Network measurement model training method and device, network measurement method and device, electronic equipment, storage medium and computer program product

Abstract

The invention relates to a training method and device of a network measurement model, a network measurement method and device, electronic equipment, a storage medium and a computer program product. The training method comprises the steps that a training sample set is obtained, the training sample set comprises input samples at multiple measurement moments and simulation network loss data, and the input sample at each measurement moment is determined based on simulation estimation information and simulation observation information of a simulation network at the corresponding measurement moment; the simulation estimation information is network state information estimated by using a congestion control model in a simulation network, and the simulation observation information is data packet transmission related information observed in the simulation network; inputting the input sample into a network measurement model to obtain estimated network loss data of the simulation network at a predetermined measurement moment, the predetermined measurement moment being a measurement moment corresponding to the input sample; and adjusting parameters of the network measurement model based on the loss of the estimated network loss data and the analog network loss data at the predetermined measurement moment.

Description

Technical Field

[0001] The present disclosure relates to the field of communications, and in particular to a network measurement model training method and device, a network measurement method and device, an electronic device, a storage medium, and a computer program product. Background Art

[0002] In recent years, the streaming media service industry has flourished, and streaming media has become the largest consumer content in the world's network traffic. Many large companies around the world have proposed their own streaming media transmission protocols and invested a lot of manpower in optimizing streaming media transmission protocols. However, the pain point of optimizing streaming media transmission protocols lies in the extreme complexity of the real Internet. Although network infrastructure has made great leaps in the past few years, the heterogeneity, policy complexity and dynamic nature of network infrastructure still make end-to-end streaming media transmission protocol optimization extremely difficult. Therefore, understanding the distribution of the real network and establishing a rich real network loss database locally are important cornerstones for streaming media transmission optimization. The real network loss database will greatly accelerate the iteration speed of the streaming media transmission protocol and reduce the cost of trial and error.

[0003] In order to understand the distribution of real networks, the industry and academia have proposed several network measurement methods, such as bandwidth measurement based on pathload, delay measurement based on ping packets, and network routing measurement based on trace routes, etc. However, these network measurement methods will bring additional bandwidth costs, and they cannot instantly reflect the real network situation when the user network quality deteriorates. Summary of the invention

[0004] The present disclosure provides a training method and device for a network measurement model, a network measurement method and device, an electronic device, a storage medium, and a computer program product, so as to at least solve the problem that the bandwidth cost consumed by related technologies is high or cannot reflect the changes in the real network.

[0005] According to a first aspect of an embodiment of the present disclosure, a method for training a network measurement model is provided, comprising: obtaining a training sample set, wherein the training sample set includes input samples and simulated network loss data at multiple measurement moments, the input samples at each measurement moment are determined based on simulated estimation information and simulated observation information of a simulated network at the corresponding measurement moment, the simulated estimation information is network status information estimated by using a congestion control model in a simulated network, the simulated observation information is information related to data packet transmission observed in a simulated network, and the simulated network loss data refers to data characterizing the real-time quality of the simulated network; inputting the input samples into the network measurement model to obtain estimated network loss data of the simulated network at a predetermined measurement moment, wherein the predetermined measurement moment is a measurement moment corresponding to the input samples; and adjusting parameters of the network measurement model based on the loss of the estimated network loss data and the simulated network loss data at the predetermined measurement moment.

[0006] Optionally, the simulation estimation information and the simulation observation information are obtained in the following manner: when the target application software is running on a simulated network, confirmation messages of multiple simulated data packets are obtained; for each confirmation message, the confirmation message is input into a congestion control model to obtain simulation estimation information corresponding to the round-trip time of the corresponding simulated data packet in the simulated network; when the target application software is running on a simulated network, the simulated network that transmits multiple simulated data packets is monitored to obtain simulation observation information corresponding to the round-trip time of each simulated data packet in the simulated network; the simulation estimation information and simulation observation information corresponding to the round-trip delay of the simulated data packet transmitted starting from each measurement moment are respectively determined as the simulation estimation information and simulation observation information at each measurement moment.

[0007] Optionally, the input samples are obtained in the following manner: the simulated estimation information and the simulated observation information corresponding to the round trip time of a predetermined number of simulated data packets transmitted starting from each measurement moment are determined as the input samples at each measurement moment.

[0008] Optionally, the simulated network is generated in the following manner: obtaining a simulated network loss data set, wherein the simulated network loss data set includes multiple simulated network loss data, and each simulated network loss data has a granularity of a predetermined time length; generating a simulated network based on multiple simulated network loss data divided from each simulated network loss data and a preset distribution curve for each simulated network loss data.

[0009] Optionally, the simulated network loss data includes at least one of simulated network loss bandwidth, simulated network loss delay and foam network loss packet loss.

[0010] According to a second aspect of an embodiment of the present disclosure, a network measurement method is provided, comprising: when a target application software is running on a target network, obtaining estimated information and observed information of the target network at a current measurement moment, wherein the estimated information is network status information estimated by using a congestion control model on the target network, and the observed information is information related to data packet transmission observed on the target network; based on the estimated information and observed information at the current measurement moment, determining input information at the current measurement moment; inputting the input information into a network measurement model to obtain network loss data of the target network at the current measurement moment, wherein the network loss data refers to data characterizing the real-time quality of the target network; wherein the network measurement model is trained using the network measurement model training method of the present disclosure as described above.

