Multi-dimensional charging method and device for computing power resources and electronic equipment

Abstract

The invention provides a multi-dimensional charging method and device for computing power resources and electronic equipment, and the method comprises the steps: converting physical measurement indexes of all resource types in an original resource ticket, and obtaining a standard resource ticket; constructing a rate configuration model based on the basic unit price and the adjustment coefficient corresponding to each resource type; the adjustment coefficient is used for reflecting the resource commercial value difference of each resource type; and finally, inputting the standard resource ticket into the rate configuration model to obtain a cost result. According to the method provided by the invention, the original resource ticket of each resource type is acquired and converted into the standard resource ticket, so that unified measurement of heterogeneous resource data is realized; by constructing the rate configuration model containing the resource performance coefficient and comprehensively considering the influence of the resource type and the performance difference on the commercial value, accurate metering and charging of various intelligent computing resources are realized, the benefits of service providers and users are effectively balanced, and the standardization and high efficiency level of intelligent computing power business charging management is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent computing power service billing technology, and in particular to a multi-dimensional billing method, apparatus and electronic equipment for computing power resources. Background Technology

[0002] Driven by the rapid development of artificial intelligence technology, the demand for intelligent computing power services is experiencing explosive growth, with applications covering multiple fields such as deep learning training, inference, and scientific computing. Billing, as a core component of the commercial operation of intelligent computing power services, directly impacts the revenue of service providers and the usage costs for users, and is a crucial support for sustainable business development. Currently, most billing models rely solely on resource usage duration or a single resource type (such as GPUs only) as the basis for billing.

[0003] However, the current billing method ignores the collaborative dependence of intelligent computing services on multiple resources. In other words, billing based on a single dimension cannot fully reflect the actual consumption of resources, which can easily lead to unfair phenomena such as multiple resource consumptions being billed as a single resource, thus harming the interests of service providers or users. Summary of the Invention

[0004] This invention provides a multi-dimensional billing method, apparatus, and electronic device for computing resources, which addresses the shortcomings of the single-dimensional billing model for computing resources in the prior art.

[0005] This invention provides a multi-dimensional billing method for computing power resources, including: Obtain the original resource call detail records for each resource type; The physical metering indicators of each resource type in the original resource call detail record are converted to obtain the standard resource call detail record. Based on the base unit price and adjustment coefficient corresponding to each resource type, a rate configuration model is constructed. The standard resource call detail record (CDR) is input into the rate configuration model to calculate the cost result; The adjustment coefficient is used to reflect the differences in the commercial value of the resources of each resource type, and the adjustment coefficient includes the resource performance coefficient.

[0006] According to the present invention, a multi-dimensional billing method for computing power resources is provided, wherein the standard resource call detail records (CDRs) include service scenarios; the step of constructing a rate configuration model based on the basic unit price and adjustment coefficient corresponding to each resource type includes: The billing type is determined based on the aforementioned business scenario; The base unit price and adjustment coefficient corresponding to the billing type and each resource type are matched to obtain the following: The rate configuration model is constructed based on the product of the base unit price and the adjustment coefficient.

[0007] According to the multidimensional billing method for computing power resources provided by the present invention, the adjustment coefficient further includes at least one of an energy efficiency coefficient and a market demand coefficient; The energy efficiency coefficient is determined based on the energy use efficiency index of each resource type. The market demand coefficient is determined based on the time period in which the business scenario occurs.

[0008] According to the multi-dimensional billing method for computing power resources provided by the present invention, the step of converting the physical metering indicators of each resource type in the original resource call detail record (CDR) to obtain a standard resource CDR includes: Parse the original resource call detail records to identify the hardware model and physical metering parameters of each resource type; Based on the physical measurement indicators and the total hardware specifications that match the hardware model, the resource utilization ratio is calculated. The resource occupancy ratios of each resource type are converted to obtain the standard resource call detail records.

[0009] According to the present invention, a multi-dimensional billing method for computing resources is provided, wherein the physical measurement indicators include at least one of the following: core occupancy, video memory usage capacity, and computing load value.

[0010] According to a multi-dimensional billing method for computing resources provided by the present invention, the step of converting the resource occupancy ratio of each resource type to obtain the standard resource call detail record includes: When the resource type is GPU, the resource usage ratio of the GPU is converted based on the hardware model of the GPU to obtain the standard resource call detail record.

[0011] According to the present invention, a multi-dimensional billing method for computing resources is provided, wherein the resource types include CPU, memory, and GPUs of various hardware models.

[0012] The present invention also provides a multi-dimensional billing device for computing power resources, comprising: The acquisition unit retrieves the original resource call detail records for each resource type. The conversion unit converts the physical metering indicators of each resource type in the original resource call detail record (CDR) to obtain a standard resource CDR. The configuration unit constructs a rate configuration model based on the base unit price and adjustment coefficient corresponding to each resource type. The billing unit inputs the standard resource call detail record into the rate configuration model and calculates the cost result; The adjustment coefficient is used to reflect the differences in the commercial value of the resources of each resource type, and the adjustment coefficient includes the resource performance coefficient.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements a multidimensional billing method for computing resources as described above.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the multidimensional billing method for computing resources as described above.

[0015] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements a multidimensional billing method for computing resources as described above.

