scheduling, multi-tenant management and control, and cross-domain computing power asset trading. Its kernel hardware abstraction layer and scheduling instruction issuance module are general-purpose underlying components, without server-specific binding logic, and can be adapted to various computing hardware equipped with operating systems; this application provides independent and detailed protection for the multi-permission hierarchical execution sub-module after the scheduling policy is issued in the system. The two share the entire PECP2.0 quantitative system for hardware loss reduction, health loss and net loss optimization, and belong to the same layered and progressive technical solution under the same PECP2.0 general inventive concept. Technical Field

[0003] This invention relates to the fields of operating system underlying resource management, hardware wear scheduling, and cross-permission software execution technology. It is particularly applicable to various host computing devices such as computing servers, personal computers, mobile terminals, and edge gateways, enabling multi-permission adaptive hierarchical execution in hardware physical wear perception scheduling scenarios. Background Technology

[0004] As the scale of intelligent computing clusters, PC terminals, mobile devices, and edge nodes expands synchronously, scheduling mechanisms based on hardware physical wear and tear perception can quantify hardware wear and tear by combining indicators such as temperature, load, and hardware errors, and generate resource partitioning, computing power allocation, and power consumption control strategies to achieve hardware life extension and computing power balance.

[0005] The existing loss-aware scheduling system has significant shortcomings in practical application: While the parent heterogeneous computing power operating system can complete hardware indicator collection, optimal partition number calculation, and global scheduling strategy generation, it lacks a layered implementation path designed for different host execution permissions. In real-world scenarios, the execution permissions of different devices and programs vary greatly: kernel modules have direct access to hardware registers; administrator services can only call the system's standard power API; ordinary user-space programs cannot modify underlying hardware parameters; and some embedded and PC devices only support indirect hardware control via UEFI / BIOS firmware.

[0006] The existing scheduling output lacks a hierarchical adaptation execution mechanism. A single scheduling strategy cannot be compatible with four types of permission environments: high, medium, low, and firmware collaborative. It can only run in high-privilege kernel server scenarios and cannot be extended to ordinary PCs, Android terminals, and edge embedded gateways. At the same time, the mixing of decision code and firmware execution code poses security risks. The scheduling scheme cannot be implemented in restricted environments such as mobile terminals and edge terminals, and the applicability of hardware-depleted smooth scheduling is limited.

[0007] Meanwhile, various smart home appliances have built-in power hardware such as MCUs, power transistors, motors, and compressors. Long-term high-frequency operation can easily generate temperature and switching losses. Existing home appliance control logic only uses static threshold regulation, lacks standardized hardware loss quantification and hierarchical execution scheduling architecture, and different home appliances have large differences in open permissions of firmware, making it impossible to uniformly issue loss balancing scheduling strategies. As a result, the devices age quickly and the overall lifespan of the machine is shortened.

[0008] CN202610982953.5 is a prior scheduling patent of the applicant, which only discloses the algorithm logic for solving the optimal number of partitions based on a four-dimensional profit and loss model. It does not study the permission hierarchical landing link after the scheduling strategy is generated, and does not have a four-level execution channel or firmware data packet isolation execution related technical solutions, which are clearly different from the technical innovation of this invention. Summary of the Invention

[0009] Purpose of the invention

[0010] To address the shortcomings of existing hardware loss-aware scheduling strategies that cannot be adapted to multi-privilege and multi-type host computing devices, and the difficulty in executing hardware control in low-privilege and firmware-restricted environments, this invention provides a hierarchical execution method and system for loss-aware scheduling decisions.

