Super-computing center network system and application method thereof

Abstract

The invention discloses a supercomputing center network system and an application method thereof, and the system comprises a high-speed interconnection module which is used for recognizing a network application scene in real time through a deep learning algorithm, querying a matched network topology according to the network application scene recognized in real time, and carrying out the calculation of the network according to a calculation task demand in the network and the dynamically matched network topology, dynamically connecting each computing node in the network through a switch or a router by adopting a high-speed internet technology; the management module is used for monitoring a network state index in real time, and predicting a network state in a period of time in the future by using a deep learning algorithm according to the real-time monitored network state index; calculating each index score of the predicted network state according to a preset network state index scoring rule, and dynamically adjusting routing control according to a preset routing adjustment strategy matched with the index corresponding to the lowest index score; and the storage module is used for storing the obtained calculation result data of each calculation node. The bandwidth utilization rate and the data transmission efficiency can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of supercomputing center network systems, and specifically to a supercomputing center network system and an application method thereof. Background Art

[0002] As an important infrastructure for high-performance computing, supercomputing centers play an increasingly important role in scientific research, engineering simulation, big data processing and other fields. As the complexity and scale of computing tasks continue to increase, the requirements for data transmission efficiency, network latency and system scalability are also getting higher and higher. Supercomputing centers must not only have powerful computing capabilities, but also have efficient network architecture to achieve seamless connection between large-scale data throughput and complex computing. This makes the optimal design of the network system one of the key factors to ensure the normal operation of the supercomputing center and improve the overall performance.

[0003] Existing supercomputing center network systems usually adopt a hierarchical structure design, which realizes mutual communication between nodes through multi-level network devices such as switches and routers. In addition, there are network design solutions based on specific protocols or custom standards, which coordinate the scheduling of data flows in the network through centralized management or distributed control mechanisms. These solutions have improved the speed of data interaction within the supercomputing center to a certain extent, but there are still certain limitations. Specifically, in the process of coping with complex computing tasks, traditional network architectures show performance bottlenecks, especially high network latency and insufficient bandwidth utilization, which directly affects the overall performance and efficiency of the supercomputing center. Summary of the invention

[0004] In order to improve bandwidth utilization and data transmission efficiency and reduce network latency, the present application provides a supercomputing center network system and its application method.

[0005] In a first aspect, the present application provides a supercomputing center network system, including: a high-speed interconnection module, a management module, and a storage module; The high-speed interconnection module is used to use a deep learning algorithm to identify network application scenarios in real time, and query the matching network topology according to the real-time identified network application scenarios; according to the computing task requirements in the network and the dynamically matched network topology, the high-speed interconnection network technology is used to dynamically connect the computing nodes in the network through switches or routers; wherein different network application scenarios correspond to different network topologies; the network application scenarios include: computing intensive and data intensive; the network topologies include: star ring, grid, tree and clustering; The management module is used to monitor network status indicators in real time, wherein the network status indicators include network traffic, delay and packet loss rate; based on the real-time monitoring of network status indicators, a deep learning algorithm is used to predict the network status in the future; the scores of each indicator of the predicted network status are calculated according to the preset network status indicator scoring rules, and the routing control is dynamically adjusted according to the preset routing adjustment strategy matching the indicator corresponding to the lowest indicator score; the preset routing adjustment strategy includes a load balancing strategy matching the network traffic indicator, a delay optimization routing strategy matching the delay indicator, and a reliability priority routing strategy matching the packet loss rate indicator; The storage module is used to obtain the calculation result data of each calculation node for storage.

[0006] By adopting the above solution, the high-speed interconnection module is used to dynamically adjust the network topology for different network application scenarios, thereby improving bandwidth utilization and data transmission efficiency; the management module is used to monitor network status indicators in real time to predict the network status, and the routing strategy is dynamically adjusted according to the prediction results, thereby achieving load balancing, latency optimization and reliability-first routing selection, ensuring efficient and stable network communications; the storage module is used to store network traffic data and its calculation results to ensure reliable data storage.

[0007] Preferably, the management module is also used to determine whether a preset state is reached based on the predicted network state, and when the preset state is reached, dynamically connect new computing nodes using high-speed Internet technology; the network state indicators also include traffic peak, average response time, and resource utilization; the preset state includes: the network traffic peak exceeds the preset traffic peak threshold, the average response time of the computing node exceeds the preset response time threshold, and the resource utilization of the computing nodes exceeding the preset proportion exceeds the preset resource utilization threshold.

