High-performance computing power resource assessment method based on AI technology

Abstract

The invention discloses a high-performance computing power resource evaluation method based on an AI technology, relates to the technical field of high-performance computing power resource evaluation, and solves the technical problems that in the prior art, influences brought by all parameters cannot be overcome during computing power resource evaluation adjustment, and the processing performance of computing power resources is reduced, in particular to computing power resource characteristic evaluation. Carrying out multi-dimensional index acquisition and detection according to various types of hardware related to the computing power resources; after multi-dimensional index detection is completed, the historical operation process of the computing power resources is detected through the AI technology; evaluating computing power parameters, performing weight analysis according to various types of hardware multi-dimensional indexes related to computing power resources, and performing division according to an influence weight proportion to obtain real-time computing power high-performance parameters and computing power low-performance parameters; and AI prediction: detecting the load of the computing power resource, inputting characteristics, and obtaining a computing power demand interval according to historical output so as to carry out real-time assessment on the computing power resource.

Description

Technical Field

[0001] The present invention relates to the technical field of high-performance computing power resource evaluation, and specifically to a high-performance computing power resource evaluation method based on AI technology. Background Art

[0002] The evaluation of high-performance computing power resources based on AI technology needs to comprehensively consider multi-dimensional indicators such as hardware performance, software adaptation, application scenarios, and energy efficiency, and combine technologies such as machine learning and deep learning to achieve dynamic and accurate evaluation.

[0003] However, in the existing technology, when using high-performance computing power resources, it is impossible to collect and detect according to multi-dimensional indicators for adaptability analysis, impossible to adjust in time as indicator parameters, and unable to monitor the influence of multi-dimensional indicators through weight analysis. It is impossible to overcome the influence brought by all parameters during the evaluation and adjustment of computing power resources, reducing the processing performance of computing power resources.

[0004] In view of the above technical defects, a solution is proposed. Summary of the Invention

[0005] The purpose of the present invention is to solve the above-mentioned problems and propose a high-performance computing power resource evaluation method based on AI technology.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] As a preferred embodiment of the present invention, a high-performance computing power resource evaluation method based on AI technology;

[0008] Step 1: Evaluate the characteristics of computing power resources. Collect and detect multi-dimensional indicators based on various types of hardware involved in computing power resources. The hardware includes hardware devices such as processors, GPUs, and memories. The multi-dimensional indicators include computing power, storage performance, and network communication;

[0009] Infer whether the characteristics of the computing power resources constructed by the operation of the hardware are efficient, so as to evaluate the characteristics of the computing power resources through hardware characteristic analysis. When the computing power resources are not adaptable, adjust the hardware indicators in time to improve the computing power performance of the computing power resources themselves;

[0010] After completing the multi-dimensional indicator detection, use AI technology to detect the historical operation process of the computing power resources, count the time nodes of the fluctuations of the corresponding multi-dimensional indicators of various types of hardware, and infer the hardware adaptation efficiency based on the time node analysis. After obtaining the analysis result of the adaptation efficiency, conduct real-time fluctuation prediction according to the regulation of real-time multi-dimensional indicator parameters to infer the real-time operation performance of the current computing power resources;

[0011] Step 2: Computing power parameter evaluation. Perform weight analysis based on multi-dimensional indicators of various types of hardware involved in computing power resources, and divide to obtain real-time high-performance computing power parameters and low-performance computing power parameters according to the proportion of influence weights; Step 3: AI prediction. Detect the load of computing power resources, input features, and obtain the computing power demand range based on historical outputs. Detect the influence of input features according to the historical load processing process, and perform influence analysis on real-time input features according to the influence detection results to perform real-time evaluation of computing power resources.

