A computing power token metering method and system

Abstract

The application discloses a kind of computing power Token measurement method and system, belong to computing power resource measurement technical field, method includes: constructing multidimensional computing power attribute vector, including six dimensions of attribute parameters of core computing capacity, text length, hardware performance, task adaptation degree, computing power electric power synergistic efficiency and running stability;According to the quantization index of each attribute parameter, each attribute parameter is standardized collected, and the quantization value of each attribute parameter is obtained;According to the quantization value of core computing capacity, the computing power Token basic value is calculated;According to the quantization value of text length, hardware performance, task adaptation degree, computing power electric power synergistic efficiency and running stability, the comprehensive correction coefficient is calculated;According to computing power Token basic value and comprehensive correction coefficient, generate computing power Token final value, and the comprehensive computing power service capability of computing power node is quantified by computing power Token final value.The method can realize the unified quantization representation of heterogeneous computing power multidimension.

Description

Technical Field

[0001] This invention relates to a computing power token measurement method and system, belonging to the field of computing power resource measurement technology. Background Technology

[0002] With the rapid development of the digital economy and artificial intelligence industry, computing power has become a core production factor. The large-scale deployment and cross-scenario application of heterogeneous computing power such as CPUs, GPUs, and NPUs are becoming increasingly widespread, which places higher demands on the accurate measurement and unified quantitative representation of computing power resources.

[0003] Current computing power measurement is based on a single dimension. Most existing technologies only measure a few hardware indicators such as computing power and memory bandwidth, without comprehensively considering key factors such as different task processing characteristics, power utilization efficiency, and long-term operating status. As a result, the measurement results deviate significantly from the actual application value of computing power and cannot truly reflect the comprehensive service capabilities of computing power nodes.

[0004] Currently, heterogeneous computing power lacks a unified quantitative representation carrier. The measurement results of computing power nodes of different types and specifications do not have a unified and comparable presentation format. In the current heterogeneous computing environment, there are many types of computing power nodes such as CPU, GPU, FPGA, and ASIC. Different computing power chips, server forms, storage forms, and network forms can be combined and configured to form hundreds or thousands of combinations. This situation makes it difficult to achieve accurate quantitative matching when scheduling computing power resources across platforms and integrating them across scenarios. The parameter acquisition methods of different types of computing power hardware are very different, and the accuracy and consistency of data are difficult to guarantee, which further restricts the uniformity of computing power quantitative representation.

[0005] The existing computing power measurement has not established a scientific multi-dimensional correction mechanism, and has not made differentiated corrections to the basic measurement results based on the actual attribute characteristics of computing power nodes, resulting in insufficient measurement accuracy. This easily leads to a mismatch between the allocation of computing power resources and actual needs, which in turn causes a waste of computing power resources or an insufficient supply of computing power.

[0006] Existing computing power measurement methods mostly focus on macroscopic measurement of the overall computing power of intelligent computing centers, lacking refined measurement schemes for individual computing nodes. This fails to meet the actual needs of granular scheduling and refined management of computing resources. Some solutions focus on constructing computing power measurement units through fixed hardware indicators, while others emphasize the optimization of computing resource scheduling. However, none of these solutions construct a multi-dimensional measurement system covering computing hardware performance, task processing capabilities, and operational status, nor do they design standardized computing power representation tokens and corresponding collection and correction logic.

[0007] Therefore, there is an urgent need for a computing power measurement technology solution that can achieve standardized collection and accurate correction of multi-dimensional attributes of heterogeneous computing power, and complete unified quantitative presentation through computing power tokens, in order to solve the above-mentioned technical pain points. Summary of the Invention

[0008] The purpose of this invention is to provide a computing power token measurement method and system that can achieve multi-dimensional unified quantitative representation of heterogeneous computing power.

