Information processing method, apparatus, device, storage medium, and computer program product

Abstract

The application discloses an information processing method, device, equipment, storage medium and computer program product. The method comprises the following steps: acquiring first information sent by a terminal; the first information is used for requesting to use computing power resources; determining a first sub-knowledge graph from a computing power knowledge graph according to the first information; the computing power knowledge graph is constructed based on entities and relationships related to heterogeneous computing power resources; the first sub-knowledge graph comprises entities and relationships related to candidate computing power resources matched with the first information; determining an embedding vector of the first sub-knowledge graph; inputting the embedding vector of the first sub-knowledge graph into a first model trained in advance to perform reasoning, and obtaining a reasoning result; the reasoning result comprises a recommended computing power resource list.

Description

Technical Field

[0001] This application relates to the field of cloud computing technology, and in particular to an information processing method, apparatus, device, storage medium, and computer program product. Background Technology

[0002] Currently, when scheduling computing resources, the system allocates resources to users based on their submitted business requirements. However, due to reasons such as the inability to obtain sufficient information about computing resources, the allocated resources may not be suitable for the user's current business needs. Summary of the Invention

[0003] In view of this, embodiments of this application aim to provide an information processing method, apparatus, device, storage medium, and computer program product.

[0004] The technical solution of this application embodiment is implemented as follows: This application provides an information processing method, the method comprising: Obtain the first information sent by the terminal; the first information is used to request the use of computing resources. Based on the first information, a first subgraph is determined from the computing power knowledge graph; the computing power knowledge graph is constructed based on entities and relationships related to heterogeneous computing power resources; the first subgraph includes entities and relationships related to candidate computing power resources that match the first information; Determine the embedding vector of the first sub-graph; The embedding vector of the first sub-graph is input into the pre-trained first model for inference to obtain the inference result; the inference result includes a recommended list of computing resources.

[0005] Furthermore, according to at least one embodiment of this application, the first information includes at least one of the following: Task type; Computing power requirements; Budget; Delay requirement; Data location.

[0006] Furthermore, according to at least one embodiment of this application, the method further includes: Acquire multi-source data; the multi-source data includes heterogeneous computing resources from multiple cloud vendors; Extract entities and relationships related to the heterogeneous computing resources from the multi-source data; Based on the entities and relationships related to the heterogeneous computing resources, the computing power knowledge graph is constructed.

[0007] Furthermore, according to at least one embodiment of this application, the entity includes at least one of the following: Cloud vendors; Data center; Compute nodes; Heterogeneous computing resources; Computing power type; Computing power inventory; Network links; user; Task type; Service quality indicators; The relationship includes at least one of the following: lie in; supply; connect; support; belong; Historical use; Performance.

[0008] Furthermore, according to at least one embodiment of this application, the method further includes: The metadata description file submitted by the first cloud vendor is obtained through the registration interface; the metadata description file is related to the heterogeneous computing resources of the first cloud vendor. Parse the metadata description file to obtain the metadata of the heterogeneous computing resources of the first cloud vendor; Based on the metadata, determine the entities and relationships related to the heterogeneous computing resources of the first cloud vendor; Update the entities and relationships related to the heterogeneous computing power resources of the first cloud vendor to the computing power knowledge graph.

[0009] Furthermore, according to at least one embodiment of this application, the method further includes: In the event that the heterogeneous computing resources of the cloud provider are updated, determine the entities and relationships related to the updated heterogeneous computing resources of the cloud provider. The computing power knowledge graph is updated based on the entities and relationships related to the updated heterogeneous computing power resources of the cloud vendor.

[0010] Furthermore, according to at least one embodiment of this application, the method further includes: Generate an explanation report; the explanation report is used to explain the reasons for the recommendation. The recommended list of computing resources and the explanation report are sent to the terminal; the heterogeneous computing resources in the recommended list of computing resources carry recommendation weights; the recommendation weights are set according to the preferences of the user corresponding to the terminal; the explanation report is used to instruct the user to select a target heterogeneous computing resource from the list of computing resources. Obtain the target heterogeneous computing resources selected by the user from the recommended list of computing resources; A first request is sent to the cloud platform of the target cloud vendor corresponding to the target heterogeneous computing power resource. The first request is used to request the completion of the creation and configuration of the target heterogeneous computing power resource.

[0011] Furthermore, according to at least one embodiment of this application, the method further includes: Obtain second and third information sent by the cloud platform of the target cloud vendor; the second information includes at least one of the following: the creation completion time of the target heterogeneous computing power resource, and whether the creation of the target heterogeneous computing power resource was successful; the third information includes the current computing power inventory of the target cloud vendor's heterogeneous computing power resources; Based on the second information, a notification message is sent to the terminal; the notification message is used to notify the user corresponding to the terminal whether to use the target heterogeneous computing resources to perform a task. The computing power knowledge graph is updated based on the third information.

[0012] This application provides an information processing apparatus, including: The acquisition module is used to acquire first information; the first information is used to request the use of computing resources. A first processing module is configured to determine a first sub-graph from the computing power knowledge graph based on the first information; the computing power knowledge graph is constructed based on entities and relationships related to heterogeneous computing power resources; the first sub-graph includes entities and relationships related to candidate computing power resources that match the first information. The second processing module is used to determine the embedding vector of the first sub-graph; input the embedding vector of the first sub-graph into a pre-trained first model for inference, and obtain the inference result; the inference result includes a recommended list of computing resources.

[0013] This application provides a network device, including a processor and a memory for storing computer programs that can run on the processor. When the processor is used to run the computer program, it performs the steps of any of the above methods.

[0014] At least one embodiment of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0015] At least one embodiment of this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the methods described above.

[0016] The information processing method, device, storage medium, and computer program product provided in this application embodiment include: acquiring first information sent by a terminal; the first information is used to request the use of computing power resources; determining a first subgraph from a computing power knowledge graph based on the first information; the computing power knowledge graph is constructed based on entities and relationships related to heterogeneous computing power resources; the first subgraph includes entities and relationships related to candidate computing power resources that match the first information; determining the embedding vector of the first subgraph; inputting the embedding vector of the first subgraph into a pre-trained first model for inference to obtain an inference result; the inference result includes a recommended list of computing power resources.

[0017] By employing the technical solution provided in this application, combining the first information sent by the terminal with a pre-constructed computing power knowledge graph, a first sub-graph is determined. Then, a pre-trained first model is used to infer the embedding vectors of the first sub-graph to obtain a recommended list of computing power resources. Since the constructed computing power knowledge graph can describe the entities and relationships related to heterogeneous computing power resources, it can achieve a unified representation of information related to a sufficient number of heterogeneous computing power resources. The recommended list of computing power resources is then used by the trained first model to select target heterogeneous computing power resources for the user. In this way, scheduling the target heterogeneous computing power resources for the user can meet the user's current business needs. Attached Figure Description

[0018] Figure 1 This is a schematic diagram illustrating the implementation flow of the information processing method according to an embodiment of this application; Figure 2 This is a schematic diagram of the composition structure of the information processing device according to an embodiment of this application; Figure 3 This is a schematic diagram of the composition structure of the network device in an embodiment of this application. Detailed Implementation

[0019] Before introducing the technical solutions of the embodiments of this application, the relevant technologies will be introduced first.

