Multipath transmission congestion control method and device

Abstract

The present application provides a multipath transmission congestion control method and device. The method includes: obtaining real-time status information of transmission substreams in multipath transmission, and calculating the network fairness value of multipath transmission at the current moment according to the data throughput in the real-time status information, wherein the real-time status information characterizes the state characteristics of the transmission substream when performing data transmission; generating a network fairness prediction result of multipath transmission under each preset parameter according to the real-time status information and the network fairness value; and determining the congestion window size of the transmission substream at the next moment according to the network fairness prediction result. The method of the present application improves the network fairness and competitiveness in multipath transmission congestion and optimizes the multipath transmission congestion control mechanism.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a multi-path transmission congestion control method and device. Background Art

[0002] Multipath transmission is a method of configuring multiple transmission paths (transmission sub-streams) between the sender and the receiver to achieve efficient and reliable data transmission. The Multi Path Transmission Control Protocol (MPTCP) is a multipath transmission protocol standard developed by the Internet Engineering Task Force. The congestion control algorithm based on the multipath transmission protocol is an important factor affecting the performance of multipath transmission.

[0003] In the prior art, the congestion control method based on MPTCP usually adopts the chain growth algorithm and the non-coupling algorithm, or estimates the network congestion state and implements the window adjustment according to the change of link delay. However, the above method needs to passively wait until the retransmission timeout, or repeated confirmation, or when the link delay increases, in order to determine the network congestion. Not only is the control strategy too conservative, it is difficult to obtain sufficient network resources in the competition, and the competitiveness is not high; and when the MPTCP transmission sub-flow competes with the traditional TCP flow for the shared bottleneck link bandwidth, the fairness of the network cannot be guaranteed.

[0004] That is, the congestion control method based on MPTCP in the prior art has the technical problem of low overall performance. Summary of the invention

[0005] The present application provides a multi-path transmission congestion control method and device to solve the problem of low comprehensive performance of the congestion control method based on MPTCP in the prior art.

[0006] On the one hand, the present application provides a multipath transmission congestion control method, comprising:

[0007] Acquire real-time status information of a transmission substream in multipath transmission, and calculate a network fairness value of the multipath transmission at a current moment according to the data throughput in the real-time status information, wherein the real-time status information represents a status feature of the transmission substream when performing data transmission;

[0008] Generate a network fairness prediction result of multipath transmission under each preset parameter according to the real-time status information and the network fairness value;

[0009] The congestion window size of the transmission sub-flow at the next moment is determined according to the network fairness prediction result.

[0010] Optionally, calculating the network fairness value of multipath transmission at the current moment according to the data throughput in the real-time status information includes:

[0011] Performing a summation operation on the data throughput of the transmission sub-flows at the current moment in the multi-path transmission to obtain summation result information;

[0012] Logarithmic operation is performed on the summation result information to obtain the network fairness value of the multipath transmission at the current moment.

[0013] Optionally, generating a network fairness prediction result of multipath transmission under each preset parameter according to the real-time status information and the network fairness value includes:

[0014] Determining a preset congestion window size of the transmission substream under each preset parameter;

[0015] Bringing the real-time status information and the preset congestion window size into a pre-trained fairness prediction model, and outputting a network fairness prediction value corresponding to multipath transmission under each preset congestion window size;

[0016] According to the network fairness prediction value and the network fairness value, a network fairness prediction result of multipath transmission under each preset parameter is calculated.

[0017] Optionally, the step of bringing the real-time status information and the preset congestion window size into a pre-trained fairness prediction model and outputting a network fairness prediction value corresponding to multipath transmission under each preset congestion window size includes:

[0018] According to the real-time status information and the preset congestion window size, obtaining a local feature vector of multipath transmission at the current moment under each preset parameter;

[0019] The local feature vector is expressed as a local feature matrix, the local feature matrix is ​​dimensionally transformed, and the transformation result is sequentially normalized and multi-head parallel attention calculated to determine the prediction target; the prediction target is linearly transformed to obtain a linear transformation result;

[0020] The linear transformation result is normalized and the network fairness prediction value is output.

