Heterogeneous computing resource unified scheduling method based on multi-instruction-set architecture

Abstract

The invention belongs to the technical field of heterogeneous computing resource scheduling, and particularly relates to a heterogeneous computing resource unified scheduling method based on a multi-instruction-set architecture. Comprising a heterogeneous resource hybrid scheduling module, a task resource complexity analysis and load real-time detection module, a technical optimization module in a credential environment, a fault domain hidden danger monitoring and recovery module, a resource scheduling optimization and adjustment module and a historical performance analysis and energy consumption monitoring module. The tasks are reasonably distributed according to the dependence degree of the tasks on different computing resources, and resource idle waste is avoided; meanwhile, the task resource complexity analysis and load real-time detection module dynamically adjusts resource allocation, efficient execution of tasks is ensured, and the resource utilization rate and the task execution efficiency are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of heterogeneous computing resource scheduling, and specifically is a method for unified scheduling of heterogeneous computing resources based on a multi-instruction set architecture. Background Art

[0002] With the rapid development of information technology, computer systems are increasingly used in various fields, and the demand for computing power is also growing. In order to meet the complex computing needs in different application scenarios, heterogeneous computing resources (such as CPU, GPU, FPGA, etc.) are widely used. However, in the management and scheduling of heterogeneous computing resources, existing technologies have many shortcomings.

[0003] 1. Heterogeneous resource mixed scheduling problem: Different computing resources have different instruction set architectures and performances. Existing scheduling methods are difficult to fully consider the differences and cannot effectively mix and schedule, resulting in low resource utilization and affected task execution efficiency.

[0004] 2. Defects in task resource analysis and load detection: The analysis of task resource complexity is inaccurate and incomplete, and the real-time load detection technology is imperfect, making it difficult to achieve dynamic matching of task requirements and resource utilization, affecting system performance.

[0005] 3. Insufficient technology in the trusted computing environment: The trusted computing environment has higher requirements on computing resource performance, which cannot be met by existing virtualization and hardware and software fusion acceleration technologies, limiting the development of system applications.

[0006] 4. Limitations of fault monitoring and recovery: Existing technologies cannot comprehensively and timely monitor and warn of faults, and recovery measures are ineffective when faults occur, affecting system stability and reliability.

[0007] To sum up, there are many problems with the existing heterogeneous computing resource scheduling methods, and there is an urgent need for a unified scheduling method for heterogeneous computing resources that can effectively solve the above problems, so as to improve resource utilization, enhance system performance, ensure stable system operation, and meet the high performance requirements in the information and innovation environment. Summary of the invention

[0008] In order to make up for the deficiencies of the prior art, the technical solution adopted by the present invention to solve its technical problems is: a unified scheduling method for heterogeneous computing resources based on a multi-instruction set architecture, including a heterogeneous resource hybrid scheduling module, a task resource complexity analysis and load real-time detection module, a technology optimization module under a trusted innovation environment, a fault domain hidden danger monitoring and recovery module, a resource scheduling optimization and adjustment module, and a historical performance analysis and energy consumption monitoring module, wherein:

[0009] The heterogeneous resource hybrid scheduling module is used to formulate a hybrid scheduling strategy;

[0010] The task resource complexity analysis and load real-time detection module is used to analyze and monitor tasks;

[0011] The technical optimization module in the trusted innovation environment is used to optimize the computing resource performance in the trusted innovation environment;

[0012] The fault domain hidden danger monitoring and recovery module is used to monitor and recover from faults;

[0013] The resource scheduling optimization and adjustment module is used to optimize and adjust resource scheduling;

[0014] The historical performance analysis and energy consumption monitoring module is used to analyze historical performance and monitor energy consumption;

[0015] The heterogeneous resource hybrid scheduling module includes a scheduling strategy formulation unit, which is used to study the characteristics of heterogeneous resources under multiple instruction set environments. Assume that different computing resource sets are R={R1, R2, ..., R n}, the task set is T = {T1, T2, ...T m}, for each task T j , and define its dependence on different computing resources as in Represents task T j For computing resources R i The degree of dependence is in the range of [0,1];

[0016] According to the instruction set architecture, performance characteristics and task requirements of different computing resources, the computing resources in the system are classified and prioritized. By analyzing the compatibility and differences of the instruction sets, the optimal scheduling method for different resource combinations is determined. The resource priority determination unit has a resource priority function P(R i ), calculated as where w j For task T j The weight of Different computing resources include any one or more combinations of CPU, GPU, and FPGA.