[0011] According to a third aspect of an embodiment of the present disclosure, a training device for a network measurement model is provided, comprising: a first acquisition unit, configured to obtain a training sample set, wherein the training sample set includes input samples and simulated network loss data at multiple measurement moments, and the input samples at each measurement moment are determined based on simulated estimation information and simulated observation information of a simulated network at the corresponding measurement moment, respectively, the simulated estimation information is network status information estimated by using a congestion control model under a simulated network, the simulated observation information is information related to data packet transmission observed under a simulated network, and the simulated network loss data refers to data characterizing the real-time quality of the simulated network; a first measurement unit, configured to input the input sample into the network measurement model to obtain estimated network loss data of the simulated network at a predetermined measurement moment, wherein the predetermined measurement moment is a measurement moment corresponding to the input sample; an adjustment unit, configured to adjust parameters of the network measurement model based on the loss of the estimated network loss data and the simulated network loss data at the predetermined measurement moment.

[0012] Optionally, the first acquisition unit is further configured to acquire simulation estimation information and simulation observation information in the following manner: when the target application software is running on a simulated network, acquiring confirmation messages of multiple simulated data packets; for each confirmation message, inputting the confirmation message into a congestion control model to obtain simulation estimation information corresponding to the round-trip time of the corresponding simulated data packet in the simulated network; when the target application software is running on a simulated network, monitoring the simulated network that transmits multiple simulated data packets, and acquiring simulation observation information corresponding to the round-trip time of each simulated data packet in the simulated network; and determining the simulation estimation information and simulation observation information corresponding to the round-trip delay of the simulated data packet transmitted starting from each measurement moment as the simulation estimation information and simulation observation information at each measurement moment, respectively.

[0013] Optionally, the first acquisition unit is configured to acquire input samples in the following manner, including: determining simulation estimation information and simulation observation information corresponding to the round-trip time of a predetermined number of simulated data packets transmitted from each measurement moment as input samples at each measurement moment.

[0014] Optionally, the simulated network is generated in the following manner: obtaining a simulated network loss data set, wherein the simulated network loss data set includes multiple simulated network loss data, and each simulated network loss data has a granularity of a predetermined time length; generating a simulated network based on multiple simulated network loss data divided from each simulated network loss data and a preset distribution curve for each simulated network loss data.

[0015] Optionally, the simulated network loss data includes at least one of simulated network loss bandwidth, simulated network loss delay and simulated network loss packet loss.

[0016] According to a fourth aspect of an embodiment of the present disclosure, a network measurement device is provided, comprising: a second acquisition unit, configured to acquire, when a target application software is running on a target network, estimated information and observed information of a target network at a current measurement moment, wherein the estimated information is network status information estimated by using a congestion control model on the target network, and the observed information is information related to data packet transmission observed on the target network; a determination unit, configured to determine input information at the current measurement moment based on the estimated information and observed information at the current measurement moment; a second measurement unit, configured to input the input information into a network measurement model to obtain network loss data of the target network at the current measurement moment, wherein the network loss data refers to data characterizing the real-time quality of the target network; wherein the network measurement model is trained using the network measurement model training method of the present disclosure as described above.

[0017] According to a third aspect of an embodiment of the present disclosure, an electronic device is provided, comprising: at least one processor; and at least one memory storing computer executable instructions, wherein the computer executable instructions, when executed by the at least one processor, prompt the at least one processor to execute the network measurement model training method and network measurement method disclosed above.

[0018] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided. When instructions in the computer-readable storage medium are executed by at least one processor, the at least one processor is prompted to execute the network measurement model training method and network measurement method disclosed above.

[0019] According to a fifth aspect of an embodiment of the present disclosure, a computer program product is provided, including computer instructions, which, when executed by a processor, implement the network measurement model training method and network measurement method of the present disclosure.

[0020] The technical solution provided by the embodiments of the present disclosure brings at least the following beneficial effects:

[0021] According to the training method and device of the network measurement model, the network measurement method and device, the electronic device, the storage medium, and the computer program product disclosed in the present invention, based on the simulated estimation information and the simulated observation information of each measurement moment of the simulated network, the input sample at each measurement moment is determined, and the simulated network loss data of the input sample is used to train the network measurement model, so that the trained network measurement model can accurately output the estimated network loss data that can characterize the real-time quality of the simulated network, so that the network loss data of any network can be accurately estimated without adding additional bandwidth costs, and a large-scale online real network loss database can be established, thereby improving the accuracy of network measurement. Moreover, the present invention can instantly reflect the real network situation when the quality of any network deteriorates. Therefore, the present invention solves the problem that the bandwidth cost consumed by the related technology is high or cannot reflect the changes in the real network.

[0022] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description are used to explain the principles of the present disclosure, and do not constitute improper limitations on the present disclosure.

[0024] Figure 1 is a schematic diagram of an implementation scenario of a network measurement model training method and a network measurement method according to an exemplary embodiment of the present disclosure;

[0025] Figure 2 is a flow chart of a method for training a network measurement model according to an exemplary embodiment of the present disclosure;

[0026] Figure 3 is a flow chart of a network measurement method according to an exemplary embodiment of the present disclosure;

[0027] Figure 4 A block diagram of a training device for a network measurement model according to an exemplary embodiment of the present disclosure;

[0028] Figure 5 A block diagram of a network measurement device according to an exemplary embodiment of the present disclosure;

[0029] Figure 6 A block diagram of an electronic device 600 is shown according to an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0030] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.

[0031] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The implementation methods described in the following examples do not represent all implementation methods consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the attached claims.

[0032] It should be noted that the phrase "at least one of the items" in the present disclosure includes three types of parallel situations: "any one of the items", "a combination of any number of the items", and "all of the items". For example, "including at least one of A and B" includes the following three parallel situations: (1) including A; (2) including B; (3) including A and B. Another example is "executing at least one of step 1 and step 2" which means the following three parallel situations: (1) executing step 1; (2) executing step 2; (3) executing step 1 and step 2.