[0021] Optionally, the acquiring, according to the real-time status information and the preset congestion window size, a local feature vector of multipath transmission at the current moment under each preset parameter includes:

[0022] Mapping the real-time status information and the preset congestion window size to form an input sequence of multipath transmission under each preset parameter;

[0023] Perform a convolution operation on each vector in each of the input sequences to obtain local information of each vector convolution kernel size k, where k is a natural number;

[0024] According to the reset gate parameter and the update gate parameter, each of the vectors and the corresponding local information are processed respectively to obtain the local information vector after each vector is processed;

[0025] Performing a linear operation on the local information vector to obtain a fitted local information vector;

[0026] A time position code representing transmission time information is added to the fitted local information vector to generate a final local feature vector of each vector, that is, a local feature vector of multipath transmission at the current moment under each preset parameter.

[0027] Optionally, determining a preset congestion window size of the transmission substream under each preset parameter includes:

[0028] According to the congestion window size of the transmission substream at the current moment and the preset relationship, the preset congestion window size of the transmission substream under each preset parameter is calculated, and the preset relationship represents the functional relationship between the congestion window size at the current moment, the preset parameters, and the preset congestion window size.

[0029] Optionally, the real-time status information includes: data throughput, transmission rate, average round-trip delay, and congestion window size.

[0030] On the other hand, the present application provides a multipath transmission congestion control device, comprising:

[0031] A data unit, used to obtain real-time status information of a transmission substream in multipath transmission, and calculate a network fairness value of the multipath transmission at a current moment according to the data throughput in the real-time status information, wherein the real-time status information represents a status feature of the transmission substream when performing data transmission;

[0032] A fairness prediction unit, configured to generate a network fairness prediction result of multipath transmission under each preset parameter according to the real-time status information and the network fairness value;

[0033] The control unit is used to determine the congestion window size of the transmission sub-flow at the next moment according to the network fairness prediction result.

[0034] Optionally, the data unit includes a first processing module and a second processing module;

[0035] The first processing module is used to perform a summation operation on the data throughput of the transmission sub-flows at the current moment in the multi-path transmission to obtain summation result information;

[0036] The second processing module is used to perform logarithmic operation on the summation result information to obtain the network fairness value of the multipath transmission at the current moment.

[0037] Optionally, the fairness prediction unit includes a determination module, a prediction module, and a result module;

[0038] The determining module is used to determine the preset congestion window size of the transmission sub-stream under each preset parameter;

[0039] The prediction module is used to bring the real-time status information and the preset congestion window size into a pre-trained fairness prediction model, and output a network fairness prediction value corresponding to multipath transmission under each preset congestion window size;

[0040] The result module is used to calculate the network fairness prediction result of multipath transmission under each preset parameter according to the network fairness prediction value and the network fairness quantity value.

[0041] Optionally, the prediction module includes a local feature extraction layer, a coding layer, and a linear regression layer;

[0042] The local feature extraction layer is used to obtain the local feature vector of multipath transmission at the current moment under each preset parameter according to the real-time status information and the preset congestion window size.

[0043] The encoding layer is used to represent the local feature vector as a local feature matrix, perform dimension conversion on the local feature matrix, and perform normalization calculation and multi-head parallel attention calculation on the conversion result in sequence to determine the prediction target; perform linear transformation on the prediction target to obtain the linear transformation result;

[0044] The linear regression layer is used to normalize the linear transformation result and output the network fairness prediction value.

[0045] Optionally, the local feature extraction layer includes a convolution operation module, a gating module, a linear change module, and a time position encoding module;

[0046] The convolution operation module is used to map the real-time status information and the preset congestion window size to form an input sequence of multipath transmission under each preset parameter; and to perform a convolution operation on each vector in each of the input sequences to obtain local information of each vector convolution kernel size k, where k is a natural number;

[0047] The gate control module is used to process each vector and the corresponding local information according to the reset gate parameter and the update gate parameter, and obtain the local information vector after each vector is processed;

[0048] The linear change module is used to perform a linear operation on the local information vector to obtain a fitted local information vector;

[0049] The time position coding module is used to add the time position coding representing the transmission time information to the fitted local information vector to generate the final local feature vector of each vector, that is, the local feature vector of the multipath transmission at the current moment under each preset parameter.

[0050] Optionally, the determination module includes a calculation module;

[0051] The calculation module is used to calculate the preset congestion window size of the transmission substream under each preset parameter according to the congestion window size of the transmission substream at the current moment and the preset relationship, wherein the preset relationship represents the functional relationship between the congestion window size at the current moment, the preset parameter, and the preset congestion window size.