[0016] The multi-dimensional billing method, apparatus, and electronic equipment for computing power resources provided by this invention achieve unified metering of heterogeneous resource data by acquiring original resource call detail records (CDRs) of various resource types and converting them into standard resource CDRs, with multi-dimensional billing as the core. By constructing a rate configuration model that includes resource performance coefficients, and comprehensively considering the impact of resource types and their performance differences on commercial value, it achieves accurate metering and billing of various intelligent computing resources, effectively balancing the interests of service providers and users, and improving the standardization and efficiency of intelligent computing power service billing management. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the multi-dimensional billing method for computing power resources provided by the present invention. Figure 2 This is a schematic diagram of the rate configuration model provided by the present invention; Figure 3 This is a schematic diagram of the conversion process of the original resource call detail records provided by the present invention; Figure 4 This is a schematic diagram of the standard resource call detail record provided by the present invention; Figure 5 This is a schematic diagram of the public model access data collection and call detail record processing flow provided by the present invention; Figure 6 This is a schematic diagram of the structure of the multidimensional billing system provided by the present invention; Figure 7This is a schematic diagram of the workflow of the unified metering and billing module provided by the present invention; Figure 8 This is a schematic diagram of the multidimensional billing device for computing power resources provided by the present invention; Figure 9 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0020] It should be noted that most current billing models rely solely on resource usage duration or a single resource type (such as GPUs only) for billing, neglecting the collaborative dependence of intelligent computing services on multiple resources. For example, deep learning tasks require GPUs for computing power, as well as CPUs for data processing and memory for data storage. In other words, single-dimensional billing cannot fully reflect the actual resource consumption, easily leading to unfair situations where multiple resources are consumed but billed as a single resource, thus harming the interests of service providers or users.

[0021] To address the aforementioned issues, this invention provides a multi-dimensional billing method for computing resources, enabling cost calculations that correspond to the actual consumption of intelligent computing resources, i.e., achieving accurate cost calculations that can cover accurate billing of intelligent computing resources such as CPUs, multiple GPU models, and memory, as well as public model services. Figure 1 This is a flowchart illustrating the multi-dimensional billing method for computing power resources provided by the present invention, as shown below. Figure 1 As shown, the method includes: Step 110: Obtain the original resource call detail records for each resource type.

[0022] Here, "resource type" refers to various hardware or service categories that provide computing power support in intelligent computing business scenarios, such as CPUs, memory, and different models of GPUs, such as NVIDIA A100 and Cambricon MLU310. "Original resource call detail records" refers to unstandardized initial usage records directly obtained from the underlying resource scheduling system or data acquisition plugins; the data format may vary depending on the vendor.

[0023] Specifically, real-time usage data for various resource types can be captured by deploying data collection plugins or directly connecting to the resource scheduling system at the underlying resource layer of intelligent computing centers or cloud computing platforms. Since intelligent computing power services involve various heterogeneous resources, the log formats output by hardware from different vendors may differ significantly. Therefore, it is necessary to collect data on all resources, including CPU, memory, and various GPUs. Understandably, by extensively collecting raw resource call detail records (CDRs) for various resource types, the problem of fragmented data sources can be solved, providing a comprehensive and granular data foundation for accurate billing covering multiple dimensions of resources.

[0024] Step 120: Convert the physical metering indicators of each resource type in the original resource call detail record (CDR) to obtain the standard resource CDR.

[0025] Here, physical metrics exist in the original resource call detail records (CDRs) and are used to characterize the specific physical quantities of resources being occupied, such as the number of cores, the size of video memory, and the computing load value. Standard resource CDRs, on the other hand, are standardized records generated after cleaning and transforming the original data, possessing unified measurement dimensions and used for subsequent billing.

[0026] Specifically, the collected raw resource call detail records (CDRs) can be parsed to extract physical metering metrics. For differences in metering units between different vendors or resource types—for example, some vendors may use core counts, while others may use computing power or GPU memory percentages—pre-defined conversion rules can be applied. For instance, by addressing these metering differences, heterogeneous physical metering metrics can be uniformly converted to standardized units, such as a percentage of the entire card or a standard computing power unit. Combined with key fields like task type and runtime, a standardized resource CDR with a unified format can be generated.

[0027] Understandably, by converting the physical metering indicators of each resource type in the original resource call detail records (CDRs) to obtain standard resource CDRs, the format barriers of the underlying hardware data are eliminated, laying the foundation for accurate billing of multiple resource types.

[0028] Step 130: Construct a rate configuration model based on the basic unit price and adjustment coefficient corresponding to each resource type.

[0029] The adjustment coefficient is used to reflect the differences in the commercial value of the resources of each resource type, and the adjustment coefficient includes the resource performance coefficient.

[0030] Here, the base unit price refers to a preset benchmark price for a certain type of resource, such as a unit fee per time or per use. The adjustment coefficient is a parameter used to adjust the base unit price to reflect the differences in the commercial value of different resources. The resource performance coefficient, as one type of adjustment coefficient, is specifically used to reflect the strength of hardware performance, such as the performance ratio between high-performance and low-performance GPUs. Furthermore, the rate configuration model here is a calculation logic or mathematical formula built based on the unit price and coefficients, used to determine the final billing rules.

[0031] Specifically, a base unit price can be pre-set for different resource types and business needs. Meanwhile, to reflect the differences in commercial value among different resources, an adjustment coefficient is introduced to construct a flexible rate configuration model.

[0032] It should be noted that the core of constructing the rate configuration model lies in using a resource performance coefficient as part of the adjustment coefficient. This coefficient is used to quantify the value ratio between resources of different performance levels, such as the value ratio between different generations or models of GPUs. For example, a higher resource performance coefficient can be configured for a more powerful GPU model. Furthermore, by associating and combining the base unit price with the adjustment coefficient, which includes the resource performance coefficient, such as through multiplication, a rate configuration model that can dynamically adapt to different resource values ​​is constructed.

[0033] Understandably, by introducing an adjustment factor that includes resource performance coefficients, the limitations of traditional single-dimensional billing are broken, enabling the billing model to accurately match the actual performance value of resources and achieving a more refined pricing strategy.

[0034] Step 140: Input the standard resource call detail record into the rate configuration model to calculate the cost result.

[0035] Here, the cost result is the final amount receivable calculated by the model.

[0036] Specifically, standardized resource call detail records (CDRs) can be imported into a pre-built rate configuration model as input data. Then, the rate configuration model locks the corresponding base unit price and adjustment coefficient based on the resource type in the CDR, and automatically executes the calculation logic in conjunction with the usage data in the CDR, such as duration or usage, and finally outputs the comprehensive cost result of all computing resources used in the order.