[0020] In related technologies, resource scheduling methods in a multi-cloud environment can involve a multi-cloud resource scheduling system matching user requests with a preset rule base, selecting target resources from multiple cloud service providers, and completing deployment. The specific implementation process includes: Step 1, the user submits task requirements; Step 2, the system obtains the current resource status from each cloud platform; Step 3, the system scores and ranks candidate resources according to a preset strategy; Step 4, the system selects the resource with the highest score for deployment.

[0021] In related technologies, computing resource scheduling methods can involve a scheduling system predicting the future resource utilization of each data center based on historical load data, and using a Deep Q-Network (DQN) to determine the optimal scheduling path in conjunction with the Quality of Service (QoS) requirements of user tasks, thereby achieving dynamic allocation of computing resources. The specific implementation process includes: Step 1, collecting historical time-series data such as CPU or GPU utilization, network latency, and task completion time for each data center; Step 2, using a Long Short-Term Memory (LSTM) model to predict resource load for future periods; Step 3, constructing a reinforcement learning environment where the state is resource load and user requests, the action is resource allocation, and the reward function comprehensively considers cost and latency; Step 4, training the DQN model for online scheduling decisions.

[0022] Currently, computing resource scheduling suffers from the following technical shortcomings: First, it is mostly based on static rules or simple load balancing strategies; second, computing resource information is fragmented, meaning that information such as computing power type, performance indicators, network latency, energy consumption, and price is scattered across different cloud vendors and data centers, lacking unified semantic modeling and correlation analysis capabilities; third, dynamic access capabilities are weak, meaning that access from new cloud vendors requires manual configuration of interfaces and policies, lacking automatic identification and capability registration mechanisms; fourth, recommendation intelligence is insufficient, meaning that users rely on manual experience or simple sorting to select computing resources, and personalized recommendations cannot be made based on multi-dimensional requirements such as task type, cost, latency, and reliability; fifth, there is a lack of global view and interpretability, meaning that the computing power scheduling process is black-box, making it difficult for users to understand the reasons for recommendations and for operations and maintenance personnel to trace the decision-making path.

[0023] Based on this, in this application, first information sent by the terminal is obtained; the first information is used to request the use of computing resources; according to the first information, a first sub-graph is determined from the computing power knowledge graph; the computing power knowledge graph is constructed based on entities and relationships related to heterogeneous computing resources; the first sub-graph includes entities and relationships related to candidate computing resources that match the first information; the embedding vector of the first sub-graph is determined; the embedding vector of the first sub-graph is input into a pre-trained first model for reasoning to obtain a reasoning result; the reasoning result includes a recommended list of computing resources.

[0024] See Figure 1 , Figure 1 This is a schematic flowchart illustrating the implementation of the information processing method according to an embodiment of this application, applied to a first system, including steps 101 to 104: Step 101: Obtain the first information sent by the terminal; the first information is used to request the use of computing resources.

[0025] Here, requesting to use computing resources can also be understood as requesting the creation or deployment of computing resources.

[0026] Here, when an end user needs to use the computing power provided by a cloud provider to perform business, the first system can obtain the computing power usage request sent by the terminal, i.e., the first request, which carries the first information.

[0027] In practical applications, the first information can describe the user's computing power requirements, specifically including at least one of the following: task type, computing power requirements, budget, latency requirements, and data location. The first information can be used to query candidate computing power resources from the constructed computing power knowledge graph.

[0028] Based on this, in some embodiments, the first information includes at least one of the following: Task type; Computing power requirements; Budget; Delay requirement; Data location.

[0029] Here, task types can include, but are not limited to, computational tasks and tasks related to artificial intelligence (AI), such as scientific computing, high-performance computing, AI training, AI inference (or described as real-time inference), etc.

[0030] Here, computing power requirements can refer to the computing power required by the user, which may include, but is not limited to, the number of CPU cores, CPU model, GPU model, Tensor Processing Unit model, and memory.

[0031] Here, the budget can refer to the estimated cost incurred by end users in using computing power.

[0032] Here, latency requirement can refer to the maximum response time for executing a task using computing power.

[0033] Here, data location can refer to the storage location of training data, the storage location of inference data, the storage location of computational data, etc.

[0034] Step 102: Based on the first information, determine the first subgraph from the computing power knowledge graph; the computing power knowledge graph is constructed based on entities and relationships related to heterogeneous computing power resources; the first subgraph includes entities and relationships related to candidate computing power resources that match the first information.

[0035] Here, the entities and relationships related to heterogeneous computing resources can be understood as the various entities related to heterogeneous computing resources and the relationships between them. The various entities related to heterogeneous computing resources can include the heterogeneous computing resources themselves and other entities related to them.

[0036] Here, heterogeneous computing resources can refer to computing resources of different types and architectures. For example, computing resources composed of different types of computing units such as CPUs, GPUs, and TPUs, or computing resources composed of GPUs of architecture 1 and GPUs of architecture 2.

[0037] Here, the computing power knowledge graph may include nodes and edges, wherein each entity related to heterogeneous computing power resources can correspond to each node in the computing power knowledge graph, and the relationship between each entity related to heterogeneous computing power resources can correspond to the edge in the computing power knowledge graph.

[0038] Here, the computing power knowledge graph can be used to describe the various entities related to heterogeneous computing power resources and the relationships between these entities. Specifically, it can refer to a semantic network that organizes and expresses knowledge in the form of triples of entities, relations, and entities.

[0039] Here, determining the first subgraph from the computing power knowledge graph may include: performing a multi-hop query in the computing power knowledge graph to filter a set of candidate computing power resources that match the first information; and obtaining the first subgraph based on the entities and relationships associated with each candidate computing power resource in the candidate computing power set.

[0040] Specifically, based on the first information, multiple starting entities can be determined from the computing power knowledge graph. For each starting entity, a multi-hop query is performed to obtain the entities related to the candidate computing power resources and the relationships between them. The entities related to the candidate computing power resources and the relationships between them constitute the first sub-graph.

[0041] Here, the entities and relationships related to the candidate computing resources that match the first information can also be described as the context related to the candidate computing resources, which can refer to the various entities related to the candidate computing resources and the relationships between them. The various entities related to the candidate computing resources can include the candidate computing resources themselves and other entities related to the candidate computing resources.