[0052] Optionally, the real-time status information includes: data throughput, transmission rate, average round-trip delay, and congestion window size.

[0053] On the other hand, the present application also provides a control device, the control device comprising: a processor, and a memory communicatively connected to the processor;

[0054] The memory stores computer-executable instructions;

[0055] The processor executes the computer-executable instructions stored in the memory to implement any of the methods described above.

[0056] On the other hand, the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement any of the methods described above.

[0057] On the other hand, the present application also provides a computer program product, including a computer program, which is used to implement any of the aforementioned methods when executed by a processor.

[0058] The multipath transmission congestion control method and device provided by the present application obtains the real-time status information of the transmission substream in the multipath transmission, and calculates the network fairness value of the multipath transmission at the current moment according to the data throughput in the real-time status information, wherein the real-time status information characterizes the state characteristics of the transmission substream when performing data transmission; then, based on the real-time status information and the network fairness value, generates the network fairness prediction result of the multipath transmission under each preset parameter; finally, based on the network fairness prediction result, determines the congestion window size of the transmission substream at the next moment. The congestion window adjustment mechanism of the present application obtains the network fairness prediction result according to the real-time status information of the transmission substream, thereby determining the congestion window size of the transmission substream at the next moment, and strongly correlates the adjustment mechanism with the real-time status information of the transmission substream, so as to obtain the best network fairness, improve network fairness and competitiveness, and optimize the multipath transmission congestion control mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0060] Figure 1 A flowchart of a multipath transmission congestion control method provided in an embodiment of the present application;

[0061] Figure 2 A flowchart of another multipath transmission congestion control method provided in an embodiment of the present application;

[0062] Figure 3 An architectural diagram of a multipath transmission congestion control device provided in an embodiment of the present application;

[0063] Figure 4 An architectural diagram of a network fairness prediction sub-device provided in an embodiment of the present application;

[0064] Figure 5 A schematic diagram of the structure of a multipath transmission congestion control device provided in an embodiment of the present application;

[0065] Figure 6 A schematic diagram of the structure of another multipath transmission congestion control device provided in an embodiment of the present application;

[0066] Figure 7 A schematic diagram of the structure of a control device provided in an embodiment of the present application.

[0067] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0068] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0069] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application 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 interchangeable where appropriate, so that the embodiments of the present invention described herein can, for example, be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0070] Multipath transmission is a method of configuring multiple transmission paths between the sender and the receiver to achieve efficient and reliable data transmission. In multipath transmission, each transmission path is also called a transmission substream. MPTCP (Multipath Transmission Control Protocol) is a multipath transmission protocol standard developed by the Internet Engineering Task Force. The congestion control algorithm based on the multipath transmission protocol is an important factor affecting the performance of multipath transmission. Among them, congestion control is a mechanism to control network congestion by adjusting the size of the congestion window.

[0071] In the congestion control algorithm, the congestion window size is usually adjusted according to factors such as the number of ACKs received, round-trip delay, and packet loss, that is, the number of data packets that can be sent before the next ACK is received is limited. Among them, ACK is the full name of Acknowledgement in English. It is a positive feedback mechanism. After receiving the data, the receiver will reply to the sender with a message. The number of ACKs is the number of confirmation packets received by the sender. The round-trip delay (RTT) is the duration in milliseconds required for a network request to reach the destination from the starting point and return to the starting point again. It is the sum of the forward delay and the backward delay. Packet loss refers to the phenomenon that one or more data packets cannot reach the destination through the network.

[0072] In one example, whether a link is congested is determined by judging whether packet loss occurs, and then congestion control is performed. This congestion control mechanism is called uncoupled TCP. This method does not consider the impact of the difference in round-trip delay between transmission sub-streams on congestion control, resulting in poor results in actual application scenarios.

[0073] In another example, based on packet loss judgment, the difference in round-trip delay between transmission sub-streams is also taken into consideration. This congestion control mechanism is called the Linked Increase Algorithm (LIA). Although this algorithm takes into account the difference in round-trip delay between transmission sub-streams, it still uses packet loss as the main basis for determining network congestion, and the occurrence of packet loss does not necessarily mean network congestion. One-sidedly attributing the cause of packet loss to network congestion will generate unnecessary congestion window adjustment actions, which will in turn affect network transmission efficiency.

