Network interface card testing method and apparatus

Abstract

The application discloses a network interface card testing method and device, and relates to the technical field of computer network testing, and comprises the following steps: generating mixed traffic simulating a real network environment dynamically through an intelligent traffic generation engine; performing all-around performance measurement from a physical layer to an application layer by using a full-stack performance analysis framework, and dynamically generating and optimizing test cases; constructing a virtualized test environment, and inputting the test cases into the virtualized test environment for parallel testing; and automatically generating a test report containing problem diagnosis suggestions, so that the technical problems in the prior art, such as low test efficiency and automation degree of network card testing technology, great limitation of test scenes, difficulty in effectively analyzing data packet contents, serious waste of hardware resources and difficulty in fully verifying compatibility, are solved, and the technical effect that the performance, stability and reliability of a network card are comprehensively tested through intelligent traffic generation and deep packet inspection technology is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer network testing, and in particular to a network interface card testing method and device. BACKGROUND

[0002] Currently, network card (NIC) testing technologies, such as single-field testing technology and sequential testing technology, can test single environmental factors such as temperature and voltage, or perform stability testing through firmware and driver upgrade and downgrade operations.

[0003] However, the existing network card testing technology still has the following technical defects:

[0004] 1. Low testing efficiency: related testing technologies rely too much on manual configuration and simple tools, and cannot simulate complex network environments, making it difficult to quickly complete comprehensive testing;

[0005] 2. Incomplete test coverage: it is difficult to simulate complex traffic patterns and abnormal scenarios in real network environments, such as high concurrency, high bandwidth, packet loss, and delay jitter;

[0006] 3. Difficult result analysis: there is a lack of deep analysis of packet content, making it difficult to locate hardware or driver level problems, and lacking automated analysis tools, making test results too dependent on manual interpretation;

[0007] 4. Low automation level: the testing process requires human intervention and cannot achieve 24-hour unattended testing;

[0008] 5. Hardware resource waste: high idle rate of testing equipment, unable to effectively reuse testing environment;

[0009] 6. Insufficient compatibility verification: difficult to comprehensively test the compatibility of network cards under different operating systems and driver versions.

[0010] In summary, related network card testing technologies have low testing efficiency and automation level, limited testing scenarios, difficulty in effectively analyzing packet content, and serious hardware resource waste, making it difficult to fully verify compatibility, which needs to be addressed. SUMMARY

[0011] The present application provides a network interface card testing method and device to at least solve the technical problems in related art that network card testing technologies have low testing efficiency and automation level, limited testing scenarios, difficulty in effectively analyzing packet content, and serious hardware resource waste, making it difficult to fully verify compatibility.

[0012] The application provides a network interface card testing method, comprising the following steps: determining the traffic generation proportion corresponding to a plurality of protocol traffic scripts based on a target network environment of a network interface card, mixing a plurality of protocol traffics according to the traffic generation proportion, and obtaining network performance disturbance parameters of the target network environment, so as to generate target mixed traffic simulating the target network environment based on the mixed plurality of protocol traffics and the network performance disturbance parameters; inputting the target mixed traffic into the network interface card, and collecting a plurality of cross-layer performance index data corresponding to different network protocol layers of the network interface card; inputting the plurality of cross-layer performance index data into a pre-constructed multi-stage intelligent testing model, so as to generate a corresponding test case combination, deploying a plurality of virtualization testing environments corresponding to the network interface card, and performing parallel testing on the test case combination in the plurality of virtualization testing environments, so as to obtain testing data of the network interface card under different virtualization testing environments, constructing a corresponding association graph based on the testing data, and generating a defect testing report of the network interface card according to the association graph.

[0013] The application also provides a network interface card testing device, comprising: a traffic mixing module, configured to determine the traffic generation proportion corresponding to a plurality of protocol traffic scripts based on a target network environment of a network interface card, mix a plurality of protocol traffics according to the traffic generation proportion, and obtain network performance disturbance parameters of the target network environment, so as to generate target mixed traffic simulating the target network environment based on the mixed plurality of protocol traffics and the network performance disturbance parameters; a data collection module, configured to input the target mixed traffic into the network interface card, and collect a plurality of cross-layer performance index data corresponding to different network protocol layers of the network interface card; and a defect testing module, configured to input the plurality of cross-layer performance index data into a pre-constructed multi-stage intelligent testing model, so as to generate a corresponding test case combination, deploy a plurality of virtualization testing environments corresponding to the network interface card, and perform parallel testing on the test case combination in the plurality of virtualization testing environments, so as to obtain testing data of the network interface card under different virtualization testing environments, construct a corresponding association graph based on the testing data, and generate a defect testing report of the network interface card according to the association graph.

[0014] The application also provides an electronic device, comprising: a memory configured to store a computer program; and a processor configured to execute the computer program to implement the steps of any of the network interface card testing methods.

[0015] The application also provides a non-volatile computer readable storage medium, wherein the non-volatile computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of any of the network interface card testing methods.

[0016] The application further provides a computer program product comprising a computer program which, when executed by a processor, implements the steps of any of the network interface card testing methods described above.

[0017] Through the application, the traffic generation ratios corresponding to the multiple protocol traffic scripts can be determined based on the target network environment of the network interface card, and the multiple protocol traffic is mixed according to the traffic generation ratios, and the network performance disturbance parameters of the target network environment are obtained, so as to generate the target mixed traffic simulating the target network environment based on the mixed multiple protocol traffic and the network performance disturbance parameters; the target mixed traffic is input into the network interface card, and the multiple cross-layer performance index data of different network protocol layers corresponding to the network interface card are collected; the multiple cross-layer performance index data are input into the pre-constructed multi-stage intelligent testing model to generate the corresponding test case combination, and the multiple virtualization testing environments corresponding to the network interface card are deployed, and the test case combination is tested in parallel in the multiple virtualization testing environments to obtain the test data of the network interface card under different virtualization testing environments, the corresponding association graph is constructed based on the test data, and the defect test report of the network interface card is generated according to the association graph, so that the technical problems that the test efficiency and the degree of automation of the network card testing technology are low, the test scene has large limitations, the data packet content cannot be effectively analyzed, and hardware resources are wasted seriously and compatibility verification cannot be fully performed in the related art can be solved, and the technical effects that the network card is comprehensively tested in performance, stability and reliability through intelligent traffic generation and deep packet detection technology are achieved. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the application, the drawings needed in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0019] Figure 1 A flowchart of a network interface card testing method according to an embodiment of the application is provided.

[0020] Figure 2 An execution logic schematic diagram of a network interface card testing method according to an embodiment of the application is provided.

[0021] Figure 3 A logic architecture schematic diagram of a network interface card testing system according to an embodiment of the application is provided.

[0022] Figure 4 An example diagram of a network interface card testing device according to an embodiment of the application is provided.

[0023] Wherein, 10-network interface card testing device, 100-flow mixing module, 200-data acquisition module, 300-defect testing module. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.

[0025] It should be noted that, in the description of the present application, the terms "comprise", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. The terms "first", "second" and the like in the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence.

[0026] In order to enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0027] In combination with the specific application environment architecture or specific hardware architecture on which the network interface card testing method is executed, the specific application environment architecture or specific hardware architecture is described herein.

[0028] The embodiments of the present application provide a network interface card testing method.

[0029] As shown in Figure 1 Fig. 1 is a flowchart of the network interface card testing method according to an embodiment of the present application, wherein the network interface card testing method comprises the following steps:

[0030] In step S101, based on the target network environment of the network interface card, the traffic generation proportions corresponding to the multiple protocol traffic scripts are determined, and the multiple protocol traffic is mixed according to the traffic generation proportions, and the network performance disturbance parameters of the target network environment are obtained, so as to generate the target mixed traffic simulating the target network environment based on the mixed multiple protocol traffic and the network performance disturbance parameters.

[0031] Those skilled in the art should appreciate that, although the testing technology of the current network interface card is rapidly evolving, the traditional testing method is still dominant. For example, in a single field test, the traditional testing method only tests a single environmental factor such as temperature and voltage, and cannot simulate the multi-physical field coupling effect in a real complex environment. In sequential testing, the traditional testing method verifies stability through firmware and driver upgrade and downgrade operations, but still uses a phased testing mode and does not achieve dynamic collaborative loading.

[0032] Therefore, the embodiments of the present application can comprehensively test the performance, stability and reliability of the network card through intelligent traffic simulation, deep performance analysis and other operations, thereby improving the efficiency of network card testing, improving the problem detection rate, reducing the comprehensive cost, and establishing a new industry testing benchmark.

[0033] In actual execution, the embodiments of the present application can first comprehensively consider factors such as traffic type, protocol distribution, traffic mode, and delay jitter, and mix TCP (Transmission Control Protocol), UDP (User Datagram Protocol), ICMP (Internet Control Message Protocol) and other protocol traffic (such as iperf3 to generate basic load and scapy to customize special packets) according to the real proportion (i.e. traffic generation proportion); secondly, the embodiments of the present application can use a Markov model to simulate traffic burstiness, and combine with the tc command to dynamically inject delay (50ms ± 10ms), packet loss (0.1% ~ 5%) and jitter; thirdly, the embodiments of the present application can replay real traffic packet capture through tcpreplay, and use NS-3 (‌Network Simulator-3‌, Discrete Event Driven Network Simulator) to build a complex topology; then, the embodiments of the present application monitor indexes such as ntopng traffic composition and ss -ti retransmission rate to achieve a high-fidelity test environment containing Web short flow, video long flow, encrypted traffic and random damage.

[0034] Therefore, the embodiments of the present application can simulate real traffic and damage, and accurately monitor the corresponding indexes, thereby improving the effectiveness of NIC testing.

[0035] Optionally, in an embodiment of the present application, based on the target network environment of the network interface card, the traffic generation proportion corresponding to a plurality of protocol traffic scripts is determined, including: performing network traffic statistical analysis on the network interface card to obtain corresponding statistical data, and establishing a corresponding protocol proportion model according to the statistical data; and generating the traffic generation proportion of the plurality of protocol traffic scripts based on the protocol proportion model.