[0017] Further, the task resource complexity analysis and load real-time detection module includes a task resource complexity requirement analysis unit, a task load real-time detection unit, and a dynamic matching unit;

[0018] The task resource complexity requirement analysis unit: Assume that task T j The amount of calculation is The amount of data is Defining Task Complexity Metrics Among them, α, β, γ iis the weight coefficient, which is set according to different application scenarios and determines the type and amount of computing resources required based on the complexity of the task;

[0019] The task load real-time detection unit: set the monitored volatile resourceR i The utilization rate is The task progress is The load change rate is Then the real-time task load information vector is

[0020] The dynamic matching unit is provided with a dynamic matching formula Where f is a dynamic matching function, which is determined according to the specific scheduling strategy and algorithm. It realizes the dynamic matching of task demand and resource utilization according to the results of task resource complexity demand analysis and task load real-time detection. When the task load changes, the resource allocation is adjusted in time.

[0021] Task complexity index The weight coefficients α, β, γ i It can be dynamically adjusted according to task type, computing resource characteristics and application scenarios.

[0022] Furthermore, the technical optimization module under the information innovation environment includes a virtualization full offloading and software and hardware fusion acceleration unit and a performance improvement unit;

[0023] The virtualization full offloading and software and hardware fusion acceleration unit: Assume that the virtualization layer optimization coefficient is 0 v , the optimization coefficient of the hardware and software interface is O s , the computing resource performance improvement ratio is E, then in It is a performance improvement function, which is related to the degree of optimization of the virtualization layer and the software and hardware interface;

[0024] The performance improvement unit: Assume that the instruction set execution path optimization coefficient is 0 i , the data transmission speed is increased by a factor of O d , the computing resource processing capacity is increased by E p , then E p =ψ(O i , O d ), where ψ is the processing power improvement function, which is related to the optimization degree of instruction set execution path and data transmission speed.

[0025] Further, the fault domain hidden danger monitoring and recovery module includes a comprehensive monitoring unit, an early warning and recovery unit;

[0026] The comprehensive monitoring unit: Assume that the monitored fault information vector is F = {f1, f2, ..., fk}, the fault diagnosis algorithm is D(F), then the fault type and location determination formula is D(F) = θ(f1, f2, ..., f k ), where θ is the fault diagnosis function, determined according to different fault diagnosis algorithms;

[0027] The warning and recovery unit is provided with a warning information issuing function W(F) and a recovery measure execution function R(F). When a fault is detected, a warning information is issued in time, and corresponding recovery measures are taken according to the fault type and location. For hardware faults, the faulty components are replaced in time; for software faults, the software is repaired or reinstalled; for network faults, the network is repaired or reconfigured, wherein W(F)=ω(f1, f2, ..., f k ), R(F)=ρ(f1, f2,…,f k ), where ω is the warning information issuing function and ρ is the recovery measure execution function, which are determined according to different warning and recovery strategies.

[0028] Furthermore, in the resource scheduling optimization and adjustment module, the task execution efficiency index is Et and the resource utilization index is U t , the resource allocation strategy adjustment function is S(E t , U t ), then S(E t , U t )=δ(E t , U t ), where δ is the resource allocation strategy adjustment function, which is determined according to the changes in task execution efficiency and resource utilization; the resource allocation strategy adjustment function S(E t , U t ) is adjusted according to different situations of task execution efficiency and resource utilization, including increasing or decreasing the allocation of certain computing resources.

