CN-121984857-A - Network configuration method and system of network card to be tested, electronic equipment and storage medium

CN121984857ACN 121984857 ACN121984857 ACN 121984857ACN-121984857-A

Abstract

The embodiment of the application relates to the technical field of chip test, and discloses a network configuration method, a system, electronic equipment and a storage medium of a network card to be tested, which generate a network data stream by acquiring network configuration parameters of the network card to be tested, utilize network performance indexes fed back by a standard network card, the network configuration parameters are adaptively adjusted through the artificial intelligent model to determine the optimal network configuration parameters, and the network configuration method and the network configuration device can automatically acquire the optimal network configuration of the network card to be tested under different network environments so as to improve the network configuration efficiency of the network card.

Inventors

ZHANG KANG
WANG ZITAO
SHA MENGMENG

Assignees

无锡大普联芯科技有限公司

Dates

Publication Date: 20260505
Application Date: 20251231

Claims (11)

1. A network configuration method of a network card to be tested, the method comprising: Acquiring network configuration parameters of the network card to be tested; Generating a network data stream based on the network configuration parameters; the network card to be tested is controlled to send the network data stream to a standard network card through a virtual network so as to acquire network performance indexes fed back by the standard network card; And based on the network configuration parameters and the network performance indexes, adaptively adjusting the network configuration parameters through an artificial intelligent model to determine optimal network configuration parameters.
2. The method of claim 1, wherein the step of determining the position of the substrate comprises, The step of adaptively adjusting the network configuration parameters through an artificial intelligent model based on the network configuration parameters and the network performance index to determine optimal network configuration parameters comprises the following steps: Constructing an artificial intelligent model based on a Markov decision process, wherein the Markov decision process defines a state space and an action space based on the network performance index and the network configuration parameter; Performing supervision pre-training on the artificial intelligent model through a supervision learning data set to obtain a pre-trained artificial intelligent model; based on online reinforcement learning, performing iterative optimization on the pre-trained artificial intelligent model to obtain a trained artificial intelligent model; And adaptively adjusting the network configuration parameters through the trained artificial intelligent model to output optimal network configuration parameters.
3. The method of claim 2, wherein the step of determining the position of the substrate comprises, The supervised learning data set comprises a mapping relation data set of a state space and an action space; Performing supervised pre-training on the artificial intelligent model through the supervised learning data set to obtain a pre-trained artificial intelligent model, wherein the method comprises the following steps of: Converting the mapping relation data set into supervised learning sample pairs, wherein each supervised learning sample pair comprises a state vector representing a network state and an optimal action required for representing the adjustment from the current network configuration parameters to the optimal network configuration parameters; Training the artificial intelligence model using the supervised learning sample pair; calculating the total loss of the artificial intelligence model in the training process; Based on the total loss, updating parameters of the artificial intelligence model through a back propagation algorithm and an optimizer to obtain a pre-trained artificial intelligence model.
4. The method of claim 2, wherein the step of determining the position of the substrate comprises, The iterative optimization is performed on the pre-trained artificial intelligent model based on online reinforcement learning to obtain a trained artificial intelligent model, which comprises the following steps: Loading model weights of the pre-trained artificial intelligent models, and synchronously creating experience playback buffer areas; Real-time sensing network state, executing network configuration parameter adjustment action based on epsilon-greedy strategy, collecting interaction data and storing the interaction data in the experience playback buffer zone; Sampling historical interaction data from the experience playback buffer zone, calculating loss through a dual-network architecture, updating model weight, and stabilizing a training process by adopting a soft update strategy; And carrying out iterative training on the pre-trained artificial intelligent model until the model performance converges or reaches the preset training round number.
5. The method of claim 4, wherein the step of determining the position of the first electrode is performed, The dual-network architecture comprises a main network for parameter synchronous update and a target network for parameter delay update, wherein the main network is used for action selection and weight parameter gradient update; The performing a network configuration parameter adjustment action based on the epsilon-greedy policy includes: and selecting an action or a random action with the highest Q value by adopting an epsilon-greedy strategy based on the Q value output by the main network.
6. The method of claim 5, wherein the step of determining the position of the probe is performed, The method further comprises the steps of: after each update of the weight parameters of the main network, the weight parameters of the main network are partially synchronized to the target network by a soft update mode.
7. The method according to any one of claims 2 to 6, wherein, The network performance indexes comprise throughput, delay and data packet loss rate; the network configuration parameters comprise the buffer space size of the network card to be tested, a priority flow control trigger threshold, a priority flow control withdrawal threshold, an explicit congestion notification trigger threshold and an explicit congestion notification withdrawal threshold; The markov decision process further defines a reward function based on the network performance indicator and the network configuration parameter, the reward function comprising: Reward = α × Throughput β × Latency γ × Loss_rate Wherein Reward is a prize value, throughput is Throughput, alpha is a super parameter corresponding to Throughput, latency is delay, beta is a super parameter corresponding to delay, loss_rate is a data packet Loss rate, and gamma is a super parameter corresponding to a data packet Loss rate.
8. A network configuration system for a network card to be tested, the system comprising: the network card to be tested is used for generating a network data stream based on the network configuration parameters; the standard network card is used for receiving the network data stream sent by the network card to be tested and feeding back the network performance index; The virtual network simulation module is connected with the network card to be tested and the standard network card and is used for simulating a network topology structure and real network conditions; The intelligent computing module is connected with the network card to be tested, the standard network card and the virtual network simulation module and is used for self-adaptively adjusting the network configuration parameters through an artificial intelligent model based on the network configuration parameters and network performance indexes fed back by the standard network card so as to determine optimal network configuration parameters and sending the optimal network configuration parameters to the network card to be tested.
9. The network configuration system of claim 8, wherein, The intelligent computing module comprises: The data acquisition module is used for acquiring the network configuration parameters and the network performance index fed back by the standard network card; The network performance optimization module is connected with the data acquisition module and is used for operating the artificial intelligent model so as to adaptively adjust the network configuration parameters based on the network configuration parameters and the network performance indexes to determine optimal network configuration parameters; The driving module is connected with the network performance optimization module and the network card to be tested and is used for converting the optimal network configuration parameters into hardware identifiable instructions so as to send the optimal network configuration parameters to the network card to be tested; the log recording module is connected with the data acquisition module and the network performance optimization module and is used for recording network performance indexes fed back by the data acquisition module and recording optimal network configuration parameters output by the network performance optimization module.
10. An electronic device, comprising: At least one processor; At least one memory for storing at least one program; when at least one of the programs is executed by at least one of the processors, the at least one of the processors is caused to implement the method of claim 1 The method of 7.
11. A non-transitory computer readable storage medium having stored therein a processor-executable program, characterized in that the processor-executable program, when executed by a processor, is for performing the method of claim 1 The method of 7.