[0009] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for measuring computing power tokens, comprising: Construct a multi-dimensional computing power attribute vector, which includes attribute parameters in six dimensions: core computing power, text length, hardware performance, task adaptability, computing power-electricity coordination efficiency, and operational stability. Based on the quantitative indicators of each attribute parameter, the attribute parameters are collected in a standardized manner to obtain the quantitative values ​​of each attribute parameter. Calculate the base value of the computing power token based on the quantified value of the core computing power; A comprehensive correction coefficient is calculated based on the quantitative values ​​of text length, hardware performance, task adaptability, computing power-electricity synergy efficiency, and operational stability. Based on the base value of the computing power token and the comprehensive correction coefficient, the final value of the computing power token is generated, and the comprehensive computing power service capability of the computing power node is quantified by the final value of the computing power token.

[0010] In conjunction with the first aspect, the quantized values ​​of each attribute parameter are periodically monitored. When the fluctuation of the quantized value of any attribute parameter exceeds the fluctuation threshold, the final value of the computing power token is updated.

[0011] In conjunction with the first aspect, further, core computing power is quantified by the number of floating-point operations per second, text length by the total number of characters input and output in the task, hardware performance by the bandwidth of video memory or memory read and write, task adaptability by the task processing latency, computing power-power synergy efficiency by the computing power provided per unit power consumption under full load operation, and operational stability by the operational status score within a preset time period; among which, the operational status score is determined based on fault-free operation time, computing power fluctuation range, or task interruption rate.

[0012] In conjunction with the first aspect, the formula for calculating the basic value of the computing power token is as follows: ; in, This represents the base value of the computing power token. A quantified value representing core computing power. This represents the basic unit of measurement for computing power tokens. This indicates the cumulative time taken for a computing node to execute computing tasks.

[0013] In conjunction with the first aspect, further, based on the quantified values ​​of text length, hardware performance, task adaptability, computing power-electricity synergy efficiency, and operational stability, a comprehensive correction coefficient is calculated, including: Based on the type of quantification index, the quantification values ​​of text length, hardware performance, task adaptability, computing power-electricity synergy efficiency, and operational stability are positively or negatively processed respectively, and the processed quantification values ​​are mapped to a unified dimensionless interval to obtain normalized values ​​of text length, hardware performance, task adaptability, computing power-electricity synergy efficiency, and operational stability. The comprehensive correction coefficient is calculated based on the normalized values ​​of text length, hardware performance, task adaptability, computing power-electricity synergy efficiency, and operational stability, as well as their corresponding weights. The formula for calculating the overall correction factor is: ; in, This represents the overall correction factor. and All are fitting constants. , , , , These represent the normalized values ​​for text length, hardware performance, task adaptability, computing power-electricity coordination efficiency, and operational stability, respectively. , , , , These represent the weights corresponding to text length, hardware performance, task adaptability, computing power-electricity coordination efficiency, and operational stability, respectively.

[0014] In conjunction with the first aspect, furthermore, the quantitative indicators of text length, hardware performance, computing power-electricity synergy efficiency, and operational stability are positive indicators, and the quantitative values ​​of text length, hardware performance, computing power-electricity synergy efficiency, and operational stability are positively processed; the quantitative indicator of task adaptability is a negative indicator, and the quantitative value of task adaptability is negatively processed.

[0015] In conjunction with the first aspect, the final formula for calculating the computing power token value is as follows: ; in, This represents the final value of the computing power token. This represents the base value of the computing power token. This represents the overall correction factor.

[0016] Secondly, the present invention provides a computing power token measurement system, comprising: The multi-dimensional computing power attribute acquisition module is used to construct a multi-dimensional computing power attribute vector, which includes attribute parameters in six dimensions: core computing power, text length, hardware performance, task adaptability, computing power-electricity coordination efficiency, and operational stability. Based on the quantitative indicators of each attribute parameter, the module performs standardized acquisition of each attribute parameter to obtain the quantitative value of each attribute parameter. The comprehensive correction coefficient calculation module is used to calculate the basic value of computing power token based on the quantified value of core computing capabilities; and to calculate the comprehensive correction coefficient based on the quantified values ​​of text length, hardware performance, task adaptability, computing power-electricity coordination efficiency, and operational stability. The computing power token measurement module is used to generate the final value of the computing power token based on the base value of the computing power token and the comprehensive correction coefficient. The final value of the computing power token is used to quantify the comprehensive computing power service capability of the computing power node.