[0011] This invention builds upon the publicly available hardware loss quantification and net profit / loss optimization framework of its parent project, designing a four-level adaptive execution channel for the scheduling strategy distribution phase. It automatically switches hardware control paths based on program execution permissions: high-permission users can directly manipulate registers, medium-permission users call standard system interfaces, and low-permission users output standardized suggestion data packets for firmware parsing and execution. This achieves bidirectional isolation between decision logic and firmware execution. The entire execution architecture is universal and adaptable to various hardware such as servers, PCs, mobile terminals, and edge gateways, as well as multiple operating systems including Windows, Linux, Android, embedded Linux, and domestic desktops, thus overcoming the shortcomings of existing scheduling systems in terms of deployment and adaptability. In addition to computing power devices, this universal hierarchical execution architecture is also adaptable to main control devices for home appliances such as smart air conditioners, refrigerators, home energy storage, and whole-house smart gateways, enabling adaptive loss control by opening permissions for different firmware versions of home appliances.

[0012] Technical solution

[0013] In a first aspect, the present invention provides a hierarchical execution method for loss-aware scheduling decisions, applied to a resource scheduling system based on hardware physical loss awareness, comprising the following steps: Decision-making steps: The decision layer runs in the user mode or kernel mode of the operating system of the host computing device, obtains the real-time hardware consumption factor and resource health gain factor, calls the preset net gain objective function to solve for the optimal number of partitions, and generates the corresponding resource allocation strategy. The resource allocation strategy includes the computing power quota, read and write rate limit value, temperature constraint parameter and hot and cold data migration indication for each logical partition. Permission determination steps: The decision layer obtains the permission level of its own host computing device, which includes at least the kernel high permission level, the permission level in the operating system interface, the user-mode low permission level, and the firmware coordination level; Channel selection steps: Select one or a combination of the following execution channels according to the permission level to implement the resource allocation strategy as hardware control actions: - High-privilege channel: The decision-making layer or kernel execution unit directly manipulates hardware registers through the kernel hardware abstraction layer to complete logical partition reconstruction and bandwidth limitation; - Middle-level access control channel: The decision-making level calls the official power management interface of the operating system to map the output of the net profit and loss objective function to the hard limit of computing power of the job object, the frequency point suggestion of the hardware acceleration chip or the power throttling strategy; - Low-privilege channel: The decision-making layer serializes the resource allocation strategy into a suggested value data packet and writes it to ACPI variables, WMI variables or standard output, without modifying the hardware registers itself; - Firmware Collaboration Channel: Low-privilege suggested value data packets are received by the UEFI-side trusted service and handed over to the BIOS or EC for final power adjustment and fan control; Execution feedback steps: After hardware execution, real-time resource health and profit / loss factor change data of each logical partition are collected and sent back to the decision-making layer to correct the subsequent net profit / loss objective function optimization weight preference.

[0014] Furthermore, the suggested value data packet is a JSON or key-value pair structure, and includes at least a logical partition identifier, a standardized computing unit weight, a suggested power value, and a read / write rate limit percentage.

[0015] Furthermore, when the host computing device is a lightweight terminal without error-correcting code memory, the statistical weight of the error-correcting code in the wear factor is automatically set to zero, and only the temperature and load indicators are fused for calculation.

[0016] Furthermore, when the hardware is a SoC that integrates the CPU and integrated graphics, the entire SoC is treated as a single logical partition object to participate in the net profit and loss objective function optimization, without separately dividing the integrated graphics.

[0017] Furthermore, under the aforementioned permission channel, in the Windows operating system environment, the partition weights output by the net profit and loss objective function are mapped to the CPU speed hard limit percentage of the job object.

[0018] Furthermore, in the firmware collaboration channel, the UEFI-side trusted service only receives suggested value data packets and translates them into standard BIOS / EC control variables, without loading decision-level code, thus achieving bidirectional isolation between decision-making and firmware execution.

[0019] Furthermore, the suggested value data packet output by the low-privilege channel carries temperature constraint parameters. These temperature constraint parameters can be consumed by the firmware layer or the operating system's temperature control abstraction layer to adjust hardware power consumption limits, processor operating frequency, or heat dissipation component output, so that the junction temperature of the corresponding logical partition hardware converges to a safe range. In practical implementation, temperature control can be achieved using RAPL power limits, Android Thermal HAL, fan PWM duty cycle, etc. These are merely exemplary implementation methods and do not limit the scope of protection of this invention.

