An identification-based wide-area heterogeneous computing power analysis method

Abstract

This invention belongs to the field of data communication network technology and discloses a wide-area heterogeneous computing power resolution method based on identifiers. The steps of this invention include: when computing power resources access the network, they send a registration request to a computing power resolution server, the request information including the attribute information of the computing power resources; the computing power resolution server abstracts the computing power resources into a hierarchical tree and encodes them according to domain name organization rules; the computing power resolution server registers and manages the computing power resources; if registration is successful, when a user initiates a task request, a dynamic and static matching algorithm is used to search for a list of available computing power resources from the computing power resolution server. This invention can effectively improve the utilization rate of computing power resources, optimize user experience, and simultaneously achieve optimized allocation and collaborative management of global computing power resources.

Description

Technical Field

[0001] This invention mainly relates to the field of data communication network technology, and in particular to an identifier-based method for resolving wide-area heterogeneous computing power. Background Technology

[0002] A Computing Power Network (CPN) is a new type of information infrastructure. Its core idea is to connect geographically distributed computing power center nodes through new network technologies, dynamically and in real time perceive the status of computing power resources, and then coordinate the allocation and scheduling of computing tasks, transmit data, and form a network that perceives, allocates, and schedules computing power globally. On this basis, computing power, data, and application resources are aggregated and shared.

[0003] Currently, computing networks present multiple development ideas and different technical routes. Specifically, they include: (1) Computation-first network, where computing resource nodes announce computing resource information such as computing type and availability in routing information by extending routing protocols such as BGP (Border Gateway Protocol) and OSPF (Open Shortest Path First). Other network nodes forward the received computing resource announcements. After receiving user requests, edge network nodes intelligently select computing resource nodes according to the computing resource view and calculate computing routes. The routing requirements meet deterministic requirements such as bandwidth, path, and latency. (2) The computing network convergence technical route based on SD-WAN (Software-Defined Wide Area Network), which builds an independent computing resource layer on top of the cloud service and network scheduling layer, and establishes an end-to-end computing power perception and unified scheduling system based on a centralized computing network collaborative scheduling platform. This system fully leverages the flexible programmability and path control capabilities of the SRv6 (Segment Routing over IPv6) protocol to achieve joint perception, dynamic orchestration, and fine-grained scheduling of computing and network resources, thereby improving overall resource utilization efficiency and ensuring business experience and service quality. (3) The core of the computing power network technology route based on domain name resolution is to introduce a computing power scheduling system into the existing Internet DNS (Domain Name System) architecture, and to enable it to query computing power information and respond intelligently by extending the DNS protocol or interface.

[0004] However, the resource forms and user needs in computing power networks exhibit diverse characteristics, specifically reflected in the heterogeneity of resource nodes, the dispersion of resource ownership, the differences in business needs, and the diversity of computing power categories. The types of computing power resources required for various computing power network scenarios also vary significantly. Existing solutions have significant limitations in identifying computing power resource attributes; their description mechanisms lack uniformity and comprehensiveness, making it difficult to systematically cover the multi-dimensional characteristics of different types of resources in terms of architecture, performance, and status. This insufficient identification capability leads to a coarse-grained resource scheduling process, unable to accurately match based on fine-grained attributes, resulting in a mismatch between resource allocation and business needs. Especially when facing business scenarios sensitive to computing power characteristics, such as high-performance computing, real-time rendering, and AI training, coarse scheduling strategies cannot meet their stringent performance requirements such as low latency, high throughput, or hardware acceleration, ultimately affecting business execution efficiency and service experience.

[0005] Therefore, there is an urgent need to design a globally unified resource identification and intelligent resolution method. This method can systematically cover the differentiated types and multi-dimensional characteristics (such as computing architecture, real-time status, performance indicators and service levels) of computing resources in different computing network scenarios, thereby achieving accurate description, efficient discovery and on-demand scheduling of heterogeneous resources, fundamentally improving resource matching accuracy and task execution efficiency, optimizing end-user experience and ensuring the service quality of various services. Summary of the Invention

[0006] To address the significant limitations of existing technologies in identifying computing resource attributes, which lack uniformity and comprehensiveness in their description mechanisms and struggle to systematically cover the multi-dimensional characteristics of different types of resources in terms of architecture, performance, and status, this invention provides an identification-based wide-area heterogeneous computing power resolution method. This method aims to effectively improve the utilization rate of computing resources, optimize user experience, and simultaneously achieve optimized allocation and collaborative management of global computing resources.