[0074] In another example, the network congestion state is estimated according to the change of link delay and the window adjustment is implemented. This method is called wVegas congestion control method. This method reduces the window operation when it finds that the data transmission delay increases. The control strategy is too conservative and is at a disadvantage in the competition with the congestion control algorithm based on packet loss. It is difficult to obtain sufficient network resources in the competition, resulting in poor network performance.

[0075] In addition, MPTCP (Multipath Transmission Control Protocol) also requires: while giving full play to the advantages of multipath bandwidth aggregation and improving session throughput, it is also necessary to ensure network fairness when MPTCP transmission sub-flows compete with traditional TCP flows for shared bottleneck link bandwidth.

[0076] Therefore, in order to solve the above problems, the present application proposes a multipath transmission congestion control method for the resource competition problem in the multipath transmission process and the fairness guarantee problem of MPTCP transmission substreams and traditional TCP streams. The multipath transmission congestion control method proposed in the present application generates a network fairness prediction result based on the real-time status information of the transmission substream in the multipath transmission and the calculated network fairness value of the multipath transmission, and determines the congestion window size of the transmission substream at the next moment based on the network fairness prediction result, so as to perform congestion control. The method of the present application optimizes the congestion window adjustment mechanism and solves the resource competition problem and network fairness problem in the multipath transmission process to a certain extent.

[0077] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0078] Figure 1 A flowchart of a multipath transmission congestion control method provided in an embodiment of the present application. The execution subject of this embodiment may be a multipath transmission congestion control device, which may be located on a server or other terminal device. The following is a detailed description using the execution subject being a multipath transmission congestion control device as an example.

[0079] like Figure 1 As shown, the multipath transmission congestion control method provided by this embodiment includes:

[0080] S101. Acquire real-time status information of transmission sub-flows in multi-path transmission, and calculate the network fairness value of multi-path transmission at the current moment according to the data throughput in the real-time status information, wherein the real-time status information represents the status characteristics of the transmission sub-flows during data transmission.

[0081] Exemplarily, in multipath transmission, the transmission path is also called a transmission sub-stream. There are as many transmission sub-streams as there are transmission paths. The present application does not limit the specific number of transmission sub-streams in multipath transmission. The number n of transmission sub-streams is a natural number greater than or equal to 2. For example, n=3 indicates that there are 3 transmission sub-streams in the multipath transmission.

[0082] Exemplarily, the multipath transmission congestion control device can collect the real-time status information of each transmission substream in the multipath transmission from the network terminal through a packet capture tool to obtain the real-time status information of all transmission substreams in the multipath transmission. Among them, the real-time status information characterizes the state characteristics of the transmission substream when performing data transmission. It can be seen from the real-time status information of the transmission substream that when the data is transmitted through multipath, each transmission substream has information such as data throughput, transmission rate, average round-trip delay, congestion window size, etc. at each moment. The more comprehensive the content contained in the real-time status information, the more accurate the analysis result when performing data analysis. In one example, the real-time status information includes: data throughput, transmission rate, average round-trip delay, and congestion window size.

[0083] After the multipath transmission congestion control device obtains the real-time status information of each transmission substream in the multipath transmission, it calculates the network fairness value of the multipath transmission at the current moment according to the data throughput at the current moment in the real-time status information of the transmission substream. Among them, the network fairness value is the network fairness value about the multipath transmission. Since there are multiple transmission substreams, it is necessary to process the data throughput of the transmission substream first and then calculate it. The present application does not restrict how to calculate the network fairness value of the multipath transmission at the current moment according to the data throughput in the real-time status information, and aims to make the calculated network fairness value better reflect the network fairness of the multipath transmission.

[0084] In one example, calculating the network fairness value of multipath transmission at the current moment according to the data throughput in the real-time status information includes:

[0085] The data throughput of the transmission sub-flows in the multi-path transmission at the current moment is summed up to obtain summed result information; and the summed result information is then logarithmically processed to obtain the network fairness value of the multi-path transmission at the current moment.

[0086] Exemplarily, in the acquired multipath transmission, the real-time status information of each transmission substream includes the data throughput of the transmission substream at the current moment. Therefore, when calculating the network fairness value of the multipath transmission at the current moment, the data throughput of all transmission substreams at the current moment can be added up first, and a sum operation can be performed to obtain the sum of the data throughput of all transmission substreams of the multipath transmission at the current moment; then, a logarithmic operation is performed on the sum result, and the result of the logarithmic operation is the network fairness value of the multipath transmission at the current moment.