[0008] By adopting the above solution, it is possible to predict in advance whether the network status will reach resource shortage or traffic peak period, and accordingly connect new computing nodes to increase computing nodes, thereby improving the scalability and flexibility of the system and ensuring that good network performance can be maintained under high load.

[0009] Preferably, the management module is further used to switch to a spare computing node or backup link when it is determined that there is an abnormality or failure in a computing node or a network link according to the predicted network status.

[0010] By adopting the above solution, when the predicted network status shows that there is an abnormality or failure in the computing node or network link, the system can automatically switch to the backup computing node or backup link, thereby improving the fault tolerance and reliability of the system and ensuring the stability and continuity of data transmission.

[0011] Preferably, the storage module is also used to match the corresponding storage method according to the real-time identification of the network application scenario, and complete the storage of the calculation result data in the matched storage method; the storage method includes: using a high-speed storage area network storage method for computationally intensive matching and using a distributed storage network storage method for data-intensive matching.

[0012] By adopting the above solution, the storage method is matched according to the real-time identified network application scenario, thereby achieving efficient storage of network traffic calculation result data.

[0013] Preferably, the management module is further used to isolate the new computing node from the original computing node using network isolation technology, and allocate the new computing node to a virtual local area network with access control and firewall policies set.

[0014] By adopting the above solution, considering that new computing nodes may cause network security problems, network isolation technology is used to logically isolate the new computing nodes from the original computing nodes and restrict unauthorized access, ensuring that good performance and stability can be maintained when the network scale is expanded.

[0015] Preferably, the high-speed interconnection module is also used to give priority to selecting computing nodes corresponding to physical machines as core computing nodes and computing nodes corresponding to virtual machines as edge computing nodes in the process of dynamically connecting each computing node using high-speed Internet technology according to computing task requirements and dynamically matched network topology; the computing nodes include computing nodes corresponding to physical machines and computing nodes corresponding to virtual machines deployed in the virtual network in the supercomputing center network.

[0016] By adopting the above solution, the computing nodes corresponding to the physical machines are preferentially selected as the core computing nodes, and the computing nodes corresponding to the virtual machines are used as the edge computing nodes, which effectively improves the processing speed and data transmission efficiency of the core computing tasks. At the same time, the edge computing nodes can share part of the computing load and improve the overall performance and response speed of the system.

[0017] Preferably, it also includes: a preview module; the preview module is used to preview the computing nodes in the network topology in real time and add or delete computing nodes according to the received user requirements.

[0018] By adopting the above solution, it is possible to preview the computing nodes in the network topology in real time, and flexibly add or delete computing nodes according to user needs, thereby enhancing the configurability and flexibility of the system, facilitating users to dynamically adjust the network structure according to actual needs, and improving the user experience and management efficiency of the supercomputing center network system.

[0019] In a second aspect, the present application provides a supercomputing center network application method, comprising: Use deep learning algorithms to identify network application scenarios in real time, and query matching network topologies based on the real-time identified network application scenarios; According to the computing task requirements and dynamically matched network topology in the network, high-speed Internet technology is used to dynamically connect the computing nodes in the network through switches or routers; Monitor network status indicators in real time, and use deep learning algorithms to predict network status in the future based on real-time monitoring of network status indicators; Calculate the scores of each indicator of the predicted network status according to the preset network status indicator scoring rules, and dynamically adjust the routing control according to the preset routing adjustment strategy matching the indicator corresponding to the lowest indicator score; Obtain the calculation result data of each calculation node for storage.

[0020] By adopting the above solution, the supercomputing center network is used to complete the rapid transmission and calculation of network traffic data, which facilitates the subsequent data analysis and optimization.

[0021] In a third aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the method as described above.

[0022] In a fourth aspect, the present application provides a computer device, comprising a memory, a processor, and a program stored and executable on the memory, wherein the program implements the steps of the above method when executed by the processor.

