CN114650260B

CN114650260B - Network packet loss type identification method and device and electronic equipment

Info

Publication number: CN114650260B
Application number: CN202210247710.9A
Authority: CN
Inventors: 李丹; 高凯辉; 李政; 陆嘉敏; 李伟适; 韩靖雯
Original assignee: Tsinghua University; Beijing Dajia Internet Information Technology Co Ltd
Current assignee: Tsinghua University; Beijing Dajia Internet Information Technology Co Ltd
Priority date: 2022-03-14
Filing date: 2022-03-14
Publication date: 2024-03-12
Anticipated expiration: 2042-03-14
Also published as: CN114650260A

Abstract

The disclosure relates to a network packet loss type identification method, a network packet loss type identification model generation method, a device, an electronic device, a computer readable storage medium and a computer program product, when a packet loss signal in a network is detected, network characteristic parameters used for representing a network state in a preset time period before a target data packet is sent are acquired, the network characteristic parameters are input into a pre-trained network packet loss type identification model for identification, so that a packet loss type corresponding to the packet loss signal is obtained, wherein the network characteristic parameters are statistical characteristic data extracted after characteristic statistics is carried out on an acquired time sequence array used for representing the network state. Because the network packet loss type identification model in the embodiment is a machine learning model obtained by training based on network characteristic sample parameters collected under a real network environment, the machine learning model can learn the packet loss characteristics under different network states, so that the packet loss type in the network can be accurately identified based on the network characteristic parameters.

Description

Network packet loss type identification method and device and electronic equipment

Technical Field

The present disclosure relates to the field of communications technologies, and in particular, to a network packet loss type identification method, a network packet loss type identification model generating method, an apparatus, an electronic device, a computer readable storage medium, and a computer program product.

Background

In the field of communication technology, a server generally transmits data through a data packet, however, during the data transmission process, the data packet may be lost, that is, the data packet may be lost. At present, packet loss is generally classified into congestion packet loss and random packet loss (also referred to as erroneous packet loss). The congestion packet loss refers to packet loss caused by insufficient network bandwidth and overflow of a buffer, and the packet loss occurs in this case, which means that the network is already congested, that is, the network bandwidth is insufficient to meet the data sending speed of the sending end. The random packet loss refers to a packet loss caused by a link error in the network, for example, a random packet loss caused by a WiFi signal interference in the wireless network, a link layer transmission error, and the like, which does not mean that the network bandwidth is insufficient.

In the related art, when distinguishing the packet loss type, one way is to distinguish the packet loss type by analyzing the packet loss rule of the router. This scheme uses a BQM (Biased queue management ) mechanism, where marked packets are selectively discarded first when router buffers overflow. Another way is to distinguish the packet loss type by the rules of TCP-Veno. The network queuing situation is presumed according to the arrival time interval of the ACK (Acknowledge character, acknowledgement character) message, and if the network is estimated to be not queued and packet loss occurs, the network is considered to be packet loss caused by link error; if the network is estimated to have queues and packet loss occurs, the network is considered to be packet loss caused by congestion.

However, the first approach requires assistance from network routing equipment, is costly, and cannot be deployed on a large scale on-site. The second way is that the real network environment is complex, and their behavior depends on a large number of random external factors, so that it is difficult to analyze and describe some packet loss through several rules, which leads to easy misjudgment. Therefore, the existing method for judging the packet loss type is high in cost, and is difficult to accurately identify the packet loss type in different network states, so that erroneous judgment is easy to cause.

Disclosure of Invention

The disclosure provides a network packet loss type identification method, a network packet loss type identification model generation method, a device, an electronic device, a computer readable storage medium and a computer program product, so as to at least solve the problem that the packet loss type in different network states is difficult to accurately identify in the related art. The technical scheme of the present disclosure is as follows:

according to a first aspect of an embodiment of the present disclosure, there is provided a network packet loss type identification method, including:

when a packet loss signal in a network is detected, acquiring network characteristic parameters for representing a network state in a preset time period before a target data packet is sent, wherein the target data packet is a lost data packet corresponding to the packet loss signal, and the network characteristic parameters are statistical characteristic data extracted after characteristic statistics is carried out on an acquired time sequence array for representing the network state;

Inputting the network characteristic parameters into a pre-trained network packet loss type identification model for identification to obtain the packet loss type corresponding to the packet loss signal.

In one embodiment, the preset time period includes at least two round trip delays; the obtaining the network characteristic parameters used for representing the network state in a preset time period before sending the target data packet includes: acquiring a time sequence array representing the network state in each round trip delay before sending a target data packet according to at least two round trip delays, wherein the round trip delay is the time length from the time when a data packet is sent by a sending end to the time when the sending end receives an acknowledgement message from a receiving end, and the acknowledgement message is returned by the receiving end when the receiving end receives the data packet; extracting statistical characteristics in a time sequence array for representing the network state in each round trip delay; acquiring difference characteristics between at least two round trip delays according to the extracted statistical characteristics; and generating corresponding network characteristic parameters according to the extracted statistical characteristics and the difference characteristics.

In one embodiment, the acquiring the timing array characterizing the network state in each round trip delay before sending the target data packet includes: acquiring an acknowledgement message returned by an opposite terminal in each round trip delay before sending a target data packet, wherein the acknowledgement message is returned by the opposite terminal for the data packet sent by a local terminal, and the acknowledgement message comprises a first receiving time for the opposite terminal to receive the data packet; acquiring the sending time of the data packet sent by the local terminal and the second receiving time of the acknowledgement message received by the local terminal; acquiring a first time sequence array of unidirectional delay between the home terminal and the opposite terminal in each round trip delay according to the sending time of the home terminal for sending the data packet and the first receiving time of the opposite terminal for receiving the data packet; and acquiring a second time sequence array of the interval time of the acknowledgement message received by the local terminal according to the second receiving time of the acknowledgement message received by the local terminal in each round trip time delay.

In one embodiment, the extracting the statistical feature in the timing array characterizing the network state in each round trip delay includes: extracting a corresponding first statistical feature in a first time sequence array according to the first time sequence array of unidirectional delay between a home terminal and an opposite terminal in each round trip delay; extracting corresponding second statistical characteristics from a second time sequence array of the interval time of the acknowledgement message received by the local terminal in each round trip delay; and obtaining the statistical characteristics used for representing the network state in each round trip delay according to the extracted first statistical characteristics and the second statistical characteristics.

In one embodiment, the obtaining the difference feature between at least two round trip delays according to the extracted statistical feature includes: according to the first statistical characteristics corresponding to the first time sequence groups in each round trip time delay, calculating a first ratio between the first statistical characteristics corresponding to the first time sequence groups in one round trip time delay and the first statistical characteristics corresponding to the first time sequence groups in the other round trip time delay in two adjacent round trip time delays; calculating a second ratio between a second statistical feature corresponding to the second time sequence array in one round trip delay and a second statistical feature corresponding to the second time sequence array in the other round trip delay in two adjacent round trip delays according to the second statistical feature corresponding to the second time sequence array in each round trip delay; the first ratio and the second ratio are determined as a characteristic of a difference between at least two round trip delays.

In one embodiment, the network packet loss type identification model includes a plurality of network packet loss type classifiers; inputting the network characteristic parameters into a pre-trained network packet loss type identification model for identification to obtain the packet loss type corresponding to the packet loss signal, wherein the method comprises the following steps: inputting the network characteristic parameters into a pre-trained network packet loss type identification model for identification, and obtaining an identification result of each classifier in the network packet loss type identification model, wherein the identification result comprises any one of congestion packet loss and error packet loss; and acquiring the number of classifiers corresponding to each type of identification result, and taking the identification result with the largest number of classifiers as the packet loss type corresponding to the packet loss signal.

In one embodiment, after the obtaining the packet loss type corresponding to the packet loss signal, the method further includes: when the packet loss type corresponding to the packet loss signal is obtained as error packet loss, obtaining error packet loss rate in a set period; calculating a target sending rate according to the error packet loss rate and a rate corresponding to a network bottleneck link, wherein the network bottleneck link is a link with the highest bandwidth utilization rate between a sending end and a receiving end of the target data packet; and transmitting the subsequent data packets by adopting the target transmission rate.

In one embodiment, the calculating the target sending rate according to the error packet loss rate and the rate corresponding to the network bottleneck link includes: obtaining a difference value between the 1 and the error packet loss rate; and calculating the product between the reciprocal of the difference value and the corresponding speed of the network bottleneck link as the target sending speed.

In one embodiment, the method further comprises: when the packet loss type corresponding to the packet loss signal is detected to be the error packet loss in the set period, a corresponding sending window is added to transmit the data packet corresponding to the packet loss signal.

In one embodiment, after the obtaining the packet loss type corresponding to the packet loss signal, the method further includes: and when the packet loss type corresponding to the packet loss signal is obtained as the congestion packet loss, reducing the sending rate of the subsequent data packet.

According to a second aspect of the embodiments of the present disclosure, there is provided a method for generating a network packet loss type identification model, including:

acquiring a plurality of groups of network characteristic sample parameters, wherein the network characteristic sample parameters are statistical characteristic sample data extracted after characteristic statistics is carried out on an acquired time sequence sample array for representing a network state, each group of network characteristic sample parameters is provided with a corresponding network state label, and the network state labels comprise labels of network congestion and labels of network errors;

And training and generating a plurality of classifiers based on a plurality of groups of network characteristic sample parameters to obtain a network packet loss type identification model.

In one embodiment, the plurality of sets of network feature sample parameters includes a first network feature sample parameter set and a second network feature sample parameter set; the training based on a plurality of groups of network characteristic sample parameters to generate a plurality of classifiers to obtain a network packet loss type identification model comprises the following steps: training based on the first network characteristic sample parameter set to generate a plurality of classifiers to obtain a network packet loss type identification model to be evaluated; verifying the network packet loss type identification model to be evaluated based on the second network characteristic sample parameter set and the corresponding network state label to obtain the false identification rate of the network packet loss type identification model to be evaluated on the target packet loss type; and when the false recognition rate is smaller than a preset threshold value, determining the network packet loss type recognition model to be evaluated as the network packet loss type recognition model.