[0029] Further, the historical performance analysis and energy consumption monitoring module includes a historical performance analysis unit and an energy saving and emission reduction unit;

[0030] The historical performance analysis unit: Assume that the historical data vector is H = {h1, h2, ..., h l}, the historical performance analysis function is A(H), then A(H)=μ(h1,h2,…,h l ), where μ is the historical performance analysis function, which is determined according to different historical data interpretation methods;

[0031] The energy saving and emission reduction unit: Assume that the energy consumption monitoring data is E d , the energy conservation and emission reduction measures execution function is M(E d ), then M(E d )=v(E d), where v is the energy-saving and emission-reduction measures execution function, which is determined according to different energy-saving and emission-reduction strategies;

[0032] The historical performance analysis function A(H) is adjusted according to the different types of historical data and the purpose of analysis, and the energy conservation and emission reduction measures execution function M(E d ) Make adjustments based on different situations of energy consumption monitoring data, including optimizing resource allocation strategies to reduce the energy consumption of computing resources; adopt energy-saving computing equipment to reduce energy consumption.

[0033] Furthermore, the performance improvement function and ψ are dynamically adjusted according to the optimization degree of the virtualization layer and the software and hardware interface, as well as the optimization degree of the instruction set execution path and data transmission speed.

[0034] Furthermore, the fault diagnosis algorithm D(F) is dynamically adjusted according to different characteristics of hardware faults, software faults and network faults.

[0035] Furthermore, the resource allocation strategy adjustment function S(E t , U t ) is dynamically adjusted according to different situations of task execution efficiency and resource utilization, including increasing or decreasing the allocation of a certain computing resource to adapt to the dynamic changes in task requirements and resource utilization.

[0036] Furthermore, the historical performance analysis function A(H) is dynamically adjusted according to the different types of historical data and the analysis purpose to more accurately reflect the performance change trend of the system; the energy saving and emission reduction measures execution function M(E d ) Make dynamic adjustments based on different situations of energy consumption monitoring data to achieve more effective energy conservation and emission reduction effects.

[0037] The beneficial effects of the present invention are as follows:

[0038] 1. The present invention uses heterogeneous resource hybrid scheduling to reasonably allocate tasks according to the degree of dependence of tasks on different computing resources, thereby avoiding idle resources and waste; at the same time, the task resource complexity analysis and load real-time detection modules dynamically adjust resource allocation to ensure efficient task execution, thereby improving resource utilization and task execution efficiency.

[0039] 2. The present invention improves computing resource performance through the technical optimization module in the information and communication environment, and the fault domain hidden danger monitoring and recovery module ensures system stability; each module promotes each other, such as the optimized resources provide a better basis for fault monitoring, and fault recovery ensures optimization results, thereby jointly improving the overall performance and stability of the system.

[0040] 3. The present invention analyzes historical data through historical performance analysis and energy consumption monitoring modules to achieve energy conservation and emission reduction; at the same time, it provides a reference for resource scheduling, makes intelligent decisions based on task load and resource utilization, optimizes resource allocation strategies, and improves system adaptability and energy utilization efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The present invention will be further described below in conjunction with the accompanying drawings.

[0042] Figure 1 It is a module schematic diagram of the present invention. DETAILED DESCRIPTION

[0043] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0044] like Figure 1 As shown, the present invention proposes a unified scheduling method for heterogeneous computing resources based on a multi-instruction set architecture, and uses an image processing task as an example to verify the technical effect of the present invention. The present invention includes a heterogeneous resource hybrid scheduling module, a task resource complexity analysis and load real-time detection module, a technology optimization module under a trusted computing environment, a fault domain hidden danger monitoring and recovery module, a resource scheduling optimization and adjustment module, and a historical performance analysis and energy consumption monitoring module;

[0045] Heterogeneous resource hybrid scheduling module:

[0046] Image processing tasks include a large number of operations that can be calculated in parallel, such as convolution operations, and also require some pre-processing and post-processing logical operations. GPU is suitable for processing parallel computing due to its powerful parallel computing capabilities, while CPU's logical processing capabilities are more suitable for pre-processing and post-processing operations. Let's assume that the image processing task depends on GPU. Dependence on GPU The weight w of the task in the task set j =0.6, computing resource priority:

[0047]

[0048]

[0049] According to the compatibility and difference of instruction sets, tasks suitable for GPU parallel computing, such as convolution operations, are assigned to the GPU, and tasks suitable for CPU processing, such as data reading, preprocessing, and result integration, are assigned to the CPU. When determining task allocation, a simple rule algorithm can be used. Let task T j Contains multiple subtasks T j，k , calculate its expected execution time t on the GPU GPU，kand the expected execution time on the CPU t CPU，k , if t GPU，k <t CPU，k , then the subtask T j，k Assign to GPU; otherwise, assign to CPU; the expected execution time can be calculated by the computational complexity C of the subtask j，k and the performance index P of the corresponding computing resources GPU (for GPU) or P CPU (for CPU) to calculate, for example, t GPU，k =C j，k / P GPU , t CPU，k =C j，k / P CPU ;

[0050] This scheduling method makes full use of the parallel computing advantages and logical processing advantages of the GPU, improves the execution efficiency of image processing tasks, and avoids idleness and waste of resources by reasonably allocating tasks to different computing resources, thereby improving resource utilization. At the same time, it considers task weights and resource dependence to determine priorities, making resource allocation more scientific and reasonable, and better meeting task requirements.

[0051] Task resource complexity analysis and load real-time detection module:

[0052] The task complexity index is defined by comprehensively considering the task's computational workload, data volume, and the degree of dependence on different computing resources. This is used to accurately assess the complexity of the task and determine the type and amount of computing resources required. For this image processing task, the amount of computing power is measured. The number of operations is estimated by counting the number of operations on the image pixels) to be 5000 operations, and the amount of data (Image file size) is 5MB;

[0053] According to the set weight coefficients α=0.3、β=0.4、γ1=0.2、γ1=0.1, according to the formula Calculating task complexity index According to the task complexity index, three GPUs and one GPU core are allocated. During the task execution, the CPU and GPU usage rates are monitored in real time. Task Progress and load change rate Assume that at a certain moment, the GPU usage is, the CPU usage is, the task progress is, and the load change rate is, then the real-time task load information vector is

[0054] According to the task complexity index Real-time task load information vector and resource priority function P(R i ), dynamically matching and adjusting resource allocation; dynamic matching and adjustment can adopt a simple linear weighted algorithm, assuming that the adjusted resource allocation vector is but where w I 、w L 、w P is the pre-set weight coefficient, Represents the real-time task load information vector The i-th element in ;

[0055] The task complexity index can accurately reflect the complexity of the task, making resource allocation more reasonable, monitoring task load information in real time and making dynamic matching adjustments. It can timely adjust resource allocation according to the actual execution of the task to ensure that the task can be executed efficiently, further improving resource utilization and task execution efficiency. This dynamic adjustment mechanism can adapt to changes in task load and avoid performance degradation caused by insufficient or excessive resources.

[0056] At the same time, the heterogeneous resource hybrid scheduling module provides a basis for task resource complexity analysis: the degree of dependence of tasks on different computing resources determined by the heterogeneous resource hybrid scheduling module is an important basis for calculating task complexity indicators; this enables task resource complexity analysis to more accurately evaluate the resources required for tasks and provide a basis for reasonable resource allocation;

[0057] In addition, the task resource complexity analysis and load real-time frontier module optimizes the mixed scheduling of heterogeneous resources: by real-time monitoring of task load and dynamic adjustment of resource allocation, the mixed scheduling of heterogeneous resources can be optimized according to the actual execution of the task. For example, when the task load changes, the resource allocation can be adjusted in time to improve the flexibility and adaptability of the mixed scheduling of heterogeneous resources and ensure efficient execution of the task.