[0020] Furthermore, the decision-making layer is decoupled from the execution layer. The decision output of the same set of net profit and loss objective functions achieves equivalent hardware loss smoothing effects through the corresponding channels in four host modes: Windows service with ordinary user privileges, administrator privilege service, Linux kernel module, and UEFI trusted service.

[0021] Furthermore, the host computing device is a personal computer or an x86 server, and the host operating system is a Windows operating system or a Linux operating system.

[0022] Furthermore, the host computing device is a mobile terminal, an edge computing gateway, or an embedded computing device, and the host operating system is Android or embedded Linux.

[0023] Secondly, the present invention provides a hierarchical execution system for loss-aware scheduling decisions, comprising: The decision-making module is used to execute the decision-making steps and permission determination steps of the aforementioned methods; The channel adaptation module is used to perform the channel selection step of the aforementioned method. It includes a high-privilege kernel interface adaptation submodule, a medium-privilege OS API adaptation submodule, a low-privilege variable writing submodule, and a firmware collaborative translation submodule. The feedback module is used to execute the execution feedback steps of the aforementioned methods; The modules work together to form a scheduling and execution entity that is decoupled from the operating system kernel and BIOS / EC.

[0024] Furthermore, the system runs as an independent user-mode service on a host operating system, including Windows and Linux; it outputs suggested value data packets through a low-privilege channel and does not directly hold write permissions to hardware registers.

[0025] Thirdly, a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the hierarchical execution method for any of the aforementioned loss-aware scheduling decisions.

[0026] Fourthly, an electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the hierarchical execution method for any of the aforementioned loss-aware scheduling decisions.

[0027] Beneficial effects

[0028] This invention sets up a four-level hierarchical execution channel, and the same hardware loss scheduling strategy can adapt to different running permissions, solving the problem that hardware management strategies cannot be implemented on low-privilege and firmware-restricted devices, and greatly expanding the range of hardware and systems applicable to loss-aware scheduling.

[0029] The firmware collaboration channel completely isolates the decision-making business code from the underlying firmware execution code. The UEFI trusted service only performs data packet parsing and translation and does not load external business logic, thus avoiding the security risk of firmware layer code injection.

[0030] The entire execution entity is decoupled from the operating system kernel and BIOS / EC firmware, eliminating the need for customized modifications to the kernel and firmware. It is compatible with existing commercial operating systems and firmware standards, resulting in low implementation and modification costs.

[0031] The architecture is versatile, adaptable to the x86 computing server scenario described in the original case, and can also be extended to ordinary PCs, Android mobile terminals, and embedded edge gateways. It is compatible with multiple operating systems and can achieve a smooth effect of equivalent hardware loss in different hardware environments.

[0032] After execution, hardware health indicators are collected in real time and fed back to the decision-making level to form a closed loop. The weights of the net profit and loss objective function are continuously optimized through iteration to reduce hardware aging and wear in the long term.

[0033] This invention provides a universal four-level execution channel that is backward compatible with various smart home appliance main control hardware. It automatically adapts and schedules the distribution methods for three permission scenarios: closed firmware, semi-open SDK, and fully open kernel of home appliance MCUs. It dynamically adjusts the operating conditions of compressors, fans, and power modules based on device losses, thereby reducing the long-term aging losses of home appliance power devices.