[0033] At present, the industry and academia have proposed several network measurement methods, such as bandwidth measurement based on pathload, delay measurement based on ping packets, and network routing measurement based on trace routes. The following briefly introduces two network measurement methods:

[0034] 1. The PathLoad-based network bandwidth measurement method is a typical method of active network measurement, that is, actively sending a packet sequence at a fixed rate, recording the arrival time and reception time of the packet, and analyzing the arrival trend of the packet interval. If the packet arrival interval has an increasing trend, it means that the sending rate is greater than the link bandwidth, and the receiving rate at this time is calculated as the link bandwidth; if the packet arrival interval has no increasing trend, the binary method is used to increase the sending rate until the packet arrival interval is observed to have an increasing trend.

[0035] However, the network bandwidth measurement method based on PathLoad has the following three disadvantages: 1) PathLoad needs to actively send packets for network measurement, which will introduce huge bandwidth costs when actually deployed in online streaming applications; 2) PathLoad uses a binary method to detect bandwidth. The binary boundary selection greatly affects the detection speed, and the bandwidth measurement efficiency is low; 3) PathLoad can only reflect the network bandwidth level when it is detecting, and cannot reflect network characteristics in other dimensions, such as delay, random packet loss, queue depth and other factors. It also cannot reflect the network characteristics when the quality of streaming applications deteriorates.

[0036] In particular, in recent years, many variant algorithms that use packet pairs and packet sequences to measure bandwidth have emerged. However, they will introduce more bandwidth costs and cannot simultaneously observe multi-dimensional network characteristics. Here, only PathLoad is listed as a typical algorithm.

[0037] 2. The network measurement method based on ICMP packets is a typical method for measuring reachability, packet loss and delay. That is, the ping command is used on the client to periodically send ICMP packets, measure the average network round-trip delay and jitter level, and observe the packet loss rate and network reachability over a period of time.

[0038] However, the network measurement method based on ICMP packets has the following two disadvantages: 1) Delay measurement using IMCP packets is generally sparse, with a typical interval of 1000ms. If it is more dense, it cannot reflect the true level of physical delay; 2) Using ICMP packets to measure cannot reflect the bandwidth level of the network, and it cannot reflect the network characteristics when the quality of streaming media applications deteriorates.

[0039] In order to solve the problems of coarse granularity, poor accuracy, and incomplete reflection of network characteristics in existing network measurement methods, the present disclosure proposes a network measurement framework based on deep learning. The framework first obtains estimated information and observation information with the round-trip time of data packets as the granularity, as well as the real label of the simulated network in a large-scale network loss simulation environment (i.e., a simulated network); then, a deep learning model is trained with the above information and real labels to obtain a network measurement model; finally, the estimated information and observation information collected online are input into the trained network measurement model to estimate the network loss data that can characterize the real-time quality of the target network without increasing additional bandwidth costs, so that more realistic online network loss data can be obtained to form a real network database.

[0040] Hereinafter, a network measurement model training method and apparatus, a network measurement method and apparatus, an electronic device, a storage medium, and a computer program product according to exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings.

[0041] Figure 1 is a schematic diagram of an implementation scenario of a network measurement model training method and a network measurement method according to an exemplary embodiment of the present disclosure, such as Figure 1 The implementation scenario includes a server 100, a user terminal 110 and a user terminal 120, wherein the user terminals are not limited to two, and include but are not limited to mobile phones, personal computers and other devices. The user terminals can install streaming media application software. The server can be a single server, or a server cluster composed of several servers, or a cloud computing platform or a virtualization center.

[0042] When the target application software runs on the user terminal 110 or 120 using a simulated network, when the target application software transmits a data packet, the server 100 can obtain a training sample set, wherein the training sample set includes input samples and simulated network loss data at multiple measurement times, and the input samples at each measurement time are determined based on simulated estimation information and simulated observation information of the simulated network at the corresponding measurement time, respectively, the simulated estimation information is network status information estimated using a congestion control model under the simulated network, the simulated observation information is information related to data packet transmission observed under the simulated network, and the simulated network loss data refers to data characterizing the real-time quality of the simulated network; then, the input sample is input into the network measurement model to obtain estimated network loss data of the simulated network at a predetermined measurement time, wherein the predetermined measurement time is a measurement time corresponding to the input sample; then, based on the loss of the estimated network loss data and the simulated network loss data at the predetermined measurement time, the parameters of the network measurement model are adjusted.

[0043] After the network measurement model is trained, the target application software can use the target network to run on the user terminal 110 or 120. When the target application software transmits a data packet, the server 100 can obtain the estimated information and observed information of the target network at the current measurement time, wherein the estimated information is the network status information estimated by using the congestion control model under the target network, and the observed information is the data packet transmission related information observed under the target network; then, based on the estimated information and observed information at the current measurement time, the input information at the current measurement time is determined; finally, the input information is input into the network measurement model to obtain the network loss data of the target network at the current measurement time, wherein the network loss data refers to the data characterizing the real-time quality of the target network.

[0044] It should be noted that the user terminal 110 or the user terminal 120 can also complete the task independently, and the present disclosure does not limit this.

[0045] Figure 2 is a flowchart of a method for training a network measurement model according to an exemplary embodiment of the present disclosure, such as Figure 2 As shown, the training method of the network measurement model includes the following steps:

[0046] In step S201, a training sample set is obtained, wherein the training sample set includes input samples and simulated network loss data at multiple measurement moments, and the input samples at each measurement moment are determined based on simulated estimation information and simulated observation information of the simulated network at the corresponding measurement moment, respectively. The simulated estimation information is network status information estimated by using a congestion control model in the simulated network, the simulated observation information is information related to data packet transmission observed in the simulated network, and the simulated network loss data refers to data characterizing the real-time quality of the simulated network.