[0058] Technical optimization module in the information innovation environment:

[0059] In the information innovation environment, the performance and efficiency of computing resources can be improved by optimizing the virtualization layer and the software and hardware interface, as well as the instruction set execution path and data transmission speed. For GPU, the virtualization layer and the software and hardware interface are optimized, and the virtualization layer optimization coefficient is set to V =0.2, software and hardware interface optimization coefficient O s =0.3, according to the performance improvement function Computing performance improvement ratio E, A simple linear combination function can be used, such as E = a × O V +b×O s, where a and b are pre-set coefficients; optimize the instruction set execution path and data transmission speed, and set the instruction set execution path optimization coefficient O i =0.3, data transmission speed improvement factor O d = 0.2, and calculate the processing capacity improvement ratio E according to the performance improvement function ψ p , ψ can adopt similar linear combination functions, such as E p =c×O i +d×O d , where c and d are pre-set coefficients;

[0060] These optimization measures can significantly improve the performance of the GPU, making it more efficient in processing image processing tasks, improving the performance and efficiency of computing resources, helping to shorten task execution time and improve the overall performance of the system; optimizing the virtualization layer and software and hardware interfaces can reduce system overhead and improve the effective utilization of resources; optimizing the instruction set execution path and data transmission speed directly improves the GPU's computing speed and data transmission capabilities, accelerating the processing of image processing tasks;

[0061] The heterogeneous resource hybrid scheduling module guides the technical optimization direction in the information and innovation environment: The heterogeneous resource hybrid scheduling module determines the computing resources that mainly undertake tasks, which points out the optimization direction for the colorful world technical optimization module in the information and innovation environment; for example, if it is determined that the image processing task is mainly undertaken by the CPU, it will guide the technical optimization module in the information and innovation environment to focus on optimizing the GPU, thereby improving the pertinence and effectiveness of the optimization;

[0062] The technical optimization module in the information innovation environment improves the mixed scheduling effect of heterogeneous resources: the performance of the optimized computing resources is improved, so that the mixed scheduling module of heterogeneous resources can play a better role. For example, the optimized GPU can handle image processing tasks more efficiently, which improves the overall effect of mixed scheduling of heterogeneous resources and enables tasks to be executed more efficiently on different computing resources.

[0063] The task resource complexity analysis and load real-time detection module provide feedback for technical optimization in the trusted innovation environment: the task complexity index and real-time task load information reflect the task's demand for and use of computing resources, and provide feedback for technical optimization in the trusted innovation environment; for example, if the task load detection shows that the image processing task has too high a load on the GPU, the technical optimization module in the trusted innovation environment can further optimize the GPU performance based on this and improve the targeted optimization;

[0064] The technical optimization module in the information and communication technology innovation environment affects the results of task resource complexity analysis and real-time load detection: the performance change of computing resources after optimization will affect the results of task resource complexity analysis and real-time load detection; for example, the improvement of GPU performance may change the degree of task dependence on the CPU, thereby affecting the calculation results of the task complexity index, and at the same time will also cause changes in information such as the utilization rate obtained by real-time load detection. This mutual influence prompts continuous adjustment and optimization between modules to achieve better task execution results.

[0065] Fault domain hidden danger monitoring and recovery module:

[0066] Set monitoring points on computing resources and network devices, collect fault-related information, use fault diagnosis algorithms to determine the fault type and location, issue warning information in a timely manner and take corresponding recovery measures. Set monitoring points on GPUs and GPUs to monitor hardware faults (such as overtemperature, chip failure, etc.), software faults (such as driver errors, algorithm errors, etc.) and network equipment faults (such as data transmission interruptions, etc.); assuming that the fault information of GPU overtemperature is monitored, determine the fault type and location through the fault diagnosis algorithm, issue a warning message, and take recovery measures such as replacing the heat sink or adjusting the GPU operating frequency. The fault diagnosis algorithm here can adopt a rule-based classification algorithm; suppose the monitored fault information vector is F = {f1, f2, ..., f k}, define a classification function D(F), for each possible fault type T m , calculate its similarity S(T m , F), for example, the cosine similarity algorithm can be used where t m，i Is the fault type T m The i-th element in the corresponding feature vector. Then the fault type with the highest similarity is selected as the diagnosis result;

[0067] Comprehensive fault monitoring and timely recovery measures can ensure the stable operation of the system, reduce task interruptions and data loss caused by faults, and improve the reliability and availability of the system. For image processing tasks, data integrity and processing continuity are crucial. Fault monitoring and recovery mechanisms can ensure that faults can be handled in a timely manner to avoid adverse effects on image processing results. At the same time, through timely warnings, measures can be taken in advance to prevent faults from occurring, further improving the stability of the system.