In one embodiment, the training based on the first network feature sample parameter set to generate a plurality of classifiers, to obtain a network packet loss type identification model to be evaluated includes: randomly extracting n groups of network characteristic sample parameters from the first network characteristic sample parameter set to obtain a training sample subset; k rounds of extraction are carried out on the first network characteristic sample parameter set, and k training sample subsets are obtained; and respectively training based on k training sample subsets to obtain k corresponding classifiers, and integrating the k classifiers to obtain a network packet loss type identification model to be evaluated.

In one embodiment, each set of network feature sample parameters in the first set of network feature sample parameters has a corresponding weight; training and generating a plurality of classifiers based on the first network characteristic sample parameter set to obtain a network packet loss type identification model to be evaluated, wherein the method comprises the following steps: performing iterative training based on the first network characteristic sample parameter set and the corresponding weight to obtain a classifier corresponding to the iterative training; when any group of network characteristic sample parameters are misclassified in the iterative training process, increasing the corresponding weight of the network characteristic sample parameters in the next iterative training process; and performing k rounds of iterative training on the first network characteristic sample parameter set to obtain k corresponding classifiers, and integrating the k classifiers to obtain a network packet loss type identification model to be evaluated.

According to a third aspect of the embodiments of the present disclosure, there is provided a network packet loss type identifying apparatus, including:

the network characteristic parameter acquisition module is configured to acquire network characteristic parameters used for representing a network state in a preset time period before a target data packet is transmitted when a packet loss signal in a network is detected, wherein the target data packet is a lost data packet corresponding to the packet loss signal, and the network characteristic parameters are statistical characteristic data extracted after characteristic statistics is carried out on an acquired time sequence array used for representing the network state;

The packet loss type identification module is configured to input the network characteristic parameters into a pre-trained network packet loss type identification model for identification, and the packet loss type corresponding to the packet loss signal is obtained.

In one embodiment, the preset time period includes at least two round trip delays; the network characteristic parameter acquisition module comprises: the system comprises a time sequence array acquisition unit, a data transmission unit and a data transmission unit, wherein the time sequence array acquisition unit is configured to execute time sequence array for representing network state in each round trip delay before a target data packet is transmitted according to at least two round trip delays, the round trip delays are the time length from the time when the data packet is transmitted from a transmitting end to the time when the transmitting end receives an acknowledgement message from a receiving end, and the acknowledgement message is returned when the receiving end receives the data packet; a statistical feature extraction unit configured to perform extraction of statistical features in a timing array characterizing a network state within each round trip delay; a difference feature acquisition unit configured to perform acquisition of a difference feature between at least two round trip delays based on the extracted statistical feature; and a network characteristic parameter generating unit configured to perform generation of a corresponding network characteristic parameter according to the extracted statistical characteristic and the difference characteristic.

In one embodiment, the timing array acquisition unit is configured to perform: acquiring an acknowledgement message returned by an opposite terminal in each round trip delay before sending a target data packet, wherein the acknowledgement message is returned by the opposite terminal for the data packet sent by a local terminal, and the acknowledgement message comprises a first receiving time for the opposite terminal to receive the data packet; acquiring the sending time of the data packet sent by the local terminal and the second receiving time of the acknowledgement message received by the local terminal; acquiring a first time sequence array of unidirectional delay between the home terminal and the opposite terminal in each round trip delay according to the sending time of the home terminal for sending the data packet and the first receiving time of the opposite terminal for receiving the data packet; and acquiring a second time sequence array of the interval time of the acknowledgement message received by the local terminal according to the second receiving time of the acknowledgement message received by the local terminal in each round trip time delay.

In one embodiment, the statistical feature extraction unit is configured to perform: extracting a corresponding first statistical feature in a first time sequence array according to the first time sequence array of unidirectional delay between a home terminal and an opposite terminal in each round trip delay; extracting corresponding second statistical characteristics from a second time sequence array of the interval time of the acknowledgement message received by the local terminal in each round trip delay; and obtaining the statistical characteristics used for representing the network state in each round trip delay according to the extracted first statistical characteristics and the second statistical characteristics.

In one embodiment, the difference feature acquisition unit is configured to perform: according to the first statistical characteristics corresponding to the first time sequence groups in each round trip time delay, calculating a first ratio between the first statistical characteristics corresponding to the first time sequence groups in one round trip time delay and the first statistical characteristics corresponding to the first time sequence groups in the other round trip time delay in two adjacent round trip time delays; calculating a second ratio between a second statistical feature corresponding to the second time sequence array in one round trip delay and a second statistical feature corresponding to the second time sequence array in the other round trip delay in two adjacent round trip delays according to the second statistical feature corresponding to the second time sequence array in each round trip delay; the first ratio and the second ratio are determined as a characteristic of a difference between at least two round trip delays.

In one embodiment, the network packet loss type identification model includes a plurality of network packet loss type classifiers; the packet loss type identification module is further configured to perform: inputting the network characteristic parameters into a pre-trained network packet loss type identification model for identification, and obtaining an identification result of each classifier in the network packet loss type identification model, wherein the identification result comprises any one of congestion packet loss and error packet loss; and acquiring the number of classifiers corresponding to each type of identification result, and taking the identification result with the largest number of classifiers as the packet loss type corresponding to the packet loss signal.

In one embodiment, the apparatus further comprises: the error packet loss rate acquisition module is configured to acquire an error packet loss rate in a set period when the packet loss type corresponding to the packet loss signal is obtained to be an error packet loss; the target sending rate calculating module is configured to calculate a target sending rate according to the error packet loss rate and a rate corresponding to a network bottleneck link, wherein the network bottleneck link is a link with the highest bandwidth utilization rate between a sending end and a receiving end of the target data packet; and the data transmission module is configured to transmit the subsequent data packets at the target sending rate.

In one embodiment, the target sending rate calculation module is configured to perform: obtaining a difference value between the 1 and the error packet loss rate; and calculating the product between the reciprocal of the difference value and the corresponding speed of the network bottleneck link as the target sending speed.

In one embodiment, the data transmission module is further configured to perform: when the packet loss type corresponding to the packet loss signal is detected to be the error packet loss in the set period, a corresponding sending window is added to transmit the data packet corresponding to the packet loss signal.

In one embodiment, the data transmission module is further configured to perform: and when the packet loss type corresponding to the packet loss signal is obtained as the congestion packet loss, reducing the sending rate of the subsequent data packet.

According to a fourth aspect of the embodiments of the present disclosure, there is provided a generating device of a network packet loss type identification model, including:

the system comprises a sample parameter acquisition module, a network state detection module and a network state detection module, wherein the sample parameter acquisition module is configured to execute acquisition of a plurality of groups of network characteristic sample parameters, the network characteristic sample parameters are statistical characteristic sample data extracted after characteristic statistics is carried out on an acquired time sequence sample array for representing a network state, each group of network characteristic sample parameters is provided with a corresponding network state label, and the network state label comprises a label of network congestion and a label of network error;

the model generation module is configured to perform training based on a plurality of groups of network characteristic sample parameters to generate a plurality of classifiers, and obtain a network packet loss type identification model.

In one embodiment, the plurality of sets of network feature sample parameters includes a first network feature sample parameter set and a second network feature sample parameter set; the model generation module comprises: the model training unit is configured to perform training based on the first network characteristic sample parameter set to generate a plurality of classifiers so as to obtain a network packet loss type identification model to be evaluated; the model verification unit is configured to perform verification on the network packet loss type identification model to be evaluated based on the second network characteristic sample parameter set and the corresponding network state label, so as to obtain the false identification rate of the network packet loss type identification model to be evaluated on the target packet loss type; and a model determining unit configured to perform determining the network packet loss type identification model to be evaluated as the network packet loss type identification model when the false identification rate is smaller than a preset threshold.

In one embodiment, the model training unit is further configured to perform: randomly extracting n groups of network characteristic sample parameters from the first network characteristic sample parameter set to obtain a training sample subset; k rounds of extraction are carried out on the first network characteristic sample parameter set, and k training sample subsets are obtained; and respectively training based on k training sample subsets to obtain k corresponding classifiers, and integrating the k classifiers to obtain a network packet loss type identification model to be evaluated.

In one embodiment, each set of network feature sample parameters in the first set of network feature sample parameters has a corresponding weight; the model training unit is further configured to perform: performing iterative training based on the first network characteristic sample parameter set and the corresponding weight to obtain a classifier corresponding to the iterative training; when any group of network characteristic sample parameters are misclassified in the iterative training process, increasing the corresponding weight of the network characteristic sample parameters in the next iterative training process; and performing k rounds of iterative training on the first network characteristic sample parameter set to obtain k corresponding classifiers, and integrating the k classifiers to obtain a network packet loss type identification model to be evaluated.

According to a fifth aspect of embodiments of the present disclosure, there is provided an electronic device, comprising: a processor; a memory for storing the processor-executable instructions; wherein the processor is configured to execute the instructions to implement the network packet loss type identification method according to the first aspect and the network packet loss type identification model generation method according to the second aspect.

According to a sixth aspect of embodiments of the present disclosure, there is provided a computer-readable storage medium, which when executed by a processor of an electronic device, causes the electronic device to perform the network packet loss type identification method according to the first aspect and the generation method of the network packet loss type identification model according to the second aspect.

According to a seventh aspect of embodiments of the present disclosure, there is provided a computer program product comprising instructions therein, which when executed by a processor of an electronic device, enable the electronic device to perform the network packet loss type identification method according to the first aspect above and the generation method of the network packet loss type identification model according to the second aspect above.

The technical scheme provided by the embodiment of the disclosure at least brings the following beneficial effects: when a packet loss signal in a network is detected, acquiring network characteristic parameters used for representing the network state in a preset time period before a target data packet is sent, inputting the network characteristic parameters into a pre-trained network packet loss type recognition model for recognition, and obtaining a packet loss type corresponding to the packet loss signal, wherein the network characteristic parameters are statistical characteristic data extracted after characteristic statistics is carried out on an acquired time sequence array used for representing the network state. Because the network packet loss type identification model in the embodiment is a machine learning model obtained by training based on network characteristic sample parameters collected under a real network environment, the machine learning model can learn the packet loss characteristics under different network states, so that the packet loss type in the network can be accurately identified based on the network characteristic parameters.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the disclosure and together with the description, serve to explain the principles of the disclosure and do not constitute an undue limitation on the disclosure.