[0012] As a preferred embodiment of the present invention, the multi-dimensional indicator acquisition and detection process is as follows:

[0013] The multi-dimensional indicators corresponding to various types of hardware involved in computing power resources are respectively computing power, storage performance, and network communication; and the multi-dimensional indicators are quantified. Among them, computing power is expressed as the numerical ratio of the execution buffer time for the hardware to process data tasks to the data task processing error rate corresponding value in the current time; storage performance is expressed as the numerical ratio of the actual occupancy rate of each storage space of the hardware to the delay duration of data calls in multiple storage spaces; network communication is expressed as the numerical ratio of the actual throughput of the network covered by the hardware to the actual network delay frequency; it should be noted that the numerical ratio is the numerical ratio of two collected data, without considering the dimension problem of the collected data, and only performing floating influence analysis on the numerical values of the data;

[0014] Compare the numerical values of the quantified multi-dimensional indicators, that is, compare the corresponding quantified parameters of the multi-dimensional indicators with the corresponding set parameter threshold ranges respectively: If the corresponding quantified parameter exceeds the set parameter threshold range, it is inferred that the hardware of the current computing power resource has high-performance characteristics, and analyze the complexity of the actual data processing task. If the data processing task complexity is high, set the current computing power resource as an adaptable resource; otherwise, if the data processing task complexity is low, set the current computing power resource as an excess resource;

[0015] If the corresponding quantified parameter is within the set parameter threshold range, it is inferred that the computing power resource is a high-cost-effective resource; otherwise, if the corresponding quantified parameter is lower than the set parameter threshold range, it is inferred that the computing power resource has low-performance characteristics, and perform regulation on the multi-dimensional indicators of the hardware of the computing power resource;

[0016] If high-performance characteristics are detected, the multi-dimensional indicator detection is qualified. If it is an excess resource, re-plan the use of the computing power resource.

[0017] As a preferred embodiment of the present invention, the historical operation detection process is as follows:

[0018] Obtain the historical operation period of the current computing power resource, and collect the quantization parameters of multi-dimensional indicators within the historical operation period. Before the computing power resource executes the data processing task, set the quantization parameters at the current moment as input features, and mark the type of the data processing task as a task feature, where the task feature is represented by feature parameters such as task type, data processing volume, and data setting buffer duration;

[0019] When the task feature is constant, record the numerical floating moment of the input feature and the abnormal moment of the data processing task. If the floating moment and the abnormal moment are in the same task processing cycle, count the numerical floating of the quantization parameters of the input feature, and use the overlapping probability between the numerical floating moment and the task abnormal moment as the division standard of the influence level to set the influence on the feature vector with numerical floating, such as feature vectors like numerical floating span, numerical floating duration, or numerical floating frequency;

[0020] If the overlapping probability exceeds the set probability threshold, set the feature vector with floating at the corresponding floating moment as a high-probability influence vector. Conversely, if the overlapping probability does not exceed the set probability threshold, set the feature vector with floating at the corresponding floating moment as a low-probability influence vector; and conduct real-time fluctuation analysis according to each type of feature vector and the corresponding vector type.

[0021] As a preferred implementation manner of the present invention, the real-time fluctuation analysis process is as follows:

[0022] Compare the task feature of the current task with the tasks in each historical period. The task features with task feature parameters in the same range are regarded as the same type of tasks. Collect the feature vectors of the quantization parameters with floating when the same type of tasks are executed, and identify the type of the feature vector according to the historical operation period of the same type of tasks. When it is a high-probability influence vector, give an early warning of the abnormality of the data processing task and use the abnormal type of the historical data processing task as the current warning object, and adjust the feature vector according to the warning object; when it is a low-probability influence vector, identify the floating trend of the feature vector within the historical operation period, and monitor the floating trend of the current same type of feature vector and synchronize the trend for timely adjustment.

[0023] As a preferred implementation manner of the present invention, the process of evaluating the computing power parameter in step two is as follows:

[0024] Analyze various types of hardware multi-dimensional indicators related to computing power resources, statistically analyze the numerical floating trends of the quantization parameters of the multi-dimensional indicators, and analyze the floating trends of the quantization parameters when the computing power resources execute data processing tasks. When the floating trend of the quantization parameters does not tend to be stable, that is, there is a floating trend and reciprocating floating, obtain the processing speed decrease frequency and the speed recovery buffer duration after the processing speed decreases during the current stage of the data processing task. If any value in the processing speed decrease frequency and the speed recovery buffer duration after the processing speed decreases exceeds the corresponding set threshold, it is inferred that the current quantization parameter is a high-risk parameter; conversely, if there is no value in the processing speed decrease frequency and the speed recovery buffer duration after the processing speed decreases that exceeds the corresponding set threshold, it is inferred that the current quantization parameter is a low-risk parameter;