[0007] To achieve the above objectives, this invention provides an identifier-based wide-area heterogeneous computing power resolution method, comprising the following steps:

[0008] Step 1: When computing resources access the network, they send a registration request to the computing power analysis server. The request information includes the attribute information of the computing resources.

[0009] Step 2: The computing power resolution server abstracts the computing power resources into a hierarchical tree and encodes them according to the domain name organization rules;

[0010] Step 3: The computing power analysis server registers and manages computing power resources;

[0011] Step 4: If registration is successful, when the user initiates a task request, a dynamic and static matching algorithm is used to search for a list of available computing resources from the computing power analysis server.

[0012] If registration fails, return to step 1 and enter the loop.

[0013] Furthermore, the attribute information of computing resources in step 1 includes computing network attribute information, computing ontology attribute information, and computing service type information.

[0014] Furthermore, if the computing resources are elastic, unloaded resources that have not yet hosted any instances, then one or more of the computing network attribute information, computing ontology attribute information, and computing service type information are allowed to be empty.

[0015] Furthermore, in step 2, the computing power resolution server abstracts the computing power resource hierarchy into a hierarchical tree. The hierarchical tree is divided into ten levels from top to bottom starting from the root node, namely: national domain, computing power domain, computing power type, computing power provider, product code, computing power hardware architecture, computing power model, computing power model, computing power level, and computing power instance name.

[0016] Based on the hierarchical tree, a globally unique computing resource identifier is generated. The naming rule for the computing resource identifier is to concatenate the identifiers in reverse order from the leaf node to the root node. The specific format is: computing instance name.computing level.computing model.computing type.computing hardware architecture.product code.computing provider.computing type.computing domain.country domain.

[0017] Furthermore, in step 2, encoding is performed according to the domain name organization rules, using a type-length-value structure. The encoding rules include:

[0018] Type field: Encoded in fixed-length bytes, used to uniquely identify the category of computing resource attributes;

[0019] Length field: Encoded in fixed-length bytes, it is used to explicitly indicate the length of bytes occupied by the value field that follows, so that the parsing server can accurately locate the start and end positions of each computing resource attribute;

[0020] Value field: Stores the specific numerical value or content of the attribute in variable-length bytes. Its length is explicitly specified by the length field. The content can be an integer, floating-point number, string, or binary sequence.

[0021] Furthermore, the dynamic-static combined matching algorithm in step 4 includes the following steps:

[0022] Step 4.1: Assign weights to each static indicator using the entropy weight method, and score the overall static performance of the node;

[0023] Step 4.2: Use a decision tree classification algorithm to divide nodes into high-performance nodes and low-performance nodes based on the comprehensive score, so as to narrow down the range of resource matching for different user needs;

[0024] Step 4.3: Use the n-dimensional Euclidean distance method to perform joint analysis of dynamic factors to select suitable computing nodes for user tasks.

[0025] Furthermore, step 4.1 uses the entropy weight method to assign weights to each static indicator and performs a comprehensive static performance score for the node, including the following steps:

[0026] Step 4.1.1: Construct the initial matrix: Assuming there are m nodes, and each node has n static indicators, construct the matrix. ,in Representing the The node of the first Item indicator value;

[0027] Step 4.1.2: Data Standardization: Normalize the matrix to eliminate the influence of dimensions;

[0028] Efficiency indicators are expressed as The values ​​of benefit-related indicators are positively correlated with static performance.

[0029] Cost-type indicators are expressed as Growth indicators are negatively correlated with static performance;

[0030] No. Information entropy of the indicator Represented as ,in, , ;

[0031] No. Entropy weight of the item Represented as ;

[0032] Step 4.1.3: Calculate the overall score: the overall static performance score of each node. Represented as ;

[0033] in, This represents the normalized value. Representing the The node of the first Item index value, Representing the The minimum value of the indicator among all nodes. Representing the The maximum value of the indicator across all nodes. Representing the The node at the th The value of an indicator is the proportion of the total value of all nodes of that indicator. ln is the natural logarithm in mathematics, and m is the number of nodes. Make The value range is , For the first Information entropy of the indicator Representing the The entropy weight of the indicator, where n is the number of static indicators. The overall static performance score for each node.