[0036] As an implementable manner, the embodiments of the present application can perform statistical analysis based on real network traffic to obtain corresponding statistical data, and then establish a protocol proportion model (TCP 62.3% / UDP 28.5% / ICMP 4.2%); then, the embodiments of the present application can control multiple protocol traffic scripts, such as iperf3 and Scapy scripts (traffic generation proportion of UDP packet size generated according to Pareto distribution) by the protocol proportion model and the dynamic scheduler.

[0037] Therefore, the embodiments of the present application can generate the traffic generation proportion of the multiple protocol traffic scripts by establishing the protocol proportion model and combining the dynamic scheduler, so as to accurately match the actual protocol distribution, provide the basis for the subsequent traffic test that conforms to the real scene, and improve the accuracy and effectiveness of the traffic simulation.

[0038] Optionally, in an embodiment of the present application, multiple protocol traffics are mixed according to the traffic generation proportion, and network performance disturbance parameters of a target network environment are obtained, so as to generate target mixed traffic simulating the target network environment based on the mixed multiple protocol traffics and the network performance disturbance parameters, including: mixing multiple protocol traffics according to the traffic generation proportion to generate corresponding initial mixed traffic; calculating a current network load index corresponding to a network interface card, and adjusting a state transition probability of a pre-constructed Markov model according to the current network load index, so as to inject network performance disturbance parameters into the initial mixed traffic based on the adjusted state transition probability, wherein the network performance disturbance parameters include delay parameters, packet loss parameters and jitter parameters; dynamically compensating and correcting the network performance disturbance parameters, and obtaining corresponding hardware time stamps according to a pre-set hardware time synchronization mechanism, and performing damage injection operation on the initial mixed traffic after injecting the network performance disturbance parameters by the hardware time stamps, to generate the target mixed traffic simulating the target network environment.

[0039] It should be noted that the embodiments of the present application can first mix multiple protocol traffics according to the determined traffic generation proportion (for example, TCP protocol traffic accounts for 60%, UDP protocol traffic accounts for 30%, and ICMP protocol traffic accounts for 10%), to generate initial mixed traffic, so as to ensure that the traffic basis constitutes can well conform to the actual network protocol distribution.

[0040] Secondly, the embodiments of the present application can calculate the current network load index (comprehensive bandwidth utilization rate, data packet queue length and other indicators) of the network interface card, and fuse the queue delay, packet loss rate and protocol proportion deviation into a comprehensive health index Q, which is used as the decision basis for the state transition of the pre-constructed Markov model, dynamically adjusts the state transition probability of the Markov model, and injects network performance disturbance parameters (delay, packet loss, jitter parameters) into the initial mixed traffic based on the adjusted Markov model, to realize the targeted adaptation of disturbance injection.

[0041] Again, the embodiments of the present application can implement the intelligent dynamic compensation algorithm by the netem module of Linux tc, and upgrade the static impairment to a closed-loop system based on integral differential control: real-time calculation of the packet loss rate change ΔL and the jitter change ΔJ, dynamic calculation of the delay compensation parameter according to the change, and autonomous adjustment of the impairment parameter to avoid flow distortion caused by single disturbance. The delay compensation parameter is as follows:

[0042] D(t) = 50 + ∫(0→t) [0.2ΔL-0.1ΔJ]dt,

[0043] Wherein, D(t) represents the delay compensation parameter.

[0044] At the same time, the embodiments of the present application can also use the hardware accelerated time synchronization mechanism to integrate PTPv2 (‌Precision Time Protocol Version-2‌,‌ Precision Time Protocol Version-2‌) hardware timestamp on the basis of conventional netem impairment injection, and improve the time disturbance accuracy from ms level to μs level. After obtaining the accurate hardware timestamp, the impairment injection operation is performed on the initial mixed flow after the disturbance.

[0045] Then, the embodiments of the present application can perform real-time monitoring on the flow after completing the impairment injection based on the 5G-SDN (Software Defined Networking, software defined networking) hybrid simulation architecture (realizing the joint simulation of 3GPP (‌3rd Generation Partnership Project,‌ 3rd Generation Partnership Project) standard millimeter wave channel model and OpenFlow 1.5 controller in NS-3, supporting end-to-end network slicing scenario simulation), and dynamically adjust the test parameters through the adaptive feedback controller, so that the flow characteristics monitored by ntopng and the target distribution error continuously remain ≤3.2%, ensuring that the flow characteristics are highly consistent with the target network environment (such as 5G slicing network), and finally generating the target mixed flow simulating the target network environment.

[0046] Therefore, the embodiments of the present application can simulate multiple flow modes at the same time through the mixed flow simulation technology, and can adaptively adjust the flow mode based on machine learning, so that the real network environment reconstruction rate is as high as 95% or more, thereby generating the target mixed flow close to the real network, and providing reliable environmental support for network performance testing.

[0047] Optionally, in an embodiment of the present application, a current network load index corresponding to the network interface card is calculated, and the state transition probability of the pre-constructed Markov model is adjusted according to the current network load index, so as to inject a network performance disturbance parameter into the initial mixed traffic based on the adjusted state transition probability, wherein the network performance disturbance parameter includes a time delay parameter, a packet loss parameter and a jitter parameter, and the method includes: sending the initial mixed traffic to the network interface card to collect the underlying performance indicators corresponding to the network interface card, wherein the underlying performance indicators include a packet queue delay and a packet loss rate; and performing a weighted fusion operation on the underlying performance indicators to calculate the current network load index, and outputting the current network load index.

[0048] In actual execution, an embodiment of the present application simulates traffic burst by using a three-state Markov chain (stationary / burst / recovery state), and the state transition probability is dynamically adjusted by a network load index Q=0.7×(queue delay / 50ms)+0.3×(packet loss rate / 5%) calculated in real time.

[0049] Specifically, an embodiment of the present application can first send the initial mixed traffic to the network interface card, and collect the underlying performance indicators in real time by using a traffic monitoring tool, including a queue delay (such as an average delay of 20ms and a peak delay of 80ms) of a packet in a network card buffer and a packet loss rate (such as 0.5% and 2.3%) per unit time.

[0050] Secondly, an embodiment of the present application can perform a weighted fusion operation on the collected underlying performance indicators to calculate a current network load index Q, as shown in the following formula:

[0051] Q=0.7×(queue delay / 50ms)+0.3×(packet loss rate / 5%).

[0052] For example, when the queue delay is 30ms and the packet loss rate is 2%, the following formula can be obtained:

[0053] Q=0.7×(30 / 50)+0.3×(2 / 5)=0.42+0.12=0.54.

[0054] Thirdly, an embodiment of the present application can input the current network load index into a preset nonlinear mapping function to dynamically adjust the state transition probability of the three-state Markov chain (stationary / burst / recovery state); when the Q value is low (such as ≤0.3), the stationary state retention probability is increased; and when the Q value is high (such as ≥0.7), the probability of transition to the burst state is increased.

[0055] Subsequently, an embodiment of the present application can dynamically inject a time delay, a packet loss and a jitter parameter matching the current network state into the initial mixed traffic based on the adjusted state transition probability, for example, a higher time delay jitter is injected in the burst state.

[0056] Therefore, the embodiments of the present application can make the performance disturbance parameter more suitable for the actual network state by performing the dynamic correlation network load and disturbance injection operation, thereby improving the authenticity and accuracy of the traffic simulation.

[0057] In step S102, the target mixed traffic is input into the network interface card, and the network interface card corresponding to a plurality of cross-layer performance index data of different network protocol layers is collected.

[0058] Furthermore, the embodiments of the present application can perform the cross-network protocol stack comprehensive performance measurement operation by monitoring the index through hierarchical design and combining the active detection and passive collection technology, so as to ensure covering the three dimensions of physical transmission quality, protocol efficiency and service experience.

[0059] For example, the three dimensions of physical transmission quality, protocol efficiency and service experience can collect hardware level indicators (such as temperature, power consumption, signal integrity), driver level indicators (such as interrupt frequency, DMA (‌Direct Memory Access, Direct Memory Access) efficiency) and protocol level indicators (such as throughput, delay, packet loss rate).

[0060] Then, the embodiments of the present application can perform real-time performance monitoring and abnormal mode recognition operation according to the hardware level indicators, the driver level indicators and the protocol level indicators through the corresponding analysis engine, so as to realize the weak link prediction and bottleneck positioning.

[0061] Therefore, the embodiments of the present application can effectively improve the implicit bottleneck recognition rate, the compatibility problem discovery rate, the hardware level problem detection rate such as high temperature frequency reduction, and the accuracy of maintenance suggestion prediction, and reduce the performance bottleneck positioning time (from hours to seconds).

[0062] Optionally, in an embodiment of the present application, the target mixed traffic is input into the network interface card, and a plurality of cross-layer performance index data of different network protocol layers corresponding to the network interface card is collected, including: in the physical layer of the network protocol layer, collecting the physical signal quality and bit error rate data in the cross-layer performance index data corresponding to the network interface card; in the data link layer of the network protocol layer, collecting the medium access control frame error rate and link reachability data in the plurality of cross-layer performance index data corresponding to the network interface card; in the network layer of the network protocol layer, collecting the routing state and packet fragmentation information in the plurality of cross-layer performance index data corresponding to the network interface card; in the transport layer of the network protocol layer, collecting the transmission control protocol retransmission rate and round-trip delay dynamic data in the plurality of cross-layer performance index data corresponding to the network interface card; in the application layer of the network protocol layer, collecting the service error rate and user experience index data in the plurality of cross-layer performance index data corresponding to the network interface card.

[0063] It should be noted that the embodiments of the present application can construct a three-dimensional monitoring system from the physical layer to the application layer based on the hierarchical probe deployment and spatiotemporal data fusion technology.