[0037] It should be noted that the calculated cost results can include detailed cost breakdowns and billing cycle information, and can be further pushed to the billing system to generate invoices. This ensures that in multi-resource collaboration scenarios, the billing results can reflect both the objective usage time and the performance value of the resources, greatly improving the accuracy and rationality of billing.

[0038] The method provided in this invention, with multi-dimensional billing as its core, acquires the original resource call detail records (CDRs) of various resource types and converts them into standard resource CDRs, thereby achieving unified metering of heterogeneous resource data. By constructing a rate configuration model that includes resource performance coefficients, it comprehensively considers the impact of resource types and their performance differences on commercial value, achieving accurate metering and billing for various intelligent computing resources, effectively balancing the interests of service providers and users, and improving the standardization and efficiency of intelligent computing power service billing management.

[0039] It should be noted that different models of intelligent computing resources have significant performance differences, and the services operate under different models such as annual / monthly subscriptions and pay-as-you-go billing. Furthermore, public model services are billed based on the number of calls made using Tokens and API keys. Existing billing systems lack the ability to adapt to the specific characteristics of resources and business models, thus failing to achieve differentiated and accurate billing. To address this issue, based on any of the above embodiments, the standard resource call detail records (CDRs) include business scenarios.

[0040] Step 130 includes: The billing type is determined based on the aforementioned business scenario; The base unit price and adjustment coefficient corresponding to the billing type and each resource type are matched to obtain the following: The rate configuration model is constructed based on the product of the base unit price and the adjustment coefficient.

[0041] Here, "business scenario" refers to the actual working conditions or service modes under which intelligent computing power is specifically applied, such as deep learning training tasks, AI inference tasks, scientific computing, or public model service calls. Different business scenarios have significantly different resource usage methods and durations. Additionally, "billing type" refers to the commercial settlement model set up for different business needs, mainly including annual / monthly subscriptions and pay-as-you-go billing models.

[0042] Specifically, firstly, when generating standard resource call detail records (CDRs), the task description or service identifier carried in the CDR can be parsed to identify the business scenario in which the current resource consumption occurs. For example, it can be identified whether the current task belongs to a long-running model training scenario or a bursty online inference scenario.

[0043] Next, by combining the user's contract information or order records in the product ordering module, the billing type selected by the user in this business scenario is determined. If the business scenario corresponds to a long-term service that the user has already purchased, the billing type is determined to be a subscription / monthly subscription; if the business scenario is an elastic task initiated by the user temporarily, the billing type is determined to be pay-as-you-go. It is understandable that mapping billing types to business scenarios can accurately distinguish different user intentions, providing a basis for differentiated pricing strategies in subsequent applications.

[0044] Furthermore, a pre-constructed multi-dimensional rate matrix can be used for matching. This matrix can be categorized horizontally by billing type, such as annual / monthly subscription or pay-as-you-go; and vertically by resource type, such as single A100 card, complete system, CPU, memory, etc. Then, based on the determined billing type and the resource type in the standard resource call detail record (CDR), a unique intersection point is located within this matrix to obtain the base unit price and corresponding adjustment coefficient for that specific combination. For example, for the same GPU resource, the base unit price under the annual / monthly subscription billing type may be lower than that under the pay-as-you-go billing type, or the adjustment coefficient settings may differ.

[0045] Understandably, by determining the base unit price and adjustment coefficient based on the billing type and each resource type, the rate configuration is refined, ensuring that each billing transaction accurately corresponds to the specific product specifications and business model, thus avoiding billing deviations caused by a single rate.

[0046] Finally, after obtaining the specific numerical values, the core formula logic of "base unit price × adjustment coefficient" can be used to construct the current rate configuration model. For example, the matched base unit price is used as the base, multiplied by the resource performance coefficient to form a dynamic calculation formula. It is understandable that if multiple adjustment coefficients exist, they are multiplied sequentially.

[0047] Once the rate configuration model is built, it can be used to perform weighted calculations on specific resource usage. Understandably, using a product-based approach to build the model allows various influencing factors to directly impact the final price as factors, ensuring transparency in the billing logic while also giving the billing system strong scalability, enabling it to quickly respond to market changes or business adjustments by adjusting any coefficient.

[0048] The method provided in this invention further refines the construction process of the rate configuration model by incorporating business scenarios. By determining the billing type based on business scenarios and accurately matching the basic unit price and adjustment coefficient using a multi-dimensional matrix, and then constructing the model using a product formula, a flexible billing system based on a multi-dimensional matrix is ​​realized. This not only solves the problem of poor resource and scenario adaptability in intelligent computing services and supports the coexistence of multiple modes such as annual / monthly subscriptions and pay-as-you-go billing, but also makes the billing strategy more aligned with resource value and actual user usage, achieving refined billing across all scenarios.

[0049] Based on any of the above embodiments, the adjustment coefficient dimension in the rate configuration model will be further enriched to achieve a more dynamic and ecologically oriented billing strategy. The adjustment coefficient also includes at least one of an energy efficiency coefficient and a market demand coefficient; The energy efficiency coefficient is determined based on the energy use efficiency index of each resource type. The market demand coefficient is determined based on the time period in which the business scenario occurs.

[0050] Here, the energy efficiency coefficient refers to a correction parameter used to reflect the efficiency level of energy consumption in a smart computing center or a specific computing cluster. Its purpose is to correlate energy consumption optimization and reflect the value or cost of green computing. The energy use efficiency index is the basis for determining this coefficient, and it can typically refer to the PUE (Power Usage Effectiveness) value or a specific energy consumption level.