[0034] Explanation of terms and symbols This application is a divisional application of the parent application, and the entire set of quantitative parameter systems is completely consistent with the parent application 202611070355.7; the following symbols are only used for embodiments and technical derivations in the specification, and the claims use only Chinese terminology and do not contain symbols: Ω: Real-time hardware degradation factor, which comprehensively represents the real-time degree of hardware degradation based on temperature, load, and hardware errors; Ψ: Resource health profit and loss factor, used to adjust the weight of computing power and scheduling partition revenue; μ(N): Net profit and loss objective function, used to solve for the optimal number of logical partitions in the hardware; CU: Standardized computing unit, a unified unit of measurement for hardware computing power. Attached Figure Description

[0035] Figure 1 is a schematic diagram of the overall architecture of the hierarchical execution system for loss-aware scheduling decision-making according to an embodiment of the present invention; wherein, [1]-decision module, [2]-channel adaptation module,

[201] -high-privilege kernel interface adaptation submodule,

[202] -medium-privilege OS API adaptation submodule,

[203] -low-privilege variable writing adaptation submodule,

[204] -firmware collaborative forwarding adaptation submodule, [3]-feedback module, [4]-operating system kernel, [5]-BIOS / EC firmware layer, [6]-hardware resource layer; The decision module outputs the scheduling strategy and sends it to the channel adaptation module. The channel adaptation module connects to the operating system kernel and BIOS / EC firmware layer through each permission submodule to complete the scheduling instruction issuance, which is applied to the hardware resource layer; The hardware resource layer collects hardware loss data and sends it to the feedback module. The feedback module sends the returned data back to the decision module to form a complete closed-loop scheduling. In addition to computing power chips and terminal SoC, the hardware resource layer may also include home appliance main control MCU, compressor, fan power devices and other home appliance hardware units.

[0036] Figure 2 is a schematic flowchart of the hierarchical execution method for loss-aware scheduling decision-making according to an embodiment of the present invention; wherein, S100-acquire standardized hardware loss data, S101-generate scheduling strategy, S102-channel matching judgment, S103-execute high-privilege kernel interface channel, S104-execute medium-privilege OS API channel, S105-execute low-privilege variable writing channel, S106-execute firmware collaborative forwarding channel, S107-collect hardware layer operation feedback data, S108-feedback iteration; This method first acquires standardized hardware loss data, generates a scheduling strategy carrying an execution permission level marker, selects the corresponding permission channel to complete the execution of scheduling instructions after channel matching judgment, and then collects hardware operation feedback data, which is fed back to the step of generating scheduling strategy to realize closed-loop iterative optimization of scheduling strategy. Detailed Implementation

[0038] Example 1: High-privilege scenario on a Linux x86 server

[0039] The host computing device is an x86 computing server running Linux. The decision-making layer is deployed as a kernel module, identified as having high kernel privileges, and a high-privilege channel is enabled. The decision-making module collects resource consumption factors and resource health gain / loss factors, calls the net gain / loss objective function to solve for the optimal number of partitions, and generates a resource allocation strategy. The channel adaptation module directly manipulates hardware registers through the kernel hardware abstraction layer, exemplarily using cgroup resource grouping, MSR, and RAPL interfaces to complete logical partition reconstruction and bandwidth limiting. After hardware operation, the feedback module collects real-time resource health gain / loss factors and sends them back, continuously optimizing the optimization weights. This embodiment corresponds to the original server computing power scheduling scenario of the parent case and belongs to the core adaptation scenario of this invention.

[0040] Example 2: Permission Scenario in Windows PC Administrator The host is a personal computer with Windows 11 operating system. The decision-making layer runs as a service process with administrator privileges, which is determined to be a medium-privilege channel.

[0041] The channel adaptation module calls the system's native power management interface to attempt to map the output of the net profit / loss objective function to the CPU rate hard cap (JobObjectCpuRateControlInformation) of the job object. Considering the limitations of the Windows hybrid architecture (P / E Core) and system power plan, this hard cap takes effect in Windows versions that support HARD_CAP and high-performance power plans; if the environment does not support it (such as being in power-saving mode or system policy restrictions), the policy layer automatically downgrades to ProcessPowerThrottling or EcoQoS (high-efficiency mode) soft rate limiting, and marks the current execution level as Hard-Conditional in the probe confidence.