[0064] Specifically, the embodiments of the present application can first send a PRBS23 (Pseudo-Random Binary Sequence, pseudo-random binary sequence) test code type through a BERT (‌Bidirectional Encoder Representations from Transformers‌, bidirectional encoder representations from transformers) device at the physical layer, collect SNR (Signal-to-Noise Ratio, signal noise ratio) data in a dynamic range of -3dBm to +2dBm using an optical power meter, and use ethtool -S to extract the error correction count of the network card FEC (Forward Error Correction, forward error correction) to establish a real-time calculation model of the bit error rate BER, as shown in the following formula:

[0065] BER(t)=∑(FEC_errors) / [2.5×10^10×t],

[0066] wherein FEC_errors is the forward error correction error count.

[0067] Secondly, the embodiments of the present application can deploy an out-of-band mirror port at the data link layer, use a FPGA (Field-Programmable Gate Array, field-programmable gate array) accelerated packet sniffer to count the MAC (‌Media Access Control, media access control) frame CRC (‌Cyclic Redundancy Check, cyclic redundancy check) error rate, and measure the layer 2 reachability through a self-defined ARP (‌Address Resolution Protocol, address resolution protocol) probe.

[0068] Thirdly, the embodiments of the present application can deploy a traffic mirror with a sFlow sampling ratio of 1:8192 at the border router of the network layer, draw a topology delay heat map by combining an improved traceroute (sending 3 ICMP+3 UDP+3 TCP SYN (Synchronize Sequence Numbers, synchronize sequence numbers) packets at each hop), and develop an IP fragmentation reassembly success rate index, as shown in the following formula:

[0069] Frag_succ_rate=1-(df_count∑frag_timeout) / total_frags,

[0070] Wherein, df_count is the number of packets with "Don't Fragment" flag set (i.e. packets that are not allowed to be fragmented but need to be fragmented); frag_timeout is the count of fragmentation reassembly timeout events (usually discarded due to not all fragments arriving within the time limit); total_frags is the total number of fragments received in the statistical period.

[0071] After that, the embodiment of the present application can insert a probe into the tcp_retransmit_skb kernel function through eBPF (Extended Berkeley Packet Filter) at the transport layer to collect RTT (‌Radio Transfers Technology) samples with a precision of microseconds, and calculate a dynamic RTO (‌Recovery Time Objective) value using the improved Jacobson algorithm.

[0072] In addition, the embodiment of the present application can also combine Chrome DevTools and Nginx log at the application layer to count HTTP (HyperText Transfer Protocol) delay (i.e. user experience index data) and business error rate.

[0073] It can be understood that the embodiment of the present application can use adaptive sFlow sampling technology to dynamically adjust the sampling ratio (1:1024 to 1:65536) according to the traffic entropy value, reduce the processing overhead while maintaining high statistical accuracy; the embodiment of the present application can also expand the traditional two-dimensional index to a three-dimensional analysis model (time x protocol layer x physical location) through a three-dimensional space-time correlation engine, which can detect the causal chain such as "light module temperature rise - MAC error increase - TCP throughput decrease". In addition, the embodiment of the present application can also use Xilinx Alveo U200 at the data link layer to implement 100Gbps line speed CRC verification, which reduces the latency by several orders of magnitude compared with the software solution.

[0074] Therefore, the embodiment of the present application can construct a three-dimensional monitoring system from the physical layer to the application layer by deploying probes and fusing space-time data in layers, so as to accurately collect key data at each layer and calculate core indicators, and fully grasp the network state, providing multi-dimensional basis for network problem positioning and performance optimization, and improving the comprehensiveness and accuracy of network monitoring.

[0075] Optionally, in an embodiment of the present application, in the application layer of the network protocol layer, the service error rate and the user experience index data in the plurality of cross-layer performance index data corresponding to the network interface card are collected, including: injecting a tracking identifier into an external request at an application layer service entrance, and performing full-link tracking by using the tracking identifier, to collect the application performance index, the server error rate and the first input delay corresponding to the application layer; and calculating the user experience index data based on the application performance index, the server error rate and the first input delay.

[0076] As an implementable manner, in an embodiment of the present application, the full-link tracking can be implemented by injecting a unique tracking identifier X-Trace-ID into an external request at an application layer service entrance through an Nginx module, and the application performance index, the server error rate and the first input delay corresponding to the application layer are collected to calculate the corresponding user experience index data, as shown in the following formula:

[0077] QoE = 0.6 x Apdex + 0.3 x (1-5xx_rate) + 0.1 x FID,

[0078] wherein, Apdex is the application performance index, 5xx_rate is the server error rate, and FID is the first input delay.

[0079] Therefore, in the embodiment of the present application, the full-link tracking is implemented by injecting a unique tracking identifier at the application layer entrance, to accurately collect the key data such as the application performance index and the server error rate, and then calculate the user experience index, so that the user experience situation can be comprehensively and accurately mastered, and reliable data support is provided for optimizing the application performance and improving the user experience.

[0080] Optionally, in an embodiment of the present application, after the plurality of cross-layer performance index data corresponding to different network protocol layers of the network interface card are collected, the method further includes: performing time delay alignment on the physical signal quality, the bit error rate data, the medium access control frame error rate, the link reachability data, the routing state, the data packet fragmentation information, the transmission control protocol retransmission rate, the round-trip delay dynamic data, the service error rate and the user experience index data by using a hardware time stamp, to obtain corresponding time delay alignment data; performing correlation analysis on the time delay alignment data to obtain corresponding correlation analysis results, and judging whether there is cross-layer performance index abnormal data meeting a preset abnormal condition in the plurality of cross-layer performance index data according to the correlation analysis results, wherein when there is cross-layer performance index abnormal data meeting the preset abnormal condition in the plurality of cross-layer performance index data, a corresponding abnormal diagnosis mechanism is triggered; performing abnormal diagnosis on the cross-layer performance index abnormal data by using the abnormal diagnosis mechanism to obtain a corresponding abnormal reason, and optimizing the network protocol layer corresponding to the cross-layer performance index abnormal data according to the abnormal reason, to re-collect the corresponding cross-layer performance index data through the optimized network protocol layer.

[0081] In actual implementation, the embodiment of the application can first realize nanosecond-level time synchronization by means of the PTPv2 hardware timestamp technology, and unify the time delay alignment of the physical signal quality and bit error rate data of the physical layer, the medium access control frame error rate and link reachability data of the data link layer, the routing state and packet fragmentation information of the network layer, the transmission control protocol retransmission rate and round-trip delay dynamic data of the transport layer, and the service error rate and user experience index data of the application layer, to generate time-delay alignment data with consistent time dimension.

[0082] Secondly, the embodiment of the application can input the time-delay alignment data into a space-time correlation engine, simultaneously build a cross-layer correlation dashboard based on Prometheus and Grafana to display the data in real time, and perform correlation analysis on the data by the engine to determine whether there is cross-layer performance index abnormal data that meets the preset abnormal condition; for example, when it is detected that the physical layer BER>10-6 and the TCP retransmission rate>0.5%, and the condition lasts for 3 sampling periods, the embodiment of the application can automatically trigger the corresponding abnormal diagnosis mechanism of the optical module BERT diagnosis mode.

[0083] After starting the abnormal diagnosis mechanism, the embodiment of the application can perform deep abnormal diagnosis on the cross-layer performance index abnormal data, locate the root cause (such as optical module signal attenuation), and generate diagnosis data (i.e. abnormal reason). Then, the embodiment of the application can optimize the network protocol layer (such as the physical layer) corresponding to the abnormality according to the diagnosis data (such as replacing the optical module), and after the optimization is completed, the cross-layer performance index data of each layer is re-collected to form a closed-loop optimization, covering the full-stack performance management in 5G, cloud native and other scenarios.

[0084] Therefore, the embodiment of the application can realize nanosecond-level alignment and correlation analysis of cross-layer data, and can automatically trigger abnormal diagnosis and closed-loop optimization, so as to automatically generate a joint treatment scheme including physical layer BERT diagnosis-transmission layer congestion control parameter adjustment-application layer degradation strategy when detecting cross-layer abnormality, thereby accurately locating the root cause and improving network performance and stability.

[0085] As an implementable way, the specific steps of the embodiment of the application for full-dimension deep performance analysis are as follows:

[0086] Step 1, cross-layer performance data collection and time alignment:

[0087] (1) Adopting the combination of active detection and passive collection, collect the performance data of network interface card and corresponding network protocol stack layer by layer: collect the bit error rate through BERT equipment and collect the optical signal strength through ethtool in the physical layer; collect the MAC frame error rate through switch port mirroring and collect the VLAN throughput through arping in the data link layer; collect the route convergence data through sFlow and collect the IP fragmentation information through customized traceroute in the network layer; capture the TCP retransmission rate and RTT dynamic data through eBPF kernel probe in the transport layer; collect the HTTP delay through ChromeDevTools and collect the business error rate through Nginx log in the application layer, to form the raw performance data set (raw_data);

[0088] (2) Perform nanosecond-level time alignment on the raw performance data set through PTP hardware timestamp, to ensure the consistency of time dimension of each layer data.

[0089] Step 2, multi-dimensional performance index calculation:

[0090] (1) Input the time-aligned raw performance data set (raw_data) output in step 1 into the performance calculation logic, to generate multi-dimensional performance indexes (metrics), as follows:

[0091] 1) Basic performance indexes: call the throughput calculation function (calculate_throughput) to get the throughput, and call the quantile latency calculation function (calculate_percentile_latency) to get the 50 / 90 / 99 quantile latency (latency);

[0092] 2) Advanced diagnostic indexes: calculate the interrupt efficiency (interrupt_efficiency) through "packets_processed / interrupt_count";

[0093] 3) Hardware health index: calculate the thermal throttling ratio (thermal_throttling) through "time_in_throttle / test_duration";

[0094] Step 3, performance bottleneck positioning and suggestion generation:

[0095] (1) With the multi-dimensional performance metrics output in step 2 as the basis for judgment, locate performance bottlenecks: if the interrupt efficiency < 1000, determine that it is a "high interrupt rate affecting performance" bottleneck; if the thermal throttling ratio > 0.1, determine that it is a "overheating throttling affecting stability" bottleneck;

[0096] (2) Call the generate_recommendations function to generate targeted optimization recommendations based on the located performance bottlenecks;

[0097] (3) Integrate multi-dimensional performance metrics, performance bottlenecks and optimization recommendations, and output full-dimensional performance analysis results.

