Container elastic computing power conveying method oriented to large computing power scene

Abstract

The embodiment of the invention provides a high-computing-power scene-oriented container elastic computing power conveying method, device and equipment and a medium. The method comprises the following steps of: acquiring a computing power task sent by a client; determining a target container matched with the computing power task in a container cluster; the container cluster comprises a plurality of containers; the container is obtained by virtualization according to computing power resources in the computing power platform; processing the computing power task through the target container to obtain a processing result; sending a processing result to the client; through a mode of acquiring the computing power task sent by the client and accurately matching the target container, computing power resources can be quickly matched with specific task requirements, and mismatching and waste of the computing power resources are avoided; as the container is generated based on the computing power resource virtualization of the computing power platform, the flexible characteristic of the container can fully adapt to the requirements of complex model and large data volume in a large computing power scene, and the utilization efficiency of the computing power resource is improved.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to a method, apparatus, equipment and medium for elastic computing power delivery to containers for high computing power scenarios. Background Technology

[0002] In recent years, the rapid development of artificial intelligence technology has greatly promoted innovation and progress in various fields. With the continuous increase in the complexity of artificial intelligence models and the explosive growth of data, the demand for high-performance computing resources has also increased. However, how to manage resources and deliver computing power more efficiently is an urgent problem to be solved. Summary of the Invention

[0003] In view of the above problems, embodiments of the present invention are proposed to provide a container elastic computing power delivery method, apparatus, device and medium for high computing power scenarios that overcomes or at least partially solves the above problems.

[0004] To address the aforementioned problems, this invention discloses a method for elastic container computing power delivery in high-computing-power scenarios, applied to a computing platform connected to a client; the computing platform deploys a container cluster, and the method includes: Obtain the computing power task sent by the client; A target container matching the computing power task is identified within the container cluster; the container cluster includes multiple containers; the container is obtained by virtualizing the computing power resources in the computing power platform. The computing power task is processed through the target container to obtain the processing result; The processing result is sent to the client.

[0005] Optionally, it also includes: Obtain the computing resources deployed in the computing platform; The computing resources are virtualized to obtain the multiple containers; The container cluster is obtained by deploying the multiple containers.

[0006] Optionally, the computing resources include computing servers, and the virtualization of the computing resources to obtain the plurality of containers includes: Collect hardware configuration parameters for each computing server, including at least one of the following: number of computing cores, computing power, memory capacity, storage type, IO performance, and network interface bandwidth. Based on the hardware configuration parameters of each computing server, the computing servers are classified to obtain multiple types of computing servers; Virtualize the various types of computing servers to determine the various containers.

[0007] Optionally, the computing power task includes at least one of computing power requirement, task priority, and task type; The step of determining the target container in the container cluster that matches the computing power task includes: In the container cluster, candidate containers that match the computing power task are identified; Determine the idle level of the candidate containers; The target container is determined from the candidate containers based on their idle status.

[0008] Optionally, the target container includes multiple computing servers, and the step of processing the computing task through the target container to obtain the processing result includes: The computing power task is divided into multiple computing power sub-tasks; The target computing server corresponding to each computing subtask is determined by matching the computing servers of the target container; The processing result is obtained by processing the multiple computing subtasks through the target computing server.

[0009] Optionally, obtaining the processing result by processing the plurality of computing power subtasks through the target computing power server includes: Determine the processing order of the multiple computing subtasks; According to the processing order, the computing power sub-results obtained by the target computing power server are sequentially sent to the target computing power server corresponding to the next computing power sub-task for processing to obtain the processing result.

[0010] Optionally, the computing server in the target container transmits data based on the v2v protocol.

[0011] This invention also discloses a container elastic computing power delivery device for high-computing-power scenarios, applied to a computing power platform connected to a client; the computing power platform deploys a container cluster, and the device includes: The task acquisition module is used to acquire computing power tasks sent by the client; A container determination module is used to determine a target container in the container cluster that matches the computing power task; the container cluster includes multiple containers; the container is obtained by virtualization based on the computing power resources in the computing power platform. The processing module is used to process the computing task through the target container to obtain the processing result; The sending module is used to send the processing result to the client.

[0012] The present invention also discloses an electronic device, comprising: a processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein when the computer program is executed by the processor, it implements the steps of the container elastic computing power delivery method for high computing power scenarios described above.

[0013] The present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the container elastic computing power delivery method for high computing power scenarios described above.

[0014] The embodiments of the present invention have the following advantages: This invention discloses a method, apparatus, device, and storage medium for elastic computing power delivery to containers in high-computing-power scenarios. By acquiring computing tasks sent by clients and accurately matching them with target containers, computing resources can be quickly matched with specific task requirements, avoiding mismatch and waste of computing resources. Since containers are generated based on the virtualization of computing resources on a computing platform, their flexibility can fully adapt to the needs of complex models and large data volumes in high-computing-power scenarios, improving the utilization efficiency of computing resources. Computing tasks can be processed directly through target containers, and the results can be fed back to the client after processing, simplifying the computing power delivery process, reducing losses in intermediate links, and thus achieving efficient management and accurate delivery of computing resources. This effectively alleviates the contradiction between the surge in computing resource demand and insufficient management and delivery efficiency in current high-computing-power scenarios. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the steps of a container elastic computing power delivery method for high-computing-power scenarios provided by an embodiment of the present invention. Figure 2 This is a flowchart illustrating a container elastic computing power delivery method for high-computing-power scenarios provided by an embodiment of the present invention; Figure 3 This is a structural block diagram of a container elastic computing power delivery device for high computing power scenarios provided by an embodiment of the present invention. Detailed Implementation

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] One of the core concepts of this invention is that by acquiring the computing power tasks sent by the client and accurately matching them with target containers, the invention enables rapid matching of computing power resources with specific task requirements, avoiding mismatch and waste of computing power resources. Since containers are generated based on the virtualization of computing power resources on a computing power platform, their flexible characteristics can fully adapt to the needs of complex models and large data volumes in high-computing power scenarios, improving the utilization efficiency of computing power resources. Computing power tasks can be processed directly through target containers, and the results can be fed back to the client after processing, simplifying the computing power delivery process, reducing losses in intermediate links, and thus achieving efficient management and rapid and accurate computing power delivery of computing power resources. This effectively alleviates the contradiction between the surge in computing power resource demand and insufficient management and delivery efficiency in current high-computing power scenarios.

[0018] Reference Figure 1 The diagram illustrates a flowchart of a method for elastic computing power delivery to containers in high-computing-power scenarios, as provided by an embodiment of the present invention. This method may include: Step 101: Obtain the computing power task sent by the client.

[0019] In this embodiment of the invention, the client encompasses various devices requiring high computing power, such as research institution terminals for scientific computing, enterprise servers for AI model training, and industrial control equipment requiring real-time data processing. After determining the computing power task input by the user, the client can send the task to the computing power platform. These tasks may include task types such as deep learning training, fluid dynamics simulation, and large-scale data rendering, and required computing resource parameters such as the number of CPU cores, GPU memory size, memory capacity, and computation time requirements. When acquiring these tasks, a preliminary legality check and format parsing are performed to ensure that the task information is complete and conforms to the system processing specifications, avoiding subsequent process interruptions due to missing task data or format errors. At the same time, a unique identifier ID is assigned to each task for subsequent full-process tracking, management, and scheduling, laying the foundation for accurate matching of computing power resources.

[0020] Step 102: Identify the target container in the container cluster that matches the computing power task; the container cluster includes multiple containers; the containers are obtained by virtualizing the computing power resources in the computing power platform.

[0021] In this embodiment of the invention, the container cluster consists of multiple containers generated by virtualizing the physical resources of a computing platform. These computing platforms can integrate physical servers distributed in different regions and with different hardware architectures, such as server clusters equipped with multiple high-performance CPUs and GPUs. Through virtualization technology, physical computing resources are divided into independent and flexibly schedulable container units. Each container has an independent operating system environment, computing resource quota, and network space, which can avoid resource contention and interference between different tasks and achieve fine-grained allocation of computing resources.

[0022] When determining the target container, the computing power task requirements parameters can be parsed first. Then, all containers in the container cluster that are idle or elastically scalable are traversed. The available resources of the containers, such as CPU computing power, GPU model and memory, memory size, and storage I / O rate, are compared with the task requirements. At the same time, the latency sensitivity of the task is also considered. For tasks with low latency requirements, containers with shorter network links to the client and more sufficient network bandwidth are preferred. For large-scale batch processing tasks, a combination of container clusters with lower resource utilization and horizontal scalability can be selected. Through such multi-dimensional screening, one or more target containers that can efficiently carry the computing power task are finally determined, achieving a precise match between computing power resources and task requirements.

[0023] Step 103: Process the computing task through the target container to obtain the processing result.

[0024] In this embodiment of the invention, after determining the target container, the input data required for the task, such as AI training datasets and initial parameter files for scientific computing, can be transmitted to the target container's local or shared storage via a high-speed network through the transmission channel between the task data and the target container. This ensures the integrity and timeliness of data transmission. Subsequently, the target container will launch the corresponding computing engine or application according to the task type. For example, for deep learning training tasks, the container can launch deep learning frameworks such as TensorFlow and PyTorch and call GPU resources for model training; for large-scale data processing tasks, the container can launch distributed computing frameworks such as Spark and Flink and use CPU clusters for parallel data computing.

[0025] During task processing, the resource usage status of the target container, such as CPU utilization, GPU memory usage, memory usage, network I / O and disk I / O rates, and task execution progress, can be monitored in real time. If a bottleneck is found in container resources, such as CPU utilization remaining at 100% for an extended period causing task lag, the target container can be elastically scaled up using container orchestration tools such as Kubernetes. This could involve increasing the number of CPU cores, expanding GPU memory, or mounting more memory. If errors occur during task execution, such as data corruption or program abnormalities, a fault tolerance mechanism can be triggered. The container's snapshot rollback function can be used to restore the task to its normal state before execution, reload the data, and continue executing the task, ensuring that the task can proceed stably and uninterruptedly.

[0026] Once the target container has completed all computation steps, the processing results can be temporarily stored in the container's output directory, awaiting subsequent feedback.

[0027] Step 104: Send the processing result to the client.

[0028] In this embodiment of the invention, after the computing power platform generates the processing result in the target container, it can send the processing result to the client.

[0029] This invention discloses a container-based elastic computing power delivery method for high-computing-power scenarios. By acquiring computing tasks sent by clients and accurately matching them with target containers, computing resources can be quickly matched with specific task requirements, avoiding mismatch and waste of computing resources. Since containers are generated based on the virtualization of computing resources on a computing platform, their flexibility can fully adapt to the needs of complex models and large data volumes in high-computing-power scenarios, improving the utilization efficiency of computing resources. Computing tasks can be processed directly through target containers, and the results can be fed back to the client after processing, simplifying the computing power delivery process, reducing losses in intermediate links, and thus achieving efficient management and fast, accurate computing power delivery. This effectively alleviates the contradiction between the surge in computing resource demand and insufficient management and delivery efficiency in current high-computing-power scenarios.

[0030] In one embodiment of the present invention, the method further includes: acquiring computing resources deployed in a computing platform; virtualizing the computing resources to obtain multiple containers; and deploying multiple containers to obtain a container cluster.

[0031] In this embodiment of the invention, computing resources encompass all hardware and software resources that constitute the computing platform. The hardware may include high-performance server clusters distributed in different physical locations, including multi-core CPUs, high-memory GPUs, large-capacity memory modules, storage arrays composed of high-speed solid-state drives and hard disk drives, and network devices such as switches and routers that support high-speed data transmission. The software involves operating systems, drivers, virtualization layer software, etc.

[0032] The system can use resource detection tools to scan and collect information on these resources in real time, recording the specific parameters of each type of resource, such as CPU model, number of cores, clock speed, GPU model, video memory capacity, computing power, total memory capacity and read / write speed, total storage device capacity, read / write speed and interface type, network device bandwidth, latency and connection status, etc. At the same time, it will also monitor the current usage status and available resources of various resources to ensure that the acquired resource information is comprehensive and accurate, providing a reliable basis for subsequent virtualization processing.

[0033] After obtaining complete computing resource information, physical computing resources can be logically divided and abstracted into multiple independent containers. Specifically, virtualization technology breaks the boundary limitations of physical hardware. According to a preset resource allocation strategy, CPU computing power is divided into multiple independently schedulable computing units, GPU resources are divided into different virtual GPU instances according to computing power requirements, and memory and storage resources are allocated into independent storage spaces. At the same time, each container is configured with an independent network namespace and operating system environment. Each container contains the complete runtime environment required to run a specific computing task, including applications, dependency libraries, configuration files, etc., but is isolated from other containers and underlying physical resources through the virtualization layer. This allows multiple containers to run in parallel on the same physical hardware without interfering with each other. This virtualization process not only preserves the high-performance computing capabilities of physical resources, but also gives each container lightweight characteristics, the ability to start and destroy quickly, and can flexibly adapt to the resource requirements of different types of computing tasks.

[0034] After creating multiple containers, a container cluster can be formed by deploying these containers. During deployment, containers can be rationally orchestrated and organized based on their resource configuration characteristics, performance indicators, and the physical topology of the computing platform. For example, containers with similar computing capabilities can be grouped into the same resource pool, and containers that require frequent data interaction can be deployed on physical nodes with shorter network links. At the same time, communication rules and resource scheduling strategies between containers can be configured. Through container orchestration tools, unified management of the entire container cluster can be achieved, including operations such as starting, stopping, migrating, scaling up, and scaling down containers. This ensures that containers in the cluster can work together to form an overall computing resource pool with elastic scalability. In addition, the cluster is equipped with monitoring and load balancing mechanisms to monitor the running status and resource load of each container in real time. When a container is overloaded, some tasks are automatically scheduled to other idle containers. When the overall computing power demand increases, new containers are quickly started and added to the cluster. Conversely, redundant containers are automatically shut down to release resources, thereby achieving efficient and stable operation of the entire container cluster.

[0035] This invention divides physical resources into multiple flexibly schedulable containers through virtualization, avoiding resource idleness and waste, and significantly improving the overall resource utilization rate; the container cluster can dynamically adjust the number of containers and resource allocation according to task requirements, quickly respond to changes in computing power demand, and meet the real-time computing power supply in different scenarios.

[0036] In one embodiment of the present invention, the computing resources include computing servers. The computing resources are virtualized to obtain multiple containers, including: collecting hardware configuration parameters of each computing server, the hardware configuration parameters including at least one of the following: number of computing cores, computing power, memory capacity, storage type, IO performance, and network interface bandwidth; classifying each computing server according to the hardware configuration parameters to obtain multiple types of computing servers; and virtualizing the multiple types of computing servers to determine multiple containers.

[0037] In this embodiment of the invention, the core hardware information of each computing server can be collected first, including the number of computing cores, i.e., the number of physical cores and threads of the CPU, the number of computing cores of the GPU and NPU, computing power such as the CPU's clock speed and operation speed, the GPU's floating-point operation power, the NPU's AI computing power, memory capacity including total capacity and available capacity that can be allocated to computing tasks, storage type such as mechanical hard disk, solid-state disk or NVMe, IO performance, i.e. the data read and write rate of the storage device, and network interface bandwidth, i.e. the network transmission capability of the server to communicate with the outside world.

[0038] After obtaining the hardware configuration parameters, all computing servers can be classified based on these parameters, forming multiple types of computing servers. The classification logic can be based on the core computing power characteristics and applicable scenarios of the servers. For example, servers with a high number of CPU cores and large memory capacity but low GPU configuration can be classified as general computing servers, suitable for running traditional data processing tasks; servers equipped with multiple high-performance GPUs and large video memory capacity can be classified as graphics computing servers, suitable for deep learning training and graphics rendering tasks; servers with integrated dedicated NPUs and outstanding AI computing power can be classified as intelligent computing servers, specifically used for AI inference and neural network computing scenarios. Through such classification, different types of servers can be accurately matched with specific computing power requirements, clarifying the direction for subsequent containerization processing.

[0039] After classifying the servers, virtualization can be performed on different types of computing servers to determine multiple containers. For general-purpose computing servers, physical resources are divided into multiple lightweight containers based on their CPU and memory resources. Each container is allocated an appropriate number of CPU cores and memory capacity to ensure efficient operation of multi-threaded data processing tasks. For graphics computing servers, GPU resources are sliced ​​or virtualized using virtualization technology. Each container is allocated an independent virtual GPU instance and corresponding video memory, while matching CPU and memory resources are configured to meet the intensive GPU computing power requirements of deep learning frameworks. For intelligent computing servers, dedicated containers are built around NPU resources, and auxiliary resources matching the NPU computing power are allocated to ensure the efficient execution of AI inference tasks. Each container is given an independent operating environment and resource quota during the virtualization process, maintaining consistency with the computing power characteristics of the underlying server type while possessing flexible scheduling and rapid deployment capabilities.

[0040] Taking the development of autonomous driving algorithms as an example, this scenario involves multiple computing tasks simultaneously, including cleaning and preprocessing large-scale road test data, training deep learning models, and real-time inference verification of trained models. Through the virtualization process described above, the system first collects the hardware parameters of all servers, classifies servers equipped with multiple GPUs as graphics computing type and virtualizes them into multiple GPU containers to carry model training tasks; classifies servers with a high number of CPU cores as general computing type and virtualizes them into CPU containers to handle data cleaning tasks; and classifies servers with integrated NPUs as intelligent computing type and virtualizes them into NPU containers to be responsible for model inference verification. Different types of containers are precisely matched to the computing power requirements of different tasks, so that computing resources are used efficiently throughout the entire development process.

[0041] This invention enables precise hardware parameter collection and classification, allowing virtualized containers to be deeply adapted to the characteristics of underlying computing resources, ensuring that containers can fully utilize the server's hardware performance. Simultaneously, differentiated virtualization based on server type allows generated containers to accurately match different types of computing tasks, improving task processing efficiency. Furthermore, this approach lays the foundation for elastic scheduling of container clusters, enabling the system to quickly call corresponding types of containers based on task type, achieving intelligent allocation and efficient utilization of computing resources.

[0042] In one embodiment of the present invention, the computing power task includes at least one of computing power requirement value, task priority, and task type; determining the target container matching the computing power task in the container cluster includes: determining candidate containers matching the computing power task in the container cluster; determining the idle level of the candidate containers; and determining the target container from the candidate containers based on the idle level of the candidate containers.

[0043] In this embodiment of the invention, the computing power requirement value is the specific value of the CPU computing power, GPU memory or NPU computing power required by the task, the task priority is the urgency and importance ranking of the task, and the task type is such as data processing, deep learning training or AI inference.

[0044] Based on this information, candidate containers that match the computing power task can be selected from the container cluster. For example, for deep learning training tasks requiring high GPU computing power, the system will select GPU containers equipped with sufficient video memory and computing power from the cluster; for high-priority real-time data processing tasks, general-purpose computing containers with high CPU core count and fast I / O performance will be given priority. During the selection process, not only must the correspondence between task type and container type be matched, but it must also be ensured that the available resources of the containers can cover the computing power requirements of the task, thus forming a preliminary set of candidate containers.

[0045] After identifying candidate containers, their idle status can be further evaluated. The evaluation dimensions of idle status include the container's current resource utilization, such as CPU utilization, memory usage, and GPU load; remaining available resources, such as the number of remaining CPU cores and available GPU memory capacity; and the length of the currently executing task queue and the estimated completion time. For example, two GPU containers that both meet the computing power requirements of a certain task may have different idle statuses. One has a current CPU utilization of 30% and no queued tasks, while the other has a CPU utilization of 80% and multiple tasks waiting to be executed. Obviously, the former has a higher idle status and is more suitable for taking on new tasks. By comprehensively analyzing these indicators, the idle status of each candidate container can be quantified, providing a basis for the final selection of the target container.

[0046] Finally, the target container is determined from the candidate containers based on their idle level. Generally, candidate containers with high idle levels can be selected first to ensure that the task can start quickly and execute efficiently, reducing waiting time.

[0047] Taking a meteorological forecast data processing scenario as an example, this scenario involves various computing tasks, including high-priority rapid analysis of real-time meteorological data and low-priority batch modeling of historical meteorological data. When a real-time meteorological data analysis task requiring a large amount of CPU computing power enters the system, the system first analyzes its computing power requirement and high-priority attribute, and filters out all general computing containers in the cluster whose CPU configuration meets the requirements and supports real-time data processing as candidate containers. Then, the system evaluates the idle level of these candidate containers and finds that container A currently has a CPU utilization of only 20% and no other tasks, while container B has a CPU utilization of 60% and a low-priority task in the queue. Based on a comprehensive consideration of the priority of high-priority tasks and the idle level, the system finally selects container A as the target container to ensure that real-time meteorological data can be processed quickly and to support the timely issuance of meteorological warnings.

[0048] This invention ensures that computing tasks are allocated to the most suitable containers through multi-dimensional matching and filtering, which not only meets the computing power requirements and priority requirements of the tasks, but also makes full use of the idle resources of the containers, thereby improving the overall utilization rate of computing power resources. At the same time, the dynamic selection mechanism based on the degree of idleness effectively balances the load of the container cluster, reduces the waiting time of tasks, improves the efficiency of task processing and the response speed of the system, and provides flexible and efficient support for diverse computing power needs.

[0049] In one embodiment of the present invention, the target container includes multiple computing power servers. The computing power task is processed by the target container to obtain the processing result, including: dividing the computing power task into multiple computing power sub-tasks; matching and determining the target computing power server corresponding to each computing power sub-task from the computing power servers of the target container; and processing the multiple computing power sub-tasks by the target computing power server to obtain the processing result.

[0050] In this embodiment of the invention, after obtaining the computing power task, the complete computing power task can be reasonably divided into multiple independent and parallel-processable computing power sub-tasks.

[0051] After the computing power tasks are broken down, a corresponding target computing power server can be matched and determined from the multiple computing power servers contained in the target container for each computing power subtask. The matching process takes into account the specific requirements of each computing power subtask, such as the type of computing cores required by the subtask (CPU, GPU, or NPU), computing power intensity, memory usage requirements, data read / write speed requirements, and network transmission requirements. At the same time, it takes into account the hardware configuration parameters of each computing power server, such as the number of computing cores, computing power capacity, memory capacity, IO performance, network interface bandwidth, and current idle status, such as resource utilization and task queue length. For example, for deep learning-related subtasks that require high GPU computing power support, computing power servers equipped with high-performance GPUs and currently with high idle levels will be matched first; for subtasks that are mainly data read / write and have high IO performance requirements, computing power servers with better storage types and stronger IO performance will be matched. Through such precise matching, it is ensured that each computing power subtask can be allocated the most suitable hardware resources, maximizing the processing efficiency of a single server.

[0052] Once the target computing server for each computing subtask is determined, the subtask processing phase begins. If there is a need for data interaction between subtasks, the high-speed communication link within the target container can be used to achieve real-time data transmission and synchronization between servers, ensuring the continuity of subtask processing and data accuracy. After all computing subtasks are processed, the subtask results generated by each server can be summarized and integrated, and data verification and format processing can be performed according to the output requirements of the original computing task to finally form a complete computing task processing result.

[0053] Taking the multi-sensor data fusion processing scenario in the field of autonomous driving as an example, a car manufacturer needs to perform real-time fusion analysis on massive amounts of LiDAR, camera, and millimeter-wave radar data collected by vehicles to generate accurate environmental perception results. This computing task involves a large amount of data, high processing time requirements, and requires support from multiple computing powers. The system first breaks down the fusion processing task into a LiDAR point cloud data noise reduction subtask, a camera image recognition subtask, a millimeter-wave radar target ranging subtask, and a multi-source data association and matching subtask. Then, it matches the target server from the computing power server of the target container and assigns the image recognition subtask that requires high GPU computing power to a server equipped with a high-performance GPU. The system assigns computationally intensive point cloud denoising subtasks to servers with more CPU cores, real-time radar ranging subtasks to servers with better I / O performance and network bandwidth, and data association and matching subtasks to servers with larger memory capacity. Subsequently, each target server processes its corresponding subtask in parallel. The LiDAR server quickly completes point cloud denoising, the camera server accurately identifies road targets, and the millimeter-wave radar server outputs target distances in real time. Finally, data synchronization is achieved through the internal communication link of the container, completing multi-source data association and matching to generate a complete perception result of the vehicle's surrounding environment, providing support for autonomous driving decisions.

[0054] This invention improves the processing efficiency of computing tasks by splitting tasks and processing them in parallel across multiple servers, effectively shortening the overall processing time of tasks, and is especially suitable for large-scale, highly complex computing tasks. It also enables the refined utilization of computing resources by matching the most suitable server to the sub-task requirements, avoiding waste caused by resource mismatch, and fully leveraging the hardware advantages of each computing server in the target container.

[0055] In one embodiment of the present invention, the processing of multiple computing sub-tasks by a target computing power server to obtain processing results includes: determining the processing order of the multiple computing sub-tasks; and according to the processing order, sequentially sending the computing sub-results obtained by the target computing power server to the target computing power server corresponding to the next computing sub-task for processing to obtain processing results.

[0056] In this embodiment of the invention, a clear processing sequence can be established based on the data dependencies, business logic associations, and task type characteristics among the computing power subtasks. For example, the output of some subtasks may be the input data of other subtasks. Subtasks with direct dependencies must be processed in the order of output first and input last. For subtasks that have no direct data dependencies but have a business logic sequence, such as data cleaning before data analysis, the processing sequence must be determined according to the business process. For completely independent subtasks, the parallel or serial processing sequence can be flexibly arranged according to resource availability and task priority to ensure the orderly progress of the overall process.

[0057] After determining the processing order, the process enters the stage of transferring and progressively processing the results of the subtasks. First, the target computing server corresponding to the computing power subtask in the first processing order starts processing. After completion, it generates the computing power sub-result of that subtask. The system can automatically send this computing power sub-result to the target computing power server corresponding to the next computing power subtask according to the preset processing order through the high-speed communication mechanism inside the target container. After obtaining the previous sub-result, the receiving server uses it as input data, processes it in combination with its own task requirements, generates a new computing power sub-result, and then transfers it to the subsequent target computing power servers in the same way. This process continues until the target computing power server corresponding to the last computing power subtask completes processing and generates the final processing result.

[0058] Taking a smart city traffic flow prediction scenario as an example, a task requires predicting road traffic flow for the next 24 hours by analyzing historical traffic data, real-time traffic data, and meteorological data. This task can be broken down into a historical data cleaning sub-task, a real-time data integration sub-task, a meteorological factor correlation sub-task, and a traffic flow model prediction sub-task. The processing order is historical data cleaning → real-time data integration → meteorological factor correlation → traffic flow model prediction. First, the target computing server responsible for historical data cleaning performs noise reduction and completion processing on the original historical traffic data, generating a cleaned historical dataset and sending it to the server responsible for real-time data integration. This server integrates the historical dataset with the real-time traffic data in a spatiotemporal alignment to generate a comprehensive traffic dataset, which is then passed to the server processing meteorological factor correlation. This server combines meteorological data to analyze the impact of factors such as rainfall and high temperature on traffic flow, generating a traffic dataset with meteorological features, which is then sent to the traffic flow model prediction server. Finally, this server runs a prediction model based on the above dataset and outputs the traffic flow prediction results for each road segment for the next 24 hours, completing the entire task processing.

[0059] This invention ensures the correct handling of data dependencies and logical relationships between computing subtasks through a clearly defined processing sequence, avoiding errors caused by disordered order. The orderly flow and progressive processing of sub-results enables the coordinated linkage of computing resources, allowing multiple target computing servers to form an efficient processing chain around a unified task. At the same time, this serial and progressive processing mode fully utilizes the specialized processing capabilities of each server while ensuring data accuracy, improving the processing quality of individual subtasks and ensuring the overall processing efficiency of the task, providing reliable support for the accurate completion of complex computing tasks.

[0060] In one embodiment of the present invention, the computing server in the target container transmits data based on the v2v protocol.

[0061] In this embodiment of the invention, the computing server in the target container can transmit data based on the v2v protocol. It can take advantage of the low latency and high bandwidth characteristics of the V2V protocol and combine it with RDMA technology to enable data to be transmitted between chips in a more efficient way, reducing data transmission latency and loss, improving the efficiency of collaborative processing between servers, and the security features and fault tolerance of the protocol provide a guarantee for the reliable flow of computing subtask results, further enhancing the stability and efficiency of the entire target container in processing computing tasks.

[0062] like Figure 2 The diagram illustrates a flowchart of a container elastic computing power delivery method for high-computing-power scenarios provided by an embodiment of the present invention. The client can send a task request to the computing power platform. The computing power platform can determine the matching target container in the container cluster according to the task request, and then process the computing power task through the target container to obtain the processing result, and then send the processing result to the client.

[0063] This invention discloses a container-based elastic computing power delivery method for high-computing-power scenarios. By acquiring computing tasks sent by clients and accurately matching them with target containers, computing resources can be quickly matched with specific task requirements, avoiding mismatch and waste of computing resources. Since containers are generated based on the virtualization of computing resources on a computing platform, their flexibility can fully adapt to the needs of complex models and large data volumes in high-computing-power scenarios, improving the utilization efficiency of computing resources. Computing tasks can be processed directly through target containers, and the results can be fed back to the client after processing, simplifying the computing power delivery process, reducing losses in intermediate links, and thus achieving efficient management and fast, accurate computing power delivery. This effectively alleviates the contradiction between the surge in computing resource demand and insufficient management and delivery efficiency in current high-computing-power scenarios.

[0064] It should be noted that, for the sake of simplicity, the method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0065] like Figure 3 This diagram illustrates a structural block diagram of a container elastic computing power delivery device for high-computing-power scenarios, provided by an embodiment of the present invention. The device is applied to a computing power platform, which is connected to a client. The computing power platform deploys a container cluster. The device includes: The task acquisition module 201 is used to acquire computing power tasks sent by the client; The container determination module 202 is used to determine the target container that matches the computing power task in the container cluster; the container cluster includes multiple containers; the container is obtained by virtualization based on the computing power resources in the computing power platform; Processing module 203 is used to process computing tasks through the target container to obtain processing results; The sending module 204 is used to send the processing results to the client.

[0066] This invention discloses a container-based elastic computing power delivery device for high-computing-power scenarios. By acquiring computing tasks sent by clients and accurately matching them with target containers, it enables rapid matching of computing resources with specific task requirements, avoiding mismatch and waste of computing resources. Since containers are generated based on the virtualization of computing resources on a computing platform, their flexibility can fully adapt to the needs of complex models and large data volumes in high-computing-power scenarios, improving the utilization efficiency of computing resources. Computing tasks can be processed directly through target containers, and the results can be fed back to the client after processing, simplifying the computing power delivery process, reducing losses in intermediate links, and thus achieving efficient management and rapid, accurate computing power delivery. This effectively alleviates the contradiction between the surge in computing resource demand and insufficient management and delivery efficiency in current high-computing-power scenarios.

[0067] In one embodiment of the present invention, it further includes: The resource acquisition module is used to acquire computing resources deployed in the computing platform; A virtualization module is used to virtualize the computing resources to obtain the multiple containers; The deployment module is used to deploy the container cluster based on the multiple containers.

[0068] In one embodiment of the present invention, the computing resources include a computing server and a virtualization module, comprising: The data acquisition submodule is used to collect the hardware configuration parameters of each computing server. The hardware configuration parameters include at least one of the following: number of computing cores, computing power, memory capacity, storage type, IO performance, and network interface bandwidth. The classification submodule is used to classify the computing power servers according to their hardware configuration parameters to obtain multiple types of computing power servers. The virtualization submodule is used to virtualize the multiple types of computing servers and determine the multiple containers.

[0069] In one embodiment of the present invention, the computing power task includes at least one of computing power requirement value, task priority, and task type; the container determination module includes: The first determining submodule is used to determine candidate containers in the container cluster that match the computing power task; The second determining submodule is used to determine the idle level of the candidate container; The third determining submodule is used to determine the target container from the candidate containers based on the idle level of the candidate containers.

[0070] In one embodiment of the present invention, the target container includes multiple computing servers, and the processing module includes: The sub-module is used to divide the computing power task into multiple computing power sub-tasks; The fourth determination submodule is used to match and determine the target computing server corresponding to each computing subtask from the computing power servers of the target container; The processing submodule is used to process the multiple computing subtasks through the target computing power server to obtain the processing result.

[0071] In one embodiment of the present invention, the processing submodule includes: The fifth determining submodule is used to determine the processing order of the multiple computing power subtasks; The sending submodule is used to sequentially send the computing power sub-results processed by the target computing power server to the target computing power server corresponding to the next computing power sub-task for processing, according to the processing order, so as to obtain the processing result.

[0072] In one embodiment of the present invention, the computing server in the target container transmits data based on the v2v protocol.

[0073] This invention discloses a container-based elastic computing power delivery device for high-computing-power scenarios. By acquiring computing tasks sent by clients and accurately matching them with target containers, it enables rapid matching of computing resources with specific task requirements, avoiding mismatch and waste of computing resources. Since containers are generated based on the virtualization of computing resources on a computing platform, their flexibility can fully adapt to the needs of complex models and large data volumes in high-computing-power scenarios, improving the utilization efficiency of computing resources. Computing tasks can be processed directly through target containers, and the results can be fed back to the client after processing, simplifying the computing power delivery process, reducing losses in intermediate links, and thus achieving efficient management and rapid, accurate computing power delivery. This effectively alleviates the contradiction between the surge in computing resource demand and insufficient management and delivery efficiency in current high-computing-power scenarios.

[0074] This invention also provides an electronic device, comprising: It includes a processor, a memory, and a computer program stored in the memory and capable of running on the processor. When the computer program is executed by the processor, it implements the various processes of the above-described embodiments of the container elastic computing power delivery method for high computing power scenarios and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0075] This invention also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the various processes of the above-described embodiment of the container elastic computing power delivery method for high-computing-power scenarios and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0076] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0077] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0078] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0079] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0080] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0081] Although preferred embodiments of the present invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present invention.

[0082] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device 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 terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes the element.

[0083] The present invention provides a detailed description of a container elastic computing power delivery method, apparatus, device, and storage medium for high-computing-power scenarios. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for elastic computing power delivery to containers for high-computing-power scenarios, characterized in that, It is applied to a computing platform, which is connected to a client; The computing platform is deployed with a container cluster, and the method includes: Obtain the computing power task sent by the client; A target container matching the computing power task is identified within the container cluster; the container cluster includes multiple containers; the container is obtained by virtualizing the computing power resources in the computing power platform. The computing power task is processed through the target container to obtain the processing result; The processing result is sent to the client.

2. The container elastic computing power delivery method for high-computing-power scenarios according to claim 1, characterized in that, Also includes: Obtain the computing resources deployed in the computing platform; The computing resources are virtualized to obtain the multiple containers; The container cluster is obtained by deploying the multiple containers.

3. The container elastic computing power delivery method for high-computing-power scenarios according to claim 2, characterized in that, The computing resources include computing servers, and the virtualization of the computing resources to obtain the plurality of containers includes: Collect hardware configuration parameters for each computing server, including at least one of the following: number of computing cores, computing power, memory capacity, storage type, IO performance, and network interface bandwidth. Based on the hardware configuration parameters of each computing server, the computing servers are classified to obtain multiple types of computing servers; Virtualize the various types of computing servers to determine the various containers.

4. The container elastic computing power delivery method for high-computing-power scenarios according to claim 1, characterized in that, The computing power task includes at least one of computing power requirement value, task priority, and task type; The step of determining the target container in the container cluster that matches the computing power task includes: In the container cluster, candidate containers that match the computing power task are identified; Determine the idle level of the candidate containers; The target container is determined from the candidate containers based on their idle status.

5. The container elastic computing power delivery method for high-computing-power scenarios according to claim 1, characterized in that, The target container includes multiple computing servers, and the process of processing the computing task through the target container to obtain the processing result includes: The computing power task is divided into multiple computing power sub-tasks; The target computing server corresponding to each computing subtask is determined by matching the computing servers of the target container; The processing result is obtained by processing the multiple computing subtasks through the target computing server.

6. The container elastic computing power delivery method for high-computing-power scenarios according to claim 5, characterized in that, The step of processing the multiple computing subtasks through the target computing power server to obtain the processing result includes: Determine the processing order of the multiple computing subtasks; According to the processing order, the computing power sub-results obtained by the target computing power server are sequentially sent to the target computing power server corresponding to the next computing power sub-task for processing to obtain the processing result.

7. The container elastic computing power delivery method for high-computing-power scenarios according to claim 6, characterized in that, The computing server in the target container transmits data based on the v2v protocol.

8. A container-based flexible computing power delivery device for high-computing-power scenarios, characterized in that, It is applied to a computing platform, which is connected to a client; The computing platform is deployed with a container cluster, and the device includes: The task acquisition module is used to acquire computing power tasks sent by the client; A container determination module is used to determine a target container in the container cluster that matches the computing power task; the container cluster includes multiple containers; the container is obtained by virtualization based on the computing power resources in the computing power platform. The processing module is used to process the computing task through the target container to obtain the processing result; The sending module is used to send the processing result to the client.

9. An electronic device, characterized in that, include: The processor, the memory, and the computer program stored in the memory and capable of running on the processor, wherein the computer program, when executed by the processor, implements the steps of the container elastic computing power delivery method for high computing power scenarios as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, it implements the steps of the container elastic computing power delivery method for high computing power scenarios as described in any one of claims 1-7.

Images