[0017] In conjunction with the second aspect, further details include: The dynamic monitoring and update module is used to periodically monitor the quantized values ​​of each attribute parameter. When the fluctuation of the quantized value of any attribute parameter exceeds the fluctuation threshold, the final value of the computing power token is updated.

[0018] Thirdly, the present invention provides a computer device, comprising: Storage medium: used to store computer programs; Processor: Used to execute the computer program to implement the computing power token measurement method described in the first aspect.

[0019] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the computing power token measurement method described in the first aspect.

[0020] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the computing power token measurement method described in the first aspect.

[0021] Compared with the prior art, the beneficial effects of the present invention are: The computing power token measurement method provided by this invention solves the problems of existing computing power measurement methods, such as single dimension, lack of unified quantitative representation of heterogeneous computing power, and insufficient measurement accuracy, by constructing a multi-dimensional attribute vector and a dynamic correction mechanism. The six-dimensional attributes cover core computing power, text length, hardware performance, task adaptability, computing power-power coordination efficiency, and operational stability, making the measurement results closer to the real service value of computing power nodes in actual application scenarios, and improving the accuracy of computing power scheduling and the utilization efficiency of computing power resources. The final value of the computing power token is updated in real time through a dynamic monitoring mechanism to ensure that the result is synchronized with the actual state of the computing power node. Attached Figure Description

[0022] Figure 1 This is a flowchart of the computing power token measurement method provided in the embodiments of the present invention. Detailed Implementation

[0023] The technical solution of the present invention will be further described in detail below with reference to specific embodiments.

[0024] Embodiments of the present invention are described in detail below. Examples of these embodiments are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. Unless otherwise specified, embodiments of the present invention and the technical features thereof can be combined with each other.

[0025] This invention provides a method for measuring computing power tokens, including: Construct a multi-dimensional computing power attribute vector, which includes attribute parameters in six dimensions: core computing power, text length, hardware performance, task adaptability, computing power-electricity coordination efficiency, and operational stability. Based on the quantitative indicators of each attribute parameter, the attribute parameters are collected in a standardized manner to obtain the quantitative values ​​of each attribute parameter. Calculate the base value of the computing power token based on the quantified value of the core computing power; A comprehensive correction coefficient is calculated based on the quantitative values ​​of text length, hardware performance, task adaptability, computing power-electricity synergy efficiency, and operational stability. Based on the base value of the computing power token and the comprehensive correction coefficient, the final value of the computing power token is generated, and the comprehensive computing power service capability of the computing power node is quantified by the final value of the computing power token.

[0026] The computing power token measurement method provided in this invention defines computing power tokens as standardized quantification carriers, mapping heterogeneous computing power nodes of different types, such as CPUs, GPUs, and NPUs, to the same measurement framework. This solves the problem of inconsistent comparison of heterogeneous computing power in existing technologies, providing a unified measurement benchmark for cross-platform computing power scheduling and cross-scenario resource integration. It overcomes the limitations of existing technologies that rely solely on a few hardware indicators, comprehensively considering six dimensions: core computing power, text length, hardware performance, task adaptability, computing power-power coordination efficiency, and operational stability. This makes the measurement results closer to the real service value of computing power nodes in actual application scenarios. Furthermore, combined with refined parameter standardization collection rules, it effectively avoids biases caused by single-indicator measurement. It is applicable to various computing power resource scheduling and management scenarios, such as integrated computing power networks, large model training and inference, computing power-power coordination, overseas computing power deployment, intelligent agents, AI factories, and cloud computing. It provides a unified quantitative basis for efficient allocation of computing power resources and computing power cost assessment, achieving multi-dimensional unified quantitative representation of heterogeneous computing power.