[0047] As an example, the present disclosure designs two major types of network information, namely, the estimated information output by the congestion control model and the observed information obtained by monitoring the network. In this embodiment, the target application software is running under a simulated network, so the estimated information is called simulated estimated information, and the observed information is called simulated observed information. After obtaining the simulated estimated information and simulated observed information of the simulated network at each measurement time, the network measurement information pair <simulated estimated information, simulated observed information> at each measurement time and the simulated network loss data at the corresponding measurement time can be obtained through the obtained simulated estimated information and simulated observed information. Based on the above information, a training sample set can be obtained to train the network measurement model.

[0048] As an example, the above congestion control model can adopt BBR (Bottleneck Bandwidth and Round-trip Propagation Time) V3, and other congestion control models can also be adopted. For example, more reasonable network measurement estimation information and observation information can be provided, which will help improve the accuracy of the network measurement model. This disclosure is not limited to this.

[0049] As an example, the granularity of the measurement time can be set as needed, such as the granularity of simulated network loss data, the granularity of RTT, or the granularity of any duration, which is not limited in this disclosure.

[0050] According to an exemplary embodiment of the present disclosure, simulation estimation information and simulation observation information can be obtained in the following manner: when the target application software is running under a simulated network, confirmation messages of multiple simulated data packets are obtained; for each confirmation message, the confirmation message is input into a congestion control model to obtain simulation estimation information corresponding to the round-trip time of the corresponding simulated data packet of the simulated network; when the target application software is running under a simulated network, the simulated network that transmits multiple simulated data packets is monitored to obtain simulation observation information corresponding to the round-trip time of each simulated data packet of the simulated network; the simulation estimation information and simulation observation information corresponding to the round-trip delay of the simulated data packet transmitted at each measurement moment are respectively determined as simulation estimation information and simulation observation information at each measurement moment. Through this embodiment, with the round-trip time of the data packet as the granularity, relatively accurate simulation estimation information and simulation observation information of the simulated network can be conveniently and quickly obtained.

[0051] As an example, the target application software may be a streaming media application software, or other application software, which is not limited in the present disclosure.

[0052] As an example, each simulated data packet will have a confirmation message when transmitted using the simulated network. Therefore, for the confirmation message of each simulated data packet, the confirmation message can be input into the congestion control model, and then the simulated estimation information corresponding to the round-trip time of the simulated data packet corresponding to the confirmation message of the simulated network can be obtained.

[0053] As an example, as mentioned above, the present disclosure designs two major types of network information, namely, simulated estimation information and simulated observation information. The simulated estimation information and simulated observation information may be based on RTT granularity, which may be as follows:

[0054] When the client runs the application software using the simulated network, the simulated estimation information output by the congestion control model can be recorded in real time at the granularity of the round-trip delay (RTT) of the transmitted data packet. i , where the subscript i represents the i-th RTT. The above simulation estimation information est1 i It may include but is not limited to the following information: Estimated maximum bandwidth bw_max1 i , estimate the EWMA smoothing bandwidth bw_ewma1 i , estimate the physical propagation delay d1 i , the current expected sending rate p1 i , the state machine M1 i , where the state machine M1 i Use 0 to represent the PROBE_RTT state, 1 to represent the PROBE_BW state, and 2 to represent the COLD_START stage.

[0055] When the client runs the application software using the simulated network, the passive simulation observation information obs1 can be recorded in real time at the granularity of the round-trip delay (RTT) of the transmitted data packet. i , where the subscript i represents the i-th RTT. The above simulated observation information obs1 i It may include but is not limited to the following information: Current RTT start timestamp t1 i , actual sending rate s1 i , the actual receiving rate r1 i , the actual packet loss rate l1 i , the maximum round-trip delay observed within the current RTT is rtt_max1 i , the minimum round-trip delay observed within the current RTT is rtt_min1 i , the average round-trip delay observed within the current RTT is rtt_avg1 i .

[0056] It should be noted that the present disclosure is not limited to obtaining the simulation estimation information and simulation observation information at the granularity of RTT, and other granularities can also be used to obtain the simulation estimation information and simulation observation information, and the present disclosure is not limited to this. The present disclosure is also not limited to the above-mentioned simulation estimation information and simulation observation information, and other simulation estimation information and simulation observation information can also be used.

[0057] According to an exemplary embodiment of the present disclosure, input samples can be obtained in the following manner: the simulated estimation information and simulated observation information corresponding to the round-trip time of a predetermined number of simulated data packets transmitted from each measurement moment are determined as input samples at each measurement moment. Through this embodiment, the simulated estimation information and simulated observation information corresponding to the round-trip time of multiple simulated data packets are determined as input samples, which can ensure that the input samples contain rich information, so as to achieve better training and subsequently obtain better network loss data.

[0058] As an example, the above-mentioned predetermined number can be set as needed, such as being set to 5.

[0059] As an example, assuming that the measurement time is based on the granularity of the simulated network loss data, the information est1 can be estimated through the above simulation i And simulated observation information obs1 i , obtain network measurement information with RTT as the granularity <est i ,obs i >, then the network measurement information pair corresponding to the round trip time of a predetermined number of simulated data packets transmitted starting from the current measurement time can be used as the input information of the current measurement time. The granularity of the above simulated network loss data can be set as needed, such as 1 millisecond or other time lengths, which is not limited by the present disclosure.

