Method and device for accurately scheduling computing power based on identification by intelligent computing cloud platform

Abstract

The application provides a method and device for accurately scheduling computing power based on identification by an intelligent computing cloud platform, and relates to the technical fields of intelligent computing centers, intelligent computing centers, computing power infrastructure and intelligent computing cloud technology.The method comprises the following steps: S1, obtaining a hypertext transfer protocol request sent by a client; S2, analyzing the hypertext transfer protocol request to obtain a target identification in a request header, wherein the target identification at least includes a service identification and an internet protocol identification; S3, determining a target computing power cluster from a plurality of computing power clusters according to the target identification; and S4, directing the hypertext transfer protocol request to the target computing power cluster.The application can realize fine-grained, low-latency and high-matching-degree computing power scheduling, avoid computing power waste, and improve computing power utilization efficiency and service quality.

Description

Technical Field

[0001] This invention relates to the fields of intelligent computing centers, smart computing centers, computing infrastructure, and smart cloud technologies, specifically to a method and apparatus for precise scheduling of computing power based on identifiers in an intelligent computing cloud platform. Background Technology

[0002] With the rapid development of artificial intelligence technology, "intelligent computing centers" and "smart computing centers" have emerged.

[0003] An "intelligent computing center" refers to a facility that provides the necessary computing power, data, and algorithms for artificial intelligence applications (such as the development, training, and inference of deep learning models) by utilizing large-scale heterogeneous computing resources, including general-purpose and intelligent computing power. Intelligent computing centers encompass facilities, hardware, and software, and can provide full-stack capabilities from underlying computing power to top-level application enablement.

[0004] "Intelligent computing center" includes, but is not limited to, "intelligent computing center".

[0005] "Intelligent computing center" or artificial intelligence computing center is a type of computing infrastructure that provides computing power services, data services, and algorithm services required for artificial intelligence applications, based on artificial intelligence theory and adopting artificial intelligence computing architecture.

[0006] "Computing power" is the core of "intelligent computing center" and "smart computing center". It is the ability of computer equipment or computing / data center to process parameters. It is the ability of computer hardware and software to work together to execute a certain computing requirement. It is the computing power to achieve the target result output by processing parameter data. It is a new type of productivity that integrates parameter computing power, network carrying capacity and data storage capacity. It mainly provides services to society through computing power infrastructure.

[0007] In existing technologies, computing power scheduling systems generally employ a cluster-level round-robin mechanism to allocate computing power. This technology uses the cluster as the smallest scheduling granularity, distributing requests in batches evenly across various computing clusters. However, in this scheduling method, the routing process relies on an external DNS / load balancer, introducing a latency of over 50 milliseconds; existing technologies also use random allocation for computing power matching, resulting in insufficient average load across heterogeneous clusters; furthermore, the lack of geographical awareness in scheduling strategies leads to a high proportion of cross-regional traffic, exacerbating network bandwidth consumption and data transmission latency.

[0008] It is evident that since the emergence of intelligent computing centers, how to achieve fine-grained, low-latency, and highly matched computing power scheduling, avoid computing power waste, and improve computing power utilization efficiency and service quality has always been a pressing problem to be solved in this field. Summary of the Invention

[0009] This invention provides a method and apparatus for precise scheduling of computing power based on identifiers in an intelligent computing cloud platform, in order to solve the problems of coarse-grained scheduling, high latency, low matching degree, as well as computing power waste, low computing power utilization efficiency and low service quality in the existing technology.

[0010] To solve the above problems, the present invention is implemented as follows: In a first aspect, the present invention provides a method for precise scheduling of computing power based on identifiers in an intelligent computing cloud platform, comprising: Step S1: Obtain the Hypertext Transfer Protocol request sent by the client; Step S2: Parse the Hypertext Transfer Protocol request to obtain the target identifier in the request header. The target identifier includes at least a service identifier and an Internet Protocol identifier. Step S3: Determine the target computing power cluster from multiple computing power clusters based on the target identifier; Step S4: Direct the Hypertext Transfer Protocol request to the target computing power cluster.

[0011] In one embodiment, step S3 includes: Step S3.1: Identify the computing power service type based on the service identifier, and obtain the geographic affiliation information based on the Internet Protocol identifier; Step S3.2: Determine the target computing power cluster from the plurality of computing power clusters based on the computing power service type and the geographical affiliation information.

[0012] In one embodiment, step S3.2 includes: Step S3.2.1: Determine multiple candidate computing power clusters from the multiple computing power clusters according to the computing power service type; Step S3.2.2: Obtain the real-time cluster load of each candidate computing power cluster; Step S3.2.3: Determine the target computing power cluster from the multiple candidate computing power clusters based on the geographical affiliation information of each candidate computing power cluster and the real-time cluster load.

[0013] In one embodiment, step S3.2.1 includes: Step S3.2.1.1: In response to the computing power service type being one of inference service, training service or fine-tuning service, determine a low-latency computing power cluster from the plurality of computing power clusters as the candidate computing power cluster; Step S3.2.1.2: In response to the computing power service type being an embedded service, determine the candidate computing power cluster from the computing power clusters based on the GPU attribute parameters of the plurality of computing power clusters; The low-latency computing cluster is the computing cluster whose latency is less than a preset latency threshold.

[0014] In one embodiment, step S3.2.3 includes: Step S3.2.3.1: Determine the client's home region based on the regional affiliation information, and determine the geographical distance between each candidate computing power cluster and the home region; Step S3.2.3.2: Weight the real-time cluster load and the corresponding geographical distance of each candidate computing power cluster to obtain the score value of each candidate computing power cluster; Step S3.2.3.3: Determine the target computing power cluster from the candidate computing power clusters based on the score value.

[0015] In one embodiment, it also includes: Step S3.2.3.4: Obtain the number of connections and throughput of each computing cluster; Step S3.2.3.5: Based on the number of connections and throughput of the computing power cluster, dynamically adjust the weighting of the real-time cluster load and geographical distance.

[0016] In one embodiment, step S3.2 includes: Step S3.2.1': Determine multiple candidate computing power clusters from the multiple computing power clusters according to the computing power service type; Step S3.2.2': Determine the client's region of origin based on the region affiliation information; Step S3.2.3': Determine the computing power cluster that is geographically closest to the home region from among the multiple candidate computing power clusters, and use it as the target computing power cluster.

[0017] In one embodiment, after step S4, the method further includes: Step S5: Obtain the health status of the computing cluster in each region; Step S6: In response to the health status of the computing power cluster in any target area being less than a preset health status threshold, the Internet Protocol segment traffic corresponding to the computing power cluster in the target area is switched to the backup cluster.

[0018] Secondly, the present invention also provides a device for precise scheduling of computing power based on identifiers in an intelligent computing cloud platform, comprising: The request retrieval module is used to retrieve Hypertext Transfer Protocol requests sent by the client; The identifier acquisition module is used to parse the Hypertext Transfer Protocol request and obtain the target identifier in the request header. The target identifier includes at least one of a service identifier and an Internet Protocol identifier. The cluster determination module is used to determine the target computing power cluster from multiple computing power clusters based on the target identifier; The computing power scheduling module is used to direct the Hypertext Transfer Protocol request to the target computing power cluster.

[0019] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, it implements the steps of the method for precise scheduling of computing power based on identifiers in the intelligent computing cloud platform described in the first aspect above.

[0020] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method for precise scheduling of computing power based on identifiers in the intelligent computing cloud platform described in the first aspect above.

[0021] Fifthly, the present invention also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps in the method for precise scheduling of computing power based on identifiers in the intelligent computing cloud platform described in the first aspect above.

[0022] In this invention, step S1 involves obtaining a Hypertext Transfer Protocol (HTTP) request sent by the client; step S2 involves parsing the HTTP request to obtain a target identifier in the request header, wherein the target identifier includes at least one of a service identifier and an Internet Protocol (IP) identifier; step S3 involves determining a target computing cluster from multiple computing clusters based on the target identifier; and step S4 involves directing the HTTP request to the target computing cluster. This invention enables fine-grained, low-latency, and highly matched computing power scheduling, avoiding computing power waste and improving computing power utilization efficiency and service quality. Attached Figure Description

[0023] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the present invention will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of a method for precise scheduling of computing power based on identifiers in an intelligent computing cloud platform, provided by the present invention. Figure 2 This invention provides a flowchart for determining a target computing power cluster from multiple computing power clusters based on a target identifier; Figure 3This invention provides a flowchart for determining a target computing power cluster from multiple computing power clusters based on computing power service type and geographic affiliation information; Figure 4 This is a structural diagram of a device for precisely scheduling computing power based on identifiers in an intelligent computing cloud platform provided by the present invention; Figure 5 This is a structural diagram of an electronic device provided by the present invention. Detailed Implementation

[0025] The technical solutions of this invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0026] The “computing power” mentioned in this invention refers to: the ability of computer equipment or computing / data center to process information; the ability of computer hardware and software to work together to perform a certain computing requirement; the computing power to achieve the target result output by processing information data; and a new type of productivity that integrates information computing power, network carrying capacity, and data storage capacity, mainly providing services to society through computing power infrastructure.

[0027] The "computational power" (CP) described in this invention refers to the ability of a data center server to process data and output results. It is a comprehensive indicator of a data center's computing power, encompassing general computing power, supercomputing power, and intelligent computing power. The commonly used unit of measurement is floating-point operations per second (FLOPS, 1 EFLOPS = 10^18 FLOPS), with higher values ​​indicating stronger overall computing power. It is estimated that 1 EFLOPS is approximately the computing power output of 5 Tianhe-2A supercomputers, 500,000 mainstream server CPUs, or 2 million mainstream laptops. The calculation formula is: CP = CP 通用 +CP 智能 +CP 超级 .

[0028] The "Network Power" (NP) mentioned in this invention refers to the performance of data transmission capability of computing facilities, which includes comprehensive capabilities such as network architecture, network bandwidth, transmission latency, intelligent management and scheduling, and involves network transmission within and between data centers. It is a comprehensive indicator for measuring network transmission scheduling capability.

[0029] The "Storage Power" (SP) described in this invention refers to the comprehensive capabilities of a data center in four aspects: data storage capacity, performance, security and reliability, and green and low-carbon operation. It is a comprehensive indicator for measuring the data storage capacity of a data center, including external storage devices such as storage arrays and internal storage devices within servers. The commonly used unit of measurement for storage capacity is exabytes (EB, 1EB = 2^60 bytes), while the commonly used unit of measurement for performance is the number of read / write operations per second (IOPS / TB). Disaster recovery ratio is an important indicator of security and reliability.

[0030] The "computing infrastructure" mentioned in this invention refers to a new type of information infrastructure that integrates information computing power, network carrying capacity, and data storage capacity, enabling centralized computing, storage, transmission, and application of information.

[0031] The "new information infrastructure" mentioned in this invention refers to network infrastructure such as 5G networks, fiber optic broadband networks, backbone networks, international communication networks, and satellite internet; computing infrastructure such as data centers, general computing centers, intelligent computing centers, and supercomputing centers; and new technology facilities such as artificial intelligence, blockchain, and quantum computing.

[0032] The “computing power” mentioned in this invention includes: general computing power, intelligent computing power, and supercomputing power.

[0033] The "general computing power" mentioned in this invention refers to the computing power provided by servers based on CPU (Central Processing Unit) chips, which is used to support basic general computing such as cloud computing and edge computing.

[0034] The "intelligent computing power" mentioned in this invention refers to: a computing platform deployed on a large scale based on dedicated chips such as GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), and ASIC (Application Specific Integrated Circuit) for various artificial intelligence innovative applications, such as natural language processing and machine vision.

[0035] The “supercomputing power” mentioned in this invention refers to the computing power provided by high-performance computing clusters such as supercomputers. It utilizes the centralized computing resources of multiple computer systems working in parallel and uses a dedicated operating system to handle extremely complex or data-intensive problems. It is mainly used for computing in cutting-edge scientific fields, such as planetary simulation, drug molecule design, and gene analysis.

[0036] The "intelligent computing center" described in this invention refers to a facility that, through the use of large-scale heterogeneous computing resources, including general-purpose computing power (CPU) and intelligent computing power (GPU, FPGA, ASIC, etc.), primarily provides the necessary computing power, data, and algorithms for artificial intelligence applications (such as the development, training, and inference of deep learning models). The intelligent computing center encompasses facilities, hardware, and software, and can provide full-stack capabilities from underlying computing power to top-level application enablement.

[0037] The "intelligent computing cloud platform" mentioned in this invention, abbreviated as "intelligent computing cloud", refers to a cloud computing platform that integrates hardware and software resources based on an intelligent computing center.

[0038] The "intelligent computing center" mentioned in this invention includes, but is not limited to, "smart computing center".

[0039] The "intelligent computing center" mentioned in this invention, also known as an artificial intelligence computing center, is a type of computing infrastructure that provides computing power services, data services, and algorithm services required for artificial intelligence applications, based on artificial intelligence theory and adopting an artificial intelligence computing architecture.

[0040] The "computing center" mentioned in this invention refers to a facility that is mainly composed of infrastructure such as wind, thermal, hydro, and electricity, and IT hardware and software equipment, and has computing power, carrying capacity, and storage capacity, including general data centers, intelligent computing centers, supercomputing centers, etc.

[0041] The "supercomputing center" mentioned in this invention refers to a supercomputing data center, which is a data center based on supercomputers or large-scale computing clusters. It can provide large-scale computing, storage and network services and is widely used in aerospace, defense, oil exploration, climate modeling and genome sequencing and other application scenarios.

[0042] The “computing resources” mentioned in this invention refer to the technologies and facilities required for the development of the digital society that have the ability to compute, transmit, store and apply information, including but not limited to computing resources such as CPUs and GPUs, network resources such as switches and routers, storage resources such as storage arrays and distributed storage, security resources such as firewalls and intrusion detection systems, and supporting and guaranteeing resources such as wind, fire, water and electricity.

[0043] The "model" mentioned in this invention includes, but is not limited to, "large language model" and "multimodal large model".

[0044] The "large language model" mentioned in this invention refers to a large-scale language model (LLM), which is a language model with a large number of parameters. It is designed to understand and generate human language, and is trained with a large amount of text data. It can perform a wide range of tasks, including text summarization, translation, and sentiment analysis.

[0045] The “Multimodal Large Models” mentioned in this invention refer to models that combine multimodal information such as text, images, videos, and audio for training, including but not limited to multimodal large language models.

[0046] The "computing power routing" described in this invention refers to a core routing technology for the convergence of computing and network technologies. Based on existing IP networks, it incorporates the status of computing resources (such as CPU / GPU load, memory, and computing service type) into routing decisions to achieve the joint optimal selection of network paths and computing nodes.

[0047] The "computing power service" mentioned in this invention refers to a service that provides users with on-demand information computing power, network carrying capacity, and data storage capacity through computing power resources, such as inference service, training service, or fine-tuning service.

[0048] The “computing power cluster” mentioned in this invention refers to a distributed computing system that interconnects a large number of computing nodes (usually including acceleration chips such as CPU / GPU / TPU) through a high-speed network and is scheduled to work together by a unified management system. It aims to provide aggregated computing power far exceeding that of a single node to support large-scale artificial intelligence applications, model training, scientific computing and complex data processing.

[0049] Please see Figure 1 , Figure 1 This is a flowchart of a method for precise scheduling of computing power based on identifiers in an intelligent computing cloud platform, as provided by the present invention. Figure 1 As shown, the method includes: Step S1: Obtain the Hypertext Transfer Protocol request sent by the client.

[0050] In one embodiment, the gateway or front-end service receives HTTP requests sent by the client in order to extract parameters such as the target identifier.

[0051] The technical effect of step S1 is to quickly obtain Hypertext Transfer Protocol requests, laying a reliable data foundation for subsequent accurate computing power scheduling.

[0052] Step S2: Parse the Hypertext Transfer Protocol request and obtain the target identifier in the request header. The target identifier includes at least the service identifier and the Internet Protocol identifier.

[0053] APISIX is a dynamic, real-time, high-performance cloud-native API gateway. In one embodiment, the APISIX gateway parses the target identifier in the request header using the traffic-split plugin. Optionally, rules for obtaining the client's real IP can be configured in APISIX, for example, configured as "If the request comes from a trusted proxy, read the client's real IP from a specified field in the HTTP request header (such as the X-Real-IP field) as the IP identifier."

[0054] Traffic-split is a core plugin in the APISIX gateway for traffic splitting. It can dynamically distribute incoming requests to different upstream services based on weights or conditions. Simply put, it's like a smart "traffic divider," not only allocating traffic proportionally but also finely controlling traffic flow based on request characteristics.

[0055] In one embodiment, a service identifier is used to indicate the type of computing power service. By parsing the service identifier field in the request header, the type of computing power service can be accurately identified with a single request. For example, the type of computing power service may include inference, training, fine-tuning, embedding, and other service types. The service identifier may also indicate priority, etc.

[0056] In one embodiment, by parsing the Internet Protocol (IP) identifier in the request header, it is possible to accurately identify the geographic region to which the Hypertext Transfer Protocol (HTTP) request belongs.

[0057] In one embodiment, the target identifier may also include other identifiers that can be used to identify parameters such as source IP address, geographic affiliation, requested resource type, or expected computing power requirements, and the present invention does not limit this.

[0058] The technical effects of step S2: The introduction of service identifiers can transform general HTTP requests into structured descriptions of computing power requirements, enabling the gateway to have "computing power requirement awareness" capabilities. This addresses the deficiency in existing technologies where gateways "only recognize traffic but not business services," and enables differentiated routing for heterogeneous computing power loads. The introduction of IP identifiers enables precise mapping between geographical location and network topology.

[0059] Step S3: Determine the target computing power cluster from multiple computing power clusters based on the target identifier.

[0060] In one embodiment, the client's IP address is resolved based on the IP identifier, and the regional affiliation and network operator characteristics are extracted to construct a client distribution profile. Based on the client distribution profile, the request can be directed to the nearest computing power cluster, thereby reducing network transmission latency.

[0061] This invention identifies computing power service types based on service identifiers, which is the core mechanism for achieving differentiated routing of heterogeneous computing power. In one embodiment, the required computing power service type is dynamically determined by parsing the service identifier carried in the request, such as high-parallel GPU computing power adapted for AI training, FPGA accelerated computing power for scientific computing, or general-purpose GPU computing power. Subsequently, the request is intelligently routed to the matching computing power cluster according to preset matching rules and resource topology.

[0062] In one embodiment, a multi-condition matching rule chain is executed based on each identifier included in the target identifier. For example, the target identifier includes a service identifier and an Internet Protocol identifier. Matching is performed based on the service identifier and the IP identifier, respectively, and the request is directed to the computing power cluster with the closest geographical location in the corresponding computing power service type, thereby achieving dual optimization of geographical proximity and computing power service type adaptation.

[0063] Optionally, based on this, differentiated routing further considers the real-time load, network latency, data affinity, and cost model of each cluster to achieve fine-grained traffic allocation. This mechanism not only improves the utilization of heterogeneous resources but also provides deterministic computing power guarantees for high real-time or high-throughput services, thereby achieving an optimal balance between performance and cost in a hybrid computing environment.

[0064] The technical effects of step S3 are as follows: The introduction of service identifiers enables differentiated routing for heterogeneous computing power loads; the introduction of IP identifiers reduces network transmission latency. Determining the target computing power cluster based on dual identifiers enables cross-validation, completing decisions in milliseconds and improving scheduling accuracy from "cluster-level polling" to "single-request-level directives."

[0065] Step S4: Direct the Hypertext Transfer Protocol request to the target computing cluster.

[0066] Efficiently and accurately directing Hypertext Transfer Protocol (HTTP) requests to the target computing power cluster is a key step in realizing computing power gateways and intelligent scheduling. In one embodiment, the target computing power cluster is the cluster to which the HTTP request points. The endpoint of the target computing power cluster is obtained, and the HTTP request is forwarded to the target computing power cluster based on the endpoint. Throughout the process, timeout retries, circuit breakers, and link tracing must be handled simultaneously to ensure end-to-end reliability, low latency, and observability of HTTP requests in a heterogeneous computing power environment.

[0067] The technical effects of step S4: Directing HTTP requests to the target computing power cluster enables fine-grained traffic distribution, ensuring high service availability. Simultaneously, dynamic routing based on the geographical location of the computing power service type significantly reduces response time. Furthermore, this invention can improve the overall utilization rate of the computing power cluster.

[0068] Please see Figure 2, Figure 2 This invention provides a flowchart for determining a target computing power cluster from multiple computing power clusters based on a target identifier, as shown below. Figure 2 As shown, the target identifier includes a service identifier and an Internet Protocol identifier. Step S3 includes: Step S3.1: Identify the computing power service type based on the service identifier and obtain the geographical affiliation information based on the Internet Protocol identifier.

[0069] In one embodiment, there is a mapping relationship between service identifiers and computing power service types, which can be used to determine the computing power service type corresponding to the current service identifier. For example, the service identifier of an AI training service corresponds to a high-parallelism computing power service type. As another example, the service identifier of a general computing service corresponds to a low-latency computing power service type.

[0070] In one embodiment, if the Hypertext Transfer Protocol request is an inference request, the computing power service type indicated by its service identifier is inference service; if the Hypertext Transfer Protocol request is a training request, the computing power service type indicated by its service identifier is training service; if the Hypertext Transfer Protocol request is a fine-tuning request, the computing power service type indicated by its service identifier is fine-tuning service; and if the Hypertext Transfer Protocol request is an embedding request, the computing power service type indicated by its service identifier is embedding service.

[0071] Embedding requests are computational requests aimed at mapping input data to points in a high-dimensional vector space using a deep learning model. These vectors semantically represent the original data and are commonly used for tasks such as similarity calculation, clustering, and retrieval. Inference requests send input data to a pre-trained model to obtain its predictions. They are characterized by relatively fixed computational cost and latency sensitivity, such as calling large models to generate text or recognize image content. Training requests involve learning all the parameters of a model from scratch or using a large dataset. They are characterized by high computational cost, long training time, and the need for large amounts of data and computing power. Fine-tuning requests incrementally train a pre-trained model using a small-scale dataset specific to a particular domain. Their purpose is to adapt the model to specialized tasks; the computational cost falls between that of inference and full training, and they can be performed online or offline to some extent.

[0072] In one embodiment, the system queries a locally maintained IP geolocation database or calls a third-party geolocation API to quickly parse out the geographic affiliation information by performing a binary search or hash lookup on the IP identifier and the network segment range in the database.

[0073] In one embodiment, the mapping relationship between IP address and geographical location information is obtained, and the geographical affiliation information corresponding to the current IP identifier is obtained based on the mapping relationship between IP address and geographical location information.

[0074] The technical effect of step S3.1: By identifying the computing service type based on the service identifier and obtaining the geographic affiliation information by combining the IP identifier, fine-grained routing can be achieved. This facilitates the accurate allocation of requests to the corresponding computing clusters, thereby optimizing computing power scheduling and reducing cross-domain latency.

[0075] Step S3.2: Determine the target computing power cluster from multiple computing power clusters based on the computing power service type and geographical affiliation information.

[0076] In one embodiment, the target identifier is a dual identifier, including a service identifier and an IP identifier. A multi-condition matching rule chain is executed. First, matching is performed based on the service identifier to identify the computing power service type and determine the set of computing power clusters mapped to the request. This set of computing power clusters includes multiple computing power clusters. Further, matching is performed based on the IP identifier to resolve the geographic location of the client's IP address and direct the request to the nearest target computing power cluster within the set of computing power clusters.

[0077] The technical effect of step S3.2: Accurately selecting the target computing power cluster based on the computing power service type and geographical affiliation information enables intelligent matching of computing power and demand. This avoids cross-domain scheduling, significantly reduces response latency, improves overall processing efficiency and service quality, and achieves dual optimization of geographical proximity and differentiated routing of heterogeneous computing power, thereby enhancing service response speed and user experience.

[0078] Please see Figure 3 , Figure 3 This invention provides a flowchart for determining a target computing power cluster from multiple computing power clusters based on computing power service type and geographic affiliation information, as shown below. Figure 3 As shown, step S3.2 includes: Step S3.2.1: Based on the computing power service type, determine multiple candidate computing power clusters from multiple computing power clusters.

[0079] In one embodiment, step S3.2.1 includes steps S3.2.1.1 and S3.2.1.2. These are further described below.

[0080] Step S3.2.1.1: In response to the computing power service type being inference service, training service, or fine-tuning service, determine low-latency computing power clusters from multiple computing power clusters as candidate computing power clusters. Low-latency computing power clusters are those with latency less than a preset latency threshold.

[0081] If the computing power service type is one of inference service, training service, or fine-tuning service, to ensure response timeliness, priority is given to routing to low-latency GPU clusters. That is, low-latency computing power clusters are selected from multiple computing power clusters as candidate computing power clusters. Optionally, the latency of each computing power cluster is obtained, and computing power clusters with latency less than a preset latency threshold are selected as candidate computing power clusters.

[0082] In one embodiment, the latency of the computing power cluster can be obtained through active probing. For example, the scheduler sends lightweight probe packets from the gateway or edge node to each computing node in the cluster and records their round-trip time (RTT). The latency of the computing power cluster is then determined based on the RTT. Alternatively, passive sampling can be performed in conjunction with the response time of service requests, and the latency of the computing power cluster can be calculated using a weighted average or a sliding window.

[0083] Step S3.2.1.2: In response to the computing power service type being embedded service, candidate computing power clusters are determined from the computing power clusters based on the GPU attribute parameters of multiple computing power clusters; If the computing power service type is embedding, to reduce GPU resource waste, requests are preferentially scheduled to GPU-intensive computing power clusters. That is, GPU-intensive computing power clusters are identified from multiple computing power clusters as candidate clusters. Optionally, GPU attribute parameters of each computing power cluster are obtained, and based on these parameters, GPU-intensive computing power clusters are identified from multiple clusters as candidate clusters.

[0084] Optionally, the GPU attribute parameters include at least one of the following: number of GPUs, GPU utilization, load data, input / output ratio, and cache hit rate.

[0085] Step S3.2.2: Obtain the real-time cluster load of each candidate computing power cluster.

[0086] In one embodiment, a configured monitoring system collects the connection count and throughput of each candidate computing power cluster, and determines the real-time cluster load based on the connection count and throughput. In another embodiment, an API interface is used to query the number of concurrent requests and the average response time, and the real-time cluster load of each candidate computing power cluster is determined based on the number of concurrent requests and the average response time.

[0087] In one embodiment, dynamic parameters such as utilization, task queue depth, memory usage, and network I / O of each candidate computing cluster can be collected periodically or triggered through active polling or passive reception, thereby generating quantitative indicators characterizing the real-time cluster load. This real-time load data enables the system to avoid overloaded or faulty computing clusters and prioritize directing Hypertext Transfer Protocol (HTTP) requests to computing clusters with sufficient idle resources. By obtaining accurate real-time cluster load data, this invention significantly improves allocation efficiency and execution stability among heterogeneous clusters.

[0088] Step S3.2.3: Determine the target computing power cluster from multiple candidate computing power clusters based on the geographical affiliation information and real-time cluster load of each candidate computing power cluster.

[0089] In one embodiment, step S3.2.3 includes steps S3.2.3.1, S3.2.3.2, and S3.2.3.3.

[0090] Step S3.2.3.1: Determine the client's home region based on the regional affiliation information, and determine the geographical distance between each candidate computing power cluster and its home region.

[0091] In one embodiment, the client's geographic location is accurately determined based on the geographic affiliation information. Next, based on a candidate computing power cluster geographic location database (e.g., the address information of each computing power cluster deployment), the geographic distance between each candidate computing power cluster and its geographic location is calculated. Optionally, this distance can also be expressed as Euclidean distance, road network distance, or other distances.

[0092] By quantifying geographical proximity, key weighting factors are provided for subsequent scheduling strategies, aiming to prioritize computing clusters with shorter physical links and lower transmission latency, thereby effectively reducing data round-trip time.

[0093] Step S3.2.3.2: Weight each candidate computing power cluster based on its real-time cluster load and corresponding geographical distance to obtain a score for each candidate computing power cluster.

[0094] In one embodiment, weighting coefficients are configured for real-time cluster load and geographical distance, respectively. For example, the weights can be equal or dynamically adjusted according to service preferences. For instance, distance weight can be increased for latency-sensitive services, while load weight can be emphasized for compute-intensive services. The real-time cluster load and corresponding geographical distance of each candidate computing power cluster are calculated using a preset scoring function (such as weighted sum, weighted product, or normalized linear combination) to obtain the score value of the candidate computing power cluster.

[0095] Specifically, lower real-time cluster load indicates more abundant computing power, and smaller geographical distance indicates lower transmission latency. Therefore, low load and small distance will jointly contribute to a higher score. This score serves as a quantitative basis for subsequent scheduling decisions. Compared to a crude allocation that solely relies on real-time cluster load or geographical distance, this invention achieves dynamic trade-offs through multi-factor weighting, systematically identifying the optimal target computing power cluster, thereby improving overall computing power utilization and task response speed.

[0096] Step S3.2.3.3: Determine the target computing power cluster from the candidate computing power clusters based on the score value.

[0097] In one embodiment, based on the comprehensive score calculated for each candidate computing power cluster in the preceding steps, a screening operation is performed according to a preset selection rule to determine the target computing power cluster. For example, the candidate computing power cluster with the highest score is selected as the target computing power cluster.

[0098] Through this scoring and selection mechanism, the system can automatically and objectively select the target computing power cluster, providing a clear addressing basis for allocating the client's Hypertext Transfer Protocol requests to the target computing cluster for execution. It is a key decision node for realizing automated and intelligent computing power scheduling.

[0099] In one embodiment, step S3.2.3 further includes steps S3.2.3.4 and S3.2.3.5.

[0100] Step S3.2.3.4: Obtain the number of connections and throughput of each computing cluster.

[0101] In one embodiment, the number of connections and throughput are two key dynamic indicators characterizing the current service pressure and data transmission capacity of the computing power cluster. The number of connections refers to the number of active network connections maintained by the computing power cluster per unit time, such as the number of sessions or the length of the request queue, reflecting the simultaneous access pressure faced by the cluster. Throughput refers to the amount of data successfully processed by the computing power cluster per unit time, such as the number of requests per second, the number of bits transmitted per second, or the number of computing tasks completed, reflecting the actual processing efficiency of the cluster.

[0102] In one embodiment, the connection count and throughput of each computing cluster are retrieved through the cluster management interface, or each computing cluster actively reports its connection count and throughput. These two metrics overlap with the aforementioned "real-time cluster load" but have different focuses. Real-time cluster load focuses more on resource utilization (CPU / GPU / memory), while connection count and throughput focus more on the saturation of the network and service dimensions. They can be used to identify clusters that have idle resources but congested networks, assisting in more refined scheduling decisions.

[0103] Step S3.2.3.5: Based on the number of connections and throughput of the computing cluster, dynamically adjust the weighted weights corresponding to the real-time cluster load and geographical distance.

[0104] In one embodiment, the congestion level of the current network link and the service saturation of the computing power cluster are assessed in real time based on the number of connections and throughput of each computing power cluster.

[0105] Optionally, when a cluster in a certain region is detected to have an excessively high number of connections or a sudden drop in throughput, it indicates that network congestion may be occurring and that nearby connections may no longer have a bandwidth advantage. Therefore, the weight of geographical distance can be reduced while the weight of real-time cluster load can be relatively increased, thereby prioritizing clusters with lower loads but potentially located further away.

[0106] Optionally, when a low overall connection count is detected in a certain region's cluster, indicating a potentially smooth network environment, the weight of geographical distance can be increased to achieve the lowest latency. This dynamic adjustment can be global, such as being triggered based on the network-wide average, or it can be a localized adjustment targeting a specific geographic region or a designated computing cluster.

[0107] This invention provides real-time statistics on the number of connections and throughput of computing clusters in each region, dynamically adjusts the mapping weights between IP segments and clusters, avoids single-point overload, and achieves load balancing. Through this adaptive weighting mechanism, the scheduling system can maintain optimal decision-making when network conditions fluctuate or traffic peaks, avoiding sending tasks to actually congested clusters due to rigid adherence to the proximity principle, thereby improving overall scheduling robustness and computing power utilization.

[0108] In one embodiment, step S3.2 includes: Step S3.2.1': Based on the computing power service type, determine multiple candidate computing power clusters from multiple computing power clusters.

[0109] Step S3.2.2': Determine the client's home region based on the regional affiliation information.

[0110] For a description of steps S3.2.1' and S3.2.2', please refer to the relevant content of the above embodiments, which will not be repeated here.

[0111] Step S3.2.3': Determine the computing power cluster that is geographically closest to the region of origin from multiple candidate computing power clusters, and use it as the target computing power cluster.

[0112] In one embodiment, based on a candidate computing power cluster geographic location database (e.g., address information of each computing power cluster deployment), the geographic distance between each candidate computing power cluster and its home region is calculated one by one. Optionally, this distance can also be expressed as Euclidean distance, road network distance, or other distances. The computing power cluster with the closest geographic distance to its home region is determined from multiple candidate computing power clusters and selected as the target computing power cluster. By quantifying geographic proximity, a key weighting factor is provided for subsequent scheduling strategies, aiming to prioritize computing power clusters with shorter physical links and lower transmission latency, thereby effectively reducing data round-trip time.

[0113] In one embodiment, the fusion of dual identifiers included in the target identifier can achieve precise routing decisions at the millisecond level.

[0114] For example, a two-condition matching rule chain can be built at the gateway layer to complete scheduling decisions in milliseconds, achieving a high-performance scheduling mechanism. First, the type of computing service, such as inference, training, or storage services, is quickly identified based on the service identifier. Second, the request's source or destination region is resolved based on the IP identifier, such as East China, North China, or overseas nodes. These two conditions are matched in parallel or sequentially within the rule chain, and the gateway ultimately routes the request to the optimal computing cluster that matches the computing service type and geographical proximity, thus meeting the comprehensive requirements of data compliance, low latency, and high computing power utilization.

[0115] In one embodiment, the computing power service type is identified as inference service based on the service identifier, and the geographic affiliation information is obtained based on the IP identifier. If the geographic affiliation information determines that the client's affiliation region is East China, then the Hypertext Transfer Protocol request is routed to "East China - Inference Low-Latency Computing Power Cluster".

[0116] In one embodiment, the computing power service type is identified as training service based on the service identifier, and the geographic affiliation information is obtained based on the IP identifier. If the geographic affiliation information determines that the client's affiliation region is North China, then the Hypertext Transfer Protocol request is routed to "North China - Training High-Throughput Computing Power Cluster".

[0117] The technical effects of step S3.2.3': This mechanism achieves precise delivery at the single-request level. Different requests on the same computing power route may be directed to different computing power clusters due to differences in service type; requests of the same service type may be allocated to nodes in different regions due to differences in IP source. Scheduling precision is reduced from "cluster level" to a three-dimensional control of "route level - region level - node level". Cross-validation of IP identifiers and service identifiers ensures that the dual goals of "proximity" and "adaptation" are achieved simultaneously. It enables fine-grained traffic distribution and load balancing, effectively avoiding single-point overload.

[0118] In one embodiment, step S4 is followed by: Step S5: Obtain the health status of the computing cluster in each region.

[0119] In one embodiment, the system continuously acquires the health status of the computing cluster in each region. Optionally, this health status is a comprehensive indicator that integrates one or more of the following data from the computing cluster: average CPU / GPU utilization per node, memory usage, network latency, packet loss rate, task queuing length, error rate, throughput, and number of connections.

[0120] In one embodiment, the gateway layer can maintain a mapping table between the computing power clusters and health status of each region through active detection or passive monitoring.

[0121] The technical effect of step S5 is that it can provide data support for achieving fault self-healing based on the health of regional computing power clusters.

[0122] Step S6: In response to the fact that the health of the computing power cluster in any target area is less than the preset health threshold, the Internet Protocol segment traffic corresponding to the computing power cluster in the target area is switched to the backup cluster.

[0123] In one embodiment, a health threshold is set as a trigger condition. When the system detects that the health of any computing cluster in a target area is lower than the health threshold, it determines that the computing cluster in that target area is in a fault state and is no longer suitable to handle new traffic. This triggers a traffic switching action, switching all or a proportion of the traffic in the Internet Protocol segment corresponding to the computing cluster in that target area to a backup cluster. The backup cluster can be another available computing cluster in the same area or a disaster recovery computing cluster in a neighboring area.

[0124] Optionally, the present invention achieves session persistence or lossless migration by quickly updating the routing table or IP segment mapping rules of the gateway layer, so as to avoid interruption of user requests.

[0125] In one embodiment, taking a health threshold of 60 points as an example, if the health of the computing cluster in target area A drops below 60 points, the Internet Protocol segment traffic corresponding to the computing cluster in target area A will be switched to the backup cluster.

[0126] The technical effect of step S6 is that it enables millisecond- to second-level fault detection and traffic switching without manual intervention, significantly improving the robustness and self-healing ability of the computing power scheduling system when faced with local node load spikes, hardware failures, or network jitter. When the health of a regional cluster declines, the corresponding IP segment traffic is automatically switched to the backup cluster, achieving regional-level fault isolation and traffic escape.

[0127] It is important to emphasize that the single-request-level targeted scheduling computing power of this invention differs fundamentally from the cluster-level polling of existing technologies in several core dimensions. The technical effects are analyzed item by item below: I. Scheduling Granularity: From “Batch Average Allocation” to “Independent Decision-Making Per Package”.

[0128] In existing technologies, computing power clusters are used as the smallest scheduling unit, and a large number of requests are distributed evenly to each computing power cluster in batches, which cannot distinguish the individual differences between requests.

[0129] This invention enables request-level packet-by-packet independent decision-making, allowing different requests on the same route to be directed to different computing power clusters due to differences in service identifiers and IP identifiers.

[0130] Technical effect: The scheduling accuracy is reduced from the cluster level to a three-dimensional control of "router level-region level-node level", which improves the accuracy by N times (N is the number of cluster nodes), laying the foundation for subsequent accurate matching.

[0131] II. Routing latency: From "external dependence" to "local decision-making".

[0132] Existing technologies rely on external DNS or load balancers, requiring multiple network interactions and resolutions, with latency typically exceeding 50ms.

[0133] This invention directly completes identifier resolution and rule matching at the gateway layer. Based on the service identifier and IP identifier, it determines the target computing power cluster from multiple computing power clusters without the need for external queries.

[0134] Technical benefits: Reduced routing latency is crucial for millisecond-level response scenarios such as real-time inference and interactive AI.

[0135] III. Computing power matching accuracy: From "random mismatch" to "precise feature matching".

[0136] In existing technologies, requests are randomly assigned to a cluster, lacking awareness of service requirements (such as GPU-intensive, low-latency), which may lead to mismatches.

[0137] This invention performs precise feature matching based on the service identifier (such as computing power type and priority) carried in the request and the IP address's geographical region.

[0138] Technical benefits: Avoids mismatches, improves accuracy, and avoids wasting computing power.

[0139] IV. Computing power utilization: From "idle waste" to "differentiated saturation".

[0140] In existing technologies, heterogeneous clusters (containing nodes with different performance) are evenly distributed in terms of load, resulting in high-performance nodes being idle and low-performance nodes being overloaded, leading to a low average load.

[0141] This invention dynamically schedules requests to the most suitable node based on the actual computing power requirements of each request, thus achieving differentiated saturation scheduling.

[0142] Technical benefits: Increases peak load, improves the utilization rate of computing clusters, allows more tasks to be carried by the same hardware, and reduces idle costs.

[0143] V. Cross-regional traffic: From "seamless detour" to "nearest access".

[0144] In existing technologies, there is no geographic awareness, and requests may be assigned to remote clusters, resulting in a high proportion of cross-regional traffic, which increases bandwidth costs and latency.

[0145] This invention prioritizes assigning IP address to the nearest node within the same region based on the requested IP address's geographic location.

[0146] Technical effects: Reduces the proportion of cross-regional traffic, lowers bandwidth costs, reduces backbone network congestion, and improves user experience.

[0147] The core breakthrough of this invention lies in upgrading scheduling decisions from a "single-dimensional cluster-level" to a three-dimensional control at the "router-region-node-level," offering the systemic advantages of three-dimensional scheduling. This transformation not only brings linear improvements to individual metrics (such as reduced latency) but also generates systemic gains. Precise matching reduces retry and compensation costs, high utilization reduces hardware procurement, and proximity-based access optimizes network expenditures. These advantages collectively constitute the technological barrier of this invention in the field of intelligent computing cloud platforms.

[0148] The application areas of this invention are described below.

[0149] In one embodiment, a small amount of traffic (e.g., 5%) is redirected to the new version of the computing power cluster based on the service identifier, and the system automatically scales up or rolls back based on error rate and latency metrics. This approach is suitable for scenarios such as new model deployment verification and algorithm iteration risk control, enabling incremental releases and rapid fault isolation.

[0150] In one embodiment, traffic weighting (such as 50 / 50 or 90 / 10) binds different versions of requests to a fixed computing power pool, ensuring data isolation and statistical validity in the experimental group. This approach is suitable for scenarios such as model performance comparison and user experience optimization, providing data support for algorithmic decisions.

[0151] In one embodiment, dual clusters run in parallel, enabling one-click switching of full traffic via identifiers, and achieving zero-downtime deployments with health checks. This is suitable for critical service updates and emergency rollback scenarios, ensuring service continuity.

[0152] Please see Figure 4 , Figure 4 This is a structural diagram of a device for precisely scheduling computing power based on identifiers in an intelligent computing cloud platform provided by the present invention, as shown below. Figure 4 As shown, the intelligent computing cloud platform based on identifier-based precise scheduling of computing power device 400 includes: The request acquisition module 410 is used to acquire the Hypertext Transfer Protocol request sent by the client; The identifier acquisition module 420 is used to parse the Hypertext Transfer Protocol request and obtain the target identifier in the request header. The target identifier includes at least one of a service identifier and an Internet Protocol identifier. Cluster determination module 430 is used to determine the target computing power cluster from multiple computing power clusters based on the target identifier; The computing power scheduling module 440 is used to direct the Hypertext Transfer Protocol request to the target computing power cluster.

[0153] In one embodiment, the cluster determination module 430 is further configured to: The computing power service type is identified based on the service identifier, and the geographic affiliation information is obtained based on the Internet Protocol identifier. The target computing power cluster is determined from the plurality of computing power clusters based on the computing power service type and the geographical affiliation information.

[0154] In one embodiment, the cluster determination module 430 is further configured to: Based on the type of computing power service, multiple candidate computing power clusters are determined from the multiple computing power clusters; Obtain the real-time cluster load of each of the candidate computing power clusters; The target computing power cluster is determined from the multiple candidate computing power clusters based on the geographical affiliation information of each candidate computing power cluster and the real-time cluster load.

[0155] In one embodiment, the cluster determination module 430 is further configured to: In response to the computing power service type being one of inference service, training service, or fine-tuning service, a low-latency computing power cluster is determined from the plurality of computing power clusters as the candidate computing power cluster; In response to the fact that the computing power service type is an embedded service, the candidate computing power cluster is determined from the computing power clusters based on the GPU attribute parameters of the plurality of computing power clusters; The low-latency computing cluster is the computing cluster whose latency is less than a preset latency threshold.

[0156] In one embodiment, the cluster determination module 430 is further configured to: The client's home region is determined based on the regional affiliation information, and the geographical distance between each candidate computing power cluster and the home region is determined. A score value for each candidate computing power cluster is obtained by weighting the real-time cluster load and the corresponding geographical distance of each candidate computing power cluster. The target computing power cluster is determined from the candidate computing power clusters based on the score value.

[0157] In one embodiment, the cluster determination module 430 is further configured to: Obtain the number of connections and throughput for each computing cluster; Based on the number of connections and throughput of the computing cluster, the weighted average of the real-time cluster load and geographical distance is dynamically adjusted.

[0158] In one embodiment, the cluster determination module 430 is further configured to: Based on the type of computing power service, multiple candidate computing power clusters are determined from the multiple computing power clusters; The client's geographic location is determined based on the geographic location information. The target computing power cluster is determined from among the multiple candidate computing power clusters that is geographically closest to the region of origin.

[0159] In one embodiment, the intelligent computing cloud platform based on the device 400 for precise scheduling of computing power according to identifiers further includes a traffic switching module 450, used for: Obtain the health status of the computing clusters in each region; In response to the health status of the computing power cluster in any target area being less than a preset health status threshold, the Internet Protocol segment traffic corresponding to the computing power cluster in the target area is switched to the backup cluster.

[0160] The device for precise scheduling of computing power based on identifiers in the intelligent computing cloud platform provided by this invention is capable of implementing each process of the various embodiments of the above-mentioned method for precise scheduling of computing power based on identifiers in the intelligent computing cloud platform. The technical features are one-to-one and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0161] The device for precise scheduling of computing power based on identifiers in the intelligent computing cloud platform provided by this invention is capable of implementing each process of the various embodiments of the above-mentioned method for precise scheduling of computing power based on identifiers in the intelligent computing cloud platform. The technical features are one-to-one and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0162] It should be noted that the device for precisely scheduling computing power based on identifiers in the intelligent computing cloud platform of this invention can be a device, or it can be a component, integrated circuit, or chip in an electronic device.

[0163] The present invention also provides an electronic device, see [link to relevant documentation]. Figure 5 , Figure 5This is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. The electronic device includes a memory 501, a processor 502, and a program or instructions stored in the memory 501 that run on the processor 502. When the program or instructions are executed by the processor 502, they can achieve the following: Figure 1 The corresponding intelligent computing cloud platform's method for accurately scheduling computing power based on identifiers, and any steps in the implementation of this method to achieve the same beneficial effect, will not be elaborated here.

[0164] The processor 502 can be a CPU, ASIC, FPGA, or GPU.

[0165] Those skilled in the art will understand that all or part of the steps of the above-described method embodiment for precise scheduling of computing power based on identifiers in intelligent computing cloud platforms can be implemented by hardware related to program instructions, and the program can be stored in a readable medium.

[0166] The present invention also provides a readable storage medium on which a computer program is stored, and which, when executed by a processor, can perform the above-described functions. Figure 1 The corresponding intelligent computing cloud platform can achieve the same technical effect by using any step in the method embodiment of precise scheduling of computing power based on identifiers. Therefore, to avoid repetition, it will not be described in detail here. The storage medium mentioned includes, for example, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0167] The present invention also provides a computer program product, including computer instructions that, when executed by a processor, implement the above-described... Figure 1 The corresponding intelligent computing cloud platform uses the method of precisely scheduling computing power based on identifiers to achieve the same technical effect, and will not be described in detail here to avoid repetition.

[0168] The terms "first," "second," etc., used in this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or apparatuses. Additionally, the use of "and / or" in this invention indicates at least one of the connected objects, such as A and / or B and / or C, representing seven possibilities: A alone, B alone, C alone, both A and B present, both B and C present, both A and C present, and A, B, and C present.

[0169] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0170] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or second terminal device, etc.) to execute the methods of the various embodiments of the present invention.

[0171] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of the present invention.

Claims

1. A method for precise scheduling of computing power based on identifiers in an intelligent computing cloud platform, characterized in that, include: Step S1: Obtain the Hypertext Transfer Protocol request sent by the client; Step S2: Parse the Hypertext Transfer Protocol request to obtain the target identifier in the request header. The target identifier includes at least a service identifier and an Internet Protocol identifier. Step S3: Determine the target computing power cluster from multiple computing power clusters based on the target identifier; Step S4: Direct the Hypertext Transfer Protocol request to the target computing power cluster.

2. The method as described in claim 1, characterized in that, Step S3 includes: Step S3.1: Identify the computing power service type based on the service identifier, and obtain the geographic affiliation information based on the Internet Protocol identifier; Step S3.2: Determine the target computing power cluster from the plurality of computing power clusters based on the computing power service type and the geographical affiliation information.

3. The method as described in claim 2, characterized in that, Step S3.2 includes: Step S3.2.1: Determine multiple candidate computing power clusters from the multiple computing power clusters according to the computing power service type; Step S3.2.2: Obtain the real-time cluster load of each candidate computing power cluster; Step S3.2.3: Determine the target computing power cluster from the multiple candidate computing power clusters based on the geographical affiliation information of each candidate computing power cluster and the real-time cluster load.

4. The method as described in claim 3, characterized in that, Step S3.2.1 includes: Step S3.2.1.1: In response to the computing power service type being one of inference service, training service or fine-tuning service, determine a low-latency computing power cluster from the plurality of computing power clusters as the candidate computing power cluster; Step S3.2.1.2: In response to the computing power service type being an embedded service, determine the candidate computing power cluster from the computing power clusters based on the GPU attribute parameters of the plurality of computing power clusters; The low-latency computing cluster is the computing cluster whose latency is less than a preset latency threshold.

5. The method as described in claim 3 or 4, characterized in that, Step S3.2.3 includes: Step S3.2.3.1: Determine the client's home region based on the regional affiliation information, and determine the geographical distance between each candidate computing power cluster and the home region; Step S3.2.3.2: Weight the real-time cluster load and the corresponding geographical distance of each candidate computing power cluster to obtain the score value of each candidate computing power cluster; Step S3.2.3.3: Determine the target computing power cluster from the candidate computing power clusters based on the score value.

6. The method as described in claim 5, characterized in that, Also includes: Step S3.2.3.4: Obtain the number of connections and throughput of each computing cluster; Step S3.2.3.5: Based on the number of connections and throughput of the computing power cluster, dynamically adjust the weighting of the real-time cluster load and geographical distance.

7. The method as described in claim 2, characterized in that, Step S3.2 includes: Step S3.2.1': Determine multiple candidate computing power clusters from the multiple computing power clusters according to the computing power service type; Step S3.2.2': Determine the client's region of origin based on the region affiliation information; Step S3.2.3': Determine the computing power cluster that is geographically closest to the home region from among the multiple candidate computing power clusters, and use it as the target computing power cluster.

8. The method according to any one of claims 1-4, characterized in that, After step S4, the method further includes: Step S5: Obtain the health status of the computing cluster in each region; Step S6: In response to the health status of the computing power cluster in any target area being less than a preset health status threshold, the Internet Protocol segment traffic corresponding to the computing power cluster in the target area is switched to the backup cluster.

9. A device for precisely scheduling computing power based on identifiers in an intelligent computing cloud platform, characterized in that, include: The request retrieval module is used to retrieve Hypertext Transfer Protocol requests sent by the client; The identifier acquisition module is used to parse the Hypertext Transfer Protocol request and obtain the target identifier in the request header. The target identifier includes at least one of a service identifier and an Internet Protocol identifier. The cluster determination module is used to determine the target computing power cluster from multiple computing power clusters based on the target identifier; The computing power scheduling module is used to direct the Hypertext Transfer Protocol request to the target computing power cluster.

10. An electronic device, characterized in that, include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of the method for precise scheduling of computing power based on identifiers in an intelligent computing cloud platform as described in any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for precise scheduling of computing power based on identifiers in an intelligent computing cloud platform as described in any one of claims 1 to 8.

12. A computer program product, characterized in that, It includes computer instructions, which, when executed by a processor, implement the steps of the method for precise scheduling of computing power based on identifiers in an intelligent computing cloud platform as described in any one of claims 1 to 8.

Images