[0042] The GPU side matches the NVIDIA Management Library (NVML) frequency and power throttling strategies, and implements control through the nvmlDeviceSetGpuLockedClocks or nvmlDeviceSetPowerManagementLimit interfaces. It indirectly controls the hardware through the operating system driver stack, avoiding direct reading and writing of the underlying hardware registers.

[0043] Example 3: Low-privilege scenario for ordinary Windows users The host is a personal computer, and the currently logged-in user is a regular user account with no hardware register access privileges, thus it is determined to be a low-privilege channel.

[0044] The channel adaptation module encapsulates the scheduling policy into a JSON-formatted suggestion value data packet. This packet includes logical partition identifiers, standardized computing unit weights (μ), suggested power, and temperature constraint parameters. Since ordinary users cannot write system-level configurations, this data packet is written to a user-writable location, preferably the current user key in the registry (HKCU\SOFTWARE\MODRIS\MtsAdvice) or the local application data directory (%LOCALAPPDATA%\MODRIS\mts_advice.json).

[0045] Subsequently, the module notifies upper-level system components (such as system daemons or power management UIs) via ETW (Event Tracing for Windows) events or WMI (Windows Management Instrumentation) instance methods. The decision-making process does not touch the underlying hardware and is marked as Advisory. It is read and parsed by the operating system power manager or WMI consumer service, and then selectively adjusts the system operating parameters.

[0046] Example 4: PC Firmware Collaborative Temperature Control Scenario

[0047] The device is equipped with standard UEFI and BIOS / EC firmware; the user-space service generates a suggestion data packet containing temperature constraint parameters and sends it to the UEFI trusted service; the trusted service only parses the data packet and translates it into BIOS-recognizable control variables, and adjusts the hardware junction temperature through RAPL PL1 power limits and fan PWM duty cycle, without loading decision code throughout the process, thus achieving hardware and software security isolation.

[0048] Example 5: Low-privilege extension scenario on Android mobile terminals

[0049] The host is an Android mobile terminal, which is a mobile terminal device. The system strictly restricts the application's hardware permissions and runs in a low-privilege channel. It outputs key-value pair format suggested data packets. Temperature constraint parameters are consumed by the Android Thermal HAL, dynamically adjusting the CPU frequency upper limit and controlling the chip junction temperature within a safe range. When the terminal lacks error-correcting memory, the loss factor calculation automatically removes error statistical weights, using only temperature and load indicators. This scenario is an extended application carrier for this general architecture. The underlying loss calculation and the core logic of the four-level channel are all recorded in the original parent document and do not constitute new core technology solutions.

[0050] Example 6: Embedded Edge Gateway Scenario

[0051] The host is an edge computing gateway, running embedded Linux, and the hardware is an integrated SoC with CPU and integrated graphics. During the optimization phase, the entire SoC is treated as a single logical partition, without separately splitting the integrated graphics computing power. The gateway adaptively switches between high / low / firmware collaborative arbitrary execution channels according to its own running permissions to complete the loss balance scheduling.

[0052] Example 7: Multi-homed equivalence verification example

[0053] The same set of net profit and loss objective function output scheduling strategies are deployed in four forms: Windows ordinary service, Windows administrator service, Linux kernel module, and UEFI trusted service. The system automatically matches the corresponding execution channel. Although the four types of host hardware and system environment are different, they can all achieve equivalent technical effects such as hardware load sharing, temperature constraint, and wear smoothing.

[0054] Example 8: Expanding Scenarios for Adapting Domestic Desktop Operating Systems

[0055] The host runs macOS, UnionTech UOS, Kylin Desktop, and HarmonyOS. The decision-making layer runs as a user-space service. Based on the system's open API capabilities, it selects medium-privilege interfaces or low-privilege data packets for deployment and scheduling strategies. The architecture is compatible without any specific modifications.