[0034] Furthermore, step 4.3 employs the n-dimensional Euclidean distance method to jointly analyze dynamic factors and select suitable computing nodes for user tasks, including the following steps:

[0035] Step 4.3.1: Map the user's dynamic requirements and the real-time dynamic state of the candidate nodes into points in an n-dimensional space. The user's task requirements can be represented as a vector. The state of a candidate node can be represented as a vector. , This represents the first user task requirement. This represents the second user task requirement. This represents the task requirement of the nth user. This represents the state of the first candidate node. This represents the state of the second candidate node. This represents the state of the nth candidate node;

[0036] Step 4.3.2: Calculate the n-dimensional Euclidean distance between the task requirements and the state of each candidate node. , ,in, Represents the task requirement of the kth user. This represents the state of the k-th candidate node;

[0037] Step 4.3.3: Select the Euclidean distance from the user's required vector. The smallest node is the node with the most suitable computing power for the user task at the current moment, based on dynamic resources.

[0038] Beneficial effects: This invention provides a wide-area heterogeneous computing power resolution method based on identifiers, which has the following technical effects: (1) The design of the computing power identifier system fully considers the core characteristics of heterogeneous and dynamic changes of computing power resources, and can uniformly and accurately describe and represent resource attributes from multiple dimensions such as type, performance, status and ownership. As a unified language for global cognition and access to computing power resources, this identifier is not only the basis for realizing the collaborative management, status synchronization and intelligent scheduling of computing power resources of the whole network, but also directly supports the efficient and accurate matching between computing power service demand and resource supply, thus providing crucial underlying support for global resolution of computing power resources, reliable discovery of services and integrated management and control of computing network. (2) The computing power identifier field and content are designed in TLV format, which can realize the accurate expression and efficient encoding of multi-dimensional attributes of heterogeneous computing power resources in a highly structured and flexibly extensible way. This format clearly defines semantics through the type field and achieves rapid location and extraction through the length field. It not only supports the seamless embedding and future evolution of new computing power attribute types, but also significantly improves the efficiency of computing power matching and resource scheduling, and effectively ensures the unambiguous resolution and cross-domain collaborative management of computing power identifiers in the global scope. (3) In the critical stage of matching user needs with computing power resources, by systematically integrating static indicators that characterize the hardware basic capabilities and dynamic indicators that reflect the real-time operating status, a multi-dimensional collaborative resource measurement system is constructed. This can not only significantly improve the overall utilization efficiency of computing power resources, but also accurately achieve the optimal matching between resources and complex and diverse needs.

[0039] In summary, this invention can effectively improve the utilization rate of computing resources and optimize user experience; at the same time, it can realize the optimized allocation and collaborative management of global computing resources. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of an identifier-based wide-area heterogeneous computing power resolution system according to an embodiment of the present invention;

[0041] Figure 2 This is a flowchart of the identifier-based wide-area heterogeneous computing power parsing method involved in the embodiments of the present invention. Detailed Implementation

[0042] like Figures 1-2 As shown, this invention provides a method for resolving wide-area heterogeneous computing power based on identifiers.

[0043] Example 1: The wide-area heterogeneous computing power resolution system based on identifiers in this embodiment of the invention includes a user terminal, a computing power resolution server, and computing power resources, such as... Figure 1 As shown in the figure. This invention proposes a method for parsing wide-area heterogeneous computing power based on identifiers, the flowchart of which is shown in the figure. Figure 2 As shown.

[0044] I. Registration and Coding of Computing Resources

[0045] When a computing resource connects to the network, it must send a registration request to the computing power resolution server. This request must include its detailed attribute information, covering computing power network attributes, computing power entity attributes, and computing power service type. If the resource is an elastic, unused resource that has not yet hosted any instances, some information may be left blank.

[0046] The computing resources to be identified are comprehensively sorted according to their attribute hierarchy to construct a structured computing resource abstract hierarchy tree. This hierarchy tree, starting from the root node, is divided into the following ten levels from top to bottom: Country Domain, Computing Domain, Computing Type, Computing Provider, Product Code, Computing Hardware Architecture, Computing Model, Computing Model, Computing Level, and Computing Instance Name. Based on this computing resource hierarchy tree, a globally unique computing resource identifier is generated. Its naming rule is the reverse concatenation from the leaf node to the root node, specifically in the format: Computing Instance Name.Computing Level.Computing Model.Computing Model.Computing Hardware Architecture.Product Code.Computing Provider.Computing Type.Computing Domain.Country Domain. If the computing resource is a flexible, idle resource that has not yet undertaken any specific computing tasks, the corresponding instance level in its resource abstract tree can be coded by default. Examples of specific value ranges for each level are as follows:

[0047] Country code: Adopting the ISO 3166-1 standard country code, including: CN (China), US (United States), UK (United Kingdom), DE (Germany), JP (Japan), etc.;

[0048] Computing power domain: Represents the main geographical regions within a country. Taking China as an example, it includes: North China, East China, South China, Central China, Southwest China, Northwest China, and Northeast China.