[0027] In this embodiment of the invention, the computing power token serves as the basic unit for measuring heterogeneous computing power resources. Its function is to comprehensively quantify six key dimensions of a single computing power node within a unit of time: core computing power, text length, hardware performance, task adaptability, computing power-power synergy efficiency, and operational stability. This enables the construction of a comprehensive computing power service capability with schedulable, billable, and evaluable characteristics. This unit serves as the core carrier for computing power resource pooling, dynamic scheduling, and market-oriented operation. The computing power token does not have a fixed numerical equivalent; its measurement value is derived from the comprehensive calculation of multi-dimensional attribute parameters of the computing power node, and the measurement result directly reflects the actual application value of the computing power node. Various heterogeneous computing power nodes can achieve unified quantitative representation and horizontal comparability through the computing power token. The computing power service capability represented by a single computing power token constitutes the smallest granularity of computing power measurement, supporting granular scheduling and configuration of computing power resources based on the computing power token.

[0028] Figure 1 This is a flowchart of the computing power token measurement method provided in this embodiment. This flowchart only shows the logical order of the method in this embodiment. Under the premise of no conflict, different methods can be used. Figure 1 Complete the steps shown or described in the order indicated.

[0029] The computing power token measurement method provided in this embodiment can be applied to a terminal and can be executed by a computing power token measurement system. This system can be implemented by software and / or hardware and can be integrated into the terminal, such as any tablet computer or computer device with communication function.

[0030] In one possible embodiment, the computing power token measurement method specifically includes the following steps: Step 1: Construct a multi-dimensional computing power attribute vector; In this embodiment, the multi-dimensional computing power attribute vector includes attribute parameters in six dimensions: core computing power, text length, hardware performance, task adaptability, computing power-power coordination efficiency, and operational stability.

[0031] Specifically, constructing multi-dimensional computing power attribute vectors ,in, This represents core computing power and can be quantified by metrics such as floating-point operations per second (Floats per Second). It is a core indicator for measuring the performance of computing devices. Text length can be quantified using metrics such as the total number of characters in the task's input and output. Hardware performance can be quantified by metrics such as video memory or system memory read / write bandwidth. It represents the amount of data a computing device can transfer per unit of time. High bandwidth can reduce data transmission bottlenecks and improve the overall performance of the computing device. Task adaptability can be quantified using metrics such as typical task processing latency, which is the time required from input data to output results. Low latency is particularly critical for real-time computing scenarios. The efficiency of computing power-electricity synergy can be quantified by indicators such as performance per watt, which represents the computing power provided for every watt of energy consumed. This quantification establishes the linkage logic between computing power and electricity, forming a two-way collaborative mechanism where computing is influenced by electricity and electricity follows computing. This indicates operational stability, corresponding to the sustained computing performance in computing power assessment, and reflects the actual computing power of computing equipment during long-term stable operation.

[0032] Relevant computing power metrics are often measured in petaflops per second (PFLOPS). All parameters must be acquired according to unified and refined data collection rules. At the same time, data collection should be standardized by combining a complete range of standard computing power measurement units from low to high, such as FLOPS (one PFLOPS), KFLOPS (one thousand PFLOPS), MFLOPS (one million PFLOPS), GFLOPS (one billion PFLOPS), TFLOPS (one trillion PFLOPS), PFLOPS (one quadrillion PFLOPS), and EFLOPS (one hundred quadrillion PFLOPS), to ensure the accuracy and consistency of the data.

[0033] Step 2: Based on the quantitative indicators of each attribute parameter, standardize the data collection of each attribute parameter and obtain the quantitative value of each attribute parameter; In this embodiment, core computing power is quantified by the number of floating-point operations per second, text length is quantified by the total number of characters input and output in the task, hardware performance is quantified by the bandwidth of video memory or memory read and write, task adaptability is quantified by the task processing latency, computing power-power coordination efficiency is quantified by the computing power provided per unit power consumption under full load operation, and operational stability is quantified by the operational status score within a preset time. Among them, the operational status score is determined based on at least one of the following: fault-free runtime of the computing device, computing power fluctuation range, or task interruption rate.