[0042] In practical applications, a computing power knowledge graph can be constructed that contains entities and relationships related to heterogeneous computing power resources from multiple cloud vendors. This enables a unified representation and associated storage of multi-source heterogeneous computing power resources, supports the construction of a global view of cross-domain resources, and allows for quick, comprehensive, and accurate querying of the required candidate computing power resources.

[0043] Based on this, in some embodiments, the method further includes: Acquire multi-source data; the multi-source data includes heterogeneous computing resources from multiple cloud vendors; Extract entities and relationships related to the heterogeneous computing resources from the multi-source data; Based on the entities and relationships related to the heterogeneous computing resources, the computing power knowledge graph is constructed.

[0044] Here, acquiring multi-source data can be understood as acquiring the multi-source data from multiple data sources. These multiple data sources include cloud vendors' cloud platforms, monitoring systems used to monitor cloud vendors' cloud platforms, configuration management databases (CMDBs), and user feedback data, etc.

[0045] Here, the relationship can refer to the relationship between various entities related to heterogeneous computing resources.

[0046] Here, extracting entities and relationships related to the heterogeneous computing resources from the multi-source data, and constructing the computing power knowledge graph based on the entities and relationships related to the heterogeneous computing resources, may include: extracting each entity and the relationships between each entity related to the heterogeneous computing resources from the multi-source data; generating multiple triples based on the extracted entities and the relationships between each entity; and constructing the computing power knowledge graph based on the multiple triples.

[0047] Here, the computing power knowledge graph may also include attribute information of each entity. For example, the attribute information of the entity corresponding to heterogeneous computing power resources may include price, region, etc. The region may refer to the resource pool where the heterogeneous computing power resources are located, and the resource pool may belong to a certain domain.

[0048] In some embodiments, the entity includes at least one of the following: Cloud vendors; Data center; Compute nodes; Heterogeneous computing resources; Computing power type; Computing power inventory; Network links; user; Task type; Service quality indicators; The relationship includes at least one of the following: lie in; supply; connect; support; belong; Historical use; Performance.

[0049] Here, cloud providers are used to offer heterogeneous computing resources.

[0050] Here, the data center is used to deploy computing nodes.

[0051] Here, compute nodes are used to store heterogeneous computing resources from cloud providers, and the resource pool containing these heterogeneous computing resources can belong to a certain domain.

[0052] Here, heterogeneous computing resources can refer to computing power resources of different types and architectures. For example, computing power resources composed of different types of computing units such as CPU, GPU, and TPU, or computing power resources composed of GPUs of architecture 1 and GPUs of architecture 2.

[0053] Here, computing power type can be described as computing power specification, including but not limited to CPU, GPU, TPU, etc.

[0054] Here, computing power inventory can refer to the current number of CPUs, GPUs, TPUs, etc. stored in the database.

[0055] Here, network links can be described as network topology, which can refer to a link between one computing node and another computing node, or a link between one data center and another data center.

[0056] Here, "user" refers to an end user who requests to use or deploy computing resources.

[0057] Here, the task type can include, but is not limited to, AI inference, AI training, computation, and other tasks.

[0058] Here, service quality metrics may include, but are not limited to, the latency of computing nodes executing tasks.

[0059] Here, constructing a computing power knowledge graph can specifically include: Step 1: Define the computing power knowledge graph ontology model, i.e., define entity types. Entity types can include at least one of the following: cloud vendor, data center, computing node, heterogeneous computing power resources, computing power type (CPU, GPU, TPU, etc.), computing power inventory, network link, user, task type, quality of service (QoS) metrics, etc.

[0060] Step 2, define the relationship type. The relationship type can include at least one of the following: located at, provide, join, support, belong to, historical use, performance characteristics, etc.

[0061] Here, taking cloud vendor 1 from among multiple cloud vendors as an example, the relationships between the various entities can include: Cloud vendor 1 provides heterogeneous computing resources 1; Computing node 1 provides heterogeneous computing resources 1 from cloud vendor 1; Compute node 1 is located in the data center, i.e., server room 1; Compute node 1 provides tasks of AI inference type; Compute node 1 is connected to compute node 2; Computing node 2 provides heterogeneous computing resources 2 from cloud vendor 2; The network link between compute node 1 and compute node 2 is Ethernet; Compute node 1 supports the user terminal user 1; Historically, computing node 1 has used a combination of CPU and GPU computing power. The performance metric for computing node 1 is that the response latency of executing a task using heterogeneous computing resources 1 is less than the value 1.

[0062] Step 3: Obtain multi-source data from various cloud vendors' cloud platforms (e.g., accessing cloud platforms via application programming interfaces (APIs), monitoring systems used to monitor cloud vendors' cloud platforms, configuration management databases (CMDBs), user feedback data, and other sources. The multi-source data includes heterogeneous computing resources from multiple cloud vendors. According to defined entity types and relationship types, extract entities and relationships related to the heterogeneous computing resources from the multi-source data. Generate multiple triples based on the extracted entities and the relationships between them. Construct an initial computing power knowledge graph based on the multiple triples.

[0063] Here, the computing power knowledge graph can be stored in a graph database (such as the single-machine graph database Neo4j or the distributed graph database JanusGraph), which supports efficient graph query and reasoning.

[0064] In practical applications, in order to support dynamic access by cloud vendors, new cloud vendors can submit metadata description files to the first system through the registration interface. Based on the metadata description files, the first system obtains the metadata of the new cloud vendor's heterogeneous computing resources, thereby obtaining the entities and relationships related to the new cloud vendor's heterogeneous computing resources according to the metadata, and updating the computing power knowledge graph, thereby realizing the unified representation and associated storage of new heterogeneous computing resources.

[0065] Based on this, in some embodiments, the method further includes: obtaining a metadata description file submitted by a first cloud vendor through a registration interface; the metadata description file is related to the heterogeneous computing resources of the first cloud vendor; Parse the metadata description file to obtain the metadata of the heterogeneous computing resources of the first cloud vendor; Based on the metadata, determine the entities and relationships related to the heterogeneous computing resources of the first cloud vendor; Update the entities and relationships related to the heterogeneous computing power resources of the first cloud vendor to the computing power knowledge graph.

[0066] Here, the first cloud vendor can refer to a new cloud vendor.

[0067] Here, the metadata description file can be in the format of related data JSON (JSON-LD), etc.

[0068] Here, the metadata description file can be used to store the metadata of the heterogeneous computing resources of the first cloud provider. The metadata can describe the various entities related to the heterogeneous computing resources of the first cloud provider and the relationships between the various entities.

[0069] Here, the metadata may also include at least one of the following: The geographical distribution of heterogeneous computing resources provided by the first cloud provider; The billing model for heterogeneous computing resources provided by the first cloud provider; API access method.