[0049] Computing power types: Classified by computing unit type, including: CPU (Central Processing Unit), GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), ASIC (Application-Specific Integrated Circuit), NPU (Neural Processing Unit), etc.

[0050] Computing power providers: The service providers or organizations that identify the resources, including: China Mobile, China Unicom, China Telecom, Huawei, Alibaba, Tencent, etc.

[0051] Product codes: based on the internal allocation rules of each manufacturer;

[0052] Computing hardware architecture: Identifies the microarchitecture or instruction set architecture of computing hardware, including: x86_64, ARMv8, MIPS64, RISC-V, PowerPC, etc.

[0053] Computing power model: Identifies the specific model of the hardware, and is assigned according to the specific hardware model;

[0054] Computing power model: Identifies the supported or optimized computing model or framework, including: ResNet-50, YOLOv5, BERT-Large, GPT-3, GCN, LSTM, etc.

[0055] Computing power level: Represents a service level agreement or performance classification. Computing power level 1 (low computing power): computing power ≤ 100 FLOPS, suitable for lightweight computing tasks and edge nodes. Computing power level 2 (medium computing power): computing power 100-1000 FLOPS, suitable for general computing and medium-load services. Computing power level 3 (high computing power): computing power ≥ 1 PFLOPS, suitable for high-performance computing, artificial intelligence training and other heavy-load tasks;

[0056] Computing instance name: A unique name for this computing resource instance, usually an automatically generated string or number.

[0057] II. Designing the fields and content of computing power identifiers using the TLV format.

[0058] The specific encoding of the computing power identifier adopts the Type-Length-Value (TLV) structure. This structure uses three fields, Type, Length, and Value, as basic units. Through the flexible combination and sequential concatenation of multiple units, a complete and scalable computing power resource description identifier is formed, realizing the structured and standardized expression of the multi-dimensional attributes of heterogeneous computing power resources.

[0059] The specific encoding rules are as follows:

[0060] 1. Type field: This field is encoded in fixed-length bytes (usually 1-2 bytes) to uniquely identify the category of computing power attribute, ensuring unambiguous parsing.

[0061] 0x01: Computing hardware type (CPU, GPU, NPU, FPGA, ASIC, etc.);

[0062] 0x02: Computational performance metrics (MIPS, FLOPS, TOPS, etc.);

[0063] 0x03: Storage capacity indicators (RAM capacity, VRAM capacity, hard disk storage capacity, etc.);

[0064] 0x04: Real-time status metrics (CPU utilization, GPU utilization, real-time load rate, power consumption, etc.);

[0065] 0x05: Network performance metrics (network bandwidth, network latency, packet loss rate, etc.);

[0066] 0x06: Location and topology information (node ​​geographic region);

[0067] 0x07: Supported frameworks and instruction sets (supports TensorFlow, PyTorch, ARMv8, x86_64, etc.); ......

[0068] 2. Length field: This field is encoded in fixed-length bytes (usually 1-4 bytes) to explicitly indicate the length of bytes occupied by the following "value field", enabling the parser to accurately locate the start and end positions of each attribute.

[0069] 3. Value Field: This field stores the specific numerical value or content of the attribute in variable-length bytes. Its length is explicitly specified by the "Length Field," and the content can be an integer, floating-point number, string, or binary sequence, etc. For example: when Type=0x02 (MIPS), Value can store a 4-byte integer value of 100000; when Type=0x05 (architecture name), Value can store a UTF-8 string such as "ARMv8."

[0070] III. Resource Matching Algorithm Based on Multidimensional Computing Power Resource Measurement

[0071] This invention's computing resource matching method, taking CPUs and GPUs as examples, constructs a multi-dimensional quantitative system encompassing both static and dynamic indicators, aiming to achieve accurate and efficient resource matching. First, a refined indicator system is built for different types of computing resources, strictly distinguishing between static and dynamic indicators: For CPUs, this embodiment uses the processor's computing power performance indicator TOPS / W and the CPU speed indicator—millions of instructions per second (MIPS)—as two factors in the static CPU indicator, and CPU idle rate as one element of the dynamic indicator. For GPUs, the conventional neural network computing power indicator GOPS and the storage performance indicator RAM are used as factors in the GPU's static indicators, and GPU idle rate is one element of the dynamic indicator.