[0051] Additionally, here, the market demand coefficient refers to a parameter used to reflect the impact of fluctuations in market supply and demand on prices, aiming to regulate the peak and trough distribution of resource usage through price levers. Here, the occurrence period refers to the specific time range of the actual operation of the business scenario recorded in the standard resource call detail record, such as the metering start time to metering end time.

[0052] Specifically, the method for determining the energy efficiency coefficient of any resource type includes: First, obtaining the corresponding energy usage efficiency index based on the physical environment or cluster attributes of the resource type. For example, some computing resources deployed in low-energy green data centers have lower PUE values, while other resources may be deployed in older data centers with higher PUE values. Then, by pre-setting the mapping relationship between energy usage efficiency indexes and energy efficiency coefficients, lower coefficients can be configured for resource types with higher energy efficiency to offer price discounts, or the coefficient can be configured based on actual electricity costs. Finally, the corresponding energy efficiency coefficient is determined using the energy usage efficiency index for that resource type.

[0053] Understandably, introducing an energy efficiency coefficient can link computing power costs to energy consumption, which can not only more accurately reflect operating costs, but also guide users to prioritize the use of high-energy-efficiency resources through pricing, thus aligning with the energy conservation and emission reduction operational goals of intelligent computing centers.

[0054] In addition, methods for determining the market demand coefficient for any resource type include: First, parsing the time period information of the business scenario in the standard resource call detail record (CDR). Peak periods (such as daytime on weekdays) and off-peak periods (such as nighttime or holidays) can be predefined. Then, the time period of the business scenario is compared with the preset peak and off-peak periods. If the occurrence period is during a peak period, a higher market demand coefficient is matched, such as greater than 1.0, to reflect the scarcity value of the scarce resource; if it is during an off-peak period, a lower market demand coefficient is matched, such as less than 1.0, to encourage users to use the resource during off-peak hours.

[0055] Understandably, by determining the market demand coefficient based on the time of occurrence, a peak-valley electricity pricing mechanism similar to electricity pricing is achieved, which can effectively regulate user behavior and improve the overall utilization rate and operating revenue of computing resources.

[0056] Therefore, after obtaining the aforementioned adjustment coefficients, a rate allocation model for each resource type can be constructed based on the base unit price, resource performance coefficient, energy efficiency coefficient, and market demand coefficient. Specifically, after obtaining the aforementioned coefficients, they can be uniformly incorporated into the rate calculation formula. The construction logic can follow the multiplication pattern of "base unit price × resource performance coefficient × energy efficiency coefficient × market demand coefficient".

[0057] In one embodiment, Figure 2 This is a schematic diagram of the rate configuration model provided by the present invention, as shown below. Figure 2 As shown, the model intuitively demonstrates the core calculation logic based on "base price × adjustment coefficient" and its multi-dimensional hierarchical structure. At the top of the model, the calculation formula is explicitly stated: the base price (BasePrice) is multiplied sequentially by the resource performance coefficient (MFU_Coeff), the energy efficiency coefficient (PUE_Coeff), and the market demand coefficient (Demand_Coeff). In terms of hierarchical structure, the rate settings are first divided into two main branches based on the billing type: "annual / monthly subscription" and "pay-as-you-go." The "annual / monthly subscription" branch is further subdivided into two granularities: "complete system" and "single card." The "single card" dimension is further broken down into three resource types: CPU, GPU, and memory. The GPU type is further refined to specific hardware models, including NVIDIA-A100, Cambricon-MLU310, Ascend-910B, etc., to accommodate the performance differences of hardware from different manufacturers. For the "pay-as-you-go" branch, the path is shown, which is directly refined to the specific hardware model at the single-card level (such as NVIDIA-A100, Cambricon-MLU310, Ascend-910B). Through this tree-like matrix structure, the system can configure the base unit price and corresponding adjustment coefficient for each specific leaf node (i.e., the specific hardware model under a specific billing mode), thereby achieving refined and differentiated pricing of intelligent computing resources.

[0058] It should be noted that through this multi-factor dynamic formula, changes in any dimension can be reflected in the final cost result in real time. Understandably, this construction method achieves a high degree of flexibility and adaptability in the billing model, enabling the static base price to be adjusted in real time based on resource performance, energy costs, and market dynamics.

[0059] The method provided in this invention, by further introducing energy efficiency coefficient and market demand coefficient into the adjustment coefficient, not only considers the performance differences of the hardware itself reflected by the resource performance coefficient, but also adapts to the peak and valley fluctuations of the business through the market demand coefficient. This multi-dimensional billing method can comprehensively cover the complex scenarios of intelligent computing business, realize the fine and dynamic configuration of rates, improve the market adaptability of the billing system, and help guide the efficient and reasonable allocation of resources.

[0060] It should be noted that the call detail record (CDR) formats output by different vendors for intelligent computing resources are not standardized. For example, some vendors calculate resource usage based on the number of cores, while others calculate it based on computing power or the ratio of video memory to total video memory. This results in CDR data that cannot be directly used for billing, further increasing the complexity of data processing. Manual intervention is required for data processing, which is not only time-consuming and labor-intensive but also prone to introducing human error, affecting billing efficiency and accuracy. To address this issue, based on any of the above embodiments, step 120 includes: Parse the original resource call detail records to identify the hardware model and physical metering parameters of each resource type; Based on the physical measurement indicators and the total hardware specifications that match the hardware model, the resource utilization ratio is calculated. The resource occupancy ratios of each resource type are converted to obtain the standard resource call detail records.

[0061] Here, "hardware model" refers to the specific device identifier of the computing resource, such as NVIDIA A100 or Cambricon MLU310. Different models represent different underlying architectures and performance benchmarks. "Total hardware specifications" refers to the total physical resources possessed by that hardware model, such as the total video memory capacity of a graphics card (e.g., 80GB) or the total number of cores (e.g., 6912 CUDA cores). This can be understood as the denominator for calculating computational utilization. Additionally, "resource utilization ratio" refers to the percentage of resources actually used by the user relative to the total hardware resources. It is a normalized relative value used to mask differences in physical units across different hardware components.