[0098] Step 4, cross-layer association monitoring and closed-loop diagnosis:

[0099] (1) Import the full-dimensional performance analysis results output in step 3 into the cross-layer association dashboard built by Prometheus and Grafana to real-time display the performance metric association relationship of each layer;

[0100] (2) When the dashboard detects cross-layer abnormal association (such as a sudden increase in physical layer error rate and excessive transmission layer TCP retransmission rate), automatically trigger the diagnosis process of the corresponding layer (such as optical module diagnosis), forming a closed loop from performance collection to problem diagnosis.

[0101] It should be noted that the related procedures of the above full-dimensional deep performance analysis process are as follows:

[0102] def performance_analysis(raw_data): # Multi-dimensional performance evaluation metrics = {} # Basic performance metrics metrics['throughput'] = calculate_throughput(raw_data) metrics['latency']= calculate_percentile_latency(raw_data, [50, 90, 99]) # Advanced diagnostic metrics metrics['interrupt_efficiency']= (raw_data['packets_processed'] / raw_data['interrupt_count']) # Hardware health evaluation metrics['thermal_throttling'] = (raw_data['time_in_throttle'] / raw_data['test_duration']) # Bottleneck localization bottlenecks = []ifmetrics['interrupt_efficiency']<1000:bottlenecks.append('High interruption rate affects performance')ifmetrics['thermal_throttling']>0.1:bottlenecks.append('Overheating and throttling affect stability')return {'metrics': metrics,'bottlenecks': bottlenecks,'recommendations':generate_recommendations(bottlenecks)}

[0103] Therefore, the embodiments of this application can accurately locate bottlenecks by collecting cross-protocol stack data in layers and aligning it with time, and by calculating indicators in multiple dimensions. Combined with cross-layer correlation monitoring and closed-loop diagnosis, the root cause of full-stack performance can be located, providing a precise basis for network optimization.

[0104] In step S103, various cross-layer performance index data are input into a pre-built multi-stage intelligent test model to generate corresponding test case combinations. Multiple virtualized test environments corresponding to the network interface card are deployed, and the test case combinations are tested in parallel in multiple virtualized test environments to obtain test data of the network interface card under different virtualized test environments. Based on the test data, a corresponding correlation graph is constructed, and a defect test report of the network interface card is generated according to the correlation graph.

[0105] Further, the embodiments of the present application also need to be based on a variety of cross-layer performance index data and a pre-constructed multi-stage intelligent test model (which includes a feature extraction model, a test strategy engine and an online learning model, etc.), dynamically generate and optimize the corresponding test cases; then, the embodiments of the present application can construct the corresponding virtual test environment through the infrastructure as code and containerized orchestration, to test the optimized test case combination for parallel elastic testing of multiple platforms and multiple configurations.

[0106] Optionally, in an embodiment of the present application, the plurality of cross-layer performance index data is input into the pre-constructed multi-stage intelligent test model to generate the corresponding test case combination, comprising: obtaining the historical test data corresponding to the network interface card, and training the pre-constructed feature extraction model through the historical test data, to identify the association rules between the fault modes corresponding to the network interface card and the business scenarios based on the plurality of cross-layer performance index data and the feature extraction model; inputting the association rules into the pre-constructed test strategy engine to dynamically generate the test case combination corresponding to the network interface card; obtaining the code coverage and defect detection rate of the test case combination in the running process, and inputting the code coverage and defect detection rate into the pre-constructed online learning model, to adjust the test weight of the test case combination through the online learning model, and optimize the test case combination through the adjusted test weight.

[0107] As a way that can be implemented, the embodiments of the present application can first collect the historical test data of the network interface card, covering the code change records (such as driver version update, hardware adaptation modification), defect reports (such as data transmission interruption, protocol compatibility problem), and traffic logs (such as data packet forwarding records under different protocols), and input these data into the pre-constructed feature extraction model (such as LSTM (‌Long Short Term Memory, long short term memory network‌) or Transformer) for training.

[0108] Secondly, after the training is completed, the embodiments of the present application can combine the plurality of cross-layer performance index data (such as the physical layer bit error rate and the transmission layer TCP retransmission rate) of the network interface card, so that the model automatically identifies the association rules between the high-frequency fault modes (such as the sudden rise of the packet loss rate under high load) and the business scenarios (such as the 5G slice data transmission scenario).

[0109] Again, the embodiment of the application can input the identified association rules into a test strategy engine based on a reinforcement learning framework (such as DQN (‌DeepQ Network, deep Q network)), and the engine dynamically generates test case combinations according to the rules; for example, the embodiment of the application can generate boundary value test cases (such as continuous transmission test under maximum bandwidth) for core business links (such as financial data transmission links); focus on basic function verification cases for ordinary function modules (such as network card basic configuration modules); for modules involving visual interaction, increase visual difference detection cases to generate corresponding test case combinations.

[0110] During the running of the test case combination, the embodiment of the application can monitor the code coverage (such as whether the coverage ratio of the target code reaches 90%) in real time through tools such as JaCoCo, and at the same time, count the defect detection rate (such as the number of faults detected per 100 cases), and input the two data into the online learning model.

[0111] After that, the online learning model can continuously optimize the test weight according to the data feedback, for example, increasing the generation probability of the case that has triggered a bug, and reducing the weight of the case that has not detected a defect and the coverage has reached the standard. In addition, the embodiment of the application can also introduce a GAN (‌Generative Adversarial Networks, generative adversarial network) to simulate abnormal input (such as a deformed TCP packet, a non-compliant API (Application Programming Interface, ‌application programming interface) call), supplement the test case library, and finally form a continuously optimized test case combination.

[0112] Therefore, the embodiment of the application can accurately identify the association between faults and scenarios, dynamically optimize test cases, thereby improving test efficiency and defect detection rate, and realizing the intelligentization and high efficiency of network interface card testing.

[0113] Optionally, in an embodiment of the application, the code coverage and defect detection rate of the test case combination in the running process are obtained, and the code coverage and defect detection rate are input into a pre-constructed online learning model to adjust the test weight of the test case combination through the online learning model, and the test case combination is optimized through the adjusted test weight, including: collecting corresponding test performance feedback data in the execution process of the test case combination, wherein the test performance feedback data includes code coverage and defect detection rate; inputting the test performance feedback data into the online learning model to output the performance score corresponding to the test case combination; sorting the performance scores of the test case combination in descending order to obtain a corresponding sorting result, and adjusting the test weight of the test case combination according to the sorting result to optimize the test case combination through the adjusted test weight.

[0114] It should be noted that during the execution of the test case combination (such as protocol compatibility test case of network interface card, high load stress test case, etc.), the embodiments of the present application can collect test efficiency feedback data such as code coverage and defect detection rate in real time through professional tools, as described below:

[0115] 1. Code coverage: the embodiments of the present application monitor the coverage of test cases on target code (such as network card driver code, data forwarding logic code) by means of JaCoCo, Cobertura and other tools, including class coverage, method coverage, line coverage and other detailed indicators (for example, after the execution of a certain test case combination, the line coverage is 85%, the method coverage is 90%);

[0116] 2. Defect detection rate: the embodiments of the present application count the number of valid defects detected in the test process (excluding repeated reporting and false positives) through the test management platform (such as TestRail), and calculate the defect detection rate in combination with the total number of test cases (for example, 100 test cases are executed, 12 valid defects are detected, and the detection rate is 12%).

[0117] Further, the embodiments of the present application can input the collected code coverage and defect detection rate into a pre-constructed online learning model (such as a scoring model based on gradient boosting tree), which can comprehensively calculate the two indicators according to the preset weight (such as code coverage accounting for 60% and defect detection rate accounting for 40%) and output the performance score corresponding to each test case combination.

[0118] For example, the code coverage of a certain test case combination is 88%, and the defect detection rate is 15%, so the performance score = 88% x 0.6 + 15% x 0.4 = 52.8% + 6% = 58.8%.

[0119] Then, the embodiments of the present application can sort the performance scores of all test case combinations in descending order (such as combination A 65%, combination B 58.8%, combination C 42% from high to low), and adjust the test weight according to the sorting result: for the top 30% of test case combinations (such as combination A), increase the execution weight by 20% (preferentially allocate test resources and increase execution frequency) in subsequent tests; for the last 20% of test case combinations (such as combination C), reduce the weight by 15% (reduce the execution proportion, or optimize the test case design in combination with the defect type). Through the optimized test weight, the test case combination of subsequent test is dynamically adjusted, so as to ensure that the high-performance test case plays a greater role.

[0120] Therefore, the embodiments of the present application dynamically optimize the test weight and test case combination based on real-time performance data to preferentially guarantee high-performance test execution, thereby significantly improving the test resource utilization rate and defect detection efficiency.

[0121] Optionally, in an embodiment of the present application, historical test data corresponding to the network interface card is acquired, and a pre-constructed feature extraction model is trained through the historical test data, to identify the association rules between the fault modes and the business scenarios corresponding to the network interface card based on the multiple cross-layer performance indicator data and the feature extraction model, including: collecting historical test data of the network interface card, wherein the historical test data includes performance indicator time series data, configuration change records, and defect report texts, and the historical test data is preprocessed and timestamped to fuse the performance indicator time series data, the configuration change records, and the defect report texts onto a target timeline to construct a corresponding multi-modal time series data set; inputting the multi-modal time series data set into the feature extraction model to extract performance evolution features and fault semantic features of the multi-modal time series data set through the feature extraction model, and fusing the performance evolution features and the fault semantic features to generate unified deep features, and training the feature extraction model through the unified deep features; inputting the multiple cross-layer performance indicator data into the trained feature extraction model to extract deep features of the multiple cross-layer performance indicator data, and identifying fault modes corresponding to the multiple cross-layer performance indicator data based on the deep features, and determining the association rules between the fault modes and the business scenarios.