Fig. 1 is an application environment diagram illustrating a network packet loss type identification method according to an exemplary embodiment.

Fig. 2 is a flow chart illustrating a network packet loss type identification method according to an exemplary embodiment.

Fig. 3 is a flow chart illustrating steps for acquiring network characteristic parameters according to an exemplary embodiment.

FIG. 4 is a flow chart illustrating the steps of acquiring a timing array according to an exemplary embodiment.

Fig. 5 is a flow chart illustrating the steps of extracting statistical features according to an exemplary embodiment.

Fig. 6 is a flow chart illustrating the steps of a model identifying packet loss type according to an exemplary embodiment.

Fig. 7 is a flowchart illustrating a network packet loss type identification method according to another exemplary embodiment.

Fig. 8 is a flowchart illustrating a method of generating a network packet loss type identification model according to an exemplary embodiment.

FIG. 9 is a flow chart illustrating model training steps according to an exemplary embodiment.

Fig. 10 is a block diagram illustrating a network packet loss type identification device according to an exemplary embodiment.

Fig. 11 is a block diagram illustrating a generation apparatus of a network packet loss type identification model according to an exemplary embodiment.

Fig. 12 is a block diagram of an electronic device, according to an example embodiment.

Detailed Description

In order to enable those skilled in the art to better understand the technical solutions of the present disclosure, the technical solutions of the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.

It should be noted that the terms "first," "second," and the like in the description and claims of the present disclosure and in the foregoing figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the disclosure described herein may be capable of operation in sequences other than those illustrated or described herein. The implementations described in the following exemplary examples are not representative of all implementations consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with some aspects of the present disclosure as detailed in the accompanying claims.

It should be further noted that, the user information (including, but not limited to, user equipment information, user personal information, etc.) and the data (including, but not limited to, data for presentation, analyzed data, etc.) related to the present disclosure are information and data authorized by the user or sufficiently authorized by each party.

The network packet loss type identification method provided by the disclosure can be applied to an application environment as shown in fig. 1. The server 110 performs data interaction with the terminal 120 through a network, where the network is generally based on a TCP (Transmission Control Protocol ) or qic (Quick UDP Internet Connection, UDP-based low latency internet transport layer protocol) protocol. The server 110 may be implemented as a stand-alone server or a server cluster including a plurality of servers. The terminal 120 may be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, internet of things devices, and portable wearable devices, and the internet of things devices may be smart speakers, smart televisions, smart air conditioners, smart car devices, etc. The portable wearable device may be a smart watch, smart bracelet, headset, or the like.

When data packet loss occurs, it is difficult for the conventional TCP or QUIC protocol to distinguish whether packet loss caused by link error or congestion occurs, which uniformly considers that congestion occurs when packet loss is encountered, and then reduces the transmission rate. However, in the case where no congestion actually occurs, this speed-down is not reasonable, resulting in a reduced link utilization, i.e. in the TCP and qic protocols having lower throughput on the wireless link than they would otherwise have been.

Based on this, the present embodiment proposes a network packet loss type identification method, and fig. 2 is a flowchart of a network packet loss type identification method according to an exemplary embodiment, and as shown in fig. 2, the method is used for the server of fig. 1 to illustrate, and includes the following steps.

In step S210, when a packet loss signal in the network is detected, network characteristic parameters for characterizing a network state in a preset period of time before the transmission of the target data packet are acquired.

The packet loss refers to a phenomenon that a data packet transmitted in a network is lost in a channel due to various reasons, and a packet loss signal is a signal capable of reflecting the loss of the data packet in the channel. Generally, based on the TCP protocol, the receiving end needs to return an ACK acknowledgement after receiving each data packet sent by the sending end. To prevent data loss, the TCP starts a retransmission timer, i.e., RTO (Retransmission Timeout, timeout retransmission mechanism), whenever the sender sends a packet, and if an ACK for the corresponding packet is received before the timer expires, the timer is revoked, and if an ACK for the corresponding packet is not received when the timer arrives, the packet is retransmitted and the timer is reset. Therefore, when the RTO of the transmitting end is detected to be timed out, the detection of the packet loss signal in the network is indicated. In addition, based on the TCP protocol, if the receiving end receives ACK repeated three times, it can be considered that a packet loss signal in the network is detected. In this embodiment, the server is a transmitting end of the data packet.

The target data packet is a data packet lost in the channel corresponding to the packet loss signal. The network characteristic parameter refers to a parameter capable of reflecting a network link state, and in this embodiment, the network characteristic parameter is statistical characteristic data extracted after characteristic statistics is performed on an acquired time sequence array for representing the network state, for example, a characteristic of counting transmission delay of a data packet, a characteristic of counting interval time of ACK, and the like. The preset time period is a preset time period for collecting network characteristic parameters.

In this embodiment, when the server detects a packet loss signal in the network, the sending time of the target data packet corresponding to the packet loss signal is determined, and the network characteristic parameter used for representing the network state in a preset time period before the sending of the target data packet is obtained. Specifically, when the RTO corresponding to a certain data packet is overtime, the data packet is a target data packet, so that the network characteristic parameters in a preset time period before the target data packet is sent are further acquired. For example, if the preset time period is 2S and the transmission time of the target data packet is K, the network characteristic parameters between (K-2) and K are obtained. Similarly, based on the TCP protocol, the corresponding target data packet may be determined according to the ACK repeated three times, and further, the network characteristic parameter in the preset time period may be obtained according to the time of transmitting the target data packet.

In step S220, the network characteristic parameters are input into a pre-trained network packet loss type recognition model for recognition, so as to obtain a packet loss type corresponding to the packet loss signal.

The network packet loss type recognition model is a machine learning model which is obtained based on training of network characteristic sample parameters collected in a real network environment. Specifically, after the network characteristic parameters are obtained based on the steps, the network characteristic parameters are input into a pre-trained network packet loss type identification model for identification, so that the packet loss type corresponding to the packet loss signal output by the model is obtained. Wherein, the packet loss type includes any one of erroneous packet loss and congestion packet loss.

In the network packet loss type identification method, when a packet loss signal in a network is detected, network characteristic parameters used for representing a network state in a preset time period before a target data packet is sent are acquired, and the network characteristic parameters are input into a pre-trained network packet loss type identification model for identification, so that a packet loss type corresponding to the packet loss signal is obtained, wherein the network characteristic parameters are statistical characteristic data extracted after characteristic statistics is carried out on an acquired time sequence array used for representing the network state. Because the network packet loss type identification model in the embodiment is a machine learning model obtained by training based on network characteristic sample parameters collected under a real network environment, the machine learning model can learn the packet loss characteristics under different network states, so that the packet loss type in the network can be accurately identified based on the network characteristic parameters.

In an exemplary embodiment, the preset Time period includes at least two Round Trip delays (RTTs), i.e., round-Trip times. The round trip delay refers to a period from when a data packet is sent from a sending end to when the sending end receives an acknowledgement message from a receiving end (acknowledgement message is sent immediately after the receiving end receives the data packet), so that the length of each round trip delay is different based on different network states. As shown in fig. 3, in step S210, the network characteristic parameters for characterizing the network state in a preset period of time before the target data packet is sent are obtained, which may be specifically implemented by the following steps:

in step S310, a timing array characterizing the network state in each round trip delay before sending the target packet is obtained according to at least two round trip delays.

The time sequence array is a data sequence in a real network environment acquired by a transmitting end, each element in the data sequence is orderly arranged based on the sequence of the generation time, and the data sequence can reflect the network state. For example, the timing array may be a sequence of ACK inter-arrival times per round trip delay, a sequence of one-way delays for packets per round trip delay, etc. Specifically, according to at least two round trip delays before sending the target data packet, a time sequence array characterizing the network state in each round trip delay is respectively obtained.

In step S320, statistical features in the timing array characterizing the network state within each round trip delay are extracted.

The statistical features are feature data which is extracted after the statistical analysis is performed on the acquired time sequence array and can further reflect the network state. Specifically, the statistical feature can reflect the distribution situation among individuals of the population, and for example, the statistical feature can be a maximum value, a minimum value, a median, a variance and the like in the time sequence array. The model can accurately identify whether the congestion packet is lost due to insufficient network bandwidth or the error packet is lost due to link errors in the network through the characteristic data.

Because one time sequence array is a data sequence in the network environment collected in one round trip delay, based on the time sequence array corresponding to each round trip delay, the statistical characteristics in each time sequence array can be extracted, so that the statistical characteristics of the time sequence array corresponding to each round trip delay can be obtained. For example, for two round trip delays RTT1 and RTT2 before the target packet is sent, a timing array A1 is collected in RTT1, and a timing array A2 is collected in RTT2, then the statistical feature X1 of the timing array A1 and the statistical feature X2 of the timing array A2 are extracted respectively.

In step S330, a difference characteristic between at least two round trip delays is obtained according to the extracted statistical characteristic.

The difference feature refers to feature data capable of reflecting the network state difference between two round trip delays. Specifically, the difference feature may be a ratio between statistical features of the timing arrays corresponding to the two adjacent round trip delays, respectively. For example, it may be a ratio between the statistical characteristic X1 of the timing array A1 and the statistical characteristic X2 of the timing array A2, where A1 is the timing array collected in RTT1, A2 is the timing array collected in RTT2, and RTT1 is adjacent to RTT2 and earlier than RTT2, that is, RTT2 is closer to the time of sending the target packet.

In step S340, corresponding network feature parameters are generated according to the extracted statistical features and the difference features.

Specifically, based on the statistical features and the difference features extracted in the steps, network feature parameters which are used for representing the network state and can identify the packet loss type by the model are obtained. For example, if the preset time period is two round trip delays, for the two round trip delays RTT1 and RTT2 before the target data packet is sent, a time sequence array A1 is collected in RTT1, and a time sequence array A2 is collected in RTT2, then the statistical feature X1 of the time sequence array A1 and the statistical feature X2 of the time sequence array A2 are extracted, and the difference feature X1/X2 between RTT1 and RTT2 is obtained, and then X1, X2 and X1/X2 are corresponding network feature parameters.