[0025] Set the influence weight of high-risk parameters to a high influence weight, and the set weight factor to a low numerical factor to avoid the influence of high-risk parameters being infinitely amplified relative to the influence of low-risk parameters, so as to ignore the influence of low-risk parameters. It should be noted that the high or low value of the weight factor is distinguished by comparing the low numerical factor with the high numerical factor; set the influence weight of high-risk parameters to a low influence weight, and the set weight factor to a high numerical factor;

[0026] Classify the quantization parameters of the multi-dimensional indicators, that is, uniformly mark the high influence weight and the corresponding weight factor as computing power high-performance parameters, and uniformly mark the low influence weight and the corresponding weight factor as computing power low-performance parameters.

[0027] As a preferred implementation manner of the present invention, the AI prediction process is as follows:

[0028] Collect the multi-dimensional indicators and corresponding quantization parameters of the current computing power resources, monitor the quantization parameters during the real-time data processing task execution stage, set a resource load analysis model, record the floating feature vector of the quantization parameters during the execution stage, and mark it as the model startup feature. When the model startup feature is generated, evaluate the computing power resource utilization rate according to the type of the model startup feature, that is, high-probability influence feature and low-probability influence feature;

[0029] Collect the high-risk parameters and low-risk parameters in the quantization parameters and the corresponding weight factors, and mark the high-risk parameters and low-risk parameters as C 高 and C 低 ; the weight factors are Q 高 and Q 低 ; Through the formula To obtain the influence coefficient S of the high-probability influence feature, it should be noted that the values of the high-risk parameter and the low-risk parameter are the values of the corresponding feature vectors, while the weight factor is artificially set through continuous analysis of historical periods. At the same time, when the units of the feature vectors are not unified, the weight factor can be directly used for dimensionless processing;

[0030] If the influence coefficient S exceeds the coefficient threshold and continues to increase, it indicates that the corresponding utilization rate of the current multi-dimensional indicators of the computing power resources continues to decrease. Based on the data obtained during the execution of the same type of tasks in the historical operation period, the abnormal type of the data processing task is obtained. After identifying the abnormal type of the current processing task, the corresponding feature vector regulation of the abnormal type of the data processing task is obtained, and the regulation range of the feature vector in the historical process is marked as the computing power demand interval corresponding to the quantization parameter of the current computing power resources. The current multi-dimensional indicators are regulated according to the computing power demand interval; conversely, if the influence coefficient S does not exceed the coefficient threshold and is constant, it indicates that the corresponding utilization rate of the current multi-dimensional indicators of the computing power resources is constant;

[0031] When the model startup feature is a high-probability influence feature, the maximum floating span of the high-risk parameter and the type quantity of the floating parameters of the low-risk parameter are collected. If any parameter among the maximum floating span of the high-risk parameter and the type quantity of the floating parameters of the low-risk parameter increases, the computing power demand interval needs to be expanded according to the regulation trend; conversely, if no parameter increases, the current computing power demand interval is used for regulation;

[0032] When the model startup feature is a low-probability influence feature, the numerical increase span of the floating frequency of the high-risk parameter and the continuous floating cumulative span of the low-risk parameter are collected. If any parameter among the numerical increase span of the floating frequency of the high-risk parameter and the continuous floating cumulative span of the low-risk parameter exceeds the set numerical threshold, the current computing power demand interval is used as the regulation peak, and the computing power demand interval is regulated step by step in stages; if any parameter among the numerical increase span of the floating frequency of the high-risk parameter and the continuous floating cumulative span of the low-risk parameter does not exceed the set numerical threshold, the current computing power demand interval is reduced to reduce the regulation span of the feature vector; it should be noted that the premise for regulating the computing power demand interval is that the floating of the quantization parameter feature vector does not reach the value of the feature vector when the data processing task is abnormal.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] 1. In the present invention, multi-dimensional index collection and detection are carried out according to various types of hardware involved in the computing power resources;

[0035] Infer whether the characteristics of the computing power resources constructed by the operation of the hardware are efficient, so as to evaluate the characteristics of the computing power resources through the analysis of the hardware characteristics, and timely adjust the hardware indicators when the computing power resources cannot be adapted, so as to improve the computing power performance of the computing power resources themselves; Infer the hardware adaptation efficiency according to the time node analysis, and after obtaining the adaptation efficiency analysis result, predict the real-time fluctuation according to the regulation of the real-time multi-dimensional index parameters, so as to infer the real-time operation performance of the current computing power resources.