[0068] The heterogeneous resource hybrid scheduling module affects the focus of fault monitoring and recovery: The computing resource usage and priority determined by the heterogeneous resource hybrid scheduling module affect the monitoring focus and recovery strategy of the fault domain hidden danger monitoring and recovery module. For example, for GPUs with higher priority in image processing tasks, the fault domain hidden danger monitoring and recovery module will pay more attention to GPU hardware and software faults, and formulate corresponding recovery measures to improve the pertinence of fault monitoring and recovery.

[0069] The fault domain hidden danger monitoring and recovery module ensures the stability of heterogeneous resource mixed scheduling: The fault domain hidden danger monitoring and recovery module ensures the stability of heterogeneous resource mixed scheduling by timely detecting and handling faults; if a GPU fails, timely recovery measures can enable the system to resume normal operation as soon as possible, avoiding confusion in heterogeneous resource mixed scheduling caused by the fault, and ensuring that tasks can continue to be executed according to the predetermined scheduling strategy.

[0070] Resource scheduling optimization and adjustment module:

[0071] Optimize and adjust resource scheduling according to the actual situation during task execution. Adjust resource allocation strategies by analyzing indicators such as task execution efficiency and resource utilization. For example, in addition to considering the utilization of GPU and CPU, the execution time distribution of different subtasks on different computing resources will also be analyzed. If it is found that the execution time of a subtask on the GPU is too long and its computational complexity is not high, it may be considered to be adjusted to the GPU for execution. At the same time, the overall progress and remaining workload of the task, as well as the load and performance bottlenecks of each computing resource, will be comprehensively considered to dynamically adjust the resource allocation ratio.

[0072] Receive resource allocation information from modules such as heterogeneous resource hybrid scheduling, task complexity indicators and real-time task load information from task resource complexity analysis and load real-time detection modules, performance improvement information from technology optimization modules under the information innovation environment, and fault information from fault domain hidden danger monitoring and recovery modules, and further optimize and adjust resource allocation strategies based on this information; for example, if fault information is received indicating that a GPU fault has occurred, the resource allocation strategy will be adjusted based on the severity of the fault and the urgency of the task, such as transferring all tasks originally assigned to the GPU to the CPU for execution, or reallocating tasks based on the remaining available resources;

[0073] The resource scheduling optimization and adjustment module can dynamically adjust resource allocation according to the actual execution of the task, further improving resource utilization and task execution efficiency; by continuously optimizing the resource allocation strategy, the system can better adapt to different task loads and computing resource states, and improve the overall performance and adaptability of the system; this dynamic adjustment mechanism can promptly respond to various changes in the task execution process, such as increase or decrease in task complexity, fluctuations in computing resource performance, etc., to ensure that the task can be executed under the optimal resource configuration.

[0074] Historical performance analysis and energy consumption monitoring module:

[0075] The analysis of historical performance and energy consumption monitoring provide an important basis for system optimization. Understanding the performance change trend of the system helps to predict possible problems in advance, such as resource bottlenecks or performance degradation, so as to take preventive measures; analyzing the energy consumption distribution under different task loads can adjust the resource allocation strategy and the working mode of computing resources in a targeted manner to reduce energy consumption; this not only helps to reduce operating costs and meet environmental protection requirements, but is also of great significance for energy-sensitive application scenarios such as large-scale data centers;

[0076] By collaborating with other modules, it is able to obtain relevant information and update analysis results in a timely manner, enabling the system to make more accurate decisions based on actual conditions; for example, based on changes in task complexity or computing resource performance improvement information fed back by other modules, re-evaluate historical performance and energy consumption, further optimize resource allocation strategies, improve the overall performance and adaptability of the system, and ensure that tasks can be executed under the optimal resource configuration while achieving energy conservation and emission reduction goals.