In the above embodiment, according to at least two round trip delays, a timing sequence array characterizing a network state in each round trip delay before sending a target data packet is obtained, statistical features in the timing sequence array characterizing the network state in each round trip delay are extracted, and according to the extracted statistical features, difference features between at least two round trip delays are obtained, and further according to the extracted statistical features and the difference features, corresponding network feature parameters are generated. Because the network characteristic parameters used for inputting the model to identify the packet loss type in the embodiment are based on the statistical characteristics of the time sequence arrays collected by each round trip delay before the target data packet is sent and the difference characteristics among the time sequence arrays, the state of the network during each round trip delay and the difference of the network among different round trip delays can be reflected, and the accuracy of identifying the packet loss type by the model can be improved.

In an exemplary embodiment, as shown in fig. 4, in step S310, acquiring a timing array characterizing a network state in each round trip delay before sending a target data packet may specifically include:

in step S410, an acknowledgement message returned from the peer in each round trip delay before the target packet is sent is obtained.

The method comprises the steps that an opposite terminal returns an acknowledgement message aiming at a data packet sent by the local terminal, wherein the acknowledgement message comprises first receiving time of the opposite terminal for receiving the data packet. Specifically, the home terminal is a sending terminal of the data packet, and in this embodiment, the sending terminal of the data packet is a server. The opposite end is the receiving end of the data packet. Based on TCP protocol, the receiving end needs to return ACK confirmation, namely acknowledgement message after receiving every time the sending end sends a data packet. Specifically, the acknowledgement message includes a first receiving time of the opposite-end received data packet, that is, a receiving time of the receiving-end received data packet.

In this embodiment, when the server detects the packet loss signal, an acknowledgement packet returned by the opposite end for the data packet sent by the home terminal in each round trip delay before the target data packet is sent is obtained, and the receiving time of the receiving end to the data packet is recorded in each acknowledgement packet.

In step S420, a sending time of the sending data packet of the local terminal and a second receiving time of the receiving acknowledgement packet of the local terminal are obtained.

Specifically, because the acknowledgement message is received by the local end sent by the opposite end, based on the acknowledgement message, the server may further obtain a second receiving time for receiving the corresponding acknowledgement message. And because the data packet corresponding to the acknowledgement message is sent by the local end, the server can also obtain the sending time of the data packet corresponding to the acknowledgement message.

In step S430, a first timing array of one-way delay between the home terminal and the peer terminal in each round trip delay is obtained according to the sending time of the home terminal sending the data packet and the first receiving time of the peer terminal receiving the data packet.

The unidirectional delay refers to a time period between when the local end sends a certain data packet to the opposite end receives the corresponding data packet, that is, a difference between a first receiving time when the opposite end receives the certain data packet and a sending time when the local end sends the data packet, that is, a transmission duration of the data packet in the network. Specifically, based on the obtained sending time of the sending data packet of the home terminal and the first receiving time of the receiving data packet of the opposite terminal, the transmission time length of the data packet corresponding to the acknowledgement message received by the home terminal in each round trip time delay can be obtained. For example, if the acknowledgement packet received by the home terminal corresponds to the data packets d1, d2, d3 and d4 within one round trip delay, and accordingly, the transmission durations of the data packets d1, d2, d3 and d4 are respectively T1, T2, T3 and T4, the first timing array of the unidirectional delay between the home terminal and the opposite terminal within the round trip delay is (T1, T2, T3, T4).

In step S440, according to the second receiving time of the acknowledgement message received by the home terminal in each round trip delay, a second timing array of the interval time of the acknowledgement message received by the home terminal is obtained.

The interval time of receiving the acknowledgement messages refers to the interval time between receiving two adjacent acknowledgement messages. Because the server has acquired the second receiving time of the corresponding acknowledgement message, the second time sequence array of the interval time of the local receiving acknowledgement message in each round trip delay can be obtained based on the second receiving time of the local receiving acknowledgement message in each round trip delay.

For example, if there are m1, m2, m3 and m4 acknowledgement messages received by the home terminal within a round trip delay, and accordingly, the receiving times of m1, m2, m3 and m4 are t1, t2, t3 and t4 respectively, the interval between receiving m1 and m2 is (t 2-t 1), the interval between receiving m2 and m3 is (t 3-t 2), the interval between receiving m3 and m4 is (t 4-t 3), and the second timing array of the interval between receiving acknowledgement messages by the home terminal within the round trip delay is ((t 2-t 1), (t 3-t 2), (t 4-t 3)).

In the above embodiment, by acquiring the acknowledgement packet returned by the opposite end for the data packet sent by the home terminal in each round trip delay before the target data packet is sent, and acquiring the sending time of the data packet sent by the home terminal and the second receiving time of the acknowledgement packet received by the home terminal, the first timing sequence array of unidirectional delay between the home terminal and the opposite end in each round trip delay is acquired according to the sending time of the data packet sent by the home terminal and the first receiving time of the opposite end, and the second timing sequence array of the interval time of the acknowledgement packet received by the home terminal is acquired according to the second receiving time of the acknowledgement packet received by the home terminal in each round trip delay. Because the first time sequence array and the second time sequence array can be obtained based on the analysis of the acknowledgement message received by the receiving end, the flexibility and the efficiency of data acquisition are improved.

In an exemplary embodiment, the statistical features may include maximum, minimum, median absolute deviation, variance, skewness factor, kurtosis factor, first order standard deviation, second order standard deviation, and the like. Then, as shown in fig. 5, in step S320, the extracting statistical features in the timing array characterizing the network state in each round trip delay may specifically include:

in step S510, according to the first timing sequence array of the unidirectional delay between the home terminal and the opposite terminal in each round trip delay, a corresponding first statistical feature in the first timing sequence array is extracted.

The first statistical characteristic is obtained by statistical analysis based on a first time sequence group of unidirectional delay between the home terminal and the opposite terminal. Because one first time sequence group is a data sequence of unidirectional delay between the local end and the opposite end collected in one round trip time delay, a corresponding first time sequence group exists for each round trip time delay, and for each first time sequence group, the corresponding first statistical characteristics such as maximum value, minimum value, median absolute deviation, variance, skewness coefficient, kurtosis coefficient, first-order standard deviation, second-order standard deviation and the like can be extracted.

In step S520, according to the second timing sequence array of the interval time of the acknowledgement message received by the local end in each round trip delay, a corresponding second statistical feature in the second timing sequence array is extracted.

The second statistical feature is obtained by performing statistical analysis on a second time sequence array based on the interval time of the receiving acknowledgement message of the local terminal. Because one second time sequence array is a data sequence of the interval time of the local end receiving the acknowledgement message, which is collected in one round trip delay, a corresponding second time sequence array exists for each round trip delay, and for each second time sequence array, the corresponding second statistical characteristics of maximum value, minimum value, median absolute deviation, variance, skewness coefficient, kurtosis coefficient, first-order standard deviation, second-order standard deviation and the like can be extracted.

In step S530, according to the extracted first statistical feature and second statistical feature, a statistical feature for characterizing a network state in each round trip delay is obtained.

Specifically, for a first time sequence array corresponding to a round trip time delay, extracting a corresponding first statistical feature according to the steps, and for a second time sequence array corresponding to the round trip time delay, extracting a corresponding second statistical feature according to the steps, wherein the corresponding first statistical feature and second statistical feature are statistical features used for representing network states in the round trip time delay. Based on this, statistics can be derived for each round trip delay that characterizes the network state.

In the above embodiment, according to the first timing sequence array of the unidirectional delay between the home terminal and the opposite terminal in each round trip delay, the corresponding first statistical feature in the first timing sequence array is extracted, according to the second timing sequence array of the interval time of the acknowledgement message received by the home terminal in each round trip delay, the corresponding second statistical feature in the second timing sequence array is extracted, and according to the extracted first statistical feature and second statistical feature, the statistical feature used for representing the network state in each round trip delay is obtained. Because the statistical features are extracted based on the first time sequence array and the second time sequence array corresponding to each round trip time delay respectively, the extracted statistical features can more comprehensively reflect the state of the network.

In an exemplary embodiment, in step S330, obtaining a difference feature between at least two round trip delays according to the extracted statistical feature may specifically include: according to the first statistical characteristics corresponding to the first time sequence groups in each round trip time delay, calculating a first ratio between the first statistical characteristics corresponding to the first time sequence groups in one round trip time delay and the first statistical characteristics corresponding to the first time sequence groups in the other round trip time delay in two adjacent round trip time delays; calculating a second ratio between a second statistical feature corresponding to the second time sequence array in one round trip delay and a second statistical feature corresponding to the second time sequence array in the other round trip delay in two adjacent round trip delays according to the second statistical feature corresponding to the second time sequence array in each round trip delay; the first ratio and the second ratio are determined as a characteristic of a difference between at least two round trip delays.

Specifically, based on the above embodiment, a first statistical feature of the first timing array corresponding to each round trip delay may be obtained, and a second statistical feature of the second timing array corresponding to each round trip delay may be obtained. Therefore, based on the first time sequence array and the second time sequence array corresponding to two adjacent round trip delays respectively, a first ratio between the first statistical characteristic of one first time sequence array and the first statistical characteristic of the other first time sequence array can be calculated, and a second ratio between the second statistical characteristic of one second time sequence array and the second statistical characteristic of the other second time sequence array can be calculated, wherein the first ratio and the second ratio are difference characteristics between two adjacent round trip delays.

For example, if the predetermined time period is two round trip delays, the difference features are the average ratio between the two round trip delays, the ratio of the maximum value, the ratio of the minimum value, the ratio of the maximum value to the minimum value, and the ratio of the minimum value to the maximum value. The obtaining the difference feature specifically includes: and obtaining a first average number corresponding to the first time sequence group in each round trip time delay and a second average number corresponding to the second time sequence group according to the extracted statistical characteristics, calculating the ratio between the first average number corresponding to the first time sequence group in the first round trip time delay and the first average number corresponding to the first time sequence group in the second round trip time delay, the ratio between the maximum value corresponding to the first time sequence group in the first round trip time delay and the maximum value corresponding to the first time sequence group in the second round trip time delay, the ratio between the minimum value corresponding to the first time sequence group in the first round trip time delay and the minimum value corresponding to the first time sequence group in the second round trip time delay, and calculating the ratio between the minimum value corresponding to the first time sequence group in the first round trip time delay and the maximum value corresponding to the first time sequence group in the second round trip time delay, wherein the obtained ratio is the first ratio.