[0070] Here, the geographical distribution can refer to the computing nodes where the heterogeneous computing resources provided by the first cloud vendor are located, or the resource pool where the heterogeneous computing resources provided by the first cloud vendor are located.

[0071] Here, the billing model may refer to the billing method for the use of heterogeneous computing resources provided by the first cloud provider.

[0072] Here, API access method can refer to the API interface provided by the cloud platform of the first cloud vendor for external information exchange.

[0073] Here, when the first system obtains the API access method, it can also verify the availability of the API interface provided by the first cloud vendor. If the API interface is available, it can then access the cloud platform of the first cloud vendor through the API interface and start a timed collection task to continuously obtain the real-time computing power inventory, the creation completion time of computing power resources and other dynamic attribute information of the first cloud vendor through the API interface, and update the real-time computing power inventory of the first cloud vendor to the computing power knowledge graph.

[0074] In practical applications, in order to support the automatic detection of vendor capability changes (such as expansion and decommissioning) and the updating of the computing power knowledge graph, when the heterogeneous computing power resources of cloud vendors are updated, the entities and relationships related to the updated heterogeneous computing power resources can be updated in the computing power knowledge graph, thereby ensuring the accuracy of the computing power knowledge graph.

[0075] Based on this, in some embodiments, the method further includes: In the event that the heterogeneous computing resources of the cloud provider are updated, determine the entities and relationships related to the updated heterogeneous computing resources of the cloud provider. The computing power knowledge graph is updated based on the entities and relationships related to the updated heterogeneous computing power resources of the cloud vendor.

[0076] Here, when the heterogeneous computing resources of the cloud vendor are updated, determining the entities and relationships related to the updated heterogeneous computing resources of the cloud vendor; updating the computing power knowledge graph based on the entities and relationships related to the updated heterogeneous computing resources of the cloud vendor may include: obtaining the metadata description file submitted by the cloud vendor through the API interface provided by the cloud vendor when the heterogeneous computing resources of the cloud vendor are updated; the metadata description file is related to the updated heterogeneous computing resources of the cloud vendor; parsing the metadata description file to obtain the metadata of the updated heterogeneous computing resources of the cloud vendor; determining the entities and relationships related to the updated heterogeneous computing resources of the cloud vendor based on the metadata; and updating the entities and relationships related to the updated heterogeneous computing resources of the cloud vendor to the computing power knowledge graph.

[0077] Here, the metadata description file can be used to store the metadata of the cloud provider's updated heterogeneous computing resources. The metadata can describe the various entities related to the cloud provider's updated heterogeneous computing resources and the relationships between these entities.

[0078] Step 103: Determine the embedding vector of the first sub-map.

[0079] Here, determining the embedding vector of the first sub-map may include: Feature extraction is performed on the text corresponding to the entities related to the candidate computing resources to obtain the first text embedding vector corresponding to the entity; Feature extraction is performed on the text corresponding to the relationship between the entities related to the candidate computing power resources to obtain the second text embedding vector corresponding to the relationship; The first text embedding vector and the second text embedding vector are fused to obtain the first fused embedding vector; The first fused embedding vector is used as the embedding vector of the first sub-map.

[0080] Here, there can be multiple entities associated with the candidate computing resources.

[0081] Here, the entities related to the candidate computing resources and the relationships between them can be described as the context related to the candidate computing resources, which can refer to the various entities related to the candidate computing resources and the relationships between them. The various entities related to the candidate computing resources can include the candidate computing resources themselves and other entities related to the candidate computing resources.

[0082] Here, fusing the first text embedding vector and the second text embedding vector can mean concatenating and fusing the first text embedding vector and the second text embedding vector.

[0083] Step 104: Input the embedding vector of the first sub-graph into the pre-trained first model for inference to obtain the inference result; the inference result includes a recommended list of computing resources.

[0084] Here, inputting the embedding vector of the first sub-graph into a pre-trained first model for inference can include: inputting the first fused embedding vector into the pre-trained first model for inference. The first fused embedding vector is obtained based on a first text embedding vector and a second text embedding vector. The first text embedding vector is obtained by extracting features from the text corresponding to the entities related to the candidate computing resources, and the second text embedding vector is obtained by extracting features from the text corresponding to the relationships between the entities related to the candidate computing resources.

[0085] Here, the first model can be a multi-objective model based on a multilayer perceptron architecture. A multi-objective model refers to a model that considers multiple task objectives in an optimization problem. A multi-objective model may include a shared feature extraction layer and multiple independent output heads. The first fused embedding vector is input into the pre-trained first model, and each independent output head outputs a predicted value corresponding to its respective task objective. The predicted values ​​are then weighted and summed to obtain a comprehensive score for the candidate computing resources. Based on this comprehensive score, the candidate computing resources are ranked to obtain a recommended list of computing resources. The multiple task objectives may include cost, latency, reliability, energy consumption, load balancing, etc.

[0086] Here, the first model can be obtained by training an initial model based on multiple task objectives and training data. The training data may include entities and relationships related to historical heterogeneous computing resources from various cloud vendors. The training process may involve: extracting features from the text corresponding to entities related to historical heterogeneous computing resources to obtain text embedding vectors corresponding to the entities; extracting features from the text corresponding to the relationships between entities related to historical heterogeneous computing resources to obtain text embedding vectors corresponding to the relationships; fusing the text embedding vectors corresponding to the entities and the text embedding vectors corresponding to the relationships, such as by concatenation, to obtain a second fused embedding vector; and using this second fused embedding vector to train the initial model to obtain the first model. Multiple task objectives may include cost, latency, reliability, energy consumption, load balancing, etc.

[0087] In practical applications, after obtaining a recommended list of computing resources by combining the computing power knowledge graph and the first model, to enable users to quickly and effectively select from multiple heterogeneous computing resources listed, the recommended list and the reasons for the recommendation can be sent to the terminal. After obtaining the target heterogeneous computing resources selected by the user, the target cloud vendor can be requested to create and configure those resources. Allowing users to participate in the decision-making process for target heterogeneous computing resources can improve user satisfaction.

[0088] Based on this, in some embodiments, the method further includes: Generate an explanation report; the explanation report is used to explain the reasons for the recommendation. The recommended list of computing resources and the explanation report are sent to the terminal; the heterogeneous computing resources in the recommended list of computing resources carry recommendation weights; the recommendation weights are set according to the preferences of the user corresponding to the terminal; the explanation report is used to instruct the user to select a target heterogeneous computing resource from the list of computing resources. Obtain the target heterogeneous computing resources selected by the user from the recommended list of computing resources; A first request is sent to the cloud platform of the target cloud vendor corresponding to the target heterogeneous computing power resource. The first request is used to request the completion of the creation and configuration of the target heterogeneous computing power resource.