[0077] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for unified scheduling of heterogeneous computing resources based on a multi-instruction set architecture, characterized in that: It includes heterogeneous resource hybrid scheduling module, task resource complexity analysis and load real-time detection module, technology optimization module under the information innovation environment, fault domain hidden danger monitoring and recovery module, resource scheduling optimization and adjustment module, historical performance analysis and energy consumption monitoring module, among which: The heterogeneous resource hybrid scheduling module is used to formulate a hybrid scheduling strategy; The task resource complexity analysis and load real-time detection module is used to analyze and monitor tasks; The technical optimization module in the trusted innovation environment is used to optimize the computing resource performance in the trusted innovation environment; The fault domain hidden danger monitoring and recovery module is used to monitor and recover from faults; The resource scheduling optimization and adjustment module is used to optimize and adjust resource scheduling; The historical performance analysis and energy consumption monitoring module is used to analyze historical performance and monitor energy consumption; The heterogeneous resource hybrid scheduling module includes a scheduling strategy formulation unit, which is used to study the characteristics of heterogeneous resources under multiple instruction set environments. Assume that different computing resource sets are R={R1, R2, ..., R n }, the task set is T = {T1, T2, ...T m }, for each task T j , and define its dependence on different computing resources as in Represents task T j For computing resources P i The degree of dependence is in the range of [0,1]; According to the instruction set architecture, performance characteristics and task requirements of different computing resources, the computing resources in the system are classified and prioritized. By analyzing the compatibility and differences of the instruction sets, the optimal scheduling method for different resource combinations is determined. The resource priority determination unit has a resource priority function P(R i ), calculated as where w j For task T j The weight of Different computing resources include any one or more combinations of CPU, GPU, and FPGA.

2. The UAV insulator inspection system based on multi-light fusion according to claim 1 is characterized by: The task resource complexity analysis and load real-time detection module includes a task resource complexity demand analysis unit, a task load real-time detection unit, and a dynamic matching unit; The task resource complexity requirement analysis unit: Assume that task T j The amount of calculation is The amount of data is Defining Task Complexity Metrics Among them, α, β, γ i is the weight coefficient, which is set according to different application scenarios and determines the type and amount of computing resources required based on the complexity of the task; The task load real-time detection unit: set the monitored volatile resourceR i The utilization rate is The task progress is The load change rate is Then the real-time task load information vector is The dynamic matching unit is provided with a dynamic matching formula Where f is a dynamic matching function, which is determined according to the specific scheduling strategy and algorithm. It realizes the dynamic matching of task demand and resource utilization according to the results of task resource complexity demand analysis and task load real-time detection. When the task load changes, the resource allocation is adjusted in time. Task complexity index The weight coefficients α, β, γ i It can be dynamically adjusted according to task type, computing resource characteristics and application scenarios.

3. The method for unified scheduling of heterogeneous computing resources based on a multi-instruction set architecture according to claim 1, characterized in that: The technical optimization modules in the information innovation environment include virtualization full offloading and software and hardware fusion acceleration units, and performance improvement units; The virtualization full offloading and software and hardware fusion acceleration unit: Assume that the virtualization layer optimization coefficient is 0 v , the optimization coefficient of the hardware and software interface is O s , the computing resource performance improvement ratio is E, then in It is a performance improvement function, which is related to the degree of optimization of the virtualization layer and the software and hardware interface; The performance improvement unit: Assume that the instruction set execution path optimization coefficient is 0 i , the data transmission speed is increased by a factor of O d , the computing resource processing capacity is increased by E p , then E p =ψ(O i , O d ), where ψ is the processing power improvement function, which is related to the optimization degree of instruction set execution path and data transmission speed.