[0036] 2. In the present invention, weight analysis is carried out according to various types of hardware multi-dimensional indexes involved in the computing power resources, and real-time high-performance computing power parameters and low-performance computing power parameters are obtained according to the influence weight ratio; Through the division of the performance parameters, the influence weight setting of the parameters can be effectively carried out, so as to ensure that the influence of each parameter can be collected and statistically analyzed in real time, avoid the influence of different parameters when using the computing power resources, avoid the large deviation of the corresponding influence of the two parameters resulting in the loss of the influence effect of the parameter value with low influence, so that there is a regulation deviation due to the lack of influence parameters during the regulation of the computing power demand interval, and the deviation cannot be minimized. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] For the convenience of those skilled in the art to understand, the present invention will be further described below with reference to the accompanying drawings.

[0038] Figure 1 is the overall method flow chart of the present invention;

[0039] Figure 2 is the method flow chart of multi-dimensional index acquisition and detection in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0041] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the invention. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive of other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0042] Please refer to Figure 1 as shown, a high-performance computing power resource evaluation method based on AI technology;

[0043] Step 1. Evaluation of computing power resource characteristics: Collect and detect multi-dimensional indicators based on various types of hardware involved in the computing power resources. The hardware includes hardware devices such as processors, GPUs, and memories; the multi-dimensional indicators include computing power, storage performance, and network communication;

[0044] Infer whether the corresponding characteristics of the computing power resources constructed by the operation of the hardware are efficient, so as to evaluate the characteristics of the computing power resources through the analysis of the hardware characteristics, and timely adjust the hardware indicators when the computing power resources cannot be adapted, so as to improve the computing power performance of the computing power resources themselves;

[0045] After completing the multi-dimensional indicator detection, use AI technology to detect the historical operation process of the computing power resources, statistically analyze the time nodes of the fluctuations of the corresponding multi-dimensional indicators of various types of hardware, and infer the hardware adaptation efficiency based on the time node analysis. After obtaining the analysis result of the adaptation efficiency, perform real-time fluctuation prediction according to the adjustment of the real-time multi-dimensional indicator parameters to infer the real-time operation performance of the current computing power resources;

[0046] Step 2. Evaluation of computing power parameters: Conduct a weight analysis based on the multi-dimensional indicators of various types of hardware involved in the computing power resources, and divide the real-time high-performance computing power parameters and low-performance computing power parameters according to the influence weight ratio;

[0047] Step 3. AI prediction: Detect the load of the computing power resources, input features, and obtain the computing power demand interval according to the historical output. Detect the influence of the input features according to the historical load processing process, and analyze the influence of the real-time input features according to the influence detection result to conduct real-time evaluation of the computing power resources.

[0048] Please refer to Figure 2 As shown in the figure, further, the process of collecting and detecting multi-dimensional indicators is as follows:

[0049] The corresponding multi-dimensional indicators of various types of hardware involved in the computing power resources are respectively computing power, storage performance, and network communication; and the multi-dimensional indicators are quantified. Among them, the computing power is expressed as the ratio of the execution buffer time for the hardware to process data tasks to the error rate corresponding value in the current time for data task processing; the storage performance is expressed as the ratio of the actual occupancy rate of each storage space of the hardware to the corresponding value of the delay time for data calls in multiple storage spaces; the network communication is expressed as the ratio of the actual throughput of the network covered by the hardware to the actual network delay frequency; it should be noted that the ratio of values is the ratio of two collected data, without considering the dimension problem of the collected data, and only analyzing the floating influence of the data values;

[0050] Perform numerical comparison on the quantified multi-dimensional metrics, that is, compare the corresponding quantization parameters of the multi-dimensional metrics with the corresponding set parameter threshold ranges respectively: If the corresponding quantization parameter exceeds the set parameter threshold range, infer that the hardware of the current computing power resource has high-performance characteristics, and analyze the complexity of the actual data processing task. If the data processing task is highly complex, set the current computing power resource as the adaptable resource; conversely, if the data processing task is less complex, set the current computing power resource as the surplus resource;

[0051] If the corresponding quantization parameter is within the set parameter threshold range, infer that the computing power resource is a high cost-performance resource; conversely, if the corresponding quantization parameter is lower than the set parameter threshold range, infer that the computing power resource has low-performance characteristics and perform multi-dimensional metric regulation on the hardware of the computing power resource;

[0052] If high-performance characteristics are detected, the multi-dimensional metrics detection is qualified. If it is a surplus resource, re-plan the use of the computing power resource.