[0122] Specifically, the specific process of identifying the association rules between the fault modes and the business scenarios by the feature extraction model is as follows:

[0123] 1. Collecting historical test data of the network interface card, which mainly covers three types of core information:

[0124] (1) Performance indicator time series data: such as optical power and bit error rate of the physical layer, MAC frame CRC error rate of the data link layer, routing switching delay of the network layer, and TCP retransmission rate of the transport layer, recorded at a frequency of 1 minute / time to form a time series data set;

[0125] (2) Configuration change records: including network card driver version update, bandwidth parameter adjustment, and protocol stack configuration modification (such as TCP window size setting) operation logs, marking the time point and specific content of each change;

[0126] (3) Defect report texts: fault records derived from a test management platform (such as JIRA), containing unstructured text data such as fault occurrence time, phenomenon description (such as packet loss under high load), root cause analysis (such as FEC error correction module response delay), and repair scheme.

[0127] 2. Preprocessing the collected historical test data:

[0128] The embodiments of the present application can eliminate abnormal values (such as sudden rise and fall of optical power caused by sensor failure) in performance indicators, and use linear interpolation method to fill in short-time missing data. Secondly, the embodiments of the present application can align the configuration change record and the defect report text according to the time stamp, associate the "configuration change-performance change-fault occurrence" data in the same time period, and fuse them into a unified timeline (such as taking "2024-05-20 14:00" as a time node, associating the bandwidth configuration, error rate data and data packet loss fault report in this period), to construct a multi-modal time series data set.

[0129] 3, input the multi-modal time series data set into the pre-constructed feature extraction model (such as LSTM or Transformer):

[0130] In the embodiments of the present application, the feature extraction model can extract performance evolution features (such as "the trend feature of the continuous 30-minute error rate rising from 10-9 to 10-6") for performance indicator time series data; for defect report text, extract fault semantic features (such as semantic vectors corresponding to keywords such as high load and data packet loss) through word embedding (such as Word2Vec) and attention mechanism; then, the embodiments of the present application can integrate the two types of features into unified deep features through a feature fusion layer (such as splicing and attention weighting), and iteratively train the model based on the features, until the recognition accuracy of the model for fault features reaches more than 90%.

[0131] 4, the embodiments of the present application can input various cross-layer performance indicator data of network interface cards (such as current optical power -2dBm, TCP retransmission rate 0.8%) into the trained model, the model extracts the deep features of the data, compares them with the learned fault feature library, and identifies the corresponding fault mode (such as error rate rising fault caused by low optical power); at the same time, combined with the business scenario label (such as 5G core network data transmission scenario, enterprise cloud storage access scenario), the occurrence frequency of each fault mode in different scenarios is counted to establish the corresponding association rule (such as in the 5G core network scenario, when the optical power is <-3dBm, the fault occurrence rate of the error rate exceeding the standard is 85%).

[0132] Therefore, the embodiments of the present application can accurately identify the fault mode of the network card and its association rule with the business scenario through multi-modal data fusion and model training, provide reliable basis for subsequent test strategy optimization and fault prediction, and improve the efficiency of network card testing and operation and maintenance.

[0133] Optionally, in an embodiment of the present application, a plurality of virtualization test environments corresponding to the network interface card are deployed, and the test case combination is tested in parallel in the plurality of virtualization test environments to obtain test data of the network interface card under different virtualization test environments, including: dynamically calling container instances of different operating system versions and middleware configurations in the plurality of virtualization test environments to obtain corresponding containerized test environments; using a pre-constructed overlay network to isolate the test domain in the containerized test environment, and connecting the preset hardware accelerator to the isolated test domain, and performing corresponding load testing on the optimized test case combination to output the corresponding test environment result; determining the corresponding label policy according to the test environment result, and distributing the test task to the nodes in the corresponding test domain based on the label policy, and collecting the performance data of the virtual machine group and the application performance data of the container instance in the corresponding test domain during the execution of the test task, to generate the test data according to the plurality of cross-layer performance index data, the performance data and the application performance data.

[0134] In actual execution process, the specific process of the virtualization test environment deployment and parallel testing performed by the embodiment of the present application is as follows:

[0135] 1. According to the preset test case requirements (such as compatibility test, performance stress test) of the network interface card, a plurality of heterogeneous virtualization test environments are deployed by Terraform or Ansible automation tools: KVM (Kernel-based Virtual Machine, kernel virtual machine), ESXi (Elastic Sky X Server, virtual machine monitor), Hyper-V virtual machine group covering x86 / ARM architecture, without manual configuration of hardware and system, realizing fast construction and standardization of test environment.

[0136] 2. In the deployed virtualization environment, different configurations of container instances are dynamically scheduled by the container orchestration capability of Kubernetes to build a containerized test environment; for example, CentOS 7+MySQL 5.7 (adapted to traditional business scenarios) and Ubuntu 22.04 combined with PostgreSQL 15 (adapted to cloud native scenarios) container combination are simultaneously scheduled to meet the test requirements of the network interface card under different operating systems and middleware environments.

[0137] 3、To avoid interference between different test tasks, embodiments of the present application can use VXLAN (‌Virtual Extensible Local Area Network, virtual extensible local area network) overlay network technology to isolate each test domain (such as compatibility test domain, performance test domain) in the containerized test environment, and ensure that the network environment of each test domain is independent. At the same time, embodiments of the present application can connect hardware accelerators (such as NVIDIA vGPU for graphical related tests, FPGA emulator for high-speed signal processing tests) to the corresponding isolated test domain according to the specific load requirements (such as high concurrency data forwarding, GPU (‌Graphic Processing Unit, graphic processing unit) acceleration processing) in the test case, to provide hardware resource support for testing.

[0138] 4、Perform load testing on the optimized network interface card test case combination (such as boundary value test case, abnormal input test case), and determine the label strategy according to the test environment result (such as the load type and hardware configuration of each test domain), for example, label the test domain with "high-load-gpu" label for "high-load-gpu", and label the "compatibility-x86" label for "compatibility-x86".

[0139] 5、Through the Jenkins Pipeline test task management tool, the test task is accurately distributed to the node in the corresponding test domain according to the label strategy: for example, the test task with the windows-11-gpu label is distributed to the Windows 11 virtual node equipped with a GPU accelerator. During the execution of the test task, embodiments of the present application can monitor the performance data (such as virtual CPU utilization, memory occupancy) of the virtual machine group in each test domain in real time through the Libvirt tool, and use Prometheus to grab the application performance data (such as MySQL query response time, network interface card data packet forwarding rate) of the container instance.

[0140] 6、Combine the various cross-layer performance index data of the network interface card (such as physical layer bit error rate, transmission layer TCP throughput) with the collected application performance data, integrate to generate complete test data, and generate a cross-platform comparison report (such as analyzing that the network interface card TCP throughput under Windows system is 12% lower than Linux, and the root cause is the difference in driver adaptation), to realize integrated test matrix management from bare metal to hybrid cloud.

[0141] Thus, embodiments of the present application can greatly improve the test efficiency and comprehensiveness through automated deployment and parallel test operation, provide a reliable basis for network card compatibility and performance optimization, and reduce the test cost.

[0142] Afterwards, embodiments of the present application can construct a corresponding association graph based on the test data, to perform defect analysis on the network interface card using the association graph, to generate a defect test report containing problem diagnosis suggestions.

[0143] Optionally, in an embodiment of the present application, based on the test data, a corresponding association graph is constructed, and a defect test report of the network interface card is generated according to the association graph, including: inputting the test data into a pre-constructed graph neural network to construct a corresponding association graph, and identifying a fault propagation path meeting a preset frequency requirement through the association graph; obtaining the original log and stack information corresponding to the network interface card, and parsing the original log and stack information to generate corresponding natural language diagnosis suggestions, and based on the natural language diagnosis suggestions and the fault propagation path, determining a corresponding similar historical solution; based on the similar historical solution, determining an executable repair verification command corresponding to the natural language diagnosis suggestion, and constructing a defect test report according to the natural language diagnosis suggestion and the executable repair verification command.

[0144] Specifically, the process of automatically generating a test report containing problem diagnosis suggestions according to embodiments of the present application is as follows:

[0145] 1. Collect test data of the network interface card, including code coverage, defect triggering stack, network packet capture data (such as TCP / UDP packet forwarding exception records), cross-layer performance indicators (physical layer bit error rate, transmission layer retransmission rate) and other real-time data streams, and input them into a pre-constructed GNN (Graph Neural Network); GNN learns the association between test items (such as high-load forwarding test), defects (such as packet loss), and code modules (such as network card driver data processing module), constructs an association graph of the interconnection of the three, and identifies a high-frequency fault propagation path meeting a preset frequency requirement (such as occurrence frequency ≥ 5 times) based on the node connection frequency in the graph, for example, the causal chain of “network card FEC error correction failure-TCP retransmission rate increase-Redis cache data synchronization timeout”.

[0146] 2. A causal reasoning-driven diagnostic engine is introduced. Based on the Pearl causal model, the causal relationships in the correlation graph are analyzed to distinguish between correlation and causality (e.g., eliminating false associations between high memory usage and packet loss, confirming that the decrease in disk IOPS is the root cause of database interaction timeout), thus avoiding misjudgment of fault paths. Simultaneously, this embodiment can input the original logs of the network interface card (e.g., driver runtime logs, system kernel logs) and defect trigger stack information into a finely tuned LLM (Large Language Model, such as CodeLlama-70B). The model uses semantic understanding to parse fault features in the data (e.g., the CRC check failure corresponding to "error: eth0: frame check sequence error"), generating natural language diagnostic suggestions (e.g., "It is recommended to check the physical interface link quality of the network card and investigate network cable or optical module faults").