[0023] In summary, this application has the following beneficial effects: 1. Through the design of flat network architecture, network delay is reduced and data transmission efficiency is improved; real-time monitoring of network status and prediction of future status are carried out, and different preset routing adjustment strategies are adopted to dynamically adjust routing, optimize data transmission paths, improve bandwidth utilization, and enable the system to better cope with complex computing tasks; 2. Support modular expansion, which is convenient for adding new computing nodes, improving the scalability of the system and facilitating future system upgrades; 3. By introducing redundant design, including backup network connection and fault detection, the fault tolerance of the system is enhanced, ensuring the stability and continuity of data transmission; 4. Combined with virtualization technology, secure isolation of newly added computing nodes is achieved, thus ensuring the security of the supercomputing center network. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a structural diagram of the supercomputing center network system described in a specific embodiment; Figure 2It is a flow chart of the supercomputing center network application method described in a specific embodiment. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. 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.

[0026] Example 1 like Figure 1 As shown, the embodiment of the present application discloses a supercomputing center network system, which specifically includes: a high-speed interconnection module 1, a management module 2, a storage module 3, and a preview module 4; through the cooperation of the above modules, the network delay is significantly reduced, the bandwidth utilization is improved, and the system scalability and fault tolerance are enhanced. The following is a further detailed description of each module of the present application.

[0027] The high-speed interconnection module 1 includes multiple components, such as: a network application scenario identification module 11, a network topology matching module 12 and a high-speed network connection module 13. As the core of the supercomputing center network, the high-speed interconnection module is mainly responsible for connecting various computing nodes, storage devices and other devices.

[0028] The network application scenario recognition module 11 is used to analyze the current network application scenario in real time using a deep learning algorithm. Specifically, a network application scenario recognition model constructed based on a deep learning algorithm is used to perform feature analysis on the input current network status data to identify the current network application scenario; the input of the model is the network status data, and the output of the model is the network application scenario. The network application scenario can be divided into a computing-intensive application scenario and a data-intensive application scenario.

[0029] The network topology matching module 12 is used to query the matching network topology according to the network application scenario identified in real time; wherein, the network topology includes: star ring topology, mesh topology, tree topology and cluster topology and other topological forms. Considering that the use of an adaptive network topology structure can better meet the needs of different application scenarios, a specific form of network topology is pre-matched for different application scenarios and stored in a topology matching database; in this embodiment, considering that in computing-intensive scenarios, the communication between computing nodes is frequent and requires high bandwidth, and parallel computing capabilities may also be required, the corresponding star ring topology or mesh topology is matched; considering that in data-intensive scenarios, a large amount of data transmission is often required, the corresponding tree topology and cluster topology are matched, which is more convenient for data aggregation and distribution, improves data transmission efficiency and reduces network latency.

[0030] The high-speed network connection module 13 is used to dynamically connect the computing nodes in the network through switches or routers using high-speed Internet technology according to the computing task requirements and the dynamically matched network topology in the network; wherein the supercomputing center network has a computing resource pool, which contains physical machines and virtualized resources for dynamic allocation and adjustment. According to the computing task requirements in the supercomputing center network, each computing node is obtained from the computing resource pool, and according to the dynamically matched network topology, high-speed Internet technology (such as InfiniBand technology, etc.) is used to connect the computing nodes through switches or routers.

[0031] In addition, in order to further ensure the efficient operation of the computing network, the high-speed network connection module 13 is also used to dynamically connect each computing node using high-speed Internet technology according to the computing task requirements and the dynamically matched network topology, and give priority to selecting the computing nodes corresponding to the physical machine as the core computing node (in the topology) and the computing nodes corresponding to the virtual machine as the edge computing node; the computing nodes include the computing nodes corresponding to the physical machine and the computing nodes corresponding to the virtual machine deployed in the virtual network in the supercomputing center network.

[0032] The management module 2 specifically includes a network status monitoring and prediction module 21, a routing control management module 22, etc. This module is mainly responsible for monitoring and managing the operating status of each computing node and related equipment, adjusting resource scheduling, expansion and data transmission paths, diagnosing and eliminating path failures, etc.

[0033] The network status monitoring and prediction module 21 is used to monitor network status indicators in real time, and use deep learning algorithms to predict the network status in the future based on the real-time monitoring of network status indicators; wherein the network status indicators include key data indicators such as network traffic, delay, packet loss rate, etc., which are monitored and obtained through equipment such as network traffic monitors, delay monitors and packet loss rate detectors; a network status prediction model is constructed using a deep learning algorithm to predict the network status in the future; the network status prediction model includes several sub-models, and the input of each sub-model is one or more specific network status indicators in the past and current time, and the output is one or more specific network status indicators in the future, and the specific time period can be pre-set manually.