[0034] Specifically, for core computing power, the number of floating-point operations per second (Floating-Point Operations per Second) needs to be collected by closing unnecessary processes, locking the hardware operating frequency, and controlling the ambient temperature between 18°C ​​and 25°C. A full-load floating-point operation task should be deployed using an adaptation tool, and data should be collected continuously for 10 minutes at a frequency of once per second. Outliers and extreme values ​​should be filtered out, and the average value should be taken, with a collection error not exceeding 3%. For text length, a standardized unit of measurement is used to characterize the length of the Token text sequence and to measure the processing space and transmission volume occupied by the Token text. Text length uses characters as the basic counting unit, including Chinese characters, English letters, Arabic numerals, and common punctuation marks. A single Chinese character, a single letter, a single number, and a single punctuation mark are all counted as one character. The length value is the total number of characters contained in the text. In general scenarios, the text length corresponding to a Token is limited to a range of 32 to 512 characters to achieve standardized text measurement and unified computing power scheduling adaptation. In specific cases, such as for local computing power production, the text length corresponding to a Token is limited to a range of no more than 2 characters to adapt to the lightweight billing and localized data processing needs of local computing power. For hardware performance, the acquisition of video memory or system memory read / write bandwidth requires clearing the cache, executing pure read / write tasks, with the data block size being twice the hardware cache size, and taking the average of the 95th percentile values ​​of the read / write bandwidth after 5 minutes of continuous acquisition. The coefficient of variation of the three acquisitions from the same node should not exceed 3%. For task adaptability, the acquisition of typical task processing latency requires selecting a standardized benchmark task according to the scenario, controlling the node load rate to 70% to 80%, conducting at least 1000 continuous tests, removing abnormal data, and taking the average value. The scenario transition error should not exceed 5%. For computing power-power coordination efficiency, the acquisition of computing power provided per unit power consumption under full load operation should use a power consumption detection device with an accuracy level of not less than 0.5, synchronously acquiring computing power and power consumption, and taking the average value after filtering out abnormal values ​​after 30 minutes of continuous acquisition. The data deviation between the hardware interface and the detection device should not exceed 5%. Regarding operational stability, the collection of operational status scores requires at least 72 hours of full-load testing. Scoring is conducted based on three dimensions: fault-free runtime, computing power fluctuation range, and task interruption rate. The scoring criteria are as follows: fault-free runtime scores 40 points if there are no faults; computing power fluctuation range scores 30 points if the coefficient of variation of computing power fluctuation does not exceed 3%; and task interruption rate scores 30 points if the task interruption rate is 0%. Points are deducted accordingly based on the actual faults, fluctuations, and interruptions.

[0035] Step 3: Calculate the base value of the computing power token based on the quantified value of the core computing power; Specifically, the formula for calculating the basic value of computing power tokens is as follows: ; in, This represents the base value of the computing power token. A quantified value representing core computing power. This represents the basic unit of measurement for computing power tokens, specifically set to one trillion half-precision floating-point operations per second. This indicates the cumulative time taken for a computing node to execute computing tasks.

[0036] Step 4: Calculate the comprehensive correction coefficient based on the quantitative values ​​of text length, hardware performance, task adaptability, computing power-electricity coordination efficiency, and operational stability; In this embodiment, based on the analytic hierarchy process, the overall goal is decomposed into six criterion-level dimensions: core computing power, text length, video memory or memory read / write bandwidth, typical task processing latency, computing power-electricity synergy efficiency, and long-term operational stability. The weight of each dimension is determined by pairwise comparisons, and then the parameters of each dimension are positively or negatively processed and normalized to calculate the comprehensive correction coefficient.

[0037] The comprehensive correction coefficient is calculated based on the quantified values ​​of text length, hardware performance, task adaptability, computing power-electricity coordination efficiency, and operational stability, specifically including the following steps: Step ①: Based on the type of quantification index, perform positive or negative processing on the quantification values ​​of text length, hardware performance, task adaptability, computing power-electricity synergy efficiency, and operational stability respectively, and map the processed quantification values ​​to a unified dimensionless interval to obtain the normalized values ​​of text length, hardware performance, task adaptability, computing power-electricity synergy efficiency, and operational stability. Specifically, the quantitative indicators for text length, hardware performance, computing power-electricity synergy efficiency, and operational stability are positive indicators, and the quantitative values ​​of text length, hardware performance, computing power-electricity synergy efficiency, and operational stability are positively processed; the quantitative indicator for task adaptability is a negative indicator, and the quantitative value of task adaptability is negatively processed.