Similarly, a ratio between a second average number corresponding to a second timing array in the first round-trip delay and a second average number corresponding to a second timing array in the second round-trip delay, a ratio between a maximum value corresponding to a second timing array in the first round-trip delay and a maximum value corresponding to a second timing array in the second round-trip delay, a ratio between a minimum value corresponding to a second timing array in the first round-trip delay and a minimum value corresponding to a second timing array in the second round-trip delay, a ratio between a maximum value corresponding to a second timing array in the first round-trip delay and a minimum value corresponding to a second timing array in the second round-trip delay, and a ratio between a minimum value corresponding to a second timing array in the first round-trip delay and a maximum value corresponding to a second timing array in the second round-trip delay may be calculated.

It should be noted that, in the above ratio calculation process, the first round trip delay as a numerator should occur earlier than the second round trip delay as a denominator, that is, the second round trip delay is closer to the time of sending the target packet, and the first round trip delay is farther from the time of sending the target packet.

In the above embodiment, based on the statistical feature corresponding to each round trip delay of two adjacent round trip delays, the ratio between the statistical feature corresponding to one round trip delay which occurs earlier and the statistical feature corresponding to the other round trip delay which occurs later is calculated and used as the difference feature between the two round trip delays, so that the network state difference between the two round trip delays can be reflected more comprehensively, and further the accuracy of identifying the packet loss type by the model is improved.

In an exemplary embodiment, the network packet loss type identification model includes a plurality of network packet loss type classifiers. Then, as shown in fig. 6, in step S220, the network characteristic parameter is input into a pre-trained network packet loss type identification model to identify, so as to obtain a packet loss type corresponding to the packet loss signal, which may specifically include:

in step S222, the network characteristic parameters are input into a pre-trained network packet loss type recognition model for recognition, so as to obtain the recognition result of each classifier in the network packet loss type recognition model.

Wherein the identification result includes any one of congestion packet loss and error packet loss. Specifically, since the network packet loss type recognition model in the embodiment is integrated by a plurality of classifiers, when a packet loss type specifically corresponding to a packet loss signal is recognized, the network packet loss type recognition model can be comprehensively determined based on the recognition result of each classifier.

Specifically, network characteristic parameters corresponding to the packet loss signals are input into the network packet loss type recognition model obtained through training for recognition, so that the recognition result of each classifier in the network packet loss type recognition model is obtained.

In step S224, the number of classifiers corresponding to each type of identification result is obtained, and the identification result with the largest number of classifiers is used as the packet loss type corresponding to the packet loss signal.

Specifically, based on the obtained identification result of each classifier, counting the number of the classifiers corresponding to each identification result, and taking the identification result with the largest number as the packet loss type corresponding to the packet loss signal.

In the above embodiment, the network characteristic parameters are input into the pre-trained network packet loss type recognition model to perform recognition, so as to obtain the recognition result of each classifier in the network packet loss type recognition model, obtain the number of the classifiers corresponding to each recognition result, and take the recognition result with the largest number as the packet loss type corresponding to the packet loss signal, thereby being beneficial to improving the classification precision of the model.

In an exemplary embodiment, link error packet loss (random packet loss) in the wireless network is unavoidable due to physical occlusion and channel interference. While the conventional TCP and quitc protocols have no mechanism to accurately distinguish between erroneous packet loss due to link errors and congestion packet loss due to network congestion, these protocols uniformly consider congestion to occur when encountering packet loss, and then reduce the transmission rate. However, in the case where no congestion actually occurs, this throttling is not reasonable, resulting in low link utilization, with the result that the throughput of the TCP and qic protocols over the wireless link is lower than they would have been, resulting in impaired end-to-end transmission performance.

Based on this, in this embodiment, as shown in fig. 7, after obtaining the packet loss type corresponding to the packet loss signal, the network packet loss type identification method may further include the following steps:

in step S230, when the packet loss type corresponding to the packet loss signal is obtained as an erroneous packet loss, an erroneous packet loss rate in the set period is obtained.

Wherein, the error packet loss rate refers to that the packet loss type identified by the network packet loss type identification model is error in a period of time Packet lossA ratio between the number of total packet losses in the period of time. The set period refers to a preset time period for counting the packet loss rate, and specifically, the set period may be a round trip delay, i.e. an RTT.

Specifically, when the model outputs that the packet loss type corresponding to a certain packet loss signal is an error packet loss, the trigger server can start counting the number of error packet loss and the number of congestion packet loss in the next RTT based on the error packet loss type. It should be noted that the packet loss signal identified as the erroneous packet loss is also counted in the statistics data of the RTT. And further calculating the ratio of the number of the packet losses in the RTT, which is the number of the error packet losses and the total packet loss (namely the sum of the number of the error packet losses and the number of the congestion packet losses in the RTT), so as to obtain the error packet loss rate in the RTT.

In step S240, a target sending rate is calculated according to the error packet loss rate and the rate corresponding to the network bottleneck link.

The network bottleneck link refers to a link with the highest bandwidth utilization rate between the sending end and the receiving end of the target data packet, and generally, the higher the bandwidth utilization rate is, the worse the corresponding network speed is, so the network bottleneck link is also the link with the worst rate between the corresponding sending end and the receiving end. For example, if there are a link a, a link B, and a link C between the transmitting end and the receiving end, where the transmission rate of the link a is a bit of information transmitted per second, the transmission rate of the link B is B bit of information transmitted per second, and the transmission rate of the link C is C bit of information transmitted per second, the link corresponding to the minimum value of the rates a, B, and C is a network bottleneck link. For example, when a is minimum, the link a corresponding to the transmission rate a is a network bottleneck link.

In particular, the rate corresponding to the network bottleneck link may be directly obtained based on the current network. The target sending rate is a rate for transmitting the subsequent data packet, which is determined based on the error packet loss rate and a rate corresponding to the network bottleneck link. The error packet loss is caused by link error, and the target sending rate obtained based on the calculation is used for compensating the throughput reduction caused by the error packet loss, so that the link utilization rate is improved, and the link utilization rate can be saturated as much as possible.

In step S250, the subsequent data packet is transmitted at the target transmission rate.

Specifically, the adjusted target sending rate is adopted to transmit the subsequent data packets, so that the end-to-end transmission performance is improved when the packet is lost in error.

In the above embodiment, when the packet loss type corresponding to the packet loss signal is an error packet loss, the error packet loss rate in the set period is obtained, the target sending rate is calculated according to the error packet loss rate and the rate corresponding to the network bottleneck link, and the subsequent data packet is transmitted by adopting the target sending rate. Because the target sending rate is determined based on the error packet loss rate and the rate corresponding to the network bottleneck link, the end-to-end transmission performance can be improved.

In an exemplary embodiment, in step S240, the calculating the target sending rate according to the error packet loss rate and the rate corresponding to the network bottleneck link may specifically include: acquiring a difference value between the 1 and the error packet loss rate; calculating the inverse of the differenceThe product of the number and the rate corresponding to the network bottleneck link is used as the target sending rate.

Specifically, taking fig. 1 as an example, if a certain probability of erroneous packet loss occurs between the server 110 and the terminal 120, for example, if the erroneous packet loss rate is E, and at this time, the rate corresponding to the bottleneck link between the server 110 and the network is U, the receiving rate of the terminal 120 is u× (1-E) at most. Therefore, if it is desired to bring the end-to-end throughput close to 100% while erroneous packet loss occurs, the transmission rate of the server 110 needs to be increased to u× (1/(1-E)), i.e., the target transmission rate. According to the embodiment, when the error packet loss is detected, the target sending rate of the data packet to be sent subsequently is increased according to the error packet loss rate and the rate corresponding to the network bottleneck link, so that the problem of throughput reduction caused by the error packet loss can be solved, the link utilization rate is increased, the link utilization rate is saturated as much as possible, and the end-to-end transmission performance is improved.

In an exemplary embodiment, due to the above-mentioned manner of adjusting the target sending rate based on the error packet loss rate, the sending end needs to count the number of error packet loss events occurring within a period of time (i.e. a set period, such as one RTT), so as to evaluate the error packet loss rate of the link. However, during the evaluation period, the transmitting end has not obtained an accurate link packet loss rate, and thus cannot adjust the target transmission rate. If the evaluation period is too short, the erroneous packet loss rate evaluation will be inaccurate, and if it is too long, the utilization ratio will be underloaded in the period. In addition, while compensating for the link utilization, the overload of the link utilization cannot be caused, otherwise, congestion and packet loss of the network can be caused.

Based on this, when it is detected that the packet loss type corresponding to the packet loss signal is an erroneous packet loss in the evaluation period, the throughput can be compensated by increasing the corresponding transmission window to transmit the data packet corresponding to the packet loss signal. Specifically, during the evaluation period, every time an erroneous packet loss is detected, 1 additional transmission window is added to compensate the throughput, so that the end-to-end throughput is improved. And after the evaluation period is over, adjusting the target sending rate based on the embodiment based on the error packet loss rate E obtained in the evaluation period. When the link packet loss rate is not obtained, the throughput is compensated in a mode of linearly increasing the corresponding transmission window based on the detected error packet loss, so that the transmission rate is improved, and the end-to-end transmission performance is improved when the error packet loss is facilitated.

In an exemplary embodiment, when the model outputs that a packet loss type corresponding to a certain packet loss signal is a congestion packet loss, the sending rate of the subsequent data packet is reduced, so that the end-to-end transmission performance is improved.

In an exemplary embodiment, as shown in fig. 8, the embodiment of the present disclosure further provides a method for generating a network packet loss type identification model, which specifically includes the following steps:

in step S810, a plurality of sets of network characteristic sample parameters are acquired.

The network characteristic sample parameters are statistical characteristic sample data extracted after characteristic statistics is carried out on an acquired time sequence sample array for representing the network state, each group of network characteristic sample parameters is provided with a corresponding network state label, and the network state labels comprise any one of labels of network congestion and labels of network errors. Specifically, the network characteristic sample parameters have the same data content as the network characteristic parameters except for the corresponding network state labels, and this embodiment will not be described in detail.