[0034] The routing control management module 22 calculates the scores of each indicator of the predicted network state according to the preset network state indicator scoring rules, and dynamically adjusts the routing control according to the preset routing adjustment strategy that matches the indicator corresponding to the lowest indicator score; wherein, a scoring standard library is set to store the scoring rules of each indicator of the network state, such as: network traffic values ​​in different preset network traffic value ranges correspond to different preset network traffic indicator scores, network delays in different preset network delay ranges correspond to different preset network delay indicator scores, and network transmission packet loss rates in different preset packet loss rate ranges correspond to different preset packet loss rate indicator scores; network performance is evaluated according to the preset indicator scoring rules, that is, each indicator of the predicted network state is scored and converted into a unified score system (such as a percentage system), so as to adaptively adjust the routing based on the evaluation results to ensure efficient and reliable network transmission, that is, according to the preset routing adjustment strategy corresponding to the indicator with the lowest score, the routing control adjustment is completed to alleviate the indicators with poor performance in the predicted network state, and adaptively achieve the optimal data transmission path adjustment.

[0035] Among them, the preset routing adjustment strategy includes a load balancing strategy matching the network traffic index, that is, using the optimal path algorithm, according to the network traffic of the predicted network state, with the load balancing of each computing node as the goal, dynamically adjusting the routing control between each computing node; a delay optimization routing strategy matching the delay index, that is, using the optimal path algorithm, according to the network delay of the predicted network state, with the minimum transmission path delay of each computing node as the goal, dynamically adjusting the routing control between each computing node; and a reliability priority routing strategy matching the packet loss rate index, that is, using the optimal path algorithm, according to the network transmission packet loss rate of the predicted network state, with the minimum network transmission data packet loss rate as the goal, dynamically adjusting the routing control between each computing node.

[0036] The storage module 3 is used to store the calculation result data of each computing node obtained to complete the calculation task requirements.

[0037] In addition, in order to enable each computing node to quickly and accurately store the calculation result data and the calculation process data, it is also used to match the corresponding storage method according to the real-time identification of the network application scenario, and complete the storage of the calculation result data with the matched storage method; wherein, the storage method includes: the use of high-speed storage area network storage method for computing-intensive matching to improve storage efficiency, and the use of distributed storage network storage method for data-intensive matching to enhance the scalability and reliability of the storage system.

[0038] The preview module 4 is used to preview the computing nodes in the network topology in real time and add or delete computing nodes according to the received user needs. Considering that the network operation and maintenance personnel of the supercomputing center need to obtain information about the network configuration, test and maintain the network in a timely manner, a preview module is set to display the computing nodes of each connection of the current network in real time through a preview interface. It is not limited to displaying computing nodes, and the routing control and stored calculation result data in the computing nodes can all be displayed in the preview module. In addition, according to the needs of users such as operation and maintenance personnel, computing nodes can be added or deleted in the preview interface configured in the preview module, and specific processing instructions are generated accordingly so that the high-speed interconnect module 1 completes the connection and disconnection of computing nodes according to the specific processing instructions.

[0039] The system described in this embodiment significantly reduces the network latency of the supercomputing center, improves bandwidth utilization, effectively copes with complex computing tasks, and improves the overall performance and efficiency of the supercomputing center through a flat network architecture design, a dynamic routing mechanism, and an intelligent management strategy.

[0040] Example 2 The difference from Example 1 is that it supports modular expansion and redundant design of computing nodes to enhance the scalability and fault tolerance of the system and ensure the stability and continuity of data transmission; the management module 2 also includes: a computing node resource expansion module 23.

[0041] The computing node resource expansion module 23 is also used to determine whether a preset state is reached according to the predicted network state, and when the preset state is reached, dynamically connect a new computing node using high-speed Internet technology.

[0042] Among them, the new computing node can be connected to multiple computing nodes synchronously, connected in the form of a computing node cluster to expand the computing resources of the supercomputing center network.

[0043] The network status indicators also include traffic peak, average response time, resource utilization, etc.; the preset status includes: the network traffic peak exceeds the preset traffic peak threshold, the average response time of the computing node exceeds the preset response time threshold, and the resource utilization of the computing nodes exceeding the preset percentage exceeds the preset resource utilization threshold, which indicates that the predicted network status is general, and new computing nodes are introduced to improve the scalability of the system. Among them, the preset percentage can be set manually, specifically the percentage of computing nodes whose resource utilization exceeds the preset resource utilization threshold to all connected computing nodes.