[0060] As an example, assuming that the measurement time is based on RTT granularity, the current measurement time can be the start timestamp of each RTT. At this time, the above simulation estimation information est1 i And simulated observation information obs1 i , you can get network measurement information with RTT as the granularity <est i ,obs i >, then use the network measurement information pairs corresponding to the round-trip times of a predetermined number of simulated data packets transmitted after the start timestamp of each RTT as the input information of the start timestamp of each RTT.

[0061] As an example, taking the measurement time as RTT as the granularity, the network measurement information with RTT as the granularity is obtained. <est1 i ,obs1 i>After that, since each measurement moment is the start timestamp of each RTT, according to obs1 i The current RTT starts at timestamp t1 i , get the current RTT start timestamp t1 in the simulated network loss data set i Corresponding simulated network loss data And use it as the start timestamp t1 i The true label value of , which is also the target label.

[0062] Then, t1 can be i The network measurement information of the last k RTTs is compared with t1 i The corresponding combination of true label values ​​forms the final training sample, which can be shown as follows:

[0063]

[0064] Wherein, k can be 5 or other values, which are not limited in the present disclosure. Then, a plurality of training samples are combined together to obtain a training sample set.

[0065] According to an exemplary embodiment of the present disclosure, a simulated network can be generated in the following manner: obtaining a simulated network loss data set, wherein the simulated network loss data set includes a plurality of simulated network loss data, and each type of simulated network loss data has a granularity of a predetermined time length; generating a simulated network based on a plurality of simulated network loss data divided from each type of simulated network loss data and a preset distribution curve for each type of simulated network loss data.

[0066] As an example, a large-scale simulated network loss data set can be generated according to the bandwidth range (such as 1-100000 kbps), network loss delay range (1-8000 ms) and network loss packet loss range (0%-80%) as needed. The simulated network loss data included is 100 million seconds in total, and the simulated network loss data at each moment can be expressed in the following form: <bw t ,dl t ,loss t >, where bw t ,dl t ,loss t They represent the simulated network loss bandwidth, simulated network loss delay and simulated network loss packet loss at time t respectively. It should be noted that the time granularity of t can be 1 millisecond (the above-mentioned predetermined time length), which is not limited in the present disclosure. In theory, the smaller the time granularity, the better, and generally speaking, the granularity is smaller than the round-trip time of the data packet.

[0067] Start the network loss simulator. Through the simulated network loss data set generated above, combined with the preset network loss bandwidth distribution curve, the preset network loss delay distribution curve and the preset network loss packet loss distribution curve, the real-time changes of the network are simulated according to the three dimensions of bandwidth, physical delay and random packet loss with a time granularity of 1 millisecond, that is, a simulated network is generated.

[0068] According to an exemplary embodiment of the present disclosure, the above-mentioned simulated network loss data may include but is not limited to: simulated network loss bandwidth, simulated network loss delay and simulated network loss packet loss.

[0069] In step S202, the input sample is input into the network measurement model to obtain estimated network loss data of the simulated network at a predetermined measurement time, wherein the predetermined measurement time is the measurement time corresponding to the input sample.

[0070] As an example, the above network measurement model may include multiple parts, each part corresponds to a type of network loss data. Assuming there are three types of network loss data, the network measurement model includes three parts, each of which can be composed of a GRU recurrent neural network and a fully connected neural network. During training, the above training samples and corresponding labels can be used to train the three parts separately. Specifically, taking the network loss bandwidth as an example, the above training sample set is normalized, and the normalized training sample set and the network loss bandwidth label (that is, the simulated network loss bandwidth) are used to train the GRU recurrent neural network model and the fully connected neural network to obtain the network measurement information and the network loss bandwidth. It should be noted that a converged network measurement model can be obtained by using a GPU for training for x days, wherein a typical value of x can be 10, which is not limited in the present disclosure.

[0071] It should be noted that the GRU recurrent neural network can also be replaced by other networks, such as long short-term memory network (Long-Short Term Memory, abbreviated as LSTM), Transformer, etc., which is not limited in the present disclosure.

[0072] As an example, a pytorch program can be started and a graphics processing unit (GPU) can be used to build an initial network measurement model. Specifically, the initial network measurement model can be composed of a gated recurrent unit (GRU) recurrent neural network and a fully connected neural network. The GRU recurrent neural network model can use 64 hidden nodes, the input feature dimension can be 13, the fully connected neural network input dimension can be 64, and the output dimension can be 1.

[0073] In step S203, the parameters of the network measurement model are adjusted based on the estimated network loss data and the loss of the simulated network loss data at the predetermined measurement time.

[0074] As an example, for each moment of network measurement information <est1 i ,obs1 i >, the network measurement information is input into the network measurement model to obtain the estimated network loss data, calculate the loss of the estimated network loss data and the target label (i.e., the simulated network loss data at the current measurement moment), and adjust the parameters of the network measurement model by minimizing the loss. Then, the next iterative training is carried out until the training termination condition is met, and the training is stopped to obtain a trained network measurement model.

[0075] Figure 3 is a flow chart of a network measurement method according to an exemplary embodiment of the present disclosure, such as Figure 3 As shown, the network measurement method includes the following steps:

[0076] In step S301, when the target application software is running on the target network, the estimated information and observed information of the target network at the current measurement time are obtained, wherein the estimated information is the network status information estimated by using the congestion control model on the target network, and the observed information is the data packet transmission related information observed on the target network.

[0077] As an example, the target application software may be a streaming media application software, or other application software, which is not limited in the present disclosure.