Description

Network configuration method and system of network card to be tested, electronic equipment and storage medium Technical Field The present application relates to the field of chip testing technologies, and in particular, to a network configuration method, a system, an electronic device, and a storage medium for a network card to be tested. Background With the rapid development of cloud computing, big data and artificial intelligence technology, the requirements of data centers and high-performance computing on network communication performance are increasingly increasing. The network card chip is used as a core component of a server and network equipment and plays a key role, and the performance of the network card chip directly influences the delay, throughput and reliability of network transmission. In the design flow of the network card chip, the testing stage before the chip streaming is a key link for ensuring the functional correctness and reliability of the chip. The FPGA prototype verification is used as an efficient hardware simulation means, actual function verification can be carried out on chip design before chip streaming, design defects can be found early, and chip streaming risk is reduced. However, the complexity of the function of the network card chip is increasing, and the network card chip has the characteristics of multi-parameter coupling and dynamic response. The traditional test method relies on manually preset static test vectors to verify in a fixed network environment, so that the functional scene is difficult to fully cover, the optimal network configuration of the network card cannot be determined, and the network configuration efficiency of the network card is insufficient. Disclosure of Invention The embodiment of the application provides a network configuration method, a system, electronic equipment and a storage medium of a network card to be tested, which are used for realizing self-adaptive matching of optimal network configuration parameters of the network card to be tested and improving network configuration efficiency of the network card. The embodiment of the application provides the following technical scheme: In a first aspect, an embodiment of the present application provides a network configuration method for a network card to be tested, where the method includes: Acquiring network configuration parameters of a network card to be tested; Generating a network data stream based on the network configuration parameters; the network card to be tested is controlled to send the network data stream to the standard network card through the virtual network so as to acquire the network performance index fed back by the standard network card; based on the network configuration parameters and the network performance indexes, the network configuration parameters are adaptively adjusted through the artificial intelligent model, so that the optimal network configuration parameters are determined. In some embodiments of the present invention, in some embodiments, Based on the network configuration parameters and the network performance indexes, the network configuration parameters are adaptively adjusted through the artificial intelligent model to determine optimal network configuration parameters, and the method comprises the following steps: Constructing an artificial intelligent model based on a Markov decision process, wherein the Markov decision process defines a state space and an action space based on network performance indexes and network configuration parameters, wherein the state space comprises the network performance indexes, and the action space comprises the network configuration parameters; performing supervision pre-training on the artificial intelligent model through a supervision learning data set to obtain a pre-trained artificial intelligent model; based on online reinforcement learning, performing iterative optimization on the pre-trained artificial intelligent model to obtain a trained artificial intelligent model; and adaptively adjusting the network configuration parameters through the trained artificial intelligent model to output the optimal network configuration parameters. In some embodiments of the present invention, in some embodiments, The supervised learning data set comprises a mapping relation data set of a state space and an action space; Performing supervised pre-training on the artificial intelligent model through a supervised learning data set to obtain a pre-trained artificial intelligent model, wherein the method comprises the following steps of: Converting the mapping relation data set into supervised learning sample pairs, wherein each supervised learning sample pair comprises a state vector representing the state of the network and an optimal action required for representing the adjustment from the current network configuration parameters to the optimal network configuration parameters; training the artificial intelligent model by using a