[0044] In addition, in order to better judge whether the predicted network status is in a good or general state, the computing node resource expansion module 23 is also used to dynamically adjust the predicted network status to reach the threshold values ​​of each indicator in the preset status according to the predicted network status by using a reinforcement learning algorithm; that is, a reinforcement learning model is constructed, and different preset status indicator threshold values ​​are used to judge under the historical network status to conduct user satisfaction feedback training on the network status after the new computing node is connected, and the preset status indicator threshold value with the highest satisfaction feedback is obtained as the threshold value of each indicator in the preset status reached by different network states.

[0045] Example 3 The difference from Example 1 is that, in order to enhance the fault tolerance of the system, a redundant design is introduced, including a backup network connection and a fault detection and recovery mechanism, and the management module 2 also includes: a computing node resource scheduling module 24.

[0046] The computing node resource scheduling module 24 is also used to switch to a spare computing node or backup link in advance when it is determined that there is an abnormality or failure in a computing node or network link based on the predicted network status; wherein a computing node and network link abnormality or failure identification model is constructed using a deep learning algorithm, and is trained and generated using historical network status marked with abnormalities or failures in computing nodes or network links.

[0047] In addition, the pre-switching to the backup computing node or backup link is set to be multiple, and the backup computing node or backup link can be randomly selected to complete the switching.

[0048] Example 4 The difference from Example 1 is that, in order to ensure the security of the newly connected computing nodes, virtual isolation technology is used to perform logical isolation and restrict unauthorized access to the new computing nodes; the management module 2 also includes: a computing node security management module 25.

[0049] The computing node security management module 25 is also used to isolate the new computing node from the original computing node using network isolation technology, and allocate the new computing node to a virtual local area network with access control and firewall policies set.

[0050] Among them, access control is to configure ACL on the router or switch connected to the computing node to limit access to the new node, such as: only allowing the IP address of the network operation and maintenance personnel to access the management interface of the new node, thereby preventing unauthorized access; or setting firewall rules to define a specific rule set for the new computing node, allowing only specific IP addresses or IP ranges to access specific ports of the new node, ensuring that only explicitly authorized traffic can reach the new computing node to complete the computing task.

[0051] like Figure 2As shown, the embodiment of the present application discloses an application method based on any of the above-mentioned supercomputing center network systems. Specifically, the application of the supercomputing center network system can realize various high-performance computing tasks, realize data analysis and mining, etc. The specific application method includes: S1. Use deep learning algorithms to identify network application scenarios in real time, and query matching network topologies based on the real-time identified network application scenarios; S2, according to the computing task requirements in the network and the dynamically matched network topology, use high-speed Internet technology to dynamically connect the computing nodes in the network through switches or routers; S3, real-time monitoring of network status indicators, based on which the deep learning algorithm is used to predict the network status in the future; S4, calculating the scores of each indicator of the predicted network status according to the preset network status indicator scoring rules, and dynamically adjusting the routing control according to the preset routing adjustment strategy matching the indicator corresponding to the lowest indicator score; S5. Obtain the calculation result data of each calculation node for storage.

[0052] In a specific embodiment, S2 in the application method further includes: previewing computing nodes in the network topology in real time and adding or deleting computing nodes according to received user requirements.

[0053] In a specific embodiment, S3 in the application method also includes: judging whether a preset state is reached according to the predicted network state, and when the preset state is reached, dynamically connecting a new computing node using high-speed Internet technology; when judging whether a computing node or a network link is abnormal or fails according to the predicted network state, switching to a backup computing node or backup link; isolating the new computing node from the original computing node using network isolation technology, and allocating the new computing node to a virtual local area network with access control and firewall policies set.

[0054] In a specific embodiment, S5 in the application method further includes: matching a corresponding storage mode according to real-time identification of a network application scenario, and completing storage of the calculation result data in the matched storage mode.

[0055] The embodiment of the present application also discloses a computer-readable storage medium.

[0056] Specifically, the computer-readable storage medium stores a computer program that can be loaded by a processor and executed such as the above-mentioned supercomputing center network application method. The computer-readable storage medium includes, for example: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0057] The embodiment of the present application also discloses a computer device.

[0058] Specifically, the computer device includes a memory and a processor, and the memory stores a computer program that can be loaded by the processor and execute the above-mentioned supercomputing center network application method.