[0078] As an example, the present disclosure designs two major types of information, namely, the estimated information output by the congestion control model and the observed information obtained by monitoring the target network. The specific implementation of this embodiment can be as follows:

[0079] When the client runs the application software on the target network, the estimated information output by the congestion control model can be recorded in real time at the granularity of the round-trip delay (RTT) of the transmitted data packet. i , where the subscript i represents the i-th RTT. The above estimated information est i It may include but is not limited to the following information: Estimated maximum bandwidth bw_max i , estimate the EWMA smoothing bandwidth bw_ewma i , estimate the physical propagation delay d i , the current expected sending rate p i , the state machine M i , where the state machine M i Use 0 to represent the PROBE_RTT state, 1 to represent the PROBE_BW state, and 2 to represent the COLD_START stage.

[0080] When the client runs the application software using the target network, the passive observation information obs can be recorded in real time at the granularity of the round-trip delay (RTT) of the transmitted data packets. i , where the subscript i represents the i-th RTT. The above observation information obs i It may include but is not limited to the following information: Current RTT start timestamp t i , actual sending rate s i , the actual receiving rate r i , the actual packet loss rate l i , the maximum round-trip delay observed within the current RTT is rtt_max i , the minimum round-trip delay observed within the current RTT rtt_min i , the average round-trip delay observed within the current RTT is rtt_avg i .

[0081] As an example, the above congestion control model can adopt BBR (Bottleneck Bandwidth and Round-trip Propagation Time) V3, and other congestion control models can also be adopted. For example, more reasonable network measurement estimation information and observation information can be provided, which will help improve the accuracy of the network measurement model. This disclosure is not limited to this.

[0082] In step S302, input information at the current measurement moment is determined based on the estimated information and the observed information at the current measurement moment.

[0083] As an example, the granularity of the measurement time can be set as needed, such as the granularity of simulated network loss data or the granularity of RTT, which is not limited in the present disclosure.

[0084] As an example, assuming that the measurement time is based on the granularity of the simulated network loss data, the above estimation information est i and observation information obs i , obtain network measurement information with RTT as the granularity <est i ,obs i >, then the network measurement information pair corresponding to the round trip time of the predetermined number of data packets transmitted starting from the current measurement time can be used as the input information of the current measurement time. The granularity of the above simulated network loss data can be set as needed, such as 1 millisecond or other time lengths, which is not limited by the present disclosure.

[0085] As an example, assuming that the measurement time is based on RTT granularity, the current measurement time can be the start timestamp of each RTT.i and observation information obs i , you can get network measurement information with RTT as the granularity <est i ,obs i >, and then use the network measurement information pair corresponding to the round-trip time of a predetermined number of data packets transmitted after the start timestamp of each RTT as the input information of the start timestamp of each RTT.

[0086] As an example, the above-mentioned predetermined number can be set as needed, such as being set to 5.

[0087] In step S303, the input information is input into the network measurement model to obtain the network loss data of the target network at the current measurement time, wherein the network loss data refers to the data characterizing the real-time quality of the target network; wherein the network measurement model is trained using the above-mentioned network measurement model training method.

[0088] According to an exemplary embodiment of the present disclosure, the above-mentioned network loss data may include but is not limited to: network loss bandwidth, network loss delay and network loss packet loss.

[0089] As an example, the above network measurement model may include multiple parts, each part corresponds to a type of network loss data. Assuming there are three types of network loss data, the network measurement model includes three parts, each of which may be composed of a GRU recurrent neural network and a fully connected neural network.

[0090] It should be noted that the GRU recurrent neural network can also be replaced by other networks, such as long short-term memory network (Long-Short Term Memory, abbreviated as LSTM), Transformer, etc., which is not limited in the present disclosure.

[0091] In summary, the present disclosure constructs a large network loss data training set for network measurement for the first time, which helps the network measurement model to fully perceive the characteristic differences under various networks, that is, to generate a simulated network locally through a large-scale network loss simulator, and run application software on the simulated network to obtain observation information, estimation information and network real labels. In addition, a neural network training mode suitable for network measurement data reasoning is also designed, that is, based on the network measurement information pairs obtained from the observation information and the estimation information, and based on the network measurement information pairs and the real network loss labels, the initial network measurement model is trained to learn the mapping relationship between the network flow measurement information pairs and the real network loss labels, and then the network measurement information pairs collected online are input into the trained network measurement model to obtain more accurate real network information, and a large-scale online real user weak network database is established, thereby greatly improving the measurement granularity and accuracy of traditional methods, and providing strong support for local restoration of online real networks.

[0092] It should be noted that the evaluation shows that the bandwidth prediction accuracy of the network measurement model of the present disclosure on the validation set is 85%, the physical delay prediction accuracy is 88%, and the packet loss prediction accuracy is 90%. The prediction accuracy is calculated using the mean percentage error (MAPE). At the same time, through the online network feature extraction of the present disclosure on the online business, a real network database with a length of about 6 million seconds was obtained. By using this real network database, targeted optimization was performed locally, reducing the on-demand freeze rate by about 10%.

[0093] Figure 4 A block diagram of a training device for a network measurement model according to an exemplary embodiment of the present disclosure is shown. Figure 4 The device includes a first acquisition unit 40, a first measurement unit 42 and an adjustment unit 44.

[0094] The first acquisition unit 40 is configured to acquire a training sample set, wherein the training sample set includes input samples and simulated network loss data at multiple measurement moments, and the input samples at each measurement moment are determined based on simulated estimation information and simulated observation information of the simulated network at the corresponding measurement moment, respectively, the simulated estimation information is network state information estimated by using a congestion control model in the simulated network, the simulated observation information is information related to data packet transmission observed in the simulated network, and the simulated network loss data refers to data characterizing the real-time quality of the simulated network; the first measurement unit 42 is configured to input the input sample into the network measurement model to obtain the estimated network loss data of the simulated network at a predetermined measurement moment, wherein the predetermined measurement moment is the measurement moment corresponding to the input sample; the adjustment unit 44 is configured to adjust the parameters of the network measurement model based on the loss of the estimated network loss data and the simulated network loss data at the predetermined measurement moment.