[0038] Taking hardware performance as an example, positive processing means dividing the difference between the actual parameter value and the minimum value of that dimension by the difference between the maximum value and the minimum value of that dimension. Taking task adaptability as an example, negative processing means dividing the difference between the maximum value of that dimension and the actual parameter value by the difference between the maximum value and the minimum value of that dimension.

[0039] Step 2: Calculate the comprehensive correction coefficient based on the normalized values ​​and corresponding weights of text length, hardware performance, task adaptability, computing power-electricity coordination efficiency, and operational stability.

[0040] Specifically, the formula for calculating the comprehensive correction factor is as follows: ; in, This represents the overall correction factor. , and All are fitting constants. , , , , , , These represent the normalized values ​​for text length, hardware performance, task adaptability, computing power-electricity coordination efficiency, and operational stability, respectively. The value range is set to 32 characters to 512 characters. The value range is set to 10GB / s to 100GB / s. The value range is set to 1ms to 200ms. The value range is set to 2 TFLOPS / kW to 50 TFLOPS / kW. The value range is set to 0 to 100 points. , , , , These represent the weights corresponding to text length, hardware performance, task adaptability, computing power-electricity coordination efficiency, and operational stability, respectively. , , , , .

[0041] Step 5: Generate the final value of the computing power token based on the base value of the computing power token and the comprehensive correction coefficient. Quantify the comprehensive computing power service capability of the computing power node through the final value of the computing power token.

[0042] Specifically, the formula for calculating the final value of the computing power token is as follows: ; in, This represents the final value of the computing power token. This represents the base value of the computing power token. This represents the overall correction factor.

[0043] The final value of the computing power token is retained to two decimal places, and the number of tokens generated by a single node in a single transaction is no less than 0.01.

[0044] In one possible embodiment, the computing power token measurement method further includes the following steps: Step 6: Periodically monitor the quantized values ​​of each attribute parameter. When the fluctuation of the quantized value of any attribute parameter exceeds the fluctuation threshold, update the final value of the computing power token.

[0045] Specifically, the update cycle is set to 1 hour / time in general scenarios and 10 minutes / time in high-frequency scenarios. The fluctuation threshold corresponding to the quantified value of core computing power is 10%, the fluctuation threshold corresponding to the quantified value of text length is 10%, the fluctuation threshold corresponding to the quantified value of hardware performance is 15%, the fluctuation threshold corresponding to the quantified value of task adaptability is 20%, the fluctuation threshold corresponding to the quantified value of computing power-power coordination efficiency is 10%, and the fluctuation threshold corresponding to the quantified value of operational stability is a decrease of 15 points. The entire process records the basic value of computing power token, the comprehensive correction coefficient, the final value of computing power token, and the collected environmental parameters, including temperature, voltage, and load, and the storage period is no less than 1 year.

[0046] The computing power token measurement method provided in this invention includes a recording and traceability mechanism for the entire measurement process, ensuring that the token generation process is verifiable and traceable. Combined with subsequent data encryption and access control, it effectively prevents the risk of data tampering, and the measurement results have high credibility and auditability.

[0047] This invention provides a computing power token measurement system, comprising: The multi-dimensional computing power attribute acquisition module is used to construct a multi-dimensional computing power attribute vector, which includes attribute parameters in six dimensions: core computing power, text length, hardware performance, task adaptability, computing power-electricity coordination efficiency, and operational stability. Based on the quantitative indicators of each attribute parameter, the module performs standardized acquisition of each attribute parameter to obtain the quantitative value of each attribute parameter. The comprehensive correction coefficient calculation module is used to calculate the basic value of computing power token based on the quantified value of core computing capabilities; and to calculate the comprehensive correction coefficient based on the quantified values ​​of text length, hardware performance, task adaptability, computing power-electricity coordination efficiency, and operational stability. The computing power token measurement module is used to generate the final value of the computing power token based on the base value of the computing power token and the comprehensive correction coefficient. The final value of the computing power token is used to quantify the comprehensive computing power service capability of the computing power node.

[0048] In one possible embodiment, the computing power token metering system further includes: The dynamic monitoring and update module is used to periodically monitor the quantized values ​​of each attribute parameter. When the fluctuation of the quantized value of any attribute parameter exceeds the fluctuation threshold, the final value of the computing power token is updated.