[0053] Furthermore, the historical operation detection process is as follows:

[0054] Obtain the historical operation period of the current computing power resource, and collect the quantization parameters of the multi-dimensional metrics during the historical operation period. Before the computing power resource executes the data processing task, set the quantization parameter at the current moment as the input feature, and mark the type of the data processing task as the task feature, where the task feature is represented by feature parameters such as task type, data processing volume, and data setting buffer duration;

[0055] When the task feature is constant, record the numerical floating moment of the input feature and the abnormal moment of the data processing task. If the floating moment and the abnormal moment are in the same task processing cycle, count the numerical floating of the quantization parameter of the input feature, and use the overlapping probability between the numerical floating moment and the task abnormal moment as the division standard of the influence level to set the influence on the feature vector with numerical floating, such as feature vectors such as numerical floating span, numerical floating duration, or numerical floating frequency;

[0056] If the overlapping probability exceeds the set probability threshold, set the feature vector with floating at the corresponding floating moment as the high-probability influence vector; conversely, if the overlapping probability does not exceed the set probability threshold, set the feature vector with floating at the corresponding floating moment as the low-probability influence vector; and perform real-time fluctuation analysis according to each type of feature vector and the corresponding vector type.

[0057] Furthermore, the real-time fluctuation analysis process is as follows:

[0058] Compare the task characteristics of the current task with those of the tasks in each historical period. Tasks with task characteristic parameters within the same range are regarded as tasks of the same type. Collect the feature vectors of the quantization parameters that fluctuate when tasks of the same type are executed, and identify the type of feature vectors according to the historical running periods of tasks of the same type. When it is a high-probability impact vector, issue an early warning for abnormal data processing tasks and use the abnormal types of historical data processing tasks as the current warning objects, and adjust the feature vectors according to the warning objects; when it is a low-probability impact vector, identify the floating trend of the feature vectors within the historical running period, and monitor the floating trend of the current feature vectors of the same type and adjust them in a timely manner when the trends are synchronized.

[0059] Further, the process of evaluating the computing power parameters in step two is as follows:

[0060] Analyze the multi-dimensional indicators of various types of hardware involved in the computing power resources, statistically analyze the numerical floating trends of the quantization parameters of the multi-dimensional indicators, and analyze the floating trends of the quantization parameters when the computing power resources execute data processing tasks. When the floating trends of the quantization parameters do not tend to be stable, that is, there are floating trends and reciprocating fluctuations, obtain the processing speed reduction frequency and the speed recovery buffer duration after the processing speed reduction of the data processing tasks in the current stage. If any value in the processing speed reduction frequency and the speed recovery buffer duration after the processing speed reduction exceeds the corresponding set threshold, it is inferred that the current quantization parameter is a high-risk parameter; otherwise, if there is no value in the processing speed reduction frequency and the speed recovery buffer duration after the processing speed reduction that exceeds the corresponding set threshold, it is inferred that the current quantization parameter is a low-risk parameter.

[0061] Set the impact weight of high-risk parameters to high impact weight, and the set weight factor is a low-value factor to avoid the impact of high-risk parameters being infinitely amplified compared to the impact of low-risk parameters, so as to ignore the impact of low-risk parameters. It should be noted that the high or low value of the weight factor is distinguished by comparing the low-value factor with the high-value factor; set the impact weight of high-risk parameters to low impact weight, and the set weight factor is a high-value factor.

[0062] Classify the quantization parameters of the multi-dimensional indicators, that is, uniformly mark the high impact weight and the corresponding weight factor as high-performance computing power parameters, and uniformly mark the low impact weight and the corresponding weight factor as low-performance computing power parameters.