[0095] According to an exemplary embodiment of the present disclosure, the first acquisition unit 40 is also configured to acquire simulation estimation information and simulation observation information in the following manner: when the target application software is running on a simulated network, confirmation messages of multiple simulated data packets are acquired; for each confirmation message, the confirmation message is input into a congestion control model to obtain simulation estimation information corresponding to the round-trip time of the corresponding simulated data packet in the simulated network; when the target application software is running on a simulated network, the simulated network that transmits multiple simulated data packets is monitored to acquire simulation observation information corresponding to the round-trip time of each simulated data packet in the simulated network; and the simulation estimation information and simulation observation information corresponding to the round-trip delay of the simulated data packet transmitted starting from each measurement moment are respectively determined as the simulation estimation information and simulation observation information at each measurement moment.

[0096] According to an exemplary embodiment of the present disclosure, the first acquisition unit 40 is configured to acquire input samples in the following manner, including: determining the simulated estimation information and simulated observation information corresponding to the round-trip time of a predetermined number of simulated data packets transmitted from each measurement moment as the input samples at each measurement moment.

[0097] According to an exemplary embodiment of the present disclosure, a simulated network is generated in the following manner: a simulated network loss data set is obtained, wherein the simulated network loss data set includes a plurality of simulated network loss data, and each type of simulated network loss data has a granularity of a predetermined time length; a simulated network is generated based on a plurality of simulated network loss data divided from each type of simulated network loss data and a preset distribution curve for each type of simulated network loss data.

[0098] According to an exemplary embodiment of the present disclosure, the simulated network loss data includes at least one of simulated network loss bandwidth, simulated network loss delay, and simulated network loss packet loss.

[0099] Figure 5 A block diagram of a network measurement device according to an exemplary embodiment of the present disclosure is shown. Figure 5 The device includes a second acquisition unit 50, a determination unit 52 and a second measurement unit 54.

[0100] The second acquisition unit 50 is configured to acquire the estimated information and observed information of the target network at the current measurement moment when the target application software is running on the target network, wherein the estimated information is the network status information estimated by using the congestion control model on the target network, and the observed information is the data packet transmission related information observed on the target network; the determination unit 52 is configured to determine the input information at the current measurement moment based on the estimated information and the observed information at the current measurement moment; the second measurement unit 54 is configured to input the input information into the network measurement model to obtain the network loss data of the target network at the current measurement moment, wherein the network loss data refers to the data characterizing the real-time quality of the target network; wherein the network measurement model is trained using the network measurement model training method disclosed above.

[0101] According to an embodiment of the present disclosure, an electronic device may be provided. Figure 6 According to an exemplary embodiment of the present disclosure, a block diagram of an electronic device 600 is shown, and the electronic device includes at least one memory 601 and at least one processor 602, wherein the at least one memory stores a set of computer executable instructions, and when the computer executable instruction set is executed by the at least one processor, a training method for a network measurement model and a network measurement method according to an embodiment of the present disclosure are executed.

[0102] As an example, the electronic device 600 may be a PC, a tablet device, a personal digital assistant, a smart phone, or other device capable of executing the above instruction set. Here, the electronic device 1000 is not necessarily a single electronic device, but may also be any device or circuit capable of executing the above instruction (or instruction set) individually or in combination. The electronic device 600 may also be part of an integrated control system or system manager, or may be configured as a portable electronic device interconnected with a local or remote (e.g., via wireless transmission) interface.

[0103] In the electronic device 600, the processor 602 may include a central processing unit (CPU), a graphics processing unit (GPU), a programmable logic device, a dedicated processor system, a microcontroller or a microprocessor. As an example and not limitation, the processor 602 may also include an analog processor, a digital processor, a microprocessor, a multi-core processor, a processor array, a network processor, etc.

[0104] The processor 602 can execute instructions or codes stored in the memory, wherein the memory 601 can also store data. Instructions and data can also be sent and received over the network via the network interface device, wherein the network interface device can use any known transmission protocol.

[0105] The memory 601 may be integrated with the processor 602, for example, by placing RAM or flash memory within an integrated circuit microprocessor or the like. In addition, the memory 601 may include a separate device, such as an external disk drive, a storage array, or any other storage device that can be used by a database system. The memory 601 and the processor 602 may be operationally coupled, or may communicate with each other, such as through an I / O port, a network connection, etc., so that the processor 602 can read files stored in the memory 601.

[0106] In addition, the electronic device 600 may also include a video display (such as a liquid crystal display) and a user interaction interface (such as a keyboard, a mouse, a touch input device, etc.) All components of the electronic device may be connected to each other via a bus and / or a network.

[0107] According to an embodiment of the present disclosure, a computer-readable storage medium may also be provided, wherein when the instructions in the computer-readable storage medium are executed by at least one processor, the at least one processor is prompted to execute the training method of the network measurement model and the network measurement method of the embodiment of the present disclosure. Examples of computer-readable storage media here include: read-only memory (ROM), random access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, non-volatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, Blu-ray or optical disk storage, hard disk drive (HDD), solid state drive (SSD), card storage (such as, multimedia card, secure digital (SD) card or extreme digital (XD) card), magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid state disk and any other device, any other device is configured to store computer programs and any associated data, data files and data structures in a non-transitory manner and provide the computer programs and any associated data, data files and data structures to a processor or computer so that the processor or computer can execute the computer program. The computer program in the above-mentioned computer-readable storage medium can be run in an environment deployed in a computer device such as a client, a host, an agent device, a server, etc. In addition, in one example, the computer program and any associated data, data files and data structures are distributed on a networked computer system, so that the computer program and any associated data, data files and data structures are stored, accessed and executed in a distributed manner by one or more processors or computers.