[0049] This invention provides a computing power token measurement system, specifically including: The hardware layer provides heterogeneous computing nodes and a hardware foundation for the standardized collection of attribute parameters. The data layer is used to store the collected quantized values ​​of the attribute parameters, intermediate data during the calculation process, and the final value of the generated computing power token. The business layer is used to perform standardized collection of attribute parameters, calculation of the basic value and comprehensive correction coefficient of computing power token, generation of the final value of computing power token, and dynamic monitoring of the entire process. The application layer is used to provide user interfaces and / or standardized interfaces to present and output measurement results to the outside world.

[0050] The computing power token metering system provided in this embodiment of the invention adopts a layered architecture to achieve fully automated metering with a metering error of ≤2.5%, and is suitable for various computing power resource scheduling and management scenarios.

[0051] In this embodiment, the business layer specifically includes: The multi-dimensional attribute acquisition module is used to collect and verify attribute parameters in parallel across various dimensions. The computing power token base value measurement module is used to calculate the computing power token base value based on the collected quantified value of the core computing power. The multi-dimensional correction coefficient calculation module is used to calculate the comprehensive correction coefficient based on the quantized values ​​of other dimension attribute parameters and preset weights. The final value measurement module for computing power tokens is used to combine the basic value of computing power tokens with the comprehensive correction coefficient to generate the final value of computing power tokens. The dynamic monitoring and update module is used to monitor the parameter status of computing power nodes and trigger the update process of the final value of computing power token according to preset rules.

[0052] Specifically, the standardized interfaces provided by the application layer include a descriptive state transfer interface and / or a remote procedure call interface, which are used to interact with external computing power scheduling platforms.

[0053] The data layer adopts a hybrid storage architecture, including a first database for storing real-time dynamic data and a second database for storing historical traceability data.

[0054] The multi-dimensional attribute acquisition module supports parallel acquisition from no less than 50 nodes, and the dynamic monitoring and update module has an abnormal alarm response time of no more than 30 seconds. The data layer is designed with local caching and data cleaning modules to ensure data security and accuracy.

[0055] The computing power token metering system provided in this embodiment of the invention controls the computing power token generation error to within 2.0%, which can meet the needs of fine-grained metering for large-scale computing power clusters. A single cluster can support parallel collection by no less than 100 nodes, with a collection response time of no more than 3 minutes and a collection accuracy of no less than 99%. The response time for multi-scenario weight switching is no more than 5 minutes, and the token value is updated synchronously with the actual status of the computing power nodes. The computing power resource utilization rate has been improved from 85% before optimization to over 96%, reducing resource waste. In comparison, the GPU computing power utilization rate of a typical 1000-card GPU cluster is about 70%, and that of a 10,000-card GPU cluster is about 50%. The computing power utilization rate optimized in this embodiment is at a higher level.

[0056] The computing power token measurement system provided in this embodiment of the invention can execute the computing power token measurement method provided in this embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0057] This invention provides a computer device, comprising: Storage medium: used to store computer programs; Processor: Used to execute computer programs to implement the computing power token measurement method provided in the embodiments of the present invention.

[0058] This invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the computing power token measurement method provided in this invention.

[0059] This invention provides a computer program product, including a computer program that, when executed by a processor, implements the computing power token measurement method provided in this invention.

[0060] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0061] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0062] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including an instruction set implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0063] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0064] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for measuring computing power tokens, characterized in that, include: Construct a multi-dimensional computing power attribute vector, which includes attribute parameters in six dimensions: core computing power, text length, hardware performance, task adaptability, computing power-electricity coordination efficiency, and operational stability. Based on the quantitative indicators of each attribute parameter, the attribute parameters are collected in a standardized manner to obtain the quantitative values ​​of each attribute parameter. Calculate the base value of the computing power token based on the quantified value of the core computing power; A comprehensive correction coefficient is calculated based on the quantitative values ​​of text length, hardware performance, task adaptability, computing power-electricity synergy efficiency, and operational stability. Based on the base value of the computing power token and the comprehensive correction coefficient, the final value of the computing power token is calculated, and the comprehensive computing power service capability of the computing power node is quantified by the final value of the computing power token.