[0147] 3. Leveraging an incremental knowledge graph (including historical fault cases and solution libraries), and based on natural language diagnostic suggestions and fault propagation paths, similar historical solutions are matched. For example, for a fault of "high TCP retransmission rate," it is associated with the historical case of "network card protocol stack anomaly caused by CVE-2023-4567 vulnerability," and the corresponding patch installation scheme is recommended. Combining similar historical solutions, reusable operational logic is extracted to generate executable repair verification commands corresponding to natural language diagnostic suggestions, such as "replay network traffic using the K6 tool and execute 'k6 run --vus 100 --duration 5m traffic-test-script.js' to verify the stability of packet forwarding after connection pool configuration optimization."

[0148] 4. During the report construction phase, this application embodiment can integrate content using Markdown format and embed an interactive D3.js visualization component to automatically highlight key indicator anomalies (such as the spatiotemporal correlation graph of a sudden increase in HTTP 500 error rate and CPU load curve, and the linkage trend graph of network card bit error rate and optical power changes). Simultaneously, through a difference-driven simplification mechanism, compared with historical benchmark reports, only statistically significant differences are retained (such as "the TCP throughput in this test is 8% lower than the historical average"), while redundant information is removed. The final result is a network interface card test report containing fault propagation paths, natural language diagnostic suggestions, similar historical solutions, and executable repair verification commands, supporting semantic search (such as entering "Kafka consumption delay" to associate with network card TCP window scaling configuration issues).

[0149] Therefore, the embodiments of this application can accurately identify the cause and propagation path of the fault to generate a verifiable diagnostic solution, thereby improving the efficiency of fault location and the accuracy of repair, and shortening the problem-solving cycle.

[0150] As an implementable way, the specific steps of the method for constructing the association graph of test items-defects-code modules and identifying high-frequency fault propagation paths according to the embodiments of the application are as follows:

[0151] Step 1, multi-dimensional real-time data stream acquisition and enhanced processing:

[0152] (1) The embodiments of the application can collect real-time data streams in the test execution phase of the network interface card, and the real-time data streams at least include code coverage data, defect trigger stack data, network packet capture data (including TCP / UDP packet forwarding exception records), cross-layer performance index data (including physical layer bit error rate, transmission layer retransmission rate);

[0153] (2) The collected real-time data streams are enhanced and processed, the code coverage data is labeled with the corresponding test item identifier (such as “high load forwarding test”), the defect trigger stack data is associated with the defect type label (such as data packet loss), and the cross-layer performance index data is added with the timestamp and code module mapping relationship (such as “physical layer bit error rate-network card driver data processing module”), and an enhanced multi-dimensional data set is output;

[0154] Step 2, dynamic weight GNN model training and association graph initialization:

[0155] (1) The enhanced multi-dimensional data set is input into the pre-constructed dynamic weight graph neural network model, the node layer of the GNN model is set as the test item node, the defect node and the code module node, and the initial value of the edge weight is determined by the data correlation degree (such as the correlation degree weight of the code coverage data and the test item node is set as 0.6);

[0156] (2) The implicit association relationship among the test item, the defect and the code module in the enhanced multi-dimensional data set is learned through the GNN model, the edge weight between the nodes is dynamically adjusted (such as when the high load forwarding test and the data packet loss defect appear simultaneously for more than or equal to 3 times, the corresponding edge weight is increased to 0.8), and an initial test item-defect-code module association graph (i.e. the initial association graph) is output;

[0157] Step 3, incremental graph iteration optimization and fault path mining:

[0158] (1) New network interface card test real-time data streams are continuously collected, the enhanced processing process of step 1 is repeated, and a new enhanced data set is obtained, and the new enhanced data set is input into the initial association graph output in step 2;

[0159] (2) Based on the node interaction frequency in the new enhanced data set, the initial association graph is iteratively optimized, the node edge weight and the node connection relationship are updated, and an optimized test item-defect-code module association graph is generated;

[0160] (3) Based on the optimized association graph, the occurrence frequency of each node connection path is counted, and the path meeting the preset frequency requirement (such as the number of occurrences ≥ 5 times) is screened out, and is identified as a high-frequency fault propagation path (such as "network card FEC error correction failure-TCP retransmission rate increase-Redis cache data synchronization timeout", "Redis timeout-payment service degradation-order status inconsistency");

[0161] Therefore, the embodiment of the application combines data flow enhancement with dynamic weight GNN, realizes dynamic optimization of the association graph, accurately mines the high-frequency fault propagation path, reduces false association interference, improves the efficiency and accuracy of network interface card fault positioning, and shortens the problem troubleshooting period.

[0162] In addition, after obtaining the corresponding test result (i.e., the initial test result), the embodiment of the application can also perform an adaptive traffic generation operation based on the initial test result, and the specific steps are as follows:

[0163] Step 1, initial test result data collection:

[0164] (1) Collect the core performance data corresponding to the initial test of the network interface card, which at least includes the packet loss rate (packet_loss) and the latency jitter (latency_jitter), to form an initial test result data set (initial_results);

[0165] Step 2, traffic mode decision:

[0166] (1) The initial test result data set output in step 1 is used as a basis for decision, and the traffic mode decision logic is executed: if the packet loss rate (packet_loss) in the data set is > 0.1, the decision is stress test mode (stress); if the latency jitter (latency_jitter) is > 50, the decision is constant flow test mode (constant); if neither is satisfied, the decision is the default comprehensive test mode;

[0167] Step 3, traffic mode parameter configuration:

[0168] (1) According to the traffic mode decision result output in step 2, configure the traffic parameters of the corresponding mode:

[0169] (2) If it is a stress test mode, the configuration parameters include test duration (duration) 300, packet size distribution (packet_size_dist) [64, 128, 512, 1518], and rate distribution (rate_distribution) "exponential";

[0170] (3) If it is a stable flow test mode, the configuration parameters include test duration (duration) 180, fixed packet size (packet_size) 1024, rate (rate) "90%";

[0171] (4) If it is a default comprehensive test mode, a preset default test configuration function (default_test_profile ()) is called to generate corresponding parameters;

[0172] (5) Output adaptive traffic configuration results containing mode identification and corresponding parameters.

[0173] It should be noted that the related program of the adaptive traffic generation strategy based on the initial test results is as follows:

[0174] def adaptive_traffic_generation(initial_results):# Based on the initial test results dynamically adjust the traffic mode if initial_results['packet_loss']>0.1:# High packet loss scenario, switch to stress test mode return {'mode':'stress','duration': 300,'packet_size_dist': [64,128, 512, 1518],'rate_distribution': 'exponential'}elif initial_results['latency_jitter']>50:# High jitter scenario, switch to constant flow test return {'mode': 'constant','packet_size': 1024,'rate': '90%','duration': 180}else:# Default comprehensive test mode return default_test_profile()

[0175] Therefore, the embodiments of the present application accurately decide the traffic mode based on the initial test results, configure the parameters pertinently, and adapt to high packet loss, high jitter and other scenarios without manual intervention, thereby improving the flexibility and adaptability of traffic generation, and providing a traffic solution for adapting to scenarios for network interface card testing.

[0176] It can be understood that the embodiments of the present application can improve the test efficiency and problem detection rate in the field of network card testing through intelligent traffic simulation and deep performance analysis strategy, reduce the comprehensive cost, establish a new benchmark for industry testing, promote the coordinated development of the industry, not only solve the long-standing testing problems in the industry, but also provide a solid technical support for the development of next-generation network technology, which can be well applied in extreme environment testing (such as wide temperature testing (-40℃ to 85℃), high altitude simulation (above 5000 meters), electromagnetic interference environment testing), safety critical field (such as vehicle network reliability verification, industrial control real-time guarantee, aerospace level stability testing) and future technology adaptation field (such as quantum network adaptation capability, 6G network pre-research support, space-air-ground integrated network testing) and other scenes, which has great technical value and economic value.

[0177] The execution logic of the network interface card testing method of the present application is described below by combining the accompanying drawings.

[0178] Figure 2 The execution logic of the network interface card testing method of the present application is described below by combining the accompanying drawings. Figure 2 As shown in the figure, the execution process of the network interface card testing method of the present application is as follows:

[0179] S201: Dynamically generating mixed traffic simulating real network environment;

[0180] S202: Full-range performance measurement on different network protocol layers from physical layer to application layer;

[0181] S203: Dynamically generating and optimizing test case combination based on multi-stage intelligent testing model;

[0182] S204: Building a virtualized testing environment to support parallel testing of test case combination on multiple platforms and multiple configurations;

[0183] S205: Automatically generating a defect test report containing problem diagnosis suggestions.

[0184] In addition, the network interface card testing system corresponding to the execution logic of the network interface card testing method can also be built according to the execution logic of the network interface card testing method.

[0185] Figure 3 The logic architecture of the network interface card testing system of the present application is shown in the figure. Figure 3 As shown in the figure, the network interface card testing system of the present application mainly includes an intelligent traffic generation engine 301, a full-stack performance analysis module 302, an adaptive testing module 303, a virtualized testing module 304 and an automated reporting module 305.

[0186] The intelligent traffic generation engine 301 can dynamically generate mixed traffic simulating real network environment;

[0187] The full-stack performance analysis module 302 can perform full-range performance measurement from the physical layer to the application layer.

[0188] The adaptive test module 303 can dynamically generate and optimize test cases.

[0189] The virtualized test module 304 can support parallel testing of multiple platforms and configurations.

[0190] The automated report module 305 can automatically generate a test report containing problem diagnosis suggestions.

[0191] Through the above description of the embodiments, those skilled in the art can clearly understand that the method according to the above embodiments can be implemented by means of software and the necessary general hardware platform, of course, it can also be implemented by hardware, but in many cases the former is a better embodiment.

[0192] The embodiments of the present application also provide a network interface card testing device.

[0193] As shown in Figure 4 The network interface card testing device 10 comprises a traffic mixing module 100, a data acquisition module 200, and a defect testing module 300.