[0056] Example 9: Extended Example of Smart Air Conditioning Appliances The host device is a smart air conditioner with a built-in main control MCU, belonging to embedded home appliance hardware. Internally, it includes compressor power devices, a cooling fan motor, and a power switch module. The decision-making layer runs on the home appliance's main control operating system, collecting real-time data on power transistor junction temperature, relay on / off frequency, and compressor start / stop frequency to calculate real-time hardware wear and tear factors and resource health profit and loss factors. It generates compressor start / stop and fan speed scheduling strategies through a net profit and loss objective function and marks the corresponding permission levels. 0050. The home appliance firmware only provides a standardized SDK interface. The system determines it to be at a medium permission level, enables a medium permission channel, and calls the official home appliance SDK to distribute scheduling constraints, dynamically limiting the compressor's continuous high-load operation time and balancing long-term power device losses. After the hardware starts running, it continuously collects power device temperature and on / off loss data and sends it back to the decision-making layer to iterate scheduling weights. 0051. This home appliance scenario is merely a lower-level extension application of the general architecture of this invention. The four-level execution channel, loss quantification, and closed-loop feedback core logic are completely consistent with the server and mobile terminal scenarios. The underlying core technologies are all recorded in the original application documents of the parent case and do not constitute new technical content.

[0057] Description of Relationship with Prior Patents and Parent Case Relationship to the parent application: This application is a divisional application of application number 202611070355.7. The parent application fully discloses the underlying scheduling foundation for hardware loss quantification, Zone partitioning, and net profit and loss optimization, and clarifies the mechanism by which the kernel hardware abstraction layer (HAL) translates standard kernel instructions into hardware-specific underlying instructions. Under the aforementioned general scheduling framework, this divisional application further refines and explicitly defines the multi-level execution sub-module after the scheduling policy is generated. In particular, it concretizes the instruction translation process implicit in the parent application into four levels of execution channels: high, medium, low, and firmware-coordinated. Although the parent application uses a server as an example, its kernel HAL architecture and scheduling logic are not limited to serving only specific hardware forms. Based on this general architecture, this application extends the execution channels to host devices such as personal computers, mobile terminals, and edge gateways. This extension is a reasonable continuation of the disclosed higher-level technical solutions of the parent application and does not exceed the scope of the original application.

[0058] The prior patent CN202610982953.5 only discloses a partitioned profit and loss optimization algorithm, without involving a four-level permission execution channel or a firmware data packet isolation execution mechanism. The innovations of the two are completely different. This application reuses its hardware loss quantification model, which belongs to the layered extension innovation of the same PECP2.0 technology system, and there is no issue of duplicate authorization.

[0059] Those skilled in the art should understand that the present invention is not limited to the above embodiments. Any equivalent substitutions or simple optimizations made to the execution channel adaptation method, data packet structure, and firmware interaction method within the scope of the present invention shall fall within the protection scope of the present invention.

Claims

1. A hierarchical execution method for loss-aware scheduling decisions, applied to a resource scheduling system based on hardware physical loss awareness, characterized in that, Includes the following steps: Decision-making steps: The decision layer runs in the user mode or kernel mode of the operating system of the host computing device, obtains the real-time hardware consumption factor and resource health gain factor, calls the preset net gain objective function to solve for the optimal number of partitions, and generates the corresponding resource allocation strategy. The resource allocation strategy includes the computing power quota, read and write rate limit value and hot and cold data migration indication for each logical partition. Permission determination steps: The decision layer obtains the permission level of its own host computing device, which includes at least the kernel high permission level, the permission level in the operating system interface, the user-mode low permission level, and the firmware coordination level; Channel selection steps: Select one or a combination of the following execution channels according to the permission level to implement the resource allocation strategy as hardware control actions: - High-privilege channel: The decision-making layer or kernel execution unit directly manipulates hardware registers through the kernel hardware abstraction layer to complete logical partition reconstruction and bandwidth limitation; - Middle-level access control channel: The decision-making level calls the official power management interface of the operating system to map the output of the net profit and loss objective function to the hard limit of computing power of the job object, the frequency point suggestion of the hardware acceleration chip or the power throttling strategy; - Low-privilege channel: The decision-making layer serializes the resource allocation policy into a suggested value data packet and writes it to ACPI variables, WMI variables, UEFI variables or standard output, without modifying the hardware registers itself; - Firmware Coordination Channel: Low-privilege suggested value data packets are received by the UEFI-side trusted service and handed over to the BIOS or EC for final power adjustment and fan control; Execution feedback steps: After hardware execution, real-time resource health and profit / loss factor change data of each logical partition are collected and sent back to the decision-making layer to correct the subsequent net profit / loss objective function optimization weight preference.