[0108] According to an embodiment of the present disclosure, a computer program product is provided, including computer instructions, which, when executed by a processor, implement the network measurement model training method and the network measurement method of the embodiment of the present disclosure.

[0109] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the appended claims.

[0110] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A method for training a network measurement model, characterized in that: include: Acquire a training sample set, wherein the training sample set includes input samples and simulated network loss data at multiple measurement moments, the input samples at each measurement moment are determined based on simulated estimation information and simulated observation information of the simulated network at the corresponding measurement moment, the simulated estimation information is network state information estimated by using a congestion control model in the simulated network, the simulated observation information is information related to data packet transmission observed in the simulated network, and the simulated network loss data refers to data characterizing the real-time quality of the simulated network; Inputting the input sample into a network measurement model to obtain estimated network loss data of the simulated network at a predetermined measurement time, wherein the predetermined measurement time is a measurement time corresponding to the input sample; Based on the loss of the estimated network loss data and the simulated network loss data at the predetermined measurement time, the parameters of the network measurement model are adjusted.

2. The training method according to claim 1, characterized in that: The simulated estimation information and simulated observation information at each measurement moment are obtained as follows: When the target application software is running on the simulated network, obtaining confirmation messages of multiple simulated data packets; for each confirmation message, inputting the confirmation message into the congestion control model to obtain simulation estimation information corresponding to the round trip time of the corresponding simulated data packet of the simulated network; When the target application software is running on the simulated network, monitoring the simulated network that transmits the plurality of simulated data packets, and obtaining simulated observation information corresponding to the round trip time of each simulated data packet of the simulated network; The simulated estimation information and the simulated observation information corresponding to the round trip delay of the simulated data packet transmitted starting from each measurement moment are respectively determined as the simulated estimation information and the simulated observation information at each measurement moment.

3. The training method according to claim 2, characterized in that: The input sample is obtained by: The simulated estimation information and the simulated observation information corresponding to the round trip time of a predetermined number of simulated data packets transmitted starting from each measurement moment are determined as input samples at each measurement moment.

4. The training method according to claim 1, characterized in that: The simulated network is generated in the following way: Acquire a simulated network loss data set, wherein the simulated network loss data set includes a plurality of simulated network loss data, and each type of simulated network loss data has a granularity of a predetermined time length; A simulated network is generated based on a plurality of simulated network loss data divided for each type of simulated network loss data and a preset distribution curve for each type of simulated network loss data.

5. The training method according to claim 1, characterized in that: The simulated network loss data includes at least one of simulated network loss bandwidth, simulated network loss delay and simulated network loss packet loss.

6. A network measurement method, characterized in that: include: When the target application software is running on the target network, obtaining estimated information and observed information of the target network at the current measurement time, wherein the estimated information is network state information estimated by using a congestion control model on the target network, and the observed information is information related to data packet transmission observed on the target network; Determining input information at the current measurement time based on the estimated information and the observed information at the current measurement time; Inputting the input information into a network measurement model to obtain network loss data of the target network at the current measurement time, wherein the network loss data refers to data characterizing the real-time quality of the target network; The network measurement model is trained using the network measurement model training method according to any one of claims 1 to 5.

7. A training device for a network measurement model, characterized in that: include: A first acquisition unit is configured to acquire a training sample set, wherein the training sample set includes input samples and simulated network loss data at multiple measurement moments, the input samples at each measurement moment are determined based on simulated estimation information and simulated observation information of the simulated network at the corresponding measurement moment, the simulated estimation information is network state information estimated using a congestion control model under the simulated network, the simulated observation information is information related to data packet transmission observed under the simulated network, and the simulated network loss data refers to data characterizing the real-time quality of the simulated network; A first measurement unit is configured to input the input sample into a network measurement model to obtain estimated network loss data of the simulated network at a predetermined measurement time, wherein the predetermined measurement time is a measurement time corresponding to the input sample; The adjusting unit is configured to adjust the parameters of the network measurement model based on the estimated network loss data and the loss of the simulated network loss data at the predetermined measurement time.

8. A network measurement device, characterized in that: include: A second acquisition unit is configured to acquire, when the target application software is running on the target network, estimated information and observed information of the target network at a current measurement moment, wherein the estimated information is network state information estimated by using a congestion control model on the target network, and the observed information is information related to data packet transmission observed on the target network; a determining unit configured to determine input information at the current measurement moment based on the estimated information and the observed information at the current measurement moment; a second measurement unit configured to input the input information into a network measurement model to obtain network loss data of the target network at the current measurement moment, wherein the network loss data refers to data characterizing the real-time quality of the target network; The network measurement model is trained using the network measurement model training method according to any one of claims 1 to 5.

9. An electronic device, characterized in that: include: at least one processor; at least one memory storing computer executable instructions, When the computer executable instructions are executed by the at least one processor, the at least one processor is prompted to execute the network measurement model training method according to any one of claims 1 to 5 and the network measurement method according to claim 6.

10. A computer-readable storage medium, characterized in that: When the instructions in the computer-readable storage medium are executed by at least one processor, the at least one processor is prompted to execute the network measurement model training method according to any one of claims 1 to 5 and the network measurement method according to claim 6.

11. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by a processor, the training method of the network measurement model according to any one of claims 1 to 5 and the network measurement method according to claim 6 are implemented.

Images