[0087] S102: Generate a network fairness prediction result of multipath transmission under each preset parameter according to the real-time status information and the network fairness value.

[0088] Exemplarily, a plurality of sets of preset parameters are set in the multipath transmission congestion control device, wherein the preset parameter can be a preset congestion window size, or a preset adjustment coefficient for determining the adjustment direction and amplitude of the congestion window. If the preset parameter is a preset congestion window size, the network fairness prediction result of the multipath transmission under each preset congestion window size is directly generated. The method is simple to calculate and fast, but it is not associated with the congestion window size at the current moment and is not timely. If the preset parameter is a preset adjustment coefficient for determining the adjustment direction and amplitude of the congestion window, it is necessary to first determine the preset congestion window size of each transmission substream under each preset adjustment coefficient according to the congestion window size of each transmission substream at the current moment and the preset adjustment coefficient, and then generate the network fairness prediction result of the multipath transmission under the preset congestion window size corresponding to each preset parameter; the method is relatively complex to calculate, but it is effectively associated with the congestion window size of each transmission substream at the current moment, not only with timeliness, but also makes the network fairness prediction result more in line with the actual application scenario, and can better cope with complex and changeable network environments. In practical applications, it can be flexibly set according to the needs of specific application scenarios.

[0089] Exemplarily, the multipath transmission congestion control device generates a network fairness prediction result of the multipath transmission under each preset parameter based on the acquired real-time status information of all transmission substreams in the multipath transmission and the calculated network fairness value. That is, the network fairness prediction result of the multipath transmission under each preset parameter, that is, under each preset congestion window size can be obtained. Among them, the network fairness prediction result represents the accuracy of the network fairness of the multipath transmission under the preset congestion window size based on the real-time status information of all transmission substreams and the network fairness value. The smaller the value of the network fairness prediction result, the higher the accuracy, indicating that the network fairness under the prediction parameter is also higher.

[0090] Among them, the present application does not limit how to generate the network fairness prediction result of multipath transmission under each preset parameter. For example, the fairness prediction value of multipath transmission at the next moment predicted based on the preset parameters and the network fairness value at the current moment can be used to calculate the root mean square error between the two, and use the root mean square error as the fairness prediction result. Then, the network fairness prediction value of multipath transmission under each preset parameter can be obtained based on the real-time status information of all transmission sub-streams in the multipath transmission obtained, and the calculated network fairness value, that is, the fairness prediction value of multipath transmission under each preset congestion window size, and then the root mean square error between the fairness prediction value corresponding to each preset parameter and the network fairness value is calculated to obtain the network fairness prediction result corresponding to each preset parameter.

[0091] S103: Determine the congestion window size of the transmission sub-flow at the next moment according to the network fairness prediction result.

[0092] Exemplarily, each preset parameter corresponds to a fairness prediction result, that is, each set of preset congestion window sizes corresponds to a fairness prediction result. The smaller the value of the fairness prediction result, the higher the accuracy, indicating that the network fairness is better. Therefore, the multipath transmission congestion control device compares the network fairness prediction results of multipath transmission under each preset parameter, selects the network fairness prediction result with the smallest value, and uses the corresponding preset parameter to determine the congestion window size of the transmission subflow at the next moment, and then controls the congestion window size of the transmission subflow at the next moment in the multipath transmission.

[0093] The multipath transmission congestion control method provided by the embodiment of the present application obtains the real-time status information of the transmission substream in the multipath transmission, and calculates the network fairness value of the multipath transmission at the current moment according to the data throughput in the real-time status information, wherein the real-time status information characterizes the state characteristics of the transmission substream when performing data transmission; then, based on the real-time status information and the network fairness value, generates the network fairness prediction result of the multipath transmission under each preset parameter; finally, based on the network fairness prediction result, determines the congestion window size of the transmission substream at the next moment. The congestion window adjustment mechanism of the present application obtains the network fairness prediction result according to the real-time status information of the transmission substream, thereby determining the congestion window size of the transmission substream at the next moment, and strongly correlates the adjustment mechanism with the real-time status information of the transmission substream, so as to obtain the best network fairness, improve network fairness and competitiveness, and optimize the multipath transmission congestion control mechanism.