[0063] Further, the AI prediction process is as follows:

[0064] Collect multi-dimensional metrics of the current computing power resources and corresponding quantization parameters, monitor the quantization parameters during the execution stage of real-time data processing tasks, set up a resource load analysis model, record the floating feature vectors of the quantization parameters during the execution stage, and mark them as model startup features. When the model startup features occur, evaluate the utilization rate of computing power resources according to the types of the model startup features, namely high-probability impact features and low-probability impact features;

[0065] Collect high-risk parameters and low-risk parameters in the quantization parameters and their corresponding weight factors, and mark the high-risk parameters and low-risk parameters as C 高 and C 低 ; the weight factors are Q 高 and Q 低 ; through the formula Obtain the impact coefficient S of the high-probability impact feature. It should be noted that the values of the high-risk parameters and low-risk parameters are the values of the corresponding feature vectors, while the weight factors are artificially set weight factors through continuous analysis of historical time periods. At the same time, when the units of the feature vectors are not unified, the weight factors can directly perform dimensionless processing;

[0066] If the impact coefficient S exceeds the coefficient threshold and continues to increase, it indicates that the corresponding utilization rate of the current multi-dimensional metrics of the computing power resources continues to decrease. Obtain the abnormal type of the data processing task based on the data during the execution process of the same type of task in the historical operation period. After identifying the abnormal type of the current processing task, obtain the feature vector regulation corresponding to the abnormal type of the data processing task, and mark the range of the feature vector regulation in the historical process as the computing power demand interval corresponding to the quantization parameters of the current computing power resources, and regulate the current multi-dimensional metrics according to the computing power demand interval; otherwise, if the impact coefficient S does not exceed the coefficient threshold and is constant, it indicates that the corresponding utilization rate of the current multi-dimensional metrics of the computing power resources is constant;

[0067] When the model startup feature is a high-probability impact feature, collect the maximum floating span of the high-risk parameter and the quantity of the parameter types where the low-risk parameter floats. If any parameter among the maximum floating span of the high-risk parameter and the quantity of the parameter types where the low-risk parameter floats increases, the computing power demand interval needs to be expanded according to the regulation trend; otherwise, if no parameter increases, regulate according to the current computing power demand interval;

[0068] When the model startup feature is a low-probability impact feature, the numerical increase span of the floating frequency of high-risk parameters and the cumulative span of continuous floating of low-risk parameters are collected. If the numerical increase span of the floating frequency of high-risk parameters and the cumulative span of continuous floating of low-risk parameters of any parameter exceed the set numerical threshold, the current computing power demand interval is taken as the regulation peak, and the computing power demand interval is regulated step by step in stages; if the numerical increase span of the floating frequency of high-risk parameters and the cumulative span of continuous floating of low-risk parameters of any parameter do not exceed the set numerical threshold, the current computing power demand interval is reduced, and the regulation span of the feature vector is reduced; it should be noted that the prerequisite for the regulation of the computing power demand interval is that the floating of the quantified parameter feature vector does not reach the value of the feature vector when the data processing task is abnormal.

[0069] When the present invention is in use, for the evaluation of the characteristics of computing power resources, multi-dimensional index collection and detection are carried out based on various types of hardware involved in the computing power resources; after the multi-dimensional index detection is completed, the historical operation process of the computing power resources is detected by AI technology; for the evaluation of computing power parameters, weight analysis is carried out according to the multi-dimensional indexes of various types of hardware involved in the computing power resources, and the real-time high-performance computing power parameters and low-performance computing power parameters are obtained by dividing according to the influence weight ratio; for AI prediction, the load of the computing power resources is detected, features are input, and the computing power demand interval is obtained according to the historical output for the real-time evaluation of the computing power resources.

[0070] The setting of the threshold or preset value, preset range, etc. is for result comparison and analysis to determine whether it is good or bad. Regarding the size value of it, it is set and stored by combining the large model analysis of sample data and manual experience, and can also be appropriately adjusted according to seasonal or regular influence conditions;

[0071] Regarding the setting of the weight ratio coefficient, influence factor, etc., specific values are assigned according to the influence degree of each parameter on the result, and finally the influence situation on the result is reflected. It is also set and stored by combining the large model analysis of sample data and manual experience, and can also be appropriately adjusted according to seasonal or regular influence conditions.