[0072] The matching algorithm that combines dynamic and static indicators includes the following steps:

[0073] First, the entropy weight method is used to assign weights to each static indicator, and then the node is given a comprehensive static performance score.

[0074] Then, a decision tree classification algorithm was used to divide the nodes into high-performance nodes and low-performance nodes based on the comprehensive score, so as to narrow the range of resource matching for different user needs.

[0075] Finally, the n-dimensional Euclidean distance method is used to jointly analyze dynamic factors to select suitable computing nodes for user tasks.

[0076] The specific steps are as follows:

[0077] 1. To eliminate subjective human factors and objectively evaluate the overall static performance of nodes, the entropy weight method is used to assign weights to each static indicator and score the overall static performance of nodes, including the following steps:

[0078] (1) Constructing the initial matrix: Assuming there are m nodes, and each node has n static indicators, construct the matrix. ,in Representing the The node of the first Item indicator value;

[0079] (2) Data standardization: Normalize the matrix to eliminate the influence of dimensions;

[0080] Efficiency indicators are expressed as The values ​​of benefit-type indicators are positively correlated with static performance, that is, the larger the value of the benefit-type indicator, the better the static performance.

[0081] Cost-type indicators are expressed as The growth index values ​​are negatively correlated with static performance; that is, the smaller the growth index value, the better the static performance.

[0082] No. Information entropy of the indicator Represented as ,in, , ;

[0083] No. Entropy weight of the item Represented as ;

[0084] (3) Calculate the overall score: the overall static performance score of each node. Represented as ;

[0085] in, This represents the normalized value. Representing the The node of the first Item index value, Representing the The minimum value of the indicator among all nodes. Representing the The maximum value of the indicator across all nodes. Representing the The node at the th The value of an indicator is the proportion of the total value of all nodes of that indicator. ln is the natural logarithm in mathematics, and m is the number of nodes. Make The value range is , For the first Information entropy of the indicator Representing the The entropy weight of the indicator, where n is the number of static indicators. The overall static performance score for each node.

[0086] 2. To obtain the overall static score of the nodes. Then, the decision tree algorithm is used to... The primary feature is performance, and other auxiliary features (such as node network location) can be introduced according to business needs to classify nodes into "high-performance" and "low-performance" categories. This classifier can learn the complex nonlinear relationship between scores and performance levels in historical data, and its classification rules can be intuitively represented as a set of judgments in the form of IF-THEN, thus providing a transparent and efficient basis for achieving preliminary, coarse-grained resource screening.

[0087] 3. Finally, after narrowing down the candidate node range through decision tree classification, the final matching decision is determined by dynamic indicators:

[0088] (1) Map the user's dynamic needs (such as the expected idle rate) and the real-time dynamic state of candidate nodes (such as CPU idle rate and GPU idle rate) to points in an n-dimensional space. The user's task needs can be represented as a vector. The state of a candidate node can be represented as a vector. , This represents the first user task requirement. This represents the second user task requirement. This represents the task requirement of the nth user. This represents the state of the first candidate node. This represents the state of the second candidate node. This represents the state of the nth candidate node;

[0089] (2) Calculate the n-dimensional Euclidean distance between the task requirements and the state of each candidate node. , ,in, Represents the task requirement of the kth user. This represents the state of the k-th candidate node;

[0090] (3) Matching strategy: Select the Euclidean distance from the user's demand vector. The smallest node is the "optimal solution" whose dynamic resource status best matches the user's needs at the current moment, thus achieving a precise and fine-grained final match.

[0091] This invention provides an identifier-based method for resolving heterogeneous computing power over a wide area, which can effectively improve the utilization rate of computing resources and optimize user experience; at the same time, it can realize the optimized allocation and collaborative management of global computing resources.