[0094] Figure 2 A flowchart of another multi-path transmission congestion control method provided by an embodiment of the present application is shown in FIG. The execution subject of this embodiment may be a multi-path transmission congestion control device, which may be located on a server or other terminal device.

[0095] In one example, the multipath transmission congestion control device may be divided into a plurality of sub-devices for implementing the following Figure 2 The multipath transmission congestion control method shown. For example, Figure 3 This is a diagram of the architecture of a multi-path transmission congestion control device provided in an embodiment of the present application. Figure 3As shown, the multipath transmission congestion control device includes: a data acquisition and processing sub-device, a fairness prediction sub-device, and a congestion control sub-device. The data acquisition and processing sub-device is used to obtain the real-time status information of the transmission sub-stream, and process it, and provide the real-time status information of the transmission sub-stream and the network fairness value to the fairness prediction sub-device. The fairness prediction sub-device is used to generate and output the network fairness prediction result according to the real-time status information, the network fairness value and the preset congestion window size corresponding to the preset parameters. The congestion control sub-device is used to provide the fairness prediction sub-device with the next moment congestion window size corresponding to the preset parameters, and output the optimal congestion window adjustment mechanism according to the network fairness prediction result to perform congestion control. In addition, each sub-device can also include multiple analysis and processing layers for completing their respective tasks.

[0096] like Figure 2 As shown, the multipath transmission congestion control method provided by this embodiment includes:

[0097] S201, acquiring real-time status information of transmission sub-flows in multi-path transmission, performing summation operation on data throughput of transmission sub-flows in multi-path transmission at a current moment to obtain summation result information; performing logarithmic operation on the summation result information to obtain a network fairness value of multi-path transmission at a current moment.

[0098] For example, the data collection and processing sub-device in the multi-path transmission congestion control device first collects the real-time status information of each transmission sub-flow from the network terminal through a packet capture tool. The real-time status information can be expressed as shown in formula (1): Formula (1) is as follows:

[0099]

[0100] In formula (1), r represents a transmission substream, r∈[1,n], and n is a natural number greater than or equal to 2; t represents the time of obtaining the real-time status information. For example, from the beginning of data transmission of the transmission substream, the real-time status information of each transmission substream is collected in seconds, then t=0, 1, 2, 3...; represents the sending rate of the transmission sub-stream r at time t; represents the data throughput of transmission sub-flow r at time t; represents the average round-trip delay of the transmission subflow r at time t; Represents the congestion window size of the transmission subflow r at time t.

[0101] For example, in this embodiment, n=3, that is, the multipath transmission includes 3 transmission sub-streams, and the data collection and processing sub-device collects the real-time status information of transmission sub-streams 1, 2, and 3. get:

[0102]

[0103]

[0104]

[0105] Then, the data acquisition and processing sub-device uses formula (2) to sum the data throughput of all transmission sub-flows in the multi-path transmission at the current moment, and obtains the sum result information goodput t , formula (2) is as follows:

[0106]

[0107] Goodput in formula (2) t That is, it is the sum of the data throughput of all transmission sub-streams in multi-path transmission at the current moment.

[0108] Then, the summation result information is processed logarithmically using formula (3) to obtain the network fairness value F of multipath transmission at the current moment. Formula (3) is as follows:

[0109] F = log2(goodput t ) (3)

[0110] The F in formula (3) is the network fairness value of multipath transmission at the current moment.

[0111] For example, this embodiment takes multipath transmission including three transmission sub-streams as an example, and the summation result information obtained by using formula (2) is:

[0112]

[0113] Then use formula (3) to get goodput t Logarithmic operation is performed to obtain the network fairness value F of the multi-path transmission including 3 transmission sub-streams.

[0114] Finally, the data acquisition and processing sub-device will collect the real-time status information of all transmission sub-flows in the multi-path transmission at the current moment, and the calculated network fairness value F of the multi-path transmission at the current moment to form a set of data sets D, and input them into the fairness prediction sub-device in the multi-path transmission congestion control device.

[0115] S202: Determine a preset congestion window size of a transmission substream under each preset parameter.

[0116] Exemplarily, the preset parameter is a preset adjustment coefficient for determining the adjustment direction and amplitude of the congestion window. There may be two preset adjustment coefficients, which are respectively used to determine the adjustment direction and amplitude of the congestion window. The congestion control sub-device in the multipath transmission congestion control device calculates the preset congestion window size of each transmission sub-stream under each preset parameter, and then sends it to the fairness prediction sub-device, so that the fairness prediction sub-device obtains the preset congestion window size of each transmission sub-stream under each set of preset parameters.