[0072] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the present invention to only the specific implementation manners. Obviously, according to the content of this specification, many modifications and changes can be made. These embodiments are selected and specifically described in this specification to better explain the principle and practical application of the present invention, so that those skilled in the art in the technical field can well understand and utilize the present invention. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. A high-performance computing power resource evaluation method based on AI technology, characterized in that, The evaluation method is as follows: Step 1: Evaluate the characteristics of computing power resources. Collect and detect multi-dimensional indicators based on various types of hardware involved in computing power resources. After completing the multi-dimensional indicator detection, use AI technology to detect the historical operation process of computing power resources, statistically analyze the floating of corresponding multi-dimensional indicators of various types of hardware at time nodes, infer the hardware adaptation efficiency based on the time node analysis, and predict real-time fluctuations according to the regulation of real-time multi-dimensional indicator parameters after obtaining the adaptation efficiency analysis results; Step 2: Evaluate computing power parameters. Conduct weight analysis based on the multi-dimensional indicators of various types of hardware involved in computing power resources, and divide to obtain real-time high-performance computing power parameters and low-performance computing power parameters according to the influence weight ratio. Step 3: AI prediction. Detect the load of computing power resources, input features, and obtain the computing power demand range based on historical outputs to conduct real-time evaluation of computing power resources.

2. The high-performance computing power resource evaluation method based on AI technology according to claim 1, wherein The process of multi-dimensional indicator collection and detection is as follows: The corresponding multi-dimensional indicators of various types of hardware involved in computing power resources are computing power, storage performance, and network communication respectively; and the multi-dimensional indicators are quantified. Among them, computing power is expressed as the ratio of the execution buffer time for the hardware to process data tasks to the data task processing error rate corresponding value in the current time; storage performance is expressed as the ratio of the actual occupancy rate of each storage space of the hardware to the corresponding value of the delay time for data calls in multiple storage spaces; network communication is expressed as the ratio of the actual throughput of the network covered by the hardware to the actual network delay frequency.

3. The high-performance computing power resource evaluation method based on AI technology according to claim 2, wherein Compare the numerical values of the quantified multi-dimensional indicators. If the corresponding quantified parameters exceed the set parameter threshold range, it is inferred that the hardware of the current computing power resources has high-performance characteristics, and analyze the complexity of the actual data processing task. If the data processing task complexity is high, set the current computing power resources as adapted resources; otherwise, if the data processing task complexity is low, set the current computing power resources as surplus resources; If the corresponding quantified parameters are within the set parameter threshold range, it is inferred that the computing power resources are high-cost-effective resources; otherwise, if the corresponding quantified parameters are lower than the set parameter threshold range, it is inferred that the computing power resources have low-performance characteristics, and conduct multi-dimensional indicator regulation of the hardware of the computing power resources.

4. The high-performance computing power resource evaluation method based on AI technology according to claim 3, wherein The process of historical operation detection is as follows: Obtain the historical operation period of the current computing power resources, and collect the quantified parameters of multi-dimensional indicators during the historical operation period. Before the computing power resources execute data processing tasks, set the quantified parameters at the current moment as input features, and mark the type of data processing task as task features; When the task features are constant, record the numerical floating moments of the input features and the abnormal moments of the data processing tasks. If the floating moment and the abnormal moment are in the same task processing cycle, statistically analyze the numerical floating of the quantified parameters of the input features, and use the overlapping probability of the numerical floating moment and the task abnormal moment as the division standard of the influence level to set the influence on the feature vector of the numerical floating. If the overlap probability exceeds the set probability threshold, the eigenvector that fluctuates at the corresponding floating moment is set as a high-probability influence vector; conversely, if the overlap probability does not exceed the set probability threshold, the eigenvector that fluctuates at the corresponding floating moment is set as a low-probability influence vector; and real-time fluctuation analysis is performed according to each type of eigenvector and the corresponding vector type.