Specifically, the network characteristic sample parameters are collected in a real network environment, and the real network environment refers to the network environment and conditions covered by the collection process of the sample data. In this embodiment, the collection of sample data encompasses as much of the various network environments and conditions as possible. In particular, the generation of sample data may be considered based on all a priori uncertainties of the network topology, user behavior and protocols. For example, specific network environments may refer to the relevant parameters of table 1 below:

In addition, because of the randomness of error packet loss and congestion packet loss, if the control is not performed in the process of collecting the sample data, an accurate network state label cannot be marked for the collected sample data. Thus, the acquisition conditions can be controlled during the acquisition of the sample data. For example, the link error packet loss rate is set to 0, burst (burst) is manufactured at the transmitting end, so that congestion packet loss is caused, namely, the network state label corresponding to the sample data collected under the condition is the label of network congestion; or setting the link error packet loss rate to be more than 0, and controlling the link utilization rate to be not more than 60% at the transmitting end, so that error packet loss is caused, namely, the network state label corresponding to the sample data collected under the condition is a label of network error. Based on the method, the accurate network state label can be marked for the acquired sample data, so that a plurality of groups of network characteristic sample parameters covering various real network environments are obtained.

In step S820, a plurality of classifiers are generated based on the training of the plurality of sets of network feature sample parameters, so as to obtain a network packet loss type recognition model.

Wherein the classifier is a classification function or classification model for classifying the data. In this embodiment, learning and training are performed based on the collected several sets of network feature sample parameters, so as to obtain a plurality of corresponding classification functions or classification models, and the network packet loss type recognition model is obtained by integrating the plurality of classification functions or classification models.

In the above embodiment, a plurality of classifiers are generated by collecting a plurality of sets of network feature sample parameters in a real network environment and training based on the plurality of sets of network feature sample parameters, and a corresponding network packet loss type identification model is obtained by integrating the plurality of classifiers.

In an exemplary embodiment, to further improve accuracy of the network packet loss type identification model, the acquired sets of network feature sample parameters may be divided into a first network feature sample parameter set and a second network feature sample parameter set. Then, as shown in fig. 9, in step S820, a plurality of classifiers are generated based on training of a plurality of sets of network feature sample parameters to obtain a network packet loss type identification model, which may specifically include:

in step S822, a plurality of classifiers are generated based on the training of the first network feature sample parameter set, so as to obtain a network packet loss type identification model to be evaluated.

The network packet loss type identification model to be evaluated is an integrated model of a plurality of classifiers obtained based on training of the first network characteristic sample parameter set. In particular, the training method may employ a supervised learning algorithm, such as decision trees, random forests, or deep learning neural networks, etc. The first set of network feature sample parameters may include a plurality of sets of network feature sample parameters, which are training data for training the model.

Specifically, in an exemplary embodiment, taking model training as an example of decision trees, a specific training process may include: randomly extracting n groups of network characteristic sample parameters from the first network characteristic sample parameter set to obtain a training sample subset; k rounds of extraction are carried out on the first network characteristic sample parameter set, and k training sample subsets are obtained; and respectively training based on k training sample subsets to obtain k corresponding classifiers, and integrating the k classifiers to obtain a network packet loss type identification model to be evaluated.

The first network characteristic sample parameter set is subjected to k rounds of extraction, and n groups of network characteristic sample parameters are randomly extracted as a training sample subset in each round, so that k training sample subsets are obtained. For each training sample subset, a corresponding classifier can be obtained through training, then k corresponding classifiers can be obtained through training k training sample subsets respectively, and finally the k classifiers are integrated to obtain the network packet loss type identification model to be evaluated. In this embodiment, the K classifiers are obtained by training in a manner of randomly extracting the K training sample subsets, and then the K classifiers are integrated to obtain the network packet loss type identification model to be evaluated.

In an exemplary embodiment, another decision tree based training process may further include: performing iterative training based on the first network characteristic sample parameter set and the corresponding weight to obtain a classifier corresponding to the iterative training; when any group of network characteristic sample parameters are classified by mistake in the iterative training process, the corresponding weight of the network characteristic sample parameters in the next round of iterative training process is increased; and performing k rounds of iterative training on the first network characteristic sample parameter set to obtain k corresponding classifiers, and integrating the k classifiers to obtain a network packet loss type identification model to be evaluated.

Specifically, each set of network feature sample parameters in the first network feature sample parameter set has a corresponding weight, respectively. And carrying out iterative training through the first network characteristic sample parameter set and the corresponding weight, wherein in the iterative training process, when any group of network characteristic sample parameters are misclassified, the weight corresponding to the group of network characteristic sample parameters in the next iterative training process is correspondingly increased, so that the misclassification of the group of network characteristic sample parameters is enhanced before, and the accuracy of the iteration of the round is improved. In this embodiment, k rounds of iterative training are performed on the first network feature sample parameter set, weights corresponding to the incorrectly classified network feature sample parameters are added after each round of iteration based on the above manner, so as to obtain k corresponding classifiers, and finally the k classifiers are weighted and fused based on Boosting (lifting algorithm) so as to obtain the network packet loss type identification model to be evaluated. The final network packet loss type identification model to be evaluated in the embodiment is obtained by weighting and fusing k classifiers based on Boosting, so that the model has higher identification accuracy.

In step S824, based on the second network feature sample parameter set and the corresponding network state label, the network packet loss type identification model to be evaluated is verified, so as to obtain the false identification rate of the network packet loss type identification model to be evaluated on the target packet loss type.

Wherein the second set of network characteristic sample parameters may also comprise a plurality of sets of network characteristic sample parameters, which are verification data for verifying the model.

Since the probability of mistaking the congestion packet loss for the erroneous packet loss (Err 1) and the probability of mistaking the erroneous packet loss for the congestion packet loss (Err 2) are different in the evaluation model. This is because when Err1 is high it causes congestion to occur and then it does not slow down, thereby compromising the TCP friendliness. Therefore, in evaluating the performance of the model, in addition to comprehensively considering the classification accuracy and recall (precision and recall), and indexes such as TCP friendliness, model classification time, memory consumption, and the like, the probability Err1 of congestion packet loss misclassification is not too high in order to maintain good use of TCP friendliness and bandwidth. If the probability Err1 of congestion packet loss being misclassified remains close to zero, the model has better TCP friendliness. In the wireless network, the higher Err1 is, the lower Err2 is, and the higher the gain obtained by the transmitting end is. Based on this, in order not to lose TCP friendliness, the error recognition rate of the target packet loss type in this embodiment is Err1, that is, the probability of mistaking the congestion packet loss for the error packet loss.

Wherein, the probability Err1 of mistaking the congestion packet loss for the error packet loss refers to the percentage of the number of the congestion packet loss mistaking the error packet loss and the total number of the samples in the second network characteristic sample parameter set, and the probability Err2 of mistaking the error packet loss for the congestion packet loss refers to the percentage of the number of the congestion packet loss mistaking the error packet loss and the total number of the samples in the second network characteristic sample parameter set.

Specifically, each group of network characteristic sample parameters in the second network characteristic sample parameter set is respectively input into a network packet loss type identification model to be evaluated, and a classification result of each group of network characteristic sample parameters output by the model is obtained. Based on the classification result of each group of network characteristic sample parameters and the corresponding network state label, the probability Err1 of mistaking congestion packet loss for error packet loss can be obtained through statistical calculation, and the error recognition rate of the network packet loss type recognition model to be evaluated on the target packet loss type can be obtained.

In step S826, when the false recognition rate is lower than the preset threshold, the network packet loss type recognition model to be evaluated is determined as the network packet loss type recognition model.

The preset threshold value can be reasonably set based on an actual application scene. For example, in a wireless network, the preset threshold may be set to 0.18.

Specifically, based on the obtained false recognition rate of the network packet loss type recognition model to be evaluated on the target packet loss type, when the false recognition rate is lower than a preset threshold, the verification of the network packet loss type recognition model to be evaluated is passed, so that the network packet loss type recognition model to be evaluated passing the verification is determined as the network packet loss type recognition model. And when the false recognition rate is higher than a preset threshold value, the false recognition rate indicates that the verification of the network packet loss type recognition model to be evaluated is not passed, the model parameters can be adjusted based on the false recognition rate, the training step is returned to carry out continuous training on the network packet loss type recognition model to be evaluated, the trained model is further verified based on the step, and the verification is passed only when the false recognition rate is lower than the preset threshold value.

In the above embodiment, training is performed based on the first network feature sample parameter set to generate a plurality of classifiers, so as to obtain a network packet loss type identification model to be evaluated, and verification is performed on the network packet loss type identification model to be evaluated based on the second network feature sample parameter set and the corresponding network state label, so as to obtain a false identification rate of the network packet loss type identification model to be evaluated on the target packet loss type, and the network packet loss type identification model to be evaluated is determined as the network packet loss type identification model only when the false identification rate is lower than a preset threshold value. Because the false recognition rate of the target packet loss type is used as the verification index of the model in the embodiment, the model has better TCP friendliness and is beneficial to improving the gain of the transmitting end.

It should be understood that, although the steps in the flowcharts of fig. 1 to 9 are shown in order as indicated by the arrows, these steps are not necessarily performed in order as indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least a portion of the steps of fig. 1-9 may include multiple steps or stages that are not necessarily performed at the same time, but may be performed at different times, nor does the order in which the steps or stages are performed necessarily occur sequentially, but may be performed alternately or alternately with at least a portion of the steps or stages in other steps or other steps.

It should be understood that the same/similar parts of the embodiments of the method described above in this specification may be referred to each other, and each embodiment focuses on differences from other embodiments, and references to descriptions of other method embodiments are only needed.

Fig. 10 is a block diagram of a network packet loss type identification device according to an exemplary embodiment. Referring to fig. 10, the apparatus includes a network characteristic parameter acquisition module 1002 and a packet loss type identification module 1004.