[0117] In one example, determining a preset congestion window size of a transmission substream under each preset parameter includes:

[0118] According to the congestion window size of the transmission substream at the current moment and the preset relationship, the preset congestion window size of the transmission substream under each preset parameter is calculated, and the preset relationship represents the functional relationship between the congestion window size at the current moment, the preset parameters, and the preset congestion window size.

[0119] Exemplarily, the congestion control sub-device in the multipath transmission congestion control device calculates the preset congestion window size of each transmission sub-stream under each set of preset parameters according to the functional relationship formula (4) of the congestion window size at the current moment, the preset parameters, and the preset congestion window size. Formula (4) is as follows:

[0120]

[0121] In formula (4), represents the congestion window size of the transmission subflow r at time t, It represents the preset congestion window size of the transmission substream r at the next moment calculated based on the preset parameters a and b. a is the multiplicative parameter of the congestion window adjustment, which is used to control the large adjustment of the congestion window. b is the additive parameter of the congestion window adjustment, which is used to control the small fine-tuning of the congestion window size when the network congestion status fluctuates slightly.

[0122] When a < 1, the window is halved; when a > 1, the congestion window size is increased exponentially so that the transmission substream can quickly improve the throughput when the bottleneck link bandwidth is not fully utilized. In the congestion control sub-device, multiple groups of preset multiplicative parameters and additive parameters can be set respectively. A multiplicative parameter and an additive parameter are combined into a group of preset parameters. Each transmission substream has a corresponding preset congestion window size based on each group of preset parameters. There are as many groups of preset parameters as there are groups of preset congestion window sizes for transmission substreams in multipath transmission. According to the number of groups of preset parameters, the number of groups of preset congestion window sizes for all transmission substreams in multipath transmission can be determined.

[0123] For example, according to experimental data, 6 groups of multiplicative parameters a and 5 groups of additive parameters b can be set in advance, which are:

[0124] a={0.5, 0.8, 1, 1.2, 1.5, 2}

[0125] b = {-2, -1, 0, 1, 2}

[0126] The congestion control sub-device generates multiple sets of preset parameters (a, b) by permutation and combination of 6 sets of multiplicative parameters and 5 sets of additive parameters, such as (0.5, -2), (0.5, -1), (0.8, -2), etc. Then, the preset congestion window size of each transmission sub-flow under each set of preset parameters is calculated according to formula (4), and multiple sets of preset congestion window sizes for multi-path transmission are obtained and input into the fairness prediction sub-device.

[0127] S203: Bring the real-time status information and the preset congestion window size into a pre-trained fairness prediction model, and output the network fairness prediction value corresponding to the multi-path transmission under each preset congestion window size.

[0128] Exemplarily, the fairness prediction sub-device in the multi-path transmission congestion control device first inputs the real-time status information input by the data acquisition and processing sub-device, and each set of preset congestion window sizes input by the congestion control sub-device, into a pre-trained fairness prediction model, and obtains the network fairness prediction value corresponding to the multi-path transmission under each set of preset congestion window sizes.

[0129] Among them, the fairness prediction model can be a Transformer model, which is a deep learning model with attention mechanism as the core, and its core lies in the structure of encoder and decoder. The encoder is responsible for mapping the input sequence to the hidden layer, and the decoder is responsible for mapping the hidden layer to the output sequence. The encoder and decoder are both stacked by N layers of sub-layers with the same structure. Among them, each layer of the encoder contains two sub-layers of multi-head self-attention mechanism and feedforward neural network, and each layer of the decoder contains three sub-layers of occlusion multi-head self-attention mechanism, encoder-decoder multi-head attention mechanism, and feedforward neural network. The self-attention mechanism uses scaled dot product attention to calculate the attention value of the feature matrix. The weight coefficient is first calculated by performing dot product and SoftMax normalization on the query matrix and the key matrix, and then the value matrix is ​​weighted and summed according to the weight coefficient. The multi-head self-attention mechanism is the splicing of multiple self-attention mechanisms. Multiple self-attention heads are used to learn the information of different representation subspaces respectively, and then multiple attention values ​​are spliced ​​and linearly transformed to obtain the final attention value, thereby realizing the modeling expression of different constraints. In addition, this application adds residual connection operations between each sub-layer of the encoder and decoder in the Transformer model to focus on the changes in the difference part before and after training, which can improve the training effect; at the same time, each sub-layer also adopts layer normalization operation to accelerate network convergence and improve network generalization ability.