[0062] Specifically, firstly, a standardized processing flow can be initiated to read the raw resource call detail records (CDRs) collected from the underlying system. Since the raw CDRs may originate from different monitoring plugins or resource scheduling systems, their field definitions vary. Therefore, it is necessary to utilize preset parsing rules to accurately extract the hardware model field that identifies the device, as well as the physical metrics that record resource consumption details. For example, from a log entry, the hardware model might be identified as "NVIDIA-A100-80GB," and the physical metrics extracted might be "40960MB of video memory usage" or "50% of computing cores used."

[0063] Understandably, accurately identifying the hardware model is a prerequisite for obtaining the correct specifications, while physical measurement indicators are the basis for calculating actual consumption.

[0064] Then, a static database or configuration table containing various intelligent computing hardware specifications can be maintained. Based on the identified hardware model, the corresponding total hardware specification parameters are retrieved from this database. Next, the physical metrics extracted from the call detail records (CDRs) are calculated using the retrieved total hardware specification parameters. For example, if the physical metric is "40GB of video memory used," and the total hardware specification parameter for this GPU model is "80GB of total video memory," then the calculated resource utilization rate is 50%. It should be noted that if the physical metric itself is already in percentage form, such as core utilization, further calculations need to be determined based on hardware characteristics.

[0065] Understandably, by calculating the resource usage ratio, absolute values, such as MB and Core count, are converted into relative values, thus unifying the percentage dimension of the entire card and solving the problem of inconsistent measurement units among different manufacturers, such as some measuring by video memory and others by cores.

[0066] Finally, the calculated resource usage ratio can be used as the core billing basis and encapsulated according to a predefined standard data format. This process can map the resource usage ratio to standard computing power units, or directly use it as the usage field in the standard resource call detail record (CDR). Simultaneously, combined with other information in the CDR, such as task ID and timestamp, the final standard resource CDR is generated. For example, a standard record containing "Resource Type: A100", "Billing Dimension: Percentage of Whole Card", and "Value: 0.5" can be generated.

[0067] Understandably, transforming complex underlying physical consumption into standardized data that can be directly read by the upper-level billing system ensures that the subsequent rate calculation model can process all types of resource usage records with a unified logic.

[0068] The method provided in this invention identifies the hardware model by parsing the original call detail records (CDRs) and calculates the resource occupancy ratio by combining the total hardware specification parameters, ultimately converting it into a standard resource CDR. It innovatively proposes a data processing link of physical quantity-relative quantity-standard quantity, effectively shielding the heterogeneity of the underlying hardware and unifying the resource consumption of different manufacturers and architectures into a comparable and calculable occupancy ratio. This not only solves the pain points of data fragmentation and inconsistent formats in intelligent computing scenarios, but also lays a solid data foundation for achieving fair and transparent multi-dimensional billing.

[0069] In one embodiment, Figure 3 This is a schematic diagram of the original resource call detail record (CDR) conversion process provided by the present invention, as follows: Figure 3As shown, the process mainly involves obtaining data from third-party systems, processing it through dedicated call detail record (CDR) collection plugins from various vendors, and finally completing the process of forming a unified unit of measurement by filling in the differences.

[0070] First, third-party systems, acting as the underlying management platform for intelligent computing resources, such as resource scheduling systems or model service gateways, are the source of data generation. To ensure compatibility with the data interface differences among different hardware manufacturers, various dedicated call detail record (CDR) collection plugins can be deployed to capture raw data in real time from resources from Cambricon, Ascend, NVIDIA, and other manufacturers. This enables initial access to multi-source heterogeneous data and solves the problem of fragmented data sources.

[0071] Next, for the raw data collected from different manufacturers, differentiated calculation logic is executed to uniformly convert it into a standard metric of the entire card's percentage: For Cambricon resources, the data acquisition plugin obtains its used video memory and total video memory data, and calculates the ratio of "used video memory / total video memory" to determine the proportion of the task's usage on the graphics card, i.e., calculating the percentage of the entire card's usage based on video memory capacity. For Ascend resources, the data acquisition plugin obtains the number of cores used, and combines this with the total core specifications of the hardware model to obtain the percentage of the entire card's usage through a "calculation by core count" method. For NVIDIA resources, the data acquisition plugin obtains its computing power value (such as the computing power specification or load value after MIG splitting), and obtains the percentage of the entire card's usage through a "calculation by computing power value" method. In addition, corresponding calculation interfaces are reserved for other types of resources to support more multi-dimensional conversion logic.

[0072] Finally, after completing the respective percentage calculations, the process enters the top-level standardization stage, which involves supplementing the call detail record (CDR) files to form a unified unit of measurement. In this stage, not only is the calculated "whole card percentage" used as the core measurement data, but fields such as time format, task ID, and user identifier in the CDRs reported by different plugins are also cleaned and supplemented to eliminate format differences. Ultimately, a standard resource CDR with a unified format and consistent unit of measurement is output and sent to the subsequent billing module.

[0073] The method provided in this invention shields the underlying physical differences at the acquisition layer and provides a standardized data interface to the upper-layer billing system. This allows the billing system to focus on the specific hardware details at the underlying level and only need to handle a uniform resource usage ratio, which greatly reduces the implementation complexity of multi-dimensional billing systems and improves the scalability and compatibility of the system.

[0074] Based on any of the above embodiments, the physical measurement indicators include at least one of the following: core usage, video memory usage, and computing load value.

[0075] Here, core usage typically refers to the number of physical cores or vCPUs of a CPU, or the number of computing units allocated in a GPU. Memory usage refers to the specific amount of GPU memory used, usually measured in MB or GB. Computational load reflects the busyness or utilization of the computing units, usually expressed as a percentage, or as a specific unit of computational power, such as TFLOPS.