[0089] Here, the explanatory report can also be described as an interpretable recommendation results report.

[0090] Here, recommendation weight can also be described as recommendation weight value.

[0091] Here, based on the entities and relationships related to each heterogeneous computing resource in the computing power resource list within the computing power knowledge graph, relevant information about each heterogeneous computing resource in the computing power resource list can be obtained, and recommendation reasons can be generated based on the obtained relevant information. The relevant information may include the response latency of using the heterogeneous computing resource to perform business operations, the GPU utilization rate of the computing node corresponding to the heterogeneous computing resource, the historical success rate of using the heterogeneous computing resource to perform tasks, etc. The generated recommendation reasons may include response latency less than value 1, GPU utilization greater than value 2, historical task success rate greater than value 3, etc.

[0092] Here, the explanation report may include the reasons for recommending each heterogeneous computing resource in the computing resource list.

[0093] Here, the method further includes: obtaining the user's preference information corresponding to the terminal; and setting the recommendation weights corresponding to each heterogeneous computing resource in the recommended computing resource list based on the preference information.

[0094] Here, user preferences can include prioritizing proximity, low latency, and so on. The closer heterogeneous computing resources match user preferences, the higher their recommendation weight. For example, the closer the deployment location of the heterogeneous computing resources is to the terminal, the higher their recommendation weight; or, if the cloud vendor to which the heterogeneous computing resources belong meets user expectations, the higher their recommendation weight.

[0095] In practical applications, after the target cloud vendor's cloud platform completes the creation and configuration of the target heterogeneous computing resources, it can also obtain the second and third information sent by the target cloud vendor's cloud platform through the API interface provided by the target cloud vendor's cloud platform, and update the computing power knowledge graph according to the second information.

[0096] Based on this, in some embodiments, the method further includes: Obtain second and third information sent by the cloud platform of the target cloud vendor; the second information includes at least one of the following: the creation completion time of the target heterogeneous computing power resource, and whether the creation of the target heterogeneous computing power resource was successful; the third information includes the current computing power inventory of the target cloud vendor's heterogeneous computing power resources; Based on the second information, a notification message is sent to the terminal; the notification message is used to notify the user corresponding to the terminal whether to use the target heterogeneous computing resources to perform a task. The computing power knowledge graph is updated based on the third information.

[0097] Here, the task includes at least one of the following: AI training task, AI inference task, and computation task.

[0098] Here, if it is determined from the second information that the target heterogeneous computing power resource has been successfully created, the notification message is used to notify the user corresponding to the terminal to use the target heterogeneous computing power resource to perform a task.

[0099] Here, if it is determined from the second information that the target heterogeneous computing power resource has not been successfully created, the notification message is used to notify the user corresponding to the terminal not to use the target heterogeneous computing power resource to perform the task, and instructs the user to reselect a new target heterogeneous computing power resource from the recommended computing power resource list. After obtaining the target heterogeneous computing power resource reselected by the user from the recommended computing power resource list, a first request is sent to the cloud platform of the target cloud vendor corresponding to the reselected target heterogeneous computing power resource. The first request is used to request the completion of the creation and configuration of the target heterogeneous computing power resource.

[0100] Here, updating the computing power knowledge graph based on the third information can ensure the accuracy of the computing power knowledge graph, thereby improving the accuracy of recommendations to users.

[0101] Here, the process of creating the target heterogeneous computing resources may include: Step 1: Construct a computing power knowledge graph.

[0102] Specifically, this may include: Step 1: Define the computing power knowledge graph ontology model, i.e., define entity types. Entity types may include at least one of the following: cloud vendor, data center, computing node, heterogeneous computing power resources, computing power type (CPU and / or GPU and / or TPU, etc.), computing power inventory, network link, user, task type, quality of service (QoS) metrics, etc.

[0103] Step 2, define the relationship type. The relationship type can include at least one of the following: located at, provide, join, support, belong to, historical use, performance characteristics, etc.

[0104] Step 3: Obtain multi-source data from various cloud vendors' cloud platforms (e.g., accessing cloud platforms via application programming interfaces (APIs), monitoring systems, configuration management databases (CMDBs), user feedback data, and other sources. The multi-source data includes heterogeneous computing resources from multiple cloud vendors. Extract the entities and relationships between these entities related to the heterogeneous computing resources from the multi-source data. Generate multiple triples based on the extracted entities and their relationships. Construct an initial computing power knowledge graph based on these triples.

[0105] Here, the computing power knowledge graph can be stored in a graph database (such as the single-machine graph database Neo4j or the distributed graph database JanusGraph), which supports efficient graph query and reasoning.

[0106] Here, dynamic access from cloud vendors is supported. New cloud vendors submit metadata description files (e.g., JSON-LD format) through a registration interface. These metadata description files are related to the new cloud vendor's heterogeneous computing resources and may include geographic distribution, billing models, API access methods, etc. The metadata is obtained by parsing the metadata description files. Based on the entities related to the new cloud vendor's heterogeneous computing resources and the relationships between these entities, the corresponding entities (nodes) and relationships (edges) in the computing power knowledge graph are automatically generated, and the availability of the API interface provided by the new cloud vendor is verified. After accessing the new cloud vendor's cloud platform through the API interface, a scheduled data collection task is initiated to continuously obtain real-time dynamic attribute information such as computing power inventory, latency, and availability from the new cloud vendor through the API interface, and the obtained real-time computing power inventory is updated in the computing power knowledge graph.

[0107] Here, automatic detection of vendor capability changes (such as expansion or decommissioning) and updates to the computing power knowledge graph are supported. When the heterogeneous computing power resources of cloud vendors are updated, the entities and relationships related to the updated heterogeneous computing power resources can be updated in the computing power knowledge graph.

[0108] Step two: The first system (computing power network orchestration system) receives a computing power usage request, i.e., a first request, sent by the user. The first request may include at least one of the following: task type (such as AI training, scientific computing, real-time inference), computing power requirements (number of CPU cores, GPU model, memory, etc.), budget, latency requirements, data location, etc. Based on the computing power knowledge graph, a multi-hop query is performed to filter a set of candidate computing power resources that meet the basic constraints, i.e., satisfy the first information. The set of candidate computing power resources includes at least one candidate computing power resource. For each candidate computing power resource, related entities and relationships can be filtered from the computing power knowledge graph. The entities and relationships related to each candidate computing power resource constitute the first subgraph in the computing power knowledge graph. The entities and relationships related to the candidate computing power resources can also be described as the context of the candidate computing power resources.

[0109] Here, determining the first subgraph from the computing power knowledge graph may include: performing a multi-hop query in the computing power knowledge graph to filter a set of candidate computing power resources that match the first information; and obtaining the first subgraph based on the entities and relationships associated with each candidate computing power resource in the candidate computing power set.

[0110] Specifically, based on the first information, multiple starting entities can be determined from the computing power knowledge graph. For each starting entity, a multi-hop query is performed to obtain the entities related to the candidate computing power resources and the relationships between them. The entities related to the candidate computing power resources and the relationships between them constitute the first sub-graph.

[0111] Step 3: The first system (computing power network orchestration system) encodes the entities and relationships related to the candidate computing power resources included in the first sub-graph into embedding vectors.

[0112] Specifically, this may include: extracting features from the text corresponding to the entities related to the candidate computing resources to obtain a first text embedding vector corresponding to the entity; extracting features from the text corresponding to the relationship between the entities related to the candidate computing resources to obtain a second text embedding vector corresponding to the relationship; fusing the first text embedding vector and the second text embedding vector to obtain a first fused embedding vector; and using the first fused embedding vector as the embedding vector of the first sub-graph.

[0113] Here, there can be multiple entities associated with the candidate computing resources.

[0114] Here, the entities associated with the candidate computing resources and the relationships between them can be described as the context related to the candidate computing resources.

[0115] Here, graph embedding technology can also be used to encode the entities and relationships related to the candidate computing resources included in the first subgraph into embedding vectors.

[0116] Step four: The first system (computing power network orchestration system) inputs the embedding vector of the first sub-graph into the multi-objective model (or describes it as a multi-objective optimization model), and generates the optimal computing power resource list by comprehensively considering factors such as cost, latency, reliability, energy consumption, and load balancing.

[0117] Here, the multi-objective model can be obtained by training an initial model based on multiple task objectives and training data. The training data may include historical heterogeneous computing resources related to entities and relationships from various cloud vendors. The training process may include feature extraction of text corresponding to entities related to historical heterogeneous computing resources to obtain text embedding vectors for those entities; feature extraction of text corresponding to relationships between those entities to obtain text embedding vectors for those relationships; fusion of the text embedding vectors for the entities and the relationships to obtain a second fused embedding vector; and training the initial model using this second fused embedding vector to obtain the first model. Multiple task objectives may include cost, latency, reliability, energy consumption, load balancing, etc.

[0118] Step 5: The first system (computing power network orchestration system) generates an interpretable recommendation result report, i.e., an interpretation report.

[0119] Here, the recommendation results can be reverse-engineered to extract key influencing factors. That is, based on the entities and relationships related to each heterogeneous computing resource in the computing power resource list in the computing power knowledge graph, relevant information about each heterogeneous computing resource in the computing power resource list can be obtained. Recommendation reasons can be generated based on the obtained relevant information, and a natural language explanation report can be generated to explain the recommendation reasons to the user. These reasons may include that the GPU utilization rate of the computing node corresponding to the heterogeneous computing resource is less than 30%, the network latency to the user end is less than 20ms, and the historical success rate of executing tasks using the heterogeneous computing resource is 99.8%.

[0120] Step six: The first system (computing power network orchestration system) sends the computing power resource list and the explanation report to the terminal; the heterogeneous computing power resources in the recommended computing power resource list carry recommendation weights; the recommendation weights are set according to the preferences of the user corresponding to the terminal; the explanation report is used to instruct the user to select a target heterogeneous computing power resource from the computing power resource list.

[0121] Here, users can adjust or set the recommendation weights of heterogeneous computing resources in the computing resource list according to their preferences. For example, assuming that the user prefers to prioritize low cost or low latency, the recommendation weight of heterogeneous computing resources with lower prices is relatively large, or the recommendation weight of heterogeneous computing resources with lower service response latency is relatively large; and the computing resource list carrying the recommendation weights and the explanation report are sent to the terminal.

[0122] Step 7: The first system (computing network orchestration system) obtains the target heterogeneous computing resources selected by the user from the list of computing resources; based on the target heterogeneous computing resources finally selected by the user, the system forwards the creation and configuration request of the target heterogeneous computing resources, i.e., the first request, to the cloud platform of the corresponding target cloud vendor through the API interface to complete the creation and configuration of the target heterogeneous computing resources.

[0123] Here, the target cloud vendor's cloud platform executes resource orchestration and feedback loop.

[0124] Here, the first system (computing power network orchestration system) monitors the task execution process and collects actual performance data, namely, obtaining the second and third information sent by the target cloud vendor's cloud platform. The second information includes at least one of the following: the completion time of the creation of the target heterogeneous computing power resource, and whether the creation of the target heterogeneous computing power resource was successful. The third information includes the current computing power inventory of the target cloud vendor's heterogeneous computing power resources. Based on the second information, a notification message is sent to the terminal. The notification message is used to notify the user corresponding to the terminal whether to use the target heterogeneous computing power resource to execute a task. For example, if it is determined from the second information that the target heterogeneous computing power resource was successfully created, the user corresponding to the terminal is notified to use the target heterogeneous computing power resource to execute a task such as an AI inference task. Based on the third information, the computing power knowledge graph is updated, for example, updating the current computing power inventory of the target cloud vendor in the computing power knowledge graph, thereby forming a continuous feedback loop.

[0125] The embodiments of this application have the following advantages: (1) A computing network orchestration method is provided that can integrate multi-source heterogeneous computing power information, support dynamic expansion, and have intelligent recommendation and interpretability capabilities.

[0126] In this application, the first information sent by the terminal and a pre-constructed computing power knowledge graph are combined to determine a first sub-graph. Then, a pre-trained first model is used to infer the embedding vectors of the first sub-graph to obtain a recommended list of computing power resources. Since the constructed computing power knowledge graph can describe the entities and relationships related to heterogeneous computing power resources, it can achieve a unified representation of information related to a sufficient number of heterogeneous computing power resources. The pre-trained first model then uses this information to obtain a recommended list of computing power resources, which is then recommended to the user. The user subsequently participates in selecting the target computing power resources, thus ensuring that the selected resources meet the user's current business needs.

[0127] This application constructs a computing power knowledge graph that includes entities and relationships such as cloud vendors, data centers, computing nodes, computing power types, network links, and service quality. This enables a unified semantic representation and associated storage of multi-source heterogeneous computing power resources, supporting the construction of a global view of cross-domain resources. In other words, by constructing a unified computing power knowledge graph, heterogeneous computing power resources can be semantically modeled and analyzed for association, achieving a unified view and intelligent matching of resources across regions and types, significantly improving resource utilization and scheduling efficiency.

[0128] This application deeply integrates a computing power knowledge graph with an artificial intelligence (AI) model, preserving data-driven intelligence while enhancing the interpretability of the decision-making process. Recommendation results can be traced back to specific paths and entity attributes within the computing power knowledge graph (e.g., recommendations based on low latency and high historical success rates), addressing the limitation of pure machine learning models in explaining why a particular resource was selected, thus improving user trust and operational transparency. Specifically, by performing reverse path tracing on the recommendation results, key influencing factors (such as low latency and high availability) are extracted, generating a natural language explanation report, thereby enhancing user trust and decision-making transparency.

[0129] In this application, dynamic access and automatic capability registration by cloud vendors are supported. New cloud vendors can submit interface capabilities and resource attributes through metadata description files. The first system automatically parses and generates nodes (entities) and edges (relationships) in the computing power knowledge graph, enabling plug-and-play access without the need for manual configuration of scheduling rules.

[0130] This application abandons the reliance on manually pre-set, difficult-to-maintain static rule bases and instead constructs a semantic, reasonable computing power knowledge graph. This graph not only uniformly represents multi-source heterogeneous computing power resources but also enables intelligent matching under complex conditions through multi-hop relationship reasoning, such as cloud vendor-data center-node-performance-user evaluation. This gives the system stronger expressive power and flexibility, allowing it to automatically adapt to diverse user needs and avoiding the problems of rule explosion and high maintenance costs.

[0131] In related technologies, resource scheduling methods in multi-cloud environments involve multi-cloud resource scheduling systems that match user requests using a pre-defined rule base, selecting target resources from multiple cloud service providers and completing deployment. However, this approach has the following drawbacks: it only stores cloud resources in a structured table format, lacking semantic relationships between entities; the pre-defined rule base is manually set and cannot be dynamically adjusted based on user behavior, task type, or environmental changes, nor does it support multi-hop inference under complex conditions; new cloud vendors require manual configuration of API interfaces and rule mappings, failing to achieve automated registration and capability identification; the system only outputs the optimal resource but cannot explain why it is optimal, and does not support causal explanations based on graph paths. In contrast, computing resource scheduling methods in related technologies can involve a computing resource scheduling system predicting the future resource utilization of each data center using historical load data, and combining this with the user task's Quality of Service (QoS) requirements to use a Deep Q Network (DQN) to determine the optimal scheduling path, thereby achieving dynamic allocation of computing resources. However, this solution has the following drawbacks: it abstracts computing resources into state vectors, fails to establish topological and semantic relationships between resources, and cannot express structured knowledge such as a node belonging to a certain vendor or a link being affected by geographical disasters; it lacks global semantic understanding capabilities, relying on historical data statistical patterns, making it difficult to handle cold start problems such as new vendor access and new task types, and also unable to perform logical reasoning; it does not support dynamic vendor access, meaning that model training depends on a fixed resource set, and the addition of a new vendor requires re-collecting data and retraining the model, resulting in long cycles and high costs; the decision-making process is a black box, meaning that the model outputs actions but does not provide explanations for the decision path, making it difficult for users to understand the reasons for recommendations, and operation and maintenance personnel cannot trace why the resource was selected, which is not conducive to troubleshooting and strategy optimization.

[0132] To implement the information processing method of this application embodiment, this application embodiment also provides an information processing device, which is installed in a network device. Figure 2 This is a schematic diagram of the composition structure of the information processing device according to an embodiment of this application, as shown below. Figure 2 As shown, the device includes: The acquisition module 21 is used to acquire first information sent by the terminal; the first information is used to request the use of computing resources. The first processing module 22 is configured to determine a first sub-graph from the computing power knowledge graph based on the first information; the computing power knowledge graph is constructed based on entities and relationships related to heterogeneous computing power resources; the first sub-graph includes entities and relationships related to candidate computing power resources that match the first information. The second processing module 23 is used to determine the embedding vector of the first sub-graph; input the embedding vector of the first sub-graph into a pre-trained first model for inference, and obtain the inference result; the inference result includes a recommended list of computing resources.

[0133] In some embodiments, the first information includes at least one of the following: Task type; Computing power requirements; Budget; Delay requirement; Data location.

[0134] In some embodiments, the acquisition module 21 is further configured to: acquire multi-source data; the multi-source data includes heterogeneous computing power resources from multiple cloud vendors; the first processing module 22 is further configured to: extract entities and relationships related to the heterogeneous computing power resources from the multi-source data; and construct the computing power knowledge graph based on the entities and relationships related to the heterogeneous computing power resources.

[0135] In some embodiments, the entity includes at least one of the following: Cloud vendors; Data center; Compute nodes; Heterogeneous computing resources; Computing power type; Computing power inventory; Network links; user; Task type; Service quality indicators; The relationship includes at least one of the following: lie in; supply; connect; support; belong; Historical use; Performance.

[0136] In some embodiments, the acquisition module 21 is further configured to: acquire a metadata description file submitted by the first cloud vendor through a registration interface; the metadata description file is related to the heterogeneous computing power resources of the first cloud vendor; the first processing module 22 is further configured to: parse the metadata description file to obtain the metadata of the heterogeneous computing power resources of the first cloud vendor; determine the entities and relationships related to the heterogeneous computing power resources of the first cloud vendor based on the metadata; and update the entities and relationships related to the heterogeneous computing power resources of the first cloud vendor to the computing power knowledge graph.

[0137] In some embodiments, the first processing module 22 is further configured to: determine the entities and relationships related to the updated heterogeneous computing resources of the cloud vendor when the heterogeneous computing resources of the cloud vendor are updated; and update the computing power knowledge graph based on the entities and relationships related to the updated heterogeneous computing resources of the cloud vendor.

[0138] In some embodiments, the second processing module 23 is further configured to: generate an explanation report; the explanation report is used to explain the reasons for the recommendation; send the recommended list of computing resources and the explanation report to the terminal; the heterogeneous computing resources in the recommended list of computing resources carry recommendation weights; the recommendation weights are set according to the preferences of the user corresponding to the terminal; the explanation report is used to instruct the user to select a target heterogeneous computing resource from the list of computing resources; obtain the target heterogeneous computing resource selected by the user from the list of recommended computing resources; send a first request to the cloud platform of the target cloud vendor corresponding to the target heterogeneous computing resource, the first request being used to request the completion of the creation and configuration of the target heterogeneous computing resource.

[0139] In some embodiments, the acquisition module 21 is further configured to: acquire second information and third information sent by the cloud platform of the target cloud vendor; the second information includes at least one of the following: the creation completion time of the target heterogeneous computing power resource, and whether the creation of the target heterogeneous computing power resource was successful; the third information includes the current computing power inventory of the target cloud vendor's heterogeneous computing power resources; the first processing module 22 is further configured to: send a notification message to the terminal according to the second information; the notification message is used to notify the user corresponding to the terminal whether to use the created target heterogeneous computing power resource to perform a task; and update the computing power knowledge graph according to the third information.

[0140] In practical applications, the acquisition module 21 can be implemented by the communication interface in the information processing device; the first processing module 22 and the second processing module 23 can be implemented by the processor in the information processing device.

[0141] It should be noted that the information processing device provided in the above embodiments is only illustrated by the division of the above program modules. In practical applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the device can be divided into different program modules to complete all or part of the processing described above. In addition, the information processing device and the information processing method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the method embodiments, which will not be repeated here.

[0142] This application also provides a network device, such as... Figure 3 As shown, it includes: Communication interface 31 enables information exchange with other devices; The processor 32, connected to the communication interface 31, is used to execute the methods provided by one or more technical solutions on the network device side when running a computer program. The computer program is stored in the memory 33.

[0143] It should be noted that the specific processing procedures of the processor 32 and the communication interface 31 are detailed in the method embodiment and will not be repeated here.

[0144] Of course, in practical applications, the various components in network device 30 are coupled together through bus system 34. It can be understood that bus system 34 is used to implement communication between these components. In addition to a data bus, bus system 34 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 3 The general labeled all buses as Bus System 34.

[0145] The memory 33 in this embodiment is used to store various types of data to support the operation of the network device 30. Examples of such data include any computer program used to operate on the network device 30.

[0146] The methods disclosed in the embodiments of this application can be applied to the processor 32, or implemented by the processor 32. The processor 32 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware or by instructions in the form of software in the processor 32. The processor 32 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor 32 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this application can be directly reflected as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, which is located in the memory 33. The processor 32 reads the information in the memory 33 and completes the steps of the aforementioned method in combination with its hardware.

[0147] In an exemplary embodiment, the network device 30 may be implemented by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers (MCUs), microprocessors, or other electronic components to perform the aforementioned method.

[0148] It is understood that the memory (memory 33) in the embodiments of this application can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM); magnetic surface memory can be disk storage or magnetic tape storage. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), SyncLink Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The memories described in the embodiments of this application are intended to include, but are not limited to, these and any other suitable types of memories.

[0149] In an exemplary embodiment, this application also provides a storage medium, namely a computer storage medium, specifically a computer-readable storage medium, such as a memory storing a computer program, which can be executed by the processor 32 of the network device 30 to complete the steps described in the aforementioned network device-side method. The computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface memory, optical disc, or CD-ROM.

[0150] For example, embodiments of this application also provide a computer program product, including a computer program that can be executed by the processor 32 of the network device 30 to perform the steps described in any of the foregoing methods.

[0151] It should be noted that terms such as "first" and "second" are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0152] Furthermore, the technical solutions described in the embodiments of this application can be combined arbitrarily without conflict.

[0153] The above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application.

Claims

1. An information processing method, characterized in that, The method includes: Obtain the first information sent by the terminal; the first information is used to request the use of computing resources. Based on the first information, a first subgraph is determined from the computing power knowledge graph; the computing power knowledge graph is constructed based on entities and relationships related to heterogeneous computing power resources; the first subgraph includes entities and relationships related to candidate computing power resources that match the first information; Determine the embedding vector of the first sub-graph; The embedding vector of the first sub-graph is input into the pre-trained first model for inference to obtain the inference result; the inference result includes a recommended list of computing resources.

2. The method according to claim 1, characterized in that, The first information includes at least one of the following: Task type; Computing power requirements; Budget; Delay requirement; Data location.

3. The method according to claim 1, characterized in that, The method further includes: Acquire multi-source data; the multi-source data includes heterogeneous computing resources from multiple cloud vendors; Extract entities and relationships related to the heterogeneous computing resources from the multi-source data; Based on the entities and relationships related to the heterogeneous computing resources, the computing power knowledge graph is constructed.

4. The method according to claim 3, characterized in that, The entity includes at least one of the following: Cloud vendors; Data center; Compute nodes; Heterogeneous computing resources; Computing power type; Computing power inventory; Network links; user; Task type; Service quality indicators; The relationship includes at least one of the following: lie in; supply; connect; support; belong; Historical use; Performance.

5. The method according to claim 3, characterized in that, The method further includes: The metadata description file submitted by the first cloud vendor is obtained through the registration interface; the metadata description file is related to the heterogeneous computing resources of the first cloud vendor. Parse the metadata description file to obtain the metadata of the heterogeneous computing resources of the first cloud vendor; Based on the metadata, determine the entities and relationships related to the heterogeneous computing resources of the first cloud vendor; Update the entities and relationships related to the heterogeneous computing power resources of the first cloud vendor to the computing power knowledge graph.

6. The method according to claim 3, characterized in that, The method further includes: In the event that the heterogeneous computing resources of the cloud provider are updated, determine the entities and relationships related to the updated heterogeneous computing resources of the cloud provider. The computing power knowledge graph is updated based on the entities and relationships related to the updated heterogeneous computing power resources of the cloud vendor.

7. The method according to claim 1, characterized in that, The method further includes: Generate an explanation report; the explanation report is used to explain the reasons for the recommendation. The recommended list of computing resources and the explanation report are sent to the terminal; the heterogeneous computing resources in the recommended list of computing resources carry recommendation weights; the recommendation weights are set according to the preferences of the user corresponding to the terminal; the explanation report is used to instruct the user to select a target heterogeneous computing resource from the list of computing resources. Obtain the target heterogeneous computing resources selected by the user from the recommended list of computing resources; A first request is sent to the cloud platform of the target cloud vendor corresponding to the target heterogeneous computing power resource. The first request is used to request the completion of the creation and configuration of the target heterogeneous computing power resource.

8. The method according to claim 7, characterized in that, The method further includes: Obtain second and third information sent by the cloud platform of the target cloud vendor; the second information includes at least one of the following: the creation completion time of the target heterogeneous computing power resource, and whether the creation of the target heterogeneous computing power resource was successful; the third information includes the current computing power inventory of the target cloud vendor's heterogeneous computing power resources; Based on the second information, a notification message is sent to the terminal; the notification message is used to notify the user corresponding to the terminal whether to use the target heterogeneous computing resources to perform a task. The computing power knowledge graph is updated based on the third information.

9. An information processing device, characterized in that, include: The acquisition module is used to acquire the initial information. The first information is used to request the use of computing resources; A first processing module is configured to determine a first sub-graph from the computing power knowledge graph based on the first information; the computing power knowledge graph is constructed based on entities and relationships related to heterogeneous computing power resources; the first sub-graph includes entities and relationships related to candidate computing power resources that match the first information. The second processing module is used to determine the embedding vector of the first sub-graph; input the embedding vector of the first sub-graph into a pre-trained first model for inference, and obtain the inference result; the inference result includes a recommended list of computing resources.

10. A network device, characterized in that, This includes a processor and memory for storing computer programs that can run on the processor. When the processor is used to run the computer program, it performs the steps of the method according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

Images