[0059] The above are all preferred embodiments of the present application, and are not intended to limit the protection scope of the present application. Any feature disclosed in this specification (including the abstract and drawings), unless otherwise stated, can be replaced by other equivalent or alternative features with similar purposes. That is, unless otherwise stated, each feature is only an example of a series of equivalent or similar features.

Claims

1. A supercomputing center network system, characterized in that: include: High-speed interconnection module, management module and storage module; The high-speed interconnection module is used to use a deep learning algorithm to identify network application scenarios in real time, and query the matching network topology according to the real-time identified network application scenarios; according to the computing task requirements in the network and the dynamically matched network topology, the high-speed interconnection network technology is used to dynamically connect the computing nodes in the network through switches or routers; wherein different network application scenarios correspond to different network topologies; the network application scenarios include: computing intensive and data intensive; the network topologies include: star ring, grid, tree and clustering; The management module is used to monitor network status indicators in real time, wherein the network status indicators include network traffic, delay and packet loss rate; based on the real-time monitoring of network status indicators, a deep learning algorithm is used to predict the network status in the future; the scores of each indicator of the predicted network status are calculated according to the preset network status indicator scoring rules, and the routing control is dynamically adjusted according to the preset routing adjustment strategy matching the indicator corresponding to the lowest indicator score; the preset routing adjustment strategy includes a load balancing strategy matching the network traffic indicator, a delay optimization routing strategy matching the delay indicator, and a reliability priority routing strategy matching the packet loss rate indicator; The storage module is used to obtain the calculation result data of each calculation node for storage.

2. The supercomputing center network system according to claim 1, characterized in that: The management module is also used to determine whether a preset state is reached according to the predicted network state, and when the preset state is reached, dynamically connect a new computing node using high-speed Internet technology; The network status indicators also include traffic peak, average response time, and resource utilization; The preset status includes: the network traffic peak exceeds the preset traffic peak threshold, the average response time of the computing node exceeds the preset response time threshold, and the resource utilization of the computing nodes exceeding the preset proportion exceeds the preset resource utilization threshold.

3. The supercomputing center network system according to claim 1, characterized in that: The management module is also used to switch to a standby computing node or backup link when it is determined that there is an abnormality or failure in a computing node or a network link according to the predicted network status.

4. The supercomputing center network system according to claim 1, characterized in that: The storage module is also used to match the corresponding storage method according to the real-time identification of the network application scenario, and complete the storage of the calculation result data in the matched storage method; the storage method includes: using a high-speed storage area network storage method for calculation-intensive matching, and using a distributed storage network storage method for data-intensive matching.

5. The supercomputing center network system according to claim 2, characterized in that: The management module is also used to isolate the new computing node from the original computing node using network isolation technology, and allocate the new computing node to a virtual local area network with access control and firewall policies set.

6. The supercomputing center network system according to claim 2, characterized in that: The high-speed interconnection module is also used to dynamically connect each computing node using high-speed Internet technology according to computing task requirements and dynamically matched network topology, and give priority to selecting computing nodes corresponding to physical machines as core computing nodes and computing nodes corresponding to virtual machines as edge computing nodes; the computing nodes include computing nodes corresponding to physical machines and computing nodes corresponding to virtual machines deployed in the virtual network of the supercomputing center network.

7. The supercomputing center network system according to claim 6, characterized in that: Also includes: Preview module; The preview module is used to preview the computing nodes in the network topology in real time and add or delete computing nodes according to the received user requirements.

8. An application method based on the system according to any one of claims 1 to 7, characterized in that: include: Use deep learning algorithms to identify network application scenarios in real time, and query matching network topologies based on the real-time identified network application scenarios; According to the computing task requirements and dynamically matched network topology in the network, high-speed Internet technology is used to dynamically connect the computing nodes in the network through switches or routers; Monitor network status indicators in real time, and use deep learning algorithms to predict network status in the future based on real-time monitoring of network status indicators; Calculate the scores of each indicator of the predicted network status according to the preset network status indicator scoring rules, and dynamically adjust the routing control according to the preset routing adjustment strategy matching the indicator corresponding to the lowest indicator score; Obtain the calculation result data of each calculation node for storage.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to claim 8.

10. A computer device, characterized in that: The computer device comprises a memory, a processor and a program stored and executable on the memory, and the program implements the steps of the method according to claim 8 when executed by the processor.

Images