[0092] Finally, it should be noted that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. However, any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for resolving wide-area heterogeneous computing power based on identifiers, characterized in that, Includes the following steps: Step 1: When computing resources access the network, they send a registration request to the computing power analysis server. The request information includes the attribute information of the computing resources. Step 2: The computing power resolution server abstracts the computing power resources into a hierarchical tree and encodes them according to the domain name organization rules; Step 3: The computing power analysis server registers and manages computing power resources; Step 4: If registration is successful, when the user initiates a task request, a dynamic and static matching algorithm is used to search for a list of available computing resources from the computing power analysis server. If registration fails, return to step 1 and enter the loop; The matching algorithm combining static and dynamic elements in step 4 includes the following steps: Step 4.1: Assign weights to each static indicator using the entropy weight method, and score the overall static performance of the node; Step 4.1.1: Construct the initial matrix: Assuming there are m nodes, and each node has n static indicators, construct the matrix. ,in Representing the The node of the first Item indicator value; Step 4.1.2: Data Standardization: Normalize the matrix to eliminate the influence of dimensions; Efficiency indicators are expressed as The values ​​of benefit-related indicators are positively correlated with static performance. Cost-type indicators are expressed as Growth indicators are negatively correlated with static performance; No. Information entropy of the indicator Represented as ,in, , ; No. Entropy weight of the item Represented as ; Step 4.1.3: Calculate the overall score: the overall static performance score of each node. Represented as ; in, This represents the normalized value. Representing the The node of the first Item index value, Representing the The minimum value of the indicator among all nodes. Representing the The maximum value of the indicator across all nodes. Representing the The node at the th The value of an indicator is the proportion of the total value of all nodes of that indicator. ln is the natural logarithm in mathematics, and m is the number of nodes. Make The value range is , For the first Information entropy of the indicator Representing the The entropy weight of the indicator, where n is the number of static indicators. The overall static performance score for each node; Step 4.2: Use a decision tree classification algorithm to divide nodes into high-performance nodes and low-performance nodes based on the comprehensive score, so as to narrow down the range of resource matching for different user needs; Step 4.3: Use the n-dimensional Euclidean distance method to perform joint analysis of dynamic factors to select suitable computing nodes for user tasks; Step 4.3.1: Map the user's dynamic requirements and the real-time dynamic state of the candidate nodes into points in an n-dimensional space. The user's task requirements can be represented as a vector. The state of a candidate node can be represented as a vector. , This represents the first user task requirement. This represents the second user task requirement. This represents the task requirement of the nth user. This represents the state of the first candidate node. This represents the state of the second candidate node. This represents the state of the nth candidate node; Step 4.3.2: Calculate the n-dimensional Euclidean distance between the task requirements and the state of each candidate node. , ,in, Represents the task requirement of the kth user. This represents the state of the k-th candidate node; Step 4.3.3: Select the Euclidean distance from the user's required vector. The smallest node is the node with the most suitable computing power for the user task at the current moment, based on dynamic resources.

2. The identifier-based wide-area heterogeneous computing power resolution method according to claim 1, characterized in that, The attribute information of computing resources in step 1 includes computing network attribute information, computing entity attribute information, and computing service type information.

3. The identifier-based wide-area heterogeneous computing power resolution method according to claim 2, characterized in that, If the computing power resource is an elastic, unloaded resource that has not yet hosted any instances, then one or more of the computing power network attribute information, computing power ontology attribute information, and computing power service type information are allowed to be empty.

4. The identifier-based wide-area heterogeneous computing power resolution method according to claim 1, characterized in that, In step 2, the computing power resolution server abstracts the computing power resource hierarchy into a hierarchical tree. The hierarchical tree is divided into ten levels from top to bottom, starting from the root node: country domain, computing power domain, computing power type, computing power provider, product code, computing power hardware architecture, computing power model, computing power model, computing power level, and computing power instance name. Based on the hierarchical tree, a globally unique computing resource identifier is generated. The naming rule for the computing resource identifier is to concatenate the identifiers in reverse order from the leaf node to the root node. The specific format is: computing instance name.computing level.computing model.computing type.computing hardware architecture.product code.computing provider.computing type.computing domain.country domain.

5. The identifier-based wide-area heterogeneous computing power resolution method according to claim 1, characterized in that, In step 2, encoding is performed according to the domain name organization rules, using a type-length-value structure. The encoding rules include: Type field: Encoded in fixed-length bytes, used to uniquely identify the category of computing resource attributes; Length field: Encoded in fixed-length bytes, it is used to explicitly indicate the length of bytes occupied by the value field that follows, so that the parsing server can accurately locate the start and end positions of each computing resource attribute; Value field: Stores the specific numerical value or content of the attribute in variable-length bytes. Its length is explicitly specified by the length field. The content can be an integer, floating-point number, string, or binary sequence.

Images