2. The method according to claim 1, characterized in that, The suggested value data packet is in JSON or key-value pair structure and includes at least logical partition identifier, standardized computing unit weight, suggested power value, and read / write rate limit percentage.

3. The method according to claim 1, characterized in that, When the host computing device is a lightweight terminal without error correction code memory, the error correction code statistical weight in the wear factor is automatically set to zero, and only the temperature and load indicators are fused for calculation.

4. The method according to claim 1, characterized in that, When the hardware is a SoC with a unified package for CPU and integrated graphics, the entire SoC is treated as a single logical partition object to participate in the net profit and loss objective function optimization, and the integrated graphics are not separately partitioned.

5. The method according to claim 1, characterized in that, Under the aforementioned access control channel, in the Windows operating system environment, the partition weights output by the net profit and loss objective function are mapped to the CPU speed hard cap percentage of the job object.

6. The method according to claim 1, characterized in that, In the firmware collaboration channel, the UEFI-side trusted service only receives suggested value data packets and translates them into standard BIOS / EC control variables, without loading decision-level code, thus achieving bidirectional isolation between decision-making and firmware execution.

7. The method according to claim 1, characterized in that, The suggested value data packet output by the low-privilege channel carries temperature constraint parameters; these temperature constraint parameters can be consumed by the firmware layer or the operating system temperature control abstraction layer to adjust hardware power consumption limits, processor operating frequency, or heat dissipation component output, so that the hardware junction temperature of the corresponding logical partition converges to a safe range.

8. The method according to claim 1, characterized in that, The decision-making layer is decoupled from the execution body. The decision output of the same set of net profit and loss objective functions achieves equivalent hardware loss smoothing effects through the corresponding channels in four host modes: Windows service with normal user privileges, administrator privilege service, Linux kernel module and UEFI trusted service.

9. The method according to claim 1, characterized in that, The host computing device is a personal computer or an x86 server, and the host operating system is a Windows operating system or a Linux operating system.

10. The method according to claim 1, characterized in that, The host computing device is a mobile terminal, an edge computing gateway, or an embedded computing device, and the host operating system is Android or embedded Linux.

11. A hierarchical execution system for loss-aware scheduling decision-making, characterized in that, include: The decision module is used to execute the decision-making steps and the authorization determination steps as described in claim 1; The channel adaptation module is used to perform the channel selection step as described in claim 1, and includes a high-privilege kernel interface adaptation submodule, a medium-privilege OS API adaptation submodule, a low-privilege variable writing submodule, and a firmware collaborative translation submodule. Feedback module, used to execute the execution feedback step as described in claim 1; The modules work together to form a scheduling and execution entity that is decoupled from the operating system kernel and BIOS / EC.

12. The system according to claim 11, characterized in that, The system runs as an independent user-mode service on a host operating system, including Windows and Linux; it outputs suggested value data packets through a low-privilege channel and does not directly hold write permissions to hardware registers.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the method according to any one of claims 1 to 10.

14. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method according to any one of claims 1 to 10.

Citation Information

Patent Citations

  • A heterogeneous computing power scheduling method and system based on dynamic logical partitioning and wear leveling algorithm

    CN122795544A