[0194] The traffic mixing module 100 is configured to determine traffic generation proportions corresponding to multiple protocol traffic scripts based on a target network environment of the network interface card, mix multiple protocol traffics according to the traffic generation proportions, and obtain network performance disturbance parameters of the target network environment, so as to generate target mixed traffic simulating the target network environment based on the mixed multiple protocol traffics and the network performance disturbance parameters.

[0195] The data acquisition module 200 is configured to input the target mixed traffic into the network interface card, and acquire multiple cross-layer performance index data of different network protocol layers corresponding to the network interface card.

[0196] The defect testing module 300 is configured to input the multiple cross-layer performance index data into a pre-constructed multi-stage intelligent testing model, to generate corresponding test case combinations, deploy multiple virtualized testing environments corresponding to the network interface card, and perform parallel testing on the test case combinations in the multiple virtualized testing environments, to obtain testing data of the network interface card under different virtualized testing environments, construct a corresponding association graph based on the testing data, and generate a defect testing report of the network interface card according to the association graph.

[0197] Optionally, in an embodiment of the present application, the traffic mixing module 100 comprises a statistical unit and a first generation unit.

[0198] The statistical unit is configured to perform network traffic statistical analysis on the network interface card to obtain corresponding statistical data, and establish a corresponding protocol proportion model according to the statistical data.

[0199] The first generating unit is configured to generate a traffic generation proportion of the multiple protocol traffic scripts based on the protocol proportion model.

[0200] Optionally, in an embodiment of the present application, the traffic mixing module 100 further comprises a second generating unit, a disturbance injection unit and a damage injection unit.

[0201] The second generating unit is configured to mix the multiple protocol traffics according to the traffic generation proportion to generate corresponding initial mixed traffic.

[0202] The disturbance injection unit is configured to calculate a current network load index corresponding to the network interface card, and adjust a state transition probability of a pre-constructed Markov model according to the current network load index, so as to inject a network performance disturbance parameter into the initial mixed traffic based on the adjusted state transition probability, wherein the network performance disturbance parameter comprises a time delay parameter, a packet loss parameter and a jitter parameter.

[0203] The damage injection unit is configured to dynamically compensate and correct the network performance disturbance parameter, and obtain a corresponding hardware timestamp according to a preset hardware time synchronization mechanism, and perform damage injection operation on the initial mixed traffic after the network performance disturbance parameter is injected, to generate target mixed traffic simulating a target network environment.

[0204] Optionally, in an embodiment of the present application, the data collection module 200 comprises a physical layer collection unit, a data link layer collection unit, a network layer collection unit, a transport layer collection unit and an application layer collection unit.

[0205] The physical layer collection unit is configured to collect physical signal quality and bit error rate data in cross-layer performance index data corresponding to the network interface card in a physical layer of a network protocol layer.

[0206] The data link layer collection unit is configured to collect medium access control frame error rate and link reachability data in multiple cross-layer performance index data corresponding to the network interface card in a data link layer of the network protocol layer.

[0207] The network layer collection unit is configured to collect routing state and data packet fragmentation information in multiple cross-layer performance index data corresponding to the network interface card in a network layer of the network protocol layer.

[0208] The transport layer collection unit is configured to collect transmission control protocol retransmission rate and round-trip delay dynamic data in multiple cross-layer performance index data corresponding to the network interface card in a transport layer of the network protocol layer.

[0209] The application layer acquisition unit is configured to acquire, in an application layer of a network protocol layer, service error rates and user experience index data in a plurality of cross-layer performance index data corresponding to a network interface card.

[0210] Optionally, in an embodiment of the present application, the network interface card testing device 10 further comprises a time delay alignment module, an association analysis module, and an anomaly diagnosis module.

[0211] The time delay alignment module is configured to perform time delay alignment on physical signal quality, bit error rate data, medium access control frame error rate, link reachability data, routing state, packet fragmentation information, transmission control protocol retransmission rate, round-trip delay dynamic data, service error rates, and user experience index data by using a hardware timestamp after acquiring a plurality of cross-layer performance index data corresponding to different network protocol layers of the network interface card, to obtain corresponding time delay alignment data.

[0212] The association analysis module is configured to perform association analysis on the time delay alignment data to obtain corresponding association analysis results, and determine whether there is cross-layer performance index anomaly data that meets a preset anomaly condition in the plurality of cross-layer performance index data according to the association analysis results, wherein when there is cross-layer performance index anomaly data that meets the preset anomaly condition in the plurality of cross-layer performance index data, a corresponding anomaly diagnosis mechanism is triggered.

[0213] The anomaly diagnosis module is configured to perform anomaly diagnosis on the cross-layer performance index anomaly data by using the anomaly diagnosis mechanism to obtain corresponding anomaly causes, and optimize a network protocol layer corresponding to the cross-layer performance index anomaly data according to the anomaly causes, to reacquire corresponding cross-layer performance index data by using the optimized network protocol layer.

[0214] Optionally, in an embodiment of the present application, the defect testing module 300 comprises an identification unit, an adjustment unit, and an optimization unit.

[0215] The identification unit is configured to acquire historical testing data corresponding to the network interface card, and train a pre-constructed feature extraction model by using the historical testing data, to identify an association rule between a fault mode and a service scenario corresponding to the network interface card based on the plurality of cross-layer performance index data and the feature extraction model.

[0216] The adjustment unit is configured to input the association rule into a pre-constructed testing strategy engine, to dynamically generate a test case combination corresponding to the network interface card.

[0217] The optimization unit is configured to acquire code coverage and defect detection rate of the test case combination in a running process, and input the code coverage and the defect detection rate into a pre-constructed online learning model, so as to adjust a test weight corresponding to the test case combination through the online learning model, and optimize the test case combination through the adjusted test weight.

[0218] Optionally, in an embodiment of the present application, the defect test module 300 further comprises a containerization unit, an isolation unit and a distribution unit.

[0219] The containerization unit is configured to dynamically call container instances of different operating system versions and middleware configurations in a plurality of virtualized test environments to obtain corresponding containerized test environments.

[0220] The isolation unit is configured to isolate a test domain in the containerized test environment by using a pre-constructed coverage network, and connect a preset hardware accelerator to the isolated test domain, and perform corresponding load testing on the optimized test case combination to output a corresponding test environment result.

[0221] The distribution unit is configured to determine a corresponding label strategy according to the test environment result, distribute test tasks to nodes in the corresponding test domain based on the label strategy, collect performance data of a virtual machine group and application performance data of a container instance in the corresponding test domain during execution of the test tasks, and generate test data according to a plurality of cross-layer performance index data, performance data and application performance data.

[0222] Optionally, in an embodiment of the present application, the defect test module 300 further comprises a construction unit, an analysis unit and a determination unit.

[0223] The construction unit is configured to input the test data into a pre-constructed graph neural network to construct a corresponding association graph, and identify a fault propagation path meeting a preset frequency requirement through the association graph.

[0224] The analysis unit is configured to acquire original logs and stack information corresponding to a network interface card, and analyze the original logs and the stack information to generate a corresponding natural language diagnosis suggestion, and determine a corresponding similar historical solution based on the natural language diagnosis suggestion and the fault propagation path.

[0225] The determination unit is configured to determine an executable repair verification command corresponding to the natural language diagnosis suggestion based on the similar historical solution, and construct a defect test report according to the natural language diagnosis suggestion and the executable repair verification command.

[0226] Optionally, in an embodiment of the present application, the perturbation injection unit comprises a sending subunit and a weighted fusion subunit.

[0227] The sending subunit is configured to send the initial mixed traffic to a network interface card to collect underlying performance indexes corresponding to the network interface card, wherein the underlying performance indexes include packet queue delay and packet loss rate.

[0228] The weighted fusion subunit is configured to perform a weighted fusion operation on the underlying performance indexes to calculate a current network load index, and input the current network load index into a pre-constructed nonlinear mapping function to calculate a state transition probability of the Markov model.

[0229] Optionally, in an embodiment of the present application, the identifying unit comprises a timestamp alignment subunit, a first extraction subunit and a second extraction subunit.

[0230] The timestamp alignment subunit is configured to collect historical test data of the network interface card, wherein the historical test data comprises performance index time series data, configuration change records and defect report texts, and pre-process and timestamp-align the historical test data to fuse the performance index time series data, the configuration change records and the defect report texts onto a target timeline to construct a corresponding multi-modal time series dataset.

[0231] The first extraction subunit is configured to input the multi-modal time series dataset into a feature extraction model to extract performance evolution features and fault semantic features of the multi-modal time series dataset through the feature extraction model, fuse the performance evolution features and the fault semantic features to generate unified deep features, and train the feature extraction model through the unified deep features.

[0232] The second extraction subunit is configured to input a plurality of cross-layer performance index data into the trained feature extraction model to extract deep features of the plurality of cross-layer performance index data, identify fault modes corresponding to the plurality of cross-layer performance index data based on the deep features, and determine association rules between the fault modes and business scenarios.

[0233] Optionally, in an embodiment of the present application, the application layer collection unit comprises a tracking subunit and a calculation subunit.

[0234] The tracking subunit is configured to inject a tracking identifier into an external request at an application layer service entrance, and perform full-link tracking by using the tracking identifier to collect application performance indexes, server error rates and first input delays corresponding to the application layer.

[0235] The calculation subunit is configured to calculate user experience index data based on the application performance indexes, the server error rates and the first input delays.

[0236] Optionally, in an embodiment of the present application, the optimization unit comprises a feedback subunit, a scoring subunit and a sorting subunit.

[0237] The feedback subunit is configured to collect test efficiency feedback data corresponding to the test case combination during execution of the test case combination, wherein the test efficiency feedback data includes code coverage and defect detection rate.

[0238] The scoring subunit is configured to input the test efficiency feedback data into an online learning model to output an efficiency score corresponding to the test case combination.

[0239] The sorting subunit is configured to sort the efficiency scores of the test case combinations in descending order to obtain a corresponding sorting result, and adjust test weights of the test case combinations according to the sorting result to optimize the test case combinations through the adjusted test weights.

[0240] The features of the embodiments of the network interface card testing device can be referred to the related descriptions of the embodiments of the network interface card testing method, which will not be repeated here.

[0241] Embodiments of the present application also provide an electronic device, comprising a memory and a processor, the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any of the above network interface card testing method embodiments.

[0242] Embodiments of the present application also provide a non-volatile computer readable storage medium, which stores a computer program, wherein the computer program is configured to perform the steps in any of the above network interface card testing method embodiments when running.

[0243] In an example embodiment, the above non-volatile computer readable storage medium can include, but is not limited to, a U disk, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and various media that can store computer programs.

[0244] Embodiments of the present application also provide a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the steps in any of the above network interface card testing method embodiments.

[0245] Embodiments of the present application also provide another computer program product, which comprises a non-volatile computer readable storage medium, and the non-volatile computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps in any of the above network interface card testing method embodiments.

[0246] Those skilled in the art will further realize that the mere concepts, teachings, and embodiments described herein are merely meant to provide an enabling description of embodiments of the present application and that various modifications can be made thereto without departing from the scope of the present application. Accordingly, the appended claims are intended to embrace all such alterations, modifications, and variations of the present application. The above detailed description has been presented for the purposes of clarity and understanding. It is not intended to be exhaustive or to limit the present application to the precise form described. Many modifications and variations are possible in the scope of the application. The described embodiments are intended to be illustrative, but not restrictive, of the present application. Alternate embodiments will become apparent to those of ordinary skill in the art, and it will be apparent that one or more modifications can be made to the described embodiments without departing from the scope of the claims of the application. Accordingly, the scope of the present application should be determined not with reference to the above description but should instead be determined with reference to the appended claims, along with the full scope of equivalents to which such claims are entitled. The disclosures of each patent, patent application, and publication cited above are hereby incorporated herein by reference, each in its entirety.

[0247] The above provides a network interface card testing method, device, equipment and medium. The principles and implementation manners of the present application are described by applying specific examples. The above description of the embodiments is only used to help understand the method of the present application and its core idea. It should be pointed out that, for those skilled in the art, without departing from the principles of the present application, some improvements and modifications can be made to the present application. These improvements and modifications also fall within the protection scope of the claims of the present application.

Claims

1. A method for testing network interface cards, characterized in that, Includes the following steps: Based on the target network environment of the network interface card, the traffic generation ratio corresponding to multiple protocol traffic scripts is determined, and multiple protocol traffic is mixed according to the traffic generation ratio. The network performance disturbance parameters of the target network environment are obtained, so as to generate target mixed traffic simulating the target network environment based on the mixed multiple protocol traffic and the network performance disturbance parameters. The target mixed traffic is input into the network interface card, and various cross-layer performance index data corresponding to different network protocol layers of the network interface card are collected. The various cross-layer performance index data are input into a pre-constructed multi-stage intelligent test model to generate corresponding test case combinations. Multiple virtualization test environments corresponding to the network interface card are deployed, and the test case combinations are tested in parallel in the multiple virtualization test environments to obtain test data of the network interface card under different virtualization test environments. Based on the test data, a corresponding correlation graph is constructed, and a defect test report of the network interface card is generated according to the correlation graph. The step of mixing multiple protocol traffic according to the traffic generation ratio and obtaining network performance disturbance parameters of the target network environment to generate target mixed traffic simulating the target network environment based on the mixed multiple protocol traffic and the network performance disturbance parameters includes: Multiple protocol traffic is mixed according to the traffic generation ratio to generate the corresponding initial mixed traffic; Calculate the current network load index corresponding to the network interface card, and adjust the state transition probability of the pre-constructed Markov model according to the current network load index. Based on the adjusted state transition probability, inject the network performance perturbation parameters into the initial mixed traffic. The network performance perturbation parameters include latency parameters, packet loss parameters, and jitter parameters. The network performance disturbance parameters are dynamically compensated and corrected, and the corresponding hardware timestamp is obtained according to the preset hardware time synchronization mechanism. The initial mixed traffic after the network performance disturbance parameters are injected is subjected to impairment injection operation through the hardware timestamp to generate target mixed traffic simulating the target network environment.

2. The network interface card testing method according to claim 1, characterized in that, The step of inputting the target mixed traffic into the network interface card and collecting various cross-layer performance index data corresponding to different network protocol layers of the network interface card includes: In the physical layer of the network protocol layer, physical signal quality and bit error rate data are collected from the cross-layer performance index data corresponding to the network interface card. In the data link layer of the network protocol layer, the media access control frame error rate and link reachability data are collected from the various cross-layer performance index data corresponding to the network interface card. In the network layer of the network protocol layer, the routing status and packet fragmentation information of the various cross-layer performance index data corresponding to the network interface card are collected; In the transport layer of the network protocol layer, dynamic data of transmission control protocol retransmission rate and round-trip delay are collected from the various cross-layer performance index data corresponding to the network interface card. In the application layer of the network protocol layer, service error rate and user experience index data are collected from the various cross-layer performance index data corresponding to the network interface card.

3. The network interface card testing method according to claim 2, characterized in that, After collecting various cross-layer performance index data corresponding to different network protocol layers of the network interface card, the process also includes: The physical signal quality, bit error rate data, medium access control frame error rate, link reachability data, routing status, packet fragmentation information, transmission control protocol retransmission rate, round-trip delay dynamic data, service error rate, and user experience metric data are time-aligned using the hardware timestamp to obtain corresponding time-aligned data. The time delay alignment data is subjected to correlation analysis to obtain the corresponding correlation analysis results. Based on the correlation analysis results, it is determined whether there are any cross-layer performance index abnormal data that meet the preset abnormal conditions among the various cross-layer performance index data. When there are cross-layer performance index abnormal data that meet the preset abnormal conditions among the various cross-layer performance index data, the corresponding abnormal diagnosis mechanism is triggered. The anomaly diagnosis mechanism is used to diagnose the abnormal data of the cross-layer performance indicators to obtain the corresponding cause of the anomaly. Based on the cause of the anomaly, the network protocol layer corresponding to the abnormal data of the cross-layer performance indicators is optimized so that the corresponding cross-layer performance indicator data can be re-collected through the optimized network protocol layer.

4. The network interface card testing method according to claim 1, characterized in that, The step of inputting the various cross-layer performance index data into a pre-constructed multi-stage intelligent testing model to generate corresponding test case combinations includes: Obtain historical test data corresponding to the network interface card, and train a pre-built feature extraction model using the historical test data, so as to identify the association rules between the fault mode and business scenario corresponding to the network interface card based on the various cross-layer performance index data and the feature extraction model; The association rules are input into a pre-built test strategy engine to dynamically generate test case combinations corresponding to the network interface card; The code coverage and defect detection rate of the test case combination during the operation are obtained, and the code coverage and defect detection rate are input into a pre-built online learning model to adjust the test weights corresponding to the test case combination through the online learning model, and optimize the test case combination through the adjusted test weights.

5. The network interface card testing method according to claim 4, characterized in that, The deployment of multiple virtualization test environments corresponding to the network interface card, and the parallel testing of the test case combinations in the multiple virtualization test environments to obtain test data of the network interface card under different virtualization test environments, includes: Container instances with different operating system versions and middleware configurations are dynamically retrieved from the various virtualization test environments to obtain the corresponding containerized test environment; The test domain in the containerized test environment is isolated by a pre-built overlay network, and a preset hardware accelerator is connected to the isolated test domain. The optimized test case combination is subjected to corresponding load testing to output the corresponding test environment results. Based on the test environment results, a corresponding tagging strategy is determined. Based on the tagging strategy, test tasks are distributed to nodes in the corresponding test domain. During the execution of the test tasks, performance data of virtual machine groups and application performance data of container instances in the corresponding test domain are collected. The test data is generated based on the various cross-layer performance index data, the performance data, and the application performance data.

6. The network interface card testing method according to claim 1, characterized in that, The step of constructing a corresponding correlation graph based on the test data and generating a defect test report for the network interface card based on the correlation graph includes: The test data is input into a pre-constructed graph neural network to build a corresponding association graph, and the fault propagation path that meets the preset frequency requirements is identified through the association graph. Obtain the original logs and stack information corresponding to the network interface card, and parse the original logs and stack information to generate corresponding natural language diagnostic suggestions. Based on the natural language diagnostic suggestions and the fault propagation path, determine the corresponding similar historical solutions. Based on the similar historical solutions, the executable repair verification command corresponding to the natural language diagnostic suggestion is determined, and the defect test report is constructed according to the natural language diagnostic suggestion and the executable repair verification command.

7. The network interface card testing method according to claim 1, characterized in that, The process involves calculating the current network load index corresponding to the network interface card, adjusting the state transition probability of the pre-built Markov model based on the current network load index, and injecting the network performance perturbation parameters into the initial mixed traffic based on the adjusted state transition probabilities. These network performance perturbation parameters include latency parameters, packet loss parameters, and jitter parameters, including: The initial mixed traffic is sent to the network interface card to collect the underlying performance indicators corresponding to the network interface card, including packet queue latency and packet loss rate. The underlying performance metrics are weighted and fused to calculate the current network load index, and the current network load index is input into a pre-constructed nonlinear mapping function to calculate the state transition probability of the Markov model.

8. The network interface card testing method according to claim 4, characterized in that, The process of obtaining the code coverage and defect detection rate of the test case combination during operation, and inputting the code coverage and defect detection rate into a pre-built online learning model to adjust the test weights corresponding to the test case combination through the online learning model, and optimizing the test case combination through the adjusted test weights, includes: During the execution of the test case combination, corresponding test performance feedback data is collected, wherein the test performance feedback data includes the code coverage and the defect detection rate; The test performance feedback data is input into the online learning model to output the performance score corresponding to the test case combination; The performance scores of the test case combinations are sorted in descending order to obtain the corresponding ranking results. The test weights of the test case combinations are then adjusted based on the ranking results to optimize the test case combinations through the adjusted test weights.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the network interface card testing method as described in any one of claims 1 to 8 when executing the computer program.

Images