The network characteristic parameter obtaining module 1002 is configured to obtain, when a packet loss signal in a network is detected, a network characteristic parameter for representing a network state in a preset time period before a target data packet is sent, where the target data packet is a lost data packet corresponding to the packet loss signal, and the network characteristic parameter is statistical characteristic data extracted after performing characteristic statistics on an acquired time sequence array for representing the network state;

the packet loss type recognition module 1004 is configured to perform recognition by inputting the network characteristic parameters into a pre-trained network packet loss type recognition model, so as to obtain a packet loss type corresponding to the packet loss signal, where the network packet loss type recognition model is obtained by training based on network characteristic sample parameters collected under a real network environment.

In an exemplary embodiment, the preset time period includes at least two round trip delays; the network characteristic parameter acquisition module comprises: the system comprises a time sequence array acquisition unit, a data transmission unit and a data transmission unit, wherein the time sequence array acquisition unit is configured to execute time sequence array for representing network state in each round trip delay before a target data packet is transmitted according to at least two round trip delays, the round trip delays are the time length from the time when the data packet is transmitted from a transmitting end to the time when the transmitting end receives an acknowledgement message from a receiving end, and the acknowledgement message is returned when the receiving end receives the data packet; a statistical feature extraction unit configured to perform extraction of statistical features in a timing array characterizing a network state within each round trip delay; a difference feature acquisition unit configured to perform acquisition of a difference feature between at least two round trip delays based on the extracted statistical feature; and a network characteristic parameter generating unit configured to perform generation of a corresponding network characteristic parameter according to the extracted statistical characteristic and the difference characteristic.

In an exemplary embodiment, the timing array acquisition unit is configured to perform: acquiring an acknowledgement message returned by an opposite terminal in each round trip delay before sending a target data packet, wherein the acknowledgement message is returned by the opposite terminal for the data packet sent by a local terminal, and the acknowledgement message comprises a first receiving time for the opposite terminal to receive the data packet; acquiring the sending time of the data packet sent by the local terminal and the second receiving time of the acknowledgement message received by the local terminal; acquiring a first time sequence array of unidirectional delay between the home terminal and the opposite terminal in each round trip delay according to the sending time of the home terminal for sending the data packet and the first receiving time of the opposite terminal for receiving the data packet; and acquiring a second time sequence array of the interval time of the acknowledgement message received by the local terminal according to the second receiving time of the acknowledgement message received by the local terminal in each round trip time delay.

In an exemplary embodiment, the statistical features include maximum, minimum, median absolute deviation, variance, skewness factor, kurtosis factor, first order standard deviation, and second order standard deviation; the statistical feature extraction unit is configured to perform: extracting a corresponding first statistical feature in a first time sequence array according to the first time sequence array of unidirectional delay between a home terminal and an opposite terminal in each round trip delay; extracting corresponding second statistical characteristics from a second time sequence array of the interval time of the acknowledgement message received by the local terminal in each round trip delay; and obtaining the statistical characteristics used for representing the network state in each round trip delay according to the extracted first statistical characteristics and the second statistical characteristics.

In an exemplary embodiment, the difference feature acquisition unit is configured to perform: according to the first statistical characteristics corresponding to the first time sequence groups in each round trip time delay, calculating a first ratio between the first statistical characteristics corresponding to the first time sequence groups in one round trip time delay and the first statistical characteristics corresponding to the first time sequence groups in the other round trip time delay in two adjacent round trip time delays; calculating a second ratio between a second statistical feature corresponding to the second time sequence array in one round trip delay and a second statistical feature corresponding to the second time sequence array in the other round trip delay in two adjacent round trip delays according to the second statistical feature corresponding to the second time sequence array in each round trip delay; the first ratio and the second ratio are determined as a characteristic of a difference between at least two round trip delays.

In an exemplary embodiment, the network packet loss type identification model includes a plurality of network packet loss type classifiers; the packet loss type identification module is further configured to perform: inputting the network characteristic parameters into a pre-trained network packet loss type identification model for identification, and obtaining an identification result of each classifier in the network packet loss type identification model, wherein the identification result comprises any one of congestion packet loss and error packet loss; and acquiring the number of classifiers corresponding to each type of identification result, and taking the identification result with the largest number of classifiers as the packet loss type corresponding to the packet loss signal.

In an exemplary embodiment, the apparatus further comprises: the error packet loss rate acquisition module is configured to acquire an error packet loss rate in a set period when the packet loss type corresponding to the packet loss signal is obtained to be an error packet loss; the target sending rate calculating module is configured to calculate a target sending rate according to the error packet loss rate and a rate corresponding to a network bottleneck link, wherein the network bottleneck link is a link with the highest bandwidth utilization rate between a sending end and a receiving end of the target data packet; and the data transmission module is configured to transmit the subsequent data packets at the target sending rate.

In an exemplary embodiment, the target sending rate calculation module is configured to perform: obtaining a difference value between the 1 and the error packet loss rate; and calculating the product between the reciprocal of the difference value and the corresponding speed of the network bottleneck link as the target sending speed.

In an exemplary embodiment, the data transmission module is further configured to perform: when the packet loss type corresponding to the packet loss signal is detected to be the error packet loss in the set period, a corresponding sending window is added to transmit the data packet corresponding to the packet loss signal.

In an exemplary embodiment, the data transmission module is further configured to perform: and when the packet loss type corresponding to the packet loss signal is obtained as the congestion packet loss, reducing the sending rate of the subsequent data packet.

Fig. 11 is a block diagram of a generating apparatus of a network packet loss type recognition model according to an exemplary embodiment. Referring to fig. 11, the apparatus includes a sample parameter acquisition module 1102 and a model generation module 1104.

A sample parameter obtaining module 1102, configured to perform obtaining a plurality of sets of network feature sample parameters, where the network feature sample parameters are statistical feature sample data extracted after feature statistics is performed on an acquired time sequence sample array for characterizing a network state, and each set of network feature sample parameters has a corresponding network state label, where the network state label includes a label of network congestion and a label of network error;

the model generation module 1104 is configured to perform training based on a plurality of sets of network feature sample parameters to generate a plurality of classifiers, resulting in a network packet loss type recognition model.

In an exemplary embodiment, the plurality of sets of network feature sample parameters includes a first network feature sample parameter set and a second network feature sample parameter set; the model generation module comprises: the model training unit is configured to perform training based on the first network characteristic sample parameter set to generate a plurality of classifiers so as to obtain a network packet loss type identification model to be evaluated; the model verification unit is configured to perform verification on the network packet loss type identification model to be evaluated based on the second network characteristic sample parameter set and the corresponding network state label, so as to obtain the false identification rate of the network packet loss type identification model to be evaluated on the target packet loss type; and a model determining unit configured to perform determining the network packet loss type identification model to be evaluated as the network packet loss type identification model when the false identification rate is smaller than a preset threshold.

In an exemplary embodiment, the model training unit is further configured to perform: randomly extracting n groups of network characteristic sample parameters from the first network characteristic sample parameter set to obtain a training sample subset; k rounds of extraction are carried out on the first network characteristic sample parameter set, and k training sample subsets are obtained; and respectively training based on k training sample subsets to obtain k corresponding classifiers, and integrating the k classifiers to obtain a network packet loss type identification model to be evaluated.

In an exemplary embodiment, each set of network feature sample parameters in the first set of network feature sample parameters has a corresponding weight; the model training unit is further configured to perform: performing iterative training based on the first network characteristic sample parameter set and the corresponding weight to obtain a classifier corresponding to the iterative training; when any group of network characteristic sample parameters are misclassified in the iterative training process, increasing the corresponding weight of the network characteristic sample parameters in the next iterative training process; and performing k rounds of iterative training on the first network characteristic sample parameter set to obtain k corresponding classifiers, and integrating the k classifiers to obtain a network packet loss type identification model to be evaluated.

The specific manner in which the various modules perform the operations in the apparatus of the above embodiments have been described in detail in connection with the embodiments of the method, and will not be described in detail herein.

Fig. 12 is a block diagram of an electronic device S00 for network packet loss type identification and for network packet loss type identification model generation, according to an example embodiment. For example, electronic device S00 may be a server. Referring to fig. 12, electronic device S00 includes a processing component S20 that further includes one or more processors, and memory resources represented by memory S22, for storing instructions, such as applications, executable by processing component S20. The application program stored in the memory S22 may include one or more modules each corresponding to a set of instructions. Further, the processing component S20 is configured to execute instructions to perform the above-described method.

The electronic device S00 may further include: the power supply assembly S24 is configured to perform power management of the electronic device S00, the wired or wireless network interface S26 is configured to connect the electronic device S00 to a network, and the input output (I/O) interface S28. The electronic device S00 may operate based on an operating system stored in the memory S22, such as Windows Server, mac OS X, unix, linux, freeBSD, or the like.

In an exemplary embodiment, a computer readable storage medium comprising instructions, such as a memory S22 comprising instructions, is also provided, the instructions being executable by a processor of the electronic device S00 to perform the above-described method. The storage medium may be a computer readable storage medium, which may be, for example, ROM, random Access Memory (RAM), CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.

In an exemplary embodiment, a computer program product is also provided, comprising instructions therein, which are executable by a processor of the electronic device S00 to perform the above method.

It should be noted that the descriptions of the foregoing apparatus, the electronic device, the computer readable storage medium, the computer program product, and the like according to the method embodiments may further include other implementations, and the specific implementation may refer to the descriptions of the related method embodiments and are not described herein in detail.

Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This disclosure is intended to cover any adaptations, uses, or adaptations of the disclosure following the general principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.

It is to be understood that the present disclosure is not limited to the precise arrangements and instrumentalities shown in the drawings, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A network packet loss type identification method, the method comprising:

inputting the network characteristic parameters into a pre-trained network packet loss type identification model for identification to obtain a packet loss type corresponding to the packet loss signal;

the preset time period comprises at least two round trip delays; the obtaining the network characteristic parameters used for representing the network state in a preset time period before sending the target data packet includes: acquiring a time sequence array representing the network state in each round trip delay before sending a target data packet according to at least two round trip delays, wherein the round trip delay is the time length from the time when a data packet is sent by a sending end to the time when the sending end receives an acknowledgement message from a receiving end, and the acknowledgement message is returned by the receiving end when the receiving end receives the data packet; extracting statistical characteristics in a time sequence array for representing the network state in each round trip delay; acquiring difference characteristics between at least two round trip delays according to the extracted statistical characteristics; and generating corresponding network characteristic parameters according to the extracted statistical characteristics and the difference characteristics.