5. The high-performance computing power resource evaluation method based on AI technology according to claim 4, wherein The process of real-time fluctuation analysis is as follows: Compare the task characteristics of the current task with those of the tasks in each historical period. The task characteristics with task characteristic parameters in the same range are regarded as the same type of tasks. Collect the eigenvectors of the quantization parameter characteristics that fluctuate when the same type of tasks are executed, and identify the eigenvector type according to the historical operation period of the same type of tasks. When it is a high-probability influence vector, an early warning of abnormal data processing tasks is given, and the abnormal type of historical data processing tasks is used as the current warning object, and eigenvector regulation is performed according to the warning object. When it is a low-probability influence vector, identify the floating trend of the eigenvector within the historical operation period, and perform timely regulation when the floating trend of the current same-type eigenvector is monitored and the trends are synchronized.

6. The high-performance computing power resource evaluation method based on AI technology according to claim 5, characterized in that The process of step two for evaluating computing power parameters is as follows: Statistically analyze the numerical floating trends of the quantization parameters of multi-dimensional indicators, and analyze the floating trends of the quantization parameters. In the stage where the floating trends of the quantization parameters do not tend to be stable, obtain the reduction frequency of the processing speed of the data processing tasks in the current stage and the speed recovery buffer duration after the processing speed drops. If there is any data with a value exceeding the corresponding set threshold, it is inferred that the current quantization parameter is a high-risk parameter; conversely, if there is no data with a value exceeding the corresponding set threshold, it is inferred that the current quantization parameter is a low-risk parameter. Set the influence weight of the high-risk parameter as a high influence weight, and the set weight factor as a low numerical factor; set the influence weight of the high-risk parameter as a low influence weight, and the set weight factor as a high numerical factor; classify the quantization parameters of multi-dimensional indicators, that is, uniformly mark the high influence weight and the corresponding weight factor as high-performance computing power parameters, and uniformly mark the low influence weight and the corresponding weight factor as low-performance computing power parameters.

7. The high-performance computing power resource evaluation method based on AI technology according to claim 6, wherein, The AI prediction process is as follows: Collect the multi-dimensional indicators and corresponding quantization parameters of the current computing power resources, set a resource load analysis model, record the floating eigenvectors of the quantization parameters during the execution stage, and mark them as model startup features. When the model startup features are generated, evaluate the utilization rate of the computing power resources according to the type of the model startup features. Collect high-risk parameters, low-risk parameters and corresponding weight factors in the quantization parameters, and mark the high-risk parameters and low-risk parameters as C 高 and C 低 ; The weight factors are Q 高 and Q 低 ; through the formula the influence coefficient S of the high-probability influence feature is obtained; If the influence coefficient S exceeds the coefficient threshold and continues to increase, it indicates that the utilization rate of the current multi-dimensional indicators of the computing power resources continues to decrease. Obtain the abnormal type of data processing tasks according to the execution process of the same type of tasks in the historical operation period. After identifying the current abnormal type of processing tasks, obtain the eigenvector regulation corresponding to the abnormal type of data processing tasks, and mark the eigenvector regulation range in the historical process as the computing power demand interval corresponding to the quantization parameters of the current computing power resources, and regulate the current multi-dimensional indicators according to the computing power demand interval. Conversely, if the influence coefficient S does not exceed the coefficient threshold and is constant, it indicates that the utilization rate of the current multi-dimensional indicators of the computing power resources is constant.

8. The high-performance computing power resource evaluation method based on AI technology according to claim 7, wherein When the model startup feature is a high-probability influencing feature, the maximum floating span of high-risk parameters and the quantity of parameter types with floating low-risk parameters are collected. If there is an increase in any parameter among the maximum floating span of high-risk parameters and the quantity of parameter types with floating low-risk parameters, the computing power demand interval needs to be expanded according to the regulation trend; conversely, if no parameter shows an increase, the current computing power demand interval is used for regulation.

9. The high-performance computing power resource evaluation method based on AI technology according to claim 8, characterized in that When the model startup feature is a low-probability influencing feature, the numerical increase span of the floating frequency of high-risk parameters and the cumulative continuous floating span of low-risk parameters are collected. If any parameter among the numerical increase span of the floating frequency of high-risk parameters and the cumulative continuous floating span of low-risk parameters exceeds the set numerical threshold, the current computing power demand interval is used as the regulation peak, and the computing power demand interval is regulated step by step in stages; if any parameter among the numerical increase span of the floating frequency of high-risk parameters and the cumulative continuous floating span of low-risk parameters does not exceed the set numerical threshold, the current computing power demand interval is narrowed to reduce the regulation span of the feature vector.

Images