[0130] The present application does not restrict how to output the network fairness prediction value corresponding to the multi-path transmission under each preset congestion window size.

[0131] In one example, the network fairness prediction sub-device may include: a local feature extraction layer, a coding layer, a linear regression layer, and a performance analysis layer. Figure 4 This is an architecture diagram of a network fairness prediction sub-device provided in an embodiment of the present application. Figure 4 As shown, the network fairness prediction sub-device includes: a local feature extraction layer, a coding layer, a linear regression layer and a performance analysis layer. Among them, the local feature extraction layer may include a convolution operation module, a gating module, a linear transformation module and a time position coding module; the coding layer may include 6 identical multi-head self-attention modules and a feedforward neural network module.

[0132] In one example, the real-time status information and the preset congestion window size are introduced into a pre-trained fairness prediction model, and the network fairness prediction value corresponding to the multi-path transmission under each preset congestion window size is output, which may include:

[0133] S2031. According to the real-time status information and the preset congestion window size, obtain a local feature vector of the multipath transmission at the current moment under each preset parameter.

[0134] For example, the local feature extraction layer is used to obtain the local feature vector of the multipath transmission at the current moment under each preset parameter based on the real-time status information and the preset congestion window size. Specifically, the local feature extraction layer maps the real-time status information of the transmission substream input by the data acquisition and processing sub-device, and the preset congestion window size corresponding to the transmission substream at the next moment under each set of preset parameters input by the congestion control sub-device, into a distributed semantic representation, that is, the real-time status information and the preset congestion window size of the transmission substream are simultaneously represented by a matrix, and then processed, and the local feature vector of the multipath transmission at the current moment under each set of preset parameters is provided to the coding layer in the network fairness prediction sub-device. Since there are multiple sets of preset parameters, there are also multiple sets of local feature vectors of the output multipath transmission at the current moment.

[0135] Exemplarily, the local feature extraction layer may include a convolution operation module, a gating module, a linear transformation module, and a time position encoding module. In one example, according to the real-time status information and the preset congestion window size, obtaining the local feature vector of the multipath transmission at the current moment under each preset parameter may include:

[0136] S11 . Mapping the real-time status information and the preset congestion window size to form an input sequence of multipath transmission under each preset parameter.

[0137] For example, the real-time status information of each transmission sub-stream in the multi-path transmission is And the preset congestion window size value of each transmission sub-flow at the next moment The data set is used as the input of the convolution operation module in the local feature extraction layer, which is represented as the input sequence x t ={x i} t Among them, x i It represents the vectors contained in the input sequence at time t, and its specific number is determined by the amount of information contained in the real-time status information and the number of transmission sub-streams.

[0138] Exemplarily, in this embodiment, multipath transmission includes three transmission sub-streams, and real-time status information includes data throughput, transmission rate, average round-trip delay, and congestion window size. The input sequence can be expressed as:

[0139]

[0140] S12. Perform a convolution operation on each vector in each input sequence to obtain local information of each vector with a convolution kernel size of k, where k is a natural number.

[0141] Exemplarily, the convolution operation module also performs the convolution operation on the input sequence x t ={x i} t Each vector x in i The convolution operation is performed according to the method shown in formula (5) to obtain the local information h of each vector convolution kernel size k i , k is a natural number, formula (5) is as follows:

[0142] h i =conv([x i-k / 2 , ..., x i , ..., x i+k / 2 ]) (5)

[0143] In formula (5), h i is the vector x i The output result of the convolution operation with a kernel size of k.

[0144] The smaller the convolution kernel k, the finer the feature extraction, the better the learning effect, but the slower the calculation speed; the larger the convolution kernel k, the coarser the feature extraction, but the faster the calculation speed. In this application, since the amount of calculation data is not large, k can be set to 2.

[0145] S13. Process each vector and the corresponding local information according to the reset gate parameter and the update gate parameter, and obtain the processed local information vector of each vector.

[0146] Exemplarily, the gating module in the local feature extraction layer can be composed of two control gates: a reset gate and an update gate. The reