Business processing method and processing system for heterogeneous intelligent computing platform

Abstract

The invention provides a business processing method and system for a heterogeneous intelligent computing platform, and relates to the technical field of data processing, and the method comprises the steps: generating a resource pooling platform according to the resource configuration of the heterogeneous intelligent computing platform; receiving a to-be-processed service for task decomposition to obtain a task node set; obtaining a plurality of key task nodes to form a key task module; a hot standby key task module is constructed and is in shared connection with the key task module, and the hot standby key task module comprises a plurality of hot standby key task nodes corresponding to the key task nodes; performing real-time operation monitoring to obtain abnormal key task nodes, and identifying corresponding identification hot standby key task nodes; and performing switching processing on the identified hot standby key task node and the abnormal key task node. According to the invention, the technical problem that the hardware cost and the maintenance cost of a product are increased due to the fact that extra equipment and resources need to be input to carry out hot standby and fault switching when two identical devices are used as master and slave backups in the prior art is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a service processing method and a processing system for a heterogeneous intelligent computing platform. Background Art

[0002] In key industrial fields, with the continuous advancement of informatization, the requirements for high availability and high reliability of systems, services, data, communication networks, etc. are becoming increasingly strong, which has promoted the development of hot standby technology. Through technical methods such as redundant backup, fault detection and switching, and real-time synchronization, seamless switching during a fault is achieved to ensure the continuity of operations and the integrity of data. The hot standby technology has been relatively maturely developed on servers, and many implementation methods (such as cluster technology, virtualization technology, etc.) can be used for reference. However, as an embedded system platform, the heterogeneous intelligent computing platform has the characteristics of limited software and hardware resources but flexible scalability and specialization, and it is necessary to design and implement the hot standby technology according to the actual situation.

[0003] In the prior art, dual-machine hot standby is the most common implementation method of hot standby technology. By deploying the same application programs and databases on two or more servers, real-time synchronization and fault switching of data are achieved. When the primary server fails, the standby server will automatically take over the service to ensure the continuous operation of the job. However, using two exactly the same devices for primary and standby backup requires investing additional devices and resources for hot standby and fault switching, increasing the hardware cost and maintenance cost of the product. Summary of the Invention

[0004] The present application provides a service processing method and a processing system for a heterogeneous intelligent computing platform, aiming to solve the technical problem that in the prior art, using two exactly the same devices for primary and standby backup requires investing additional devices and resources for hot standby and fault switching, increasing the hardware cost and maintenance cost of the product.

[0005] In the first aspect disclosed by the present application, a service processing method for a heterogeneous intelligent computing platform is provided. The method includes: generating a resource pooling platform according to the resource configuration of the heterogeneous intelligent computing platform; when the resource pooling platform receives a service to be processed, decomposing the service to be processed to obtain a set of required task nodes; obtaining a plurality of key task nodes in the set of task nodes to form a key task module; constructing a hot standby key task module, sharing a connection between the hot standby key task module and the key task module, where the hot standby key task module includes a plurality of hot standby key task nodes corresponding to the plurality of key task nodes; performing real-time operation monitoring on the key task module to obtain abnormal key task nodes, and identifying an identified hot standby key task node corresponding to the abnormal key task node in the hot standby key task module; and performing switching processing on the identified hot standby key task node and the abnormal key task node.

[0006] In a second aspect disclosed in the present application, a service processing system for a heterogeneous intelligent computing platform is provided. The system is used for the service processing method of the above-mentioned heterogeneous intelligent computing platform, and the system includes: a resource pooling platform generation module, configured to generate a resource pooling platform according to the resource configuration of the heterogeneous intelligent computing platform; a task decomposition module, configured to decompose a to-be-processed service when the resource pooling platform receives the to-be-processed service, and obtain a required task node set; a first module construction module, configured to obtain multiple key task nodes in the task node set to form a key task module; a second module construction module, configured to construct a hot standby key task module, and share a connection between the hot standby key task module and the key task module, where the hot standby key task module includes multiple hot standby key task nodes corresponding to the multiple key task nodes; a key task node identification module, configured to perform real-time operation monitoring on the key task module, obtain abnormal key task nodes, and identify the marked hot standby key task nodes corresponding to the abnormal key task nodes in the hot standby key task module; a switching processing module, configured to perform switching processing between the marked hot standby key task nodes and the abnormal key task nodes.

[0007] One or more technical solutions provided in the present application have at least the following beneficial effects:

[0008] By creating a resource pooling platform, different types of resources can be managed and scheduled more effectively. This centralized resource management method improves resource utilization efficiency and reduces resource redundancy and waste; splitting complex to-be-processed services into multiple task nodes makes task management clearer and also facilitates precise allocation and optimization of resources for specific tasks. Such a structured processing method improves the flexibility of task processing and the system response speed; by constructing a key task module and a corresponding hot standby key task module and ensuring real-time synchronization between the two, the fault tolerance of the system is greatly enhanced. When a key task node fails, it can quickly switch to the hot standby node to ensure service continuity and data integrity; real-time monitoring of the running status of the key task module can promptly detect and handle abnormalities, further ensuring the stable operation of the system. Through the preset fault transfer mechanism, it can quickly respond to problems and minimize the impact of faults on services. This not only reduces downtime caused by faults or maintenance but also optimizes resource allocation and improves the overall operation efficiency of the system.

[0009] The above description is only an overview of the technical solutions of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features, and advantages of the present application more obvious and understandable, the following specific embodiments of the present application are specifically given. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a schematic flow diagram of a service processing method for a heterogeneous intelligent computing platform provided by an embodiment of the present application.

[0011] Figure 2 It is a schematic structural diagram of a service processing system for a heterogeneous intelligent computing platform provided by an embodiment of the present application.

[0012] Explanation of reference numerals: Resource pooling platform generation module 10, task decomposition module 20, first module construction module 30, second module construction module 40, key task node identification module 50, switching processing module 60. Detailed implementation manners

[0013] By providing a service processing method and a processing system for a heterogeneous intelligent computing platform in an embodiment of the present application, the technical problem in the prior art that two completely identical devices are used for master-slave backup, and additional devices and resources need to be invested for hot standby and fault switching, increasing the hardware cost and maintenance cost of the product, is solved.

[0014] After introducing the basic principle of the present application, various non-limiting implementation manners of the present application will be specifically introduced below with reference to the accompanying drawings of the specification. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0015] Embodiment 1, as Figure 1 shown, an embodiment of the present application provides a service processing method for a heterogeneous intelligent computing platform, and the method includes:

[0016] S100: Generate a resource pooling platform according to the resource configuration of the heterogeneous intelligent computing platform;

[0017] A heterogeneous intelligent computing platform is an embedded system development platform that includes various software and hardware resources and can be flexibly cut and combined to adapt to generate its own required dedicated embedded device. Obtain the resource configuration information of the heterogeneous intelligent computing platform. The resource configuration includes hardware resources and software resources. According to the resource type, such as computing power resources, memory resources, storage resources, network resources, etc., the resource configuration information is divided and virtualized to form multiple sub-resource pools, and a resource pooling platform is generated, which can provide resource sharing and flexible dynamic scheduling capabilities. The main control system of the resource pooling platform is responsible for managing these virtualized resource pools, can monitor the status of each resource pool in real time, and dynamically schedule the resources in the resource pool according to the task requirements to ensure the efficient use of the resource pool in service processing.

[0018] S200: When the resource pooling platform receives a service to be processed, decompose the service to be processed to obtain a set of required task nodes;

[0019] The resource pooling platform receives the services to be processed, which come from the upper-layer application requests or scheduling requests and can be different types of tasks, such as those requiring computing, data storage, communication transmission, etc. When receiving the services to be processed, task decomposition is performed with the aim of breaking down complex services into multiple subtasks. Each subtask can be executed independently, reducing the complexity of system processing. After task decomposition, one or more resource nodes are assigned to each subtask. Each subtask represents a task node, and the obtained set of task nodes consists of multiple task nodes, which will be scheduled and executed in subsequent steps. For example, if a certain task node requires a large amount of computing power and memory resources, the resource pool will allocate the corresponding computing and storage resources.

[0020] S300: Obtain multiple key task nodes in the set of task nodes to form a key task module;

[0021] When analyzing the set of task nodes, some task nodes are designated as key task nodes due to their importance in business processing, such as the core algorithms for data processing, key data storage, etc. These key task nodes are usually those that are crucial for business continuity, data integrity, and system stability. Integrate these key task nodes into a module to form a key task module. The task nodes in the key task module usually have higher resource guarantees, such as preferential processing capabilities, highly reliable storage, and enhanced security measures.

[0022] S400: Construct a hot standby key task module and share the connection between the hot standby key task module and the key task module, where the hot standby key task module includes multiple hot standby key task nodes corresponding to the multiple key task nodes;

[0023] To ensure a quick and seamless switch when a key task node fails, a hot standby key task module is constructed. The hot standby key task module contains hot standby key task nodes corresponding to each task node in the key task module. These hot standby nodes have the same functions and configurations as the original key task nodes but do not process actual data under normal circumstances. A shared connection is established between the key task module and the hot standby key task module to ensure real-time status synchronization between the two, including configuration information, operating status, critical resources, etc., ensuring that when a key task node fails, the hot standby node can immediately take over and continue task processing with little impact on the overall performance and stability of the system.

[0024] When a key task node fails, it automatically switches to the corresponding hot standby key task node. Through this design, not only is the high reliability and high availability of critical operations guaranteed, but it can also be achieved without increasing costs because no additional hardware resources are required.

[0025] S500: Monitor the real-time operation of the critical task module, obtain abnormal critical task nodes, and identify the corresponding identified hot standby critical task nodes in the hot standby critical task module;

[0026] Continuously monitor the operation status of each task node in the critical task module, including but not limited to performance metrics, error logs, system warnings, etc., aiming to capture any abnormalities that may affect the normal operation of the system in real time. When it is detected that the behavior of a task node deviates from the normal operation range, or signs of errors and failures appear, mark this node as an abnormal critical task node. The abnormal identification is based on multiple metrics, such as response time delay, processing capacity decline, unexpected restart, etc. Identify the corresponding hot standby node in the hot standby critical task module as the identified hot standby critical task node, ensuring that the configuration, data, and status of the hot standby node are completely synchronized with the abnormal node. This identification process ensures that when a switch is needed, the correct backup node can be quickly found to take over the task.

[0027] S600: Perform a switching process on the identified hot standby critical task node and the abnormal critical task node.

[0028] Perform a switching process on the identified hot standby critical task node and the abnormal critical task node. Specifically, stop all activities of the abnormal critical task node to prevent data corruption or further system instability. At the same time, activate the corresponding identified hot standby critical task node to let it take over all tasks and responsibilities of the original node. During the switching process, ensure that all critical data and operation status are synchronized from the abnormal node to the hot standby node to maintain business continuity and data integrity. The synchronization operation ensures that the hot standby node can seamlessly continue task processing.

[0029] Through such a monitoring and switching mechanism, the interruption of critical tasks caused by node failures can be minimized to the greatest extent. This not only improves the reliability of the system but also ensures data security and integrity during the execution of critical tasks. This design allows faults to be processed invisibly in the background, greatly improving the user experience and the overall performance of the system.

[0030] Furthermore, after obtaining the required set of task nodes, the method further includes:

[0031] S210: Obtain multiple non-critical task nodes in the set of task nodes to form a non-critical task module;

[0032] S220: Monitor the real-time operation of the non-critical task module to obtain abnormal non-critical task nodes;

[0033] S230: Recycle the abnormal non-critical task node to the resource pooling platform, perform resource scheduling in the resource pooling platform to form identification task nodes, and execute the tasks of the abnormal non-critical task node according to the identification task nodes.

[0034] In the task node set, task nodes are divided into two major categories: critical and non-critical. Non-critical task nodes involve operations that are important for the overall operation of the system but not immediately critical, such as background data processing, logging, etc. The selected non-critical task nodes are organized into an independent module to form a non-critical task module. This module allows the system to more organizedly manage and monitor tasks that do not directly affect the core business operation, while simplifying resource allocation and fault handling.

[0035] Similar to the critical task module, continuous monitoring of the running state of the non-critical task module is also carried out to ensure the normal operation of all task nodes. The monitoring includes performance metrics, error detection, efficiency tracking, etc., and automatically detects and marks non-critical task nodes with abnormal performance or failures to obtain abnormal non-critical task nodes. Among them, the determination of abnormality is based on task execution delay, failed operations, or other system alarms, etc.

[0036] Remove the abnormal non-critical task node from the normally running task queue and recycle it to the resource pooling platform, aiming to quickly reduce the impact of the faulty node on other nodes or the overall system. After recycling, re-schedule resources in the resource pooling platform, utilize its dynamic resource management ability, and quickly re-allocate resources to new or existing healthy nodes according to the current resource usage situation and priorities. After the resource re-scheduling, identify one or more task nodes to take over the tasks of the recycled node. These new task nodes form identification task nodes and start to execute the tasks of the original abnormal non-critical task node, which ensures business continuity and system efficiency and reduces potential service interruptions.

[0037] Furthermore, the method for constructing a hot standby critical task module includes:

[0038] S410: Identify multiple resource combinations corresponding to the multiple critical task nodes in the critical task module;

[0039] S420: Send the multiple resource combinations to the resource pooling platform, and the resource pooling platform performs resource matching according to the multiple resource combinations to obtain multiple matching resource combinations;

[0040] S430: Construct the hot standby critical task module according to the multiple matching resource combinations.

[0041] Identify the resource combinations on which each critical task node in the critical task module depends. The resource combinations include computing resources, storage resources, network resources, and other necessary software resources. The purpose of identifying these resource combinations is to ensure that the hot standby critical task module can replicate and maintain the same operating environment and performance as the original task node.

[0042] Send the identified multiple resource combinations to the resource pooling platform. The resource pooling platform performs resource matching based on the submitted resource combinations, that is, searches for available resources to ensure that each hot standby task node can obtain the required resource configuration. Resource matching ensures that the hot standby node can replicate the configuration of the original task node, so as to seamlessly take over the task when switching, and obtain multiple matching resource combinations according to the resource matching results.

[0043] Based on the multiple matching resource combinations, construct a hot standby critical task module. The construction of the hot standby critical task module includes initializing the hot standby node, configuring necessary resources, and setting up a synchronization mechanism, etc. After the construction is completed, the nodes of the hot standby module will synchronize the status with the original critical task node to ensure the consistency of data and application status during switching. By constructing a matching hot standby module, it can quickly switch when a critical node fails, ensuring business continuity and data integrity, and at the same time avoiding potential service interruptions caused by task node failures.

[0044] Furthermore, perform a switching process on the identified hot standby critical task node and the abnormal critical task node. The method includes:

[0045] S610: Replace the identified hot standby critical task node into the critical task module;

[0046] S620: Recycle the abnormal critical task node to the resource pooling platform, and the resource pooling platform performs resource optimization on the abnormal critical task node to output an optimized task node;

[0047] S630: Supplement the optimized task node into the hot standby critical task module.

[0048] When an abnormal situation occurs in the critical task node, immediately activate the corresponding identified hot standby critical task node from the hot standby critical task module. This hot standby node has been configured to seamlessly take over the tasks of the original node, automatically replace the abnormal critical task node with the healthy identified hot standby critical task node, and the identified hot standby critical task node takes over all tasks to ensure the continuity of critical operations and the stability of the system.

[0049] Recover the abnormal critical task nodes to the resource pooling platform, which helps isolate the problem nodes and prevent the abnormal nodes from affecting other parts of the system. The resource pooling platform will analyze and optimize the resources of the abnormal critical task nodes. The optimization process includes reconfiguring resources, updating software, fixing errors, or replacing damaged hardware components. Through resource optimization, optimized task nodes are output, which means that the nodes are ready to be activated and take over tasks again when needed.

[0050] Supplement the optimized task nodes back to the hot standby critical task module to restore the integrity of the hot standby resources, which ensures that the hot standby module always maintains sufficient standby nodes to prepare for any future failures.

[0051] By implementing the above steps, it is possible to quickly respond to the abnormalities of critical task nodes, ensure the uninterrupted operation of critical services through hot standby nodes. At the same time, the optimization processing and reuse of abnormal nodes improve the resource utilization efficiency and the overall reliability of the system.

[0052] Furthermore, for real-time operation monitoring of the critical task module to obtain abnormal critical task nodes, the method further includes:

[0053] S510: Train an operation anomaly monitoring model, where the operation anomaly monitoring model is obtained by training on a historical operation anomaly sample data set and label data indicating the degree of anomaly until convergence;

[0054] S520: Download the trained operation anomaly monitoring model to the resource pooling platform for storage, and perform real-time operation monitoring on the critical task module to obtain an operation monitoring data set;

[0055] S530: Invoke the operation anomaly monitoring model to analyze the operation monitoring data set to obtain the abnormal critical task nodes in the critical task module.

[0056] Collect historical operation data sets from the system's log files, monitoring systems, and fault reports, especially instance data where anomalies are known to have occurred. The data set contains various types of operation anomaly data, such as abnormal system performance metrics, error messages in log files, problems reported by users, etc. These operation anomaly data form a historical operation anomaly sample data set. Each sample in the historical operation anomaly sample data set is labeled with a label to indicate the degree of anomaly. The label data reflects the severity of the anomaly and helps the trained model accurately judge its importance and urgency when encountering similar situations.

[0057] According to the data characteristics and anomaly detection requirements, select an appropriate model for training, including decision trees, random forests, neural networks, etc. Use the historical running anomaly sample dataset and label data indicating the degree of anomaly to train the model. During the training process, continuously adjust the model parameters, such as the learning rate, number of iterations, etc., and adopt techniques such as cross-validation to verify the generalization ability and accuracy of the model until the model converges. After training is completed, obtain a running anomaly monitoring model for automatically identifying and marking abnormal behaviors for timely response.

[0058] Download the trained running anomaly monitoring model to the resource pooling platform for storage. The resource pooling platform is responsible for managing all computing and storage resources. Therefore, deploying the model to this platform can ensure that the model can efficiently access the required data and resources. When the model deployment is completed, start real-time running monitoring of the key task modules, including monitoring performance metrics, system logs, user feedback, etc., to obtain a running monitoring dataset and provide the latest information for anomaly detection.

[0059] Invoke the deployed running anomaly monitoring model to analyze the collected running monitoring dataset. During the model analysis process, evaluate whether the current running state is normal according to the features and criteria defined during training. If the model analysis result shows that there are abnormal behaviors in some key task nodes, mark these nodes as abnormal key task nodes for subsequent processing of abnormal nodes. The deployment of the model not only improves the accuracy of fault diagnosis but also speeds up the response speed, thus significantly enhancing the overall reliability of the system and user satisfaction.

[0060] Furthermore, according to the resource configuration of the heterogeneous intelligent computing platform, generate a resource pooling platform. The method includes:

[0061] S110: Obtain the resource configuration information of the heterogeneous intelligent computing platform, where the resource configuration information includes hardware resource configuration information and software resource configuration information;

[0062] S120: Divide the resource configuration information according to the resource type to generate a resource pooling platform, and the resource pooling platform is monitored and controlled through a main control system;

[0063] Among them, the resource pooling platform includes at least task resources, computing power resources, memory resources, storage resources, and network resources.

[0064] Obtain the hardware resource configuration information and software resource configuration information of the heterogeneous intelligent computing platform. Generally, the hardware resources of the heterogeneous intelligent computing platform include multiple functional embedded boards (MCUs with different architectures), such as computing power resource embedded boards, control processing resource embedded boards, storage resource embedded boards, communication resource embedded boards, etc.; its software resources include the PC-side development EDK environment, multiple embedded operating systems, operation and maintenance management plugins, scheduling management plugins, etc.

[0065] Organize and classify the resource configuration information according to the resource type, including grouping the hardware resources and software resources, so as to manage and schedule more effectively. The classification of resource types helps to optimize resource utilization and improve efficiency. For example, allocate compute-intensive tasks to servers with stronger computing power resources, or allocate data-intensive tasks to systems with rich storage resources. After classification, a resource pooling platform is generated. This platform aggregates all the classified resources and is uniformly managed and monitored by the main control system to ensure that resources are allocated on demand and at the same time handle various resource conflicts and dependency issues.

[0066] Among them, the resource pooling platform includes at least task resources, computing power resources, memory resources, storage resources, and network resources. Task resources refer to computable tasks available for allocation. Computing power resources refer to processing capabilities. Memory resources and storage resources refer to physical and virtual storage devices required for data storage and processing. Network resources refer to network bandwidth and interfaces used for data transmission.

[0067] Furthermore, obtain multiple key task nodes in the set of task nodes to form a key task module. The method further includes:

[0068] S310: Monitor the resource load of the key task module to obtain resource load indicators;

[0069] S320: If the resource load indicator is greater than the preset resource load indicator, send an expansion instruction to the resource pooling platform to generate expansion resources, and configure the resources of the key task module according to the expansion resources.

[0070] Regularly monitor the resource load of the key task module, such as monitoring CPU utilization, memory usage, storage space occupancy, and network bandwidth usage, and convert the collected data into resource load indicators. These indicators can reflect the current operating state and resource utilization efficiency of the system.

[0071] Compare the resource load index with the preset resource load index. The preset resource load index is a pre-set threshold value, and these preset threshold values define the safe or ideal upper limit of resource usage, ensuring that the system will not experience performance degradation due to resource overload. If the resource load index exceeds the preset resource load index, a resource expansion instruction is automatically sent to the resource pooling platform. After receiving the expansion instruction, the resource pooling platform dynamically allocates additional resources to meet the needs of critical task modules, including adding more computing power, additional memory, larger storage space, or enhanced network bandwidth. After the expanded resources are successfully allocated, these resources are configured into the critical task modules. This configuration process ensures that the critical task modules can still operate stably and efficiently when the resource requirements increase, improving the reliability and availability of the system and reducing performance bottlenecks caused by insufficient resources.

[0072] In summary, the business processing method of a heterogeneous intelligent computing platform provided by the embodiments of the present application has the following technical effects:

[0073] By creating a resource pooling platform, different types of resources can be managed and scheduled more effectively. This centralized resource management method improves resource utilization efficiency and reduces resource redundancy and waste; splitting complex business to be processed into multiple task nodes makes task management clearer and also facilitates precise allocation and optimization of resources for specific tasks. Such a structured processing method improves the flexibility of task processing and the system response speed; by constructing critical task modules and corresponding hot standby critical task modules and ensuring real-time synchronization between the two, the fault tolerance of the system is greatly enhanced. When a critical task node fails, it can quickly switch to the hot standby node to ensure business continuity and data integrity; real-time monitoring of the operating status of critical task modules, timely discovery and handling of abnormalities, further ensures the stable operation of the system. Through the preset failover mechanism, problems can be quickly responded to, minimizing the impact of faults on the business. This not only reduces downtime caused by faults or maintenance, but also optimizes resource allocation and improves the overall operating efficiency of the system.

[0074] Embodiment 2, based on the same inventive concept as the business processing method of a heterogeneous intelligent computing platform in the foregoing embodiment, as Figure 2 shown, the embodiments of the present application provide a business processing system of a heterogeneous intelligent computing platform, and the system includes:

[0075] A resource pooling platform generation module 10 for generating a resource pooling platform according to the resource configuration of a heterogeneous intelligent computing platform; a task decomposition module 20 for decomposing a to-be-processed service when the resource pooling platform receives the to-be-processed service to obtain a required set of task nodes; a first module construction module 30 for obtaining multiple key task nodes in the set of task nodes to form a key task module; a second module construction module 40 for constructing a hot standby key task module and sharing a connection between the hot standby key task module and the key task module, where the hot standby key task module includes multiple hot standby key task nodes corresponding to the multiple key task nodes; a key task node identification module 50 for performing real-time operation monitoring on the key task module to obtain abnormal key task nodes and identifying, in the hot standby key task module, the marked hot standby key task nodes corresponding to the abnormal key task nodes; and a switching processing module 60 for performing switching processing between the marked hot standby key task nodes and the abnormal key task nodes.

[0076] Furthermore, the task decomposition module 20 includes:

[0077] A non-key task node acquisition unit for obtaining multiple non-key task nodes in the set of task nodes to form a non-key task module; a real-time operation monitoring unit for performing real-time operation monitoring on the non-key task module to obtain abnormal non-key task nodes; and a task execution unit for recycling the abnormal non-key task nodes to the resource pooling platform, performing resource scheduling in the resource pooling platform to form marked task nodes, and executing the tasks of the abnormal non-key task nodes according to the marked task nodes.

[0078] Furthermore, the second module construction module 40 includes:

[0079] A resource combination identification unit for identifying multiple resource combinations corresponding to the multiple key task nodes in the key task module; a resource matching unit for sending the multiple resource combinations to the resource pooling platform, and the resource pooling platform performing resource matching according to the multiple resource combinations to obtain multiple matching resource combinations; and a task module construction unit for constructing the hot standby key task module according to the multiple matching resource combinations.

[0080] Furthermore, the switching processing module 60 includes:

[0081] A replacement unit for replacing the identified hot standby critical task node to the critical task module; a resource optimization unit for recycling the abnormal critical task node to the resource pooling platform, where the resource pooling platform optimizes the resources of the abnormal critical task node and outputs an optimized task node; a supplement unit for supplementing the optimized task node to the hot standby critical task module.

[0082] Furthermore, the critical task node identification module 50 includes:

[0083] A monitoring model training unit for training an operation anomaly monitoring model, where the operation anomaly monitoring model is trained to convergence by using a historical operation anomaly sample data set and label data indicating the anomaly degree; a real-time operation monitoring unit for downloading the trained operation anomaly monitoring model to the resource pooling platform for storage, and performing real-time operation monitoring on the critical task module to obtain an operation monitoring data set; a data analysis unit for calling the operation anomaly monitoring model to analyze the operation monitoring data set to obtain abnormal critical task nodes in the critical task module.

[0084] Furthermore, the resource pooling platform generation module 10 includes:

[0085] A resource configuration information acquisition unit for acquiring the resource configuration information of the heterogeneous intelligent computing platform, where the resource configuration information includes hardware resource configuration information and software resource configuration information; a resource pooling platform generation unit for dividing the resource configuration information according to resource types to generate a resource pooling platform, and the resource pooling platform is monitored and controlled through a main control system; wherein, the resource pooling platform at least includes task resources, computing power resources, memory resources, storage resources, and network resources.

[0086] Furthermore, the first module construction module 30 includes:

[0087] A resource load monitoring unit for monitoring the resource load of the critical task module to obtain resource load indicators; a resource configuration unit for, if the resource load indicator is greater than a preset resource load indicator, sending an expansion instruction to the resource pooling platform to generate expansion resources, and configuring the resources of the critical task module according to the expansion resources.

[0088] Through the foregoing detailed description of a service processing method for a heterogeneous intelligent computing platform in this specification, those skilled in the art can clearly know a service processing system for a heterogeneous intelligent computing platform in this embodiment. Since it corresponds to the method disclosed in the embodiment, it is described relatively simply, and for related parts, reference may be made to the description in the method part.

[0089] The foregoing description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A service processing method for a heterogeneous intelligent computing platform, characterized in that: The method comprises: Generate a resource pooling platform based on the resource configuration of the heterogeneous intelligent computing platform; When the resource pooling platform receives the business to be processed, it decomposes the business to be processed into tasks and obtains a required task node set; Acquire multiple key task nodes in the task node set to form a key task module; Constructing a hot standby critical task module, and sharing a connection between the hot standby critical task module and the critical task module, wherein the hot standby critical task module includes a plurality of hot standby critical task nodes corresponding to the plurality of critical task nodes; Performing real-time operation monitoring on the critical task module, obtaining abnormal critical task nodes, and identifying the hot standby critical task node corresponding to the abnormal critical task node in the hot standby critical task module; The identified hot standby critical task node and the abnormal critical task node are switched.

2. The method according to claim 1, characterized in that After obtaining the required task node set, the method further includes: Acquire multiple non-critical task nodes in the task node set to form a non-critical task module; Performing real-time operation monitoring on the non-critical task modules to obtain abnormal non-critical task nodes; The abnormal non-critical task node is recycled to the resource pooling platform, resource scheduling is performed in the resource pooling platform to form an identified task node, and the task of the abnormal non-critical task node is executed according to the identified task node.

3. The method according to claim 1, characterized in that Building a hot standby critical mission module, the method includes: Identifying multiple resource combinations corresponding to the multiple key task nodes in the key task module; Sending the multiple resource combinations to the resource pooling platform, the resource pooling platform performs resource matching according to the multiple resource combinations to obtain multiple matching resource combinations; The hot standby critical task module is constructed according to the multiple matching resource combinations.

4. The method according to claim 3, characterized in that The method of performing switching processing on the identified hot standby critical task node and the abnormal critical task node includes: Replacing the identified hot standby critical task node with the critical task module; Recycling the abnormal critical task node to the resource pooling platform, and having the resource pooling platform optimize resources for the abnormal critical task node and output an optimized task node; The optimized task node is added to the hot standby critical task module.

5. The method according to claim 1, characterized in that Performing real-time operation monitoring on the key task module to obtain abnormal key task nodes, the method further includes: Training an operation anomaly monitoring model, wherein the training operation anomaly monitoring model is obtained by training a historical operation anomaly sample data set and label data identifying the degree of anomaly until convergence; Downloading the trained operation abnormality monitoring model to the resource pooling platform for storage, and performing real-time operation monitoring on the key task module to obtain an operation monitoring data set; The operation abnormality monitoring model is called to analyze the operation monitoring data set to obtain abnormal key task nodes in the key task module.

6. The method according to claim 1, characterized in that Generate a resource pooling platform based on the resource configuration of the heterogeneous intelligent computing platform. The method includes: Obtain resource configuration information of the heterogeneous intelligent computing platform, wherein the resource configuration information includes hardware resource configuration information and software resource configuration information; The resource configuration information is divided according to resource types to generate a resource pooling platform, and the resource pooling platform is monitored and controlled by a main control system; Among them, the resource pooling platform includes at least task resources, computing resources, memory resources, storage resources and network resources.

7. The method according to claim 1, characterized in that Acquire multiple key task nodes in the task node set to form a key task module, and the method further includes: Perform resource load monitoring on the key task module to obtain resource load indicators; If the resource load index is greater than a preset resource load index, an extension instruction is sent to the resource pooling platform to generate extended resources, and the resources of the key task module are configured according to the extended resources.

8. A business processing system for a heterogeneous intelligent computing platform, characterized in that: A system for implementing a business processing method for a heterogeneous intelligent computing platform according to any one of claims 1 to 7, comprising: The resource pooling platform generation module is used to generate a resource pooling platform based on the resource configuration of the heterogeneous intelligent computing platform; A task decomposition module is used to decompose the pending business into tasks when the resource pooling platform receives the pending business and obtain the required task node set; A first module construction module is used to obtain a plurality of key task nodes in the task node set to form a key task module; A second module construction module is used to construct a hot standby critical task module, and share a connection between the hot standby critical task module and the critical task module, wherein the hot standby critical task module includes a plurality of hot standby critical task nodes corresponding to the plurality of critical task nodes; A critical task node identification module is used to perform real-time operation monitoring on the critical task module, obtain abnormal critical task nodes, and identify the hot standby critical task node corresponding to the abnormal critical task node in the hot standby critical task module; The switching processing module is used to perform switching processing on the identified hot standby critical task node and the abnormal critical task node.

Images