4. The method for unified scheduling of heterogeneous computing resources based on a multi-instruction set architecture according to claim 1, characterized in that: The fault domain hidden danger monitoring and recovery module includes a comprehensive monitoring unit, an early warning and recovery unit; The comprehensive monitoring unit: Assume that the monitored fault information vector is F = {f1, f2, ..., f k }, the fault diagnosis algorithm is D(F), then the fault type and location determination formula is D(F) = θ(f1, f2, ..., f k ), where θ is the fault diagnosis function, determined according to different fault diagnosis algorithms; The warning and recovery unit is provided with a warning information issuing function W(F) and a recovery measure execution function R(F). When a fault is detected, a warning information is issued in time, and corresponding recovery measures are taken according to the fault type and location. For hardware faults, the faulty components are replaced in time. For software failures, repair or reinstall the software; For network failures, network repair or reconfiguration is performed, where W(F) = ω(f1, f2, ..., f k ), R(F)=ρ(f1, f2,..., f k ), where ω is the warning information issuing function and ρ is the recovery measure execution function, which are determined according to different warning and recovery strategies.

5. The method for unified scheduling of heterogeneous computing resources based on a multi-instruction set architecture according to claim 1, characterized in that: The resource scheduling optimization and adjustment module sets the task execution efficiency index as E t , the resource utilization index is U t , the resource allocation strategy adjustment function is S(E t , U t ), then S(E t , U t )=δ(E t , U t ), where δ is the resource allocation strategy adjustment function, which is determined according to the changes in task execution efficiency and resource utilization; the resource allocation strategy adjustment function S(E t , U t ) is adjusted according to different situations of task execution efficiency and resource utilization, including increasing or decreasing the allocation of certain computing resources.

6. The method for unified scheduling of heterogeneous computing resources based on a multi-instruction set architecture according to claim 1, characterized in that: The historical performance analysis and energy consumption monitoring module includes a historical performance analysis unit and an energy saving and emission reduction unit; The historical performance analysis unit: Assume that the historical data vector is H = {h1, h2, ..., h l }, the historical performance analysis function is A(H), then A(H)=μ(h1,h2,…,h l ), where μ is the historical performance analysis function, which is determined according to different historical data interpretation methods; The energy saving and emission reduction unit: Assume that the energy consumption monitoring data is E d , the energy conservation and emission reduction measures execution function is M(E d ), then M(E d )=v(E d ), where v is the energy-saving and emission-reduction measures execution function, which is determined according to different energy-saving and emission-reduction strategies; The historical performance analysis function A(H) is adjusted according to the different types of historical data and the purpose of analysis, and the energy conservation and emission reduction measures execution function M(E d ) Make adjustments based on different situations of energy consumption monitoring data, including optimizing resource allocation strategies to reduce the energy consumption of computing resources; adopt energy-saving computing equipment to reduce energy consumption.

7. The method for unified scheduling of heterogeneous computing resources based on a multi-instruction set architecture according to claim 3, characterized in that: The performance improvement function and ψ are dynamically adjusted according to the optimization degree of the virtualization layer and the software and hardware interface, as well as the optimization degree of the instruction set execution path and data transmission speed.

8. The method for unified scheduling of heterogeneous computing resources based on a multi-instruction set architecture according to claim 4, characterized in that: The fault diagnosis algorithm D(F) is dynamically adjusted according to the different characteristics of hardware faults, software faults and network faults.

9. The method for unified scheduling of heterogeneous computing resources based on a multi-instruction set architecture according to claim 5, characterized in that: The resource allocation strategy adjustment function S(E t , U t ) is dynamically adjusted according to different situations of task execution efficiency and resource utilization, including increasing or decreasing the allocation of a certain computing resource to adapt to the dynamic changes in task requirements and resource utilization.

10. A method for unified scheduling of heterogeneous computing resources based on a multi-instruction set architecture according to claim 6, characterized in that the historical performance analysis function A(H) is dynamically adjusted according to different types of historical data and analysis purposes to more accurately reflect the performance change trend of the system; the energy-saving and emission reduction measures execution function M(E d ) Make dynamic adjustments based on different situations of energy consumption monitoring data to achieve more effective energy conservation and emission reduction effects.

Images