[0076] It is understandable that physical metrics may differ for different resource types. However, regardless of which physical metric or combination of the above-mentioned metrics exists, they are all standardized into a resource utilization ratio. For example, for the same A100 graphics card, task A consumes 50% of the video memory, while task B consumes 30% of the cores. Therefore, based on preset rules, such as taking the maximum value or a weighted average, the resource utilization ratio can be determined and written into the standard resource call detail record.

[0077] Understandably, supporting multiple physical metering metrics gives the billing system great flexibility, enabling it to adapt to the metering needs of different vendors (such as NVIDIA, Cambricon, and Ascend) and different business models (such as memory-intensive and compute-intensive), truly achieving full-scenario coverage.

[0078] Based on any of the above embodiments, the resource occupancy ratios of each resource type are converted to obtain the standard resource call detail record (CDR), including: When the resource type is GPU, the resource usage ratio of the GPU is converted based on the hardware model of the GPU to obtain the standard resource call detail record.

[0079] Specifically, before finalizing the calculated resource usage ratio, the type of resource being processed is first determined. When the resource type is identified as GPU, the device identifier in the original call detail record is further parsed to clarify the specific hardware model of the GPU, for example, distinguishing between "A100-40GB" and "A100-80GB".

[0080] Furthermore, based on the GPU hardware model, the resource usage ratio of the resource model is converted to obtain a standard resource call detail record (CDR). For example, based on the hardware model and a preset performance coefficient mapping table, the GPU resource usage ratio can be converted to obtain a corrected resource usage ratio under a standardized rule. Then, the converted corrected resource usage ratio, along with key fields such as business scenario, resource usage duration, metering start and end time, and customer identifier, can be packaged into a standard resource CDR.

[0081] The method provided in this invention, in addition to converting the resource usage ratio between different hardware types, also converts the resource usage ratio of different resource models based on the GPU hardware model to obtain a more refined standard resource call detail record. This further effectively avoids unfair billing caused by ignoring hardware performance differences and ensures reasonable revenue for service providers when providing high-performance computing services.

[0082] In one embodiment, Figure 4 This is a schematic diagram of the standard resource call detail record (CDR) provided by the present invention, such as... Figure 4 As shown, the unified call detail record (CDR) for intelligent computing tasks corresponds to the standard resource CDR and is a key middleware connecting the underlying resource data collection and the upper-layer cost calculation. That is, whether it's an intelligent computing resource-related service, such as using GPUs for training, or a public model service-related service, such as calling an API, after preprocessing, it will ultimately be mapped to this standard format with unified dimensions. The standard resource CDR includes: task type, task description, billing type, whether it is exclusive, resources used, runtime, metering start time, metering end time, and customer.

[0083] It should be noted that the collection frequencies of public model service call detail records (CDRs) and computing resource CDRs are inconsistent (the former requires high-frequency collection, while the latter can be low-frequency), further increasing the complexity of data processing. This necessitates manual intervention, which is not only time-consuming and labor-intensive but also prone to introducing human error, affecting billing efficiency and accuracy. To address this issue, in one embodiment, Figure 5 This is a schematic diagram of the public model access data collection and call detail record processing flow provided by the present invention, as shown below. Figure 5 As shown, this process primarily demonstrates the high-frequency data collection and standardization process for public model services (such as large model API calls). First, the process begins at the underlying model gateway, which is the traffic entry and aggregation point for all model access requests, recording the raw logs of each call. To capture detailed model usage under instantaneous high concurrency, the middle-layer model raw call detail record (CDR) collection module executes a high-frequency collection strategy, as indicated in the figure, directly connecting to the model gateway every 6 seconds to capture raw data, ensuring the real-time and completeness of billing data. The collected raw data is then sent to the top-layer unified CDR processing module, which is responsible for parsing, cleaning, and standardizing the data, extracting key fields such as model service (identifying the specific model type called), API-KEY (determining user identity or account credentials), and model tokens (statistically calculating specific resource consumption). Through this process, heterogeneous gateway logs can be transformed into standardized billing CDRs with a unified structure, thereby achieving accurate metering and billing for public model services.

[0084] It should be noted that by segmenting GPU models, supporting annual / monthly / pay-as-you-go billing, and being compatible with public model service billing, it can adapt to different scenarios such as long-term stable use by enterprises, short-term temporary use by individuals, and high-frequency model calls, meeting diverse business needs and enhancing market competitiveness.

[0085] Based on any of the above embodiments, the resource types include CPU, memory, and GPUs of various hardware models.

[0086] It should be noted that by explicitly expanding resource types to include CPU, memory, and various hardware models of GPUs, a billing system covering all elements of intelligent computing services has been built. This effectively overcomes the limitations of the single-dimensional model in the traditional model. It can not only accurately price according to the performance differences of GPU models, but also reasonably charge for CPU and memory resources that have been occupied for free for a long time. This ensures the revenue of service providers while guiding users to apply for supporting resources according to their actual needs, avoiding resource waste and improving the overall utilization efficiency of computing resources.

[0087] Based on any of the above embodiments Figure 6 This is a schematic diagram of the structure of the multi-dimensional billing system provided by the present invention, as shown below. Figure 6 As shown, the system is centered on a multi-dimensional model construction module, integrating eight functional modules including customer management, product management, product ordering, unified call detail records (CDRs), metering and billing, accounting services, and raw CDR collection.

[0088] Specifically, the multi-dimensional model building module is used to build a billing framework adapted to intelligent computing services, sort out the metering attributes of CPU, memory and specific GPU models, and establish the underlying conversion logic.

[0089] The product management module serves as the foundational data support, responsible for creating, classifying, and maintaining intelligent computing power product information. Product attributes must clearly define resource types (CPU, memory, GPU, with GPUs further subdivided into models such as NVIDIA A100, Cambricon MLU310, and Ascend 910B), billing types (annual / monthly subscriptions or pay-as-you-go billing), resource specifications (such as CPU core count, GPU computing power, and memory capacity), and a product rate setting module to ensure accurate matching of product attributes and user needs during subsequent billing.

[0090] The customer management module works in tandem with the product ordering module. On the one hand, it manages basic customer information and exclusive discounts, and on the other hand, it records users' order requests, such as annual or monthly subscriptions or pay-as-you-go billing. The ordering data is then synchronized to the subsequent billing process, namely to the metering and billing module and the accounting service module, as the basis for subsequent billing and accounting reconciliation, thus achieving data interoperability between "ordering-billing-accounting".

[0091] The raw call detail record (CDR) acquisition module directly connects to underlying resources through plug-in capabilities, ensuring compatibility with CDR formats from different vendors. It directly connects to the resource scheduling system and model service gateway, capturing raw usage records in real time. This provides a comprehensive, fine-grained data source for billing, resolving the fragmentation issue associated with heterogeneous hardware data acquisition. The CDR unification module standardizes the collected raw data, eliminating format differences and generating unified metering inputs. The metering and billing module, as the core engine, pulls standardized CDRs and subscription information, executes the "resource locking - rule matching - cost calculation" process, and uses multi-dimensional rate formulas to calculate accurate costs.

[0092] Finally, the billing service module receives the payment results and generates invoices, automatically producing customer bills and distinguishing between annual / monthly subscriptions and pay-as-you-go billing details. It supports multi-dimensional accounting by resource type and business scenario. Built-in overdue payment monitoring logic tracks customer payment status in real time, triggering tiered alerts such as overdue payment warnings and collection reminders. It also integrates payment channels such as online payment and direct debit, recording payment transactions and linking them to invoice reconciliation. Linking with the product ordering module, when a customer's overdue payment exceeds their credit limit, it triggers resource usage permission control, such as suspending access to high-performance computing resources. Simultaneously, it outputs accounting reports, such as revenue statistics and overdue payment analysis, providing financial data support for operational decisions and achieving collaborative governance of billing, accounting, and resource management.

[0093] Figure 7 This is a schematic diagram of the workflow of the unified metering and billing module provided by the present invention, as shown below. Figure 7 As shown, this process demonstrates the complete automated billing logic from scheduled triggering to bill generation, specifically including the following steps: Scheduled task triggering: The entire process is controlled by a timer, set to execute hourly. This means the billing system will automatically start the billing task on an hourly cycle to ensure timely billing.

[0094] Obtaining User Information: After the task starts, the system first performs the operation of pulling the tenants to be billed, and obtains the list of users who need to be billed and settled in the current period and related information from the database or user management module.

[0095] Resource Status Locking: To ensure data consistency during billing, the system performs a step to lock tenant resources. This step aims to lock the tenant's current resource usage status, preventing duplicate or missed billing due to resource status changes such as sudden release or addition during billing calculation, thus ensuring data accuracy.

[0096] Executing Billing Rules: This is the core step of the process, specifically corresponding to the detailed logic block diagram on the right side of the image. This step first receives two types of standardized input data: call detail records (CDRs) for computing resources (corresponding to hardware resources such as CPU and GPU) and call detail records for model services (corresponding to API calls, token usage, etc.). For these two types of CDRs, the system matches the product unit price, i.e., finds the corresponding basic fee based on the configuration in the product management module. Next, the process enters the judgment stage, determining whether product discounts apply. This means the system checks whether the current product has specific promotions or tiered discount strategies. Subsequently, the matched product unit price is reconfirmed or adjusted, and finally, the customer discount rate is calculated based on the user's exclusive benefits to determine the actual amount payable for this period.

[0097] Generate billing records: After completing the cost calculation, the system executes the step of generating billing time, recording the specific time period of this billing (such as the start and end time of the billing) and generating a timestamp, providing time-dimensional evidence for the bill.

[0098] Bill Update: Finally, the system executes the bill update operation, writing the calculated cost details and summary results into the accounting system, updating the tenant's monthly or periodic bills, and completing the closed loop from call detail record collection to cost settlement.

[0099] Based on any of the above embodiments Figure 8 This is a schematic diagram of the multi-dimensional billing device for computing power resources provided by the present invention, as shown below. Figure 8 As shown, the device includes: Get unit 810 to obtain the original resource call detail records for each resource type; The conversion unit 820 converts the physical metering indicators of each resource type in the original resource call detail record (CDR) to obtain a standard resource CDR. Configuration unit 830 constructs a rate configuration model based on the basic unit price and adjustment coefficient corresponding to each resource type; The billing unit 840 inputs the standard resource call detail record into the rate configuration model and calculates the cost result; The adjustment coefficient is used to reflect the differences in the commercial value of the resources of each resource type, and the adjustment coefficient includes the resource performance coefficient.

[0100] The device provided in this invention, with multi-dimensional billing as its core, acquires original resource call detail records (CDRs) of various resource types and converts them into standard resource CDRs, thereby achieving unified metering of heterogeneous resource data. By constructing a rate configuration model that includes resource performance coefficients, it comprehensively considers the impact of resource types and their performance differences on commercial value, achieving accurate metering and billing of various intelligent computing resources, effectively balancing the interests of service providers and users, and improving the standardization and efficiency of intelligent computing power service billing management.

[0101] Based on any of the above embodiments, the standard resource call detail record (CDR) includes service scenarios; the configuration unit is specifically used for: The billing type is determined based on the aforementioned business scenario; The base unit price and adjustment coefficient corresponding to the billing type and each resource type are matched to obtain the following: The rate configuration model is constructed based on the product of the base unit price and the adjustment coefficient.

[0102] Based on any of the above embodiments, the adjustment coefficient further includes at least one of an energy efficiency coefficient and a market demand coefficient; The energy efficiency coefficient is determined based on the energy use efficiency index of each resource type. The market demand coefficient is determined based on the time period in which the business scenario occurs.

[0103] Based on any of the above embodiments, the conversion unit is specifically used for: Parse the original resource call detail records to identify the hardware model and physical metering parameters of each resource type; Based on the physical measurement indicators and the total hardware specifications that match the hardware model, the resource utilization ratio is calculated. The resource occupancy ratios of each resource type are converted to obtain the standard resource call detail records.

[0104] Based on any of the above embodiments, the physical measurement indicators include at least one of the following: core usage, video memory usage, and computing load value.

[0105] Based on any of the above embodiments, the conversion unit is further specifically used for: When the resource type is GPU, the resource usage ratio of the GPU is converted based on the hardware model of the GPU to obtain the standard resource call detail record.

[0106] Based on any of the above embodiments, the resource types include CPU, memory, and GPUs of various hardware models.

[0107] Figure 9 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 9As shown, the electronic device may include: a processor 910, a communication interface 920, a memory 930, and a communication bus 940, wherein the processor 910, the communication interface 920, and the memory 930 communicate with each other through the communication bus 940. The processor 910 can call logical instructions in the memory 930 to execute a multi-dimensional billing method for computing resources. This method includes: obtaining original resource call details records (CDRs) for each resource type; converting the physical metering indicators of each resource type in the original resource CDRs to obtain standard resource CDRs; constructing a rate configuration model based on the basic unit price and adjustment coefficient corresponding to each resource type; inputting the standard resource CDRs into the rate configuration model to calculate the cost result; the adjustment coefficient is used to reflect the difference in the commercial value of resources of each resource type, and the adjustment coefficient includes a resource performance coefficient.

[0108] Furthermore, the logical instructions in the aforementioned memory 930 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0109] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the multi-dimensional billing method for computing power resources provided by the above methods. The method includes: obtaining original resource call detail records (CDRs) for each resource type; converting the physical metering indicators of each resource type in the original resource CDRs to obtain standard resource CDRs; constructing a rate configuration model based on the basic unit price and adjustment coefficient corresponding to each resource type; inputting the standard resource CDRs into the rate configuration model to calculate the cost result; the adjustment coefficient is used to reflect the difference in the commercial value of resources for each resource type, and the adjustment coefficient includes a resource performance coefficient.

[0110] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a multi-dimensional billing method for computing power resources provided by the methods described above. The method includes: obtaining original resource call detail records (CDRs) for each resource type; converting the physical metering indicators of each resource type in the original resource CDRs to obtain standard resource CDRs; constructing a rate configuration model based on the base unit price and adjustment coefficient corresponding to each resource type; inputting the standard resource CDRs into the rate configuration model to calculate the cost result; wherein the adjustment coefficient is used to reflect the difference in the commercial value of each resource type, and the adjustment coefficient includes a resource performance coefficient.

[0111] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0112] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-dimensional billing method for computing power resources, characterized in that, include: Obtain the original resource call detail records for each resource type; The physical metering indicators of each resource type in the original resource call detail record are converted to obtain the standard resource call detail record. Based on the base unit price and adjustment coefficient corresponding to each resource type, a rate configuration model is constructed. The standard resource call detail record (CDR) is input into the rate configuration model to calculate the cost result; The adjustment coefficient is used to reflect the differences in the commercial value of the resources of each resource type, and the adjustment coefficient includes the resource performance coefficient.

2. The multi-dimensional billing method for computing power resources according to claim 1, characterized in that, The standard resource call detail records (CDRs) include business scenarios; the construction of a rate configuration model based on the basic unit price and adjustment coefficient corresponding to each resource type includes: The billing type is determined based on the aforementioned business scenario; The base unit price and adjustment coefficient corresponding to the billing type and each resource type are matched to obtain the following: The rate configuration model is constructed based on the product of the base unit price and the adjustment coefficient.

3. The multi-dimensional billing method for computing power resources according to claim 2, characterized in that, The adjustment coefficient also includes at least one of the energy efficiency coefficient and the market demand coefficient; The energy efficiency coefficient is determined based on the energy use efficiency index of each resource type. The market demand coefficient is determined based on the time period in which the business scenario occurs.

4. The multi-dimensional billing method for computing power resources according to any one of claims 1 to 3, characterized in that, The process of converting the physical metering indicators of each resource type in the original resource call detail record (CDR) to obtain a standard resource CDR includes: Parse the original resource call detail records to identify the hardware model and physical metering parameters of each resource type; Based on the physical measurement indicators and the total hardware specifications that match the hardware model, the resource utilization ratio is calculated. The resource occupancy ratios of each resource type are converted to obtain the standard resource call detail records.

5. The multi-dimensional billing method for computing power resources according to claim 4, characterized in that, The physical metrics include at least one of the following: core usage, video memory usage, and computing load.

6. The multi-dimensional billing method for computing power resources according to claim 4, characterized in that, The conversion of the resource occupancy ratios for each resource type to obtain the standard resource call detail record (CDR) includes: When the resource type is GPU, the resource usage ratio of the GPU is converted based on the hardware model of the GPU to obtain the standard resource call detail record.

7. The multi-dimensional billing method for computing power resources according to any one of claims 1 to 3, characterized in that, Resource types include CPU, memory, and GPUs of various hardware models.

8. A multi-dimensional billing device for computing power resources, characterized in that, include: The acquisition unit retrieves the original resource call detail records for each resource type. The conversion unit converts the physical metering indicators of each resource type in the original resource call detail record (CDR) to obtain a standard resource CDR. The configuration unit constructs a rate configuration model based on the base unit price and adjustment coefficient corresponding to each resource type. The billing unit inputs the standard resource call detail record into the rate configuration model and calculates the cost result; The adjustment coefficient is used to reflect the differences in the commercial value of the resources of each resource type, and the adjustment coefficient includes the resource performance coefficient.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the multidimensional billing method for computing resources as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the multidimensional billing method for computing resources as described in any one of claims 1 to 7.

Images