2. The computing power token measurement method according to claim 1, characterized in that, The quantized values ​​of each attribute parameter are periodically monitored. When the fluctuation of the quantized value of any attribute parameter exceeds the fluctuation threshold, the final value of the computing power token is updated.

3. The computing power token measurement method according to claim 1, characterized in that, Core computing power is quantified by the number of floating-point operations per second; text length is quantified by the total number of characters input and output in the task; hardware performance is quantified by the bandwidth of video memory or memory read and write; task adaptability is quantified by the task processing latency; computing power-power coordination efficiency is quantified by the computing power provided per unit power consumption under full load operation; and operational stability is quantified by the operational status score within a preset time. Among these, the operational status score is determined based on the fault-free operation time, the fluctuation range of computing power, or the task interruption rate.

4. The computing power token measurement method according to claim 1, characterized in that, The formula for calculating the basic value of computing power tokens is: ； in, This represents the base value of the computing power token. A quantified value representing core computing power. This represents the basic unit of measurement for computing power tokens. This indicates the cumulative time taken for a computing node to execute computing tasks.

5. The computing power token measurement method according to claim 1, characterized in that, Based on the quantified values ​​of text length, hardware performance, task adaptability, computing power-electricity synergy efficiency, and operational stability, the comprehensive correction coefficient is calculated as follows: Based on the type of quantification index, the quantification values ​​of text length, hardware performance, task adaptability, computing power-electricity synergy efficiency, and operational stability are positively or negatively processed respectively, and the processed quantification values ​​are mapped to a unified dimensionless interval to obtain normalized values ​​of text length, hardware performance, task adaptability, computing power-electricity synergy efficiency, and operational stability. The comprehensive correction coefficient is calculated based on the normalized values ​​of text length, hardware performance, task adaptability, computing power-electricity synergy efficiency, and operational stability, as well as their corresponding weights. The formula for calculating the overall correction factor is: ； in, This represents the overall correction factor. and All are fitting constants. , , , , These represent the normalized values ​​for text length, hardware performance, task adaptability, computing power-electricity coordination efficiency, and operational stability, respectively. , , , , These represent the weights corresponding to text length, hardware performance, task adaptability, computing power-electricity coordination efficiency, and operational stability, respectively.

6. The computing power token measurement method according to claim 5, characterized in that, The quantitative indicators for text length, hardware performance, computing power-electricity synergy efficiency, and operational stability are positive indicators, and the quantitative values ​​of text length, hardware performance, computing power-electricity synergy efficiency, and operational stability are positively processed; the quantitative indicator for task adaptability is a negative indicator, and the quantitative value of task adaptability is negatively processed.

7. The computing power token measurement method according to claim 1, characterized in that, The formula for calculating the final value of computing power tokens is: ； in, This represents the final value of the computing power token. This represents the base value of the computing power token. This represents the overall correction factor.

8. A computing power token measurement system, characterized in that, include: The multi-dimensional computing power attribute acquisition module is used to construct a multi-dimensional computing power attribute vector, which includes attribute parameters in six dimensions: core computing power, text length, hardware performance, task adaptability, computing power-electricity coordination efficiency, and operational stability. Based on the quantitative indicators of each attribute parameter, the module performs standardized acquisition of each attribute parameter to obtain the quantitative value of each attribute parameter. The comprehensive correction coefficient calculation module is used to calculate the basic value of computing power token based on the quantified value of core computing capabilities; and to calculate the comprehensive correction coefficient based on the quantified values ​​of text length, hardware performance, task adaptability, computing power-electricity coordination efficiency, and operational stability. The computing power token metering module is used to calculate the final value of the computing power token based on the base value and the comprehensive correction coefficient, and to quantify the comprehensive computing power service capability of the computing power node through the final value of the computing power token.

9. A computer device, characterized in that, include: Storage medium: used to store computer programs; Processor: for executing the computer program to implement the computing power token measurement method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the computing power token measurement method according to any one of claims 1 to 7.

Images