2. The method of claim 1, wherein the obtaining a timing array characterizing network conditions within each round trip delay before sending the target packet comprises:

acquiring an acknowledgement message returned by an opposite terminal in each round trip delay before sending a target data packet, wherein the acknowledgement message is returned by the opposite terminal for the data packet sent by a local terminal, and the acknowledgement message comprises a first receiving time for the opposite terminal to receive the data packet;

acquiring the sending time of the data packet sent by the local terminal and the second receiving time of the acknowledgement message received by the local terminal;

acquiring a first time sequence array of unidirectional delay between the home terminal and the opposite terminal in each round trip delay according to the sending time of the home terminal for sending the data packet and the first receiving time of the opposite terminal for receiving the data packet;

and acquiring a second time sequence array of the interval time of the acknowledgement message received by the local terminal according to the second receiving time of the acknowledgement message received by the local terminal in each round trip time delay.

3. The method of claim 2, wherein extracting statistical features in the timing array characterizing the network state for each round trip delay comprises:

extracting a corresponding first statistical feature in a first time sequence array according to the first time sequence array of unidirectional delay between a home terminal and an opposite terminal in each round trip delay;

Extracting corresponding second statistical characteristics from a second time sequence array of the interval time of the acknowledgement message received by the local terminal in each round trip delay;

and obtaining the statistical characteristics used for representing the network state in each round trip delay according to the extracted first statistical characteristics and the second statistical characteristics.

4. A method according to claim 3, wherein said obtaining a difference characteristic between at least two round trip delays based on said extracted statistical characteristic comprises:

according to the first statistical characteristics corresponding to the first time sequence groups in each round trip time delay, calculating a first ratio between the first statistical characteristics corresponding to the first time sequence groups in one round trip time delay and the first statistical characteristics corresponding to the first time sequence groups in the other round trip time delay in two adjacent round trip time delays;

calculating a second ratio between a second statistical feature corresponding to the second time sequence array in one round trip delay and a second statistical feature corresponding to the second time sequence array in the other round trip delay in two adjacent round trip delays according to the second statistical feature corresponding to the second time sequence array in each round trip delay;

the first ratio and the second ratio are determined as a characteristic of a difference between at least two round trip delays.

5. The method of claim 1, wherein the network packet loss type identification model comprises a plurality of network packet loss type classifiers; inputting the network characteristic parameters into a pre-trained network packet loss type identification model for identification to obtain the packet loss type corresponding to the packet loss signal, wherein the method comprises the following steps:

inputting the network characteristic parameters into a pre-trained network packet loss type identification model for identification, and obtaining an identification result of each classifier in the network packet loss type identification model, wherein the identification result comprises any one of congestion packet loss and error packet loss;

and acquiring the number of classifiers corresponding to each type of identification result, and taking the identification result with the largest number of classifiers as the packet loss type corresponding to the packet loss signal.

6. The method according to any one of claims 1 to 5, wherein after the obtaining the packet loss type corresponding to the packet loss signal, the method further includes:

when the packet loss type corresponding to the packet loss signal is obtained as error packet loss, obtaining error packet loss rate in a set period;

calculating a target sending rate according to the error packet loss rate and a rate corresponding to a network bottleneck link, wherein the network bottleneck link is a link with the highest bandwidth utilization rate between a sending end and a receiving end of the target data packet;

And transmitting the subsequent data packets by adopting the target transmission rate.

7. The method of claim 6, wherein calculating the target sending rate according to the error packet loss rate and the rate corresponding to the network bottleneck link comprises:

obtaining a difference value between the 1 and the error packet loss rate;

and calculating the product between the reciprocal of the difference value and the corresponding speed of the network bottleneck link as the target sending speed.

8. The method of claim 6, wherein the method further comprises:

when the packet loss type corresponding to the packet loss signal is detected to be the error packet loss in the set period, a corresponding sending window is added to transmit the data packet corresponding to the packet loss signal.

9. The method according to any one of claims 1 to 5, wherein after the obtaining the packet loss type corresponding to the packet loss signal, the method further includes:

and when the packet loss type corresponding to the packet loss signal is obtained as the congestion packet loss, reducing the sending rate of the subsequent data packet.

10. The method according to any one of claims 1 to 5, wherein the generating of the network packet loss type identification model comprises:

11. The method of claim 10, wherein the plurality of sets of network feature sample parameters comprises a first set of network feature sample parameters and a second set of network feature sample parameters; the training based on a plurality of groups of network characteristic sample parameters to generate a plurality of classifiers to obtain a network packet loss type identification model comprises the following steps:

training based on the first network characteristic sample parameter set to generate a plurality of classifiers to obtain a network packet loss type identification model to be evaluated;

verifying the network packet loss type identification model to be evaluated based on the second network characteristic sample parameter set and the corresponding network state label to obtain the false identification rate of the network packet loss type identification model to be evaluated on the target packet loss type;

And when the false recognition rate is smaller than a preset threshold value, determining the network packet loss type recognition model to be evaluated as the network packet loss type recognition model.

12. The method of claim 11, wherein training based on the first network feature sample parameter set to generate a plurality of classifiers, to obtain a network packet loss type identification model to be evaluated, comprises:

randomly extracting n groups of network characteristic sample parameters from the first network characteristic sample parameter set to obtain a training sample subset;

k rounds of extraction are carried out on the first network characteristic sample parameter set, and k training sample subsets are obtained;

and respectively training based on k training sample subsets to obtain k corresponding classifiers, and integrating the k classifiers to obtain a network packet loss type identification model to be evaluated.

13. The method of claim 11, wherein each set of network feature sample parameters in the first set of network feature sample parameters has a corresponding weight; training and generating a plurality of classifiers based on the first network characteristic sample parameter set to obtain a network packet loss type identification model to be evaluated, wherein the method comprises the following steps:

performing iterative training based on the first network characteristic sample parameter set and the corresponding weight to obtain a classifier corresponding to the iterative training;

When any group of network characteristic sample parameters are misclassified in the iterative training process, increasing the corresponding weight of the network characteristic sample parameters in the next iterative training process;

and performing k rounds of iterative training on the first network characteristic sample parameter set to obtain k corresponding classifiers, and integrating the k classifiers to obtain a network packet loss type identification model to be evaluated.

14. A network packet loss type identification device, comprising:

the packet loss type identification module is configured to input the network characteristic parameters into a pre-trained network packet loss type identification model for identification to obtain a packet loss type corresponding to the packet loss signal;

the preset time period comprises at least two round trip delays; the network characteristic parameter acquisition module comprises:

The system comprises a time sequence array acquisition unit, a data transmission unit and a data transmission unit, wherein the time sequence array acquisition unit is configured to execute time sequence array for representing network state in each round trip delay before a target data packet is transmitted according to at least two round trip delays, the round trip delays are the time length from the time when the data packet is transmitted from a transmitting end to the time when the transmitting end receives an acknowledgement message from a receiving end, and the acknowledgement message is returned when the receiving end receives the data packet;

a statistical feature extraction unit configured to perform extraction of statistical features in a timing array characterizing a network state within each round trip delay;

a difference feature acquisition unit configured to perform acquisition of a difference feature between at least two round trip delays based on the extracted statistical feature;

and a network characteristic parameter generating unit configured to perform generation of a corresponding network characteristic parameter according to the extracted statistical characteristic and the difference characteristic.

15. The apparatus of claim 14, wherein the timing array acquisition unit is configured to perform:

16. The apparatus of claim 15, wherein the statistical feature extraction unit is configured to perform:

17. The apparatus according to claim 16, wherein the difference feature acquisition unit is configured to perform:

18. The apparatus of claim 14, wherein the network packet loss type identification model comprises a plurality of network packet loss type classifiers; the packet loss type identification module is further configured to perform:

19. The apparatus according to any one of claims 14 to 18, further comprising:

the error packet loss rate acquisition module is configured to acquire an error packet loss rate in a set period when the packet loss type corresponding to the packet loss signal is obtained to be an error packet loss;

the target sending rate calculating module is configured to calculate a target sending rate according to the error packet loss rate and a rate corresponding to a network bottleneck link, wherein the network bottleneck link is a link with the highest bandwidth utilization rate between a sending end and a receiving end of the target data packet;

and the data transmission module is configured to transmit the subsequent data packets at the target sending rate.

20. The apparatus of claim 19, wherein the target transmission rate calculation module is configured to perform:

obtaining a difference value between the 1 and the error packet loss rate;

21. The apparatus of claim 19, wherein the data transmission module is further configured to perform:

22. The apparatus of claim 19, wherein the data transmission module is further configured to perform:

23. The apparatus according to any one of claims 14 to 18, further comprising means for generating a network packet loss type identification model, the generating means comprising:

24. The apparatus of claim 23, wherein the plurality of sets of network feature sample parameters comprises a first set of network feature sample parameters and a second set of network feature sample parameters; the model generation module comprises:

the model training unit is configured to perform training based on the first network characteristic sample parameter set to generate a plurality of classifiers so as to obtain a network packet loss type identification model to be evaluated;

the model verification unit is configured to perform verification on the network packet loss type identification model to be evaluated based on the second network characteristic sample parameter set and the corresponding network state label, so as to obtain the false identification rate of the network packet loss type identification model to be evaluated on the target packet loss type;

and a model determining unit configured to perform determining the network packet loss type identification model to be evaluated as the network packet loss type identification model when the false identification rate is smaller than a preset threshold.

25. The apparatus of claim 24, wherein the model training unit is further configured to perform:

26. The apparatus of claim 24, wherein each set of network feature sample parameters in the first set of network feature sample parameters has a corresponding weight; the model training unit is further configured to perform:

27. An electronic device, comprising:

a processor;

a memory for storing the processor-executable instructions;

wherein the processor is configured to execute the instructions to implement the network packet loss type identification method of any of claims 1 to 13.

28. A computer readable storage medium, characterized in that instructions in the computer readable storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the network packet loss type identification method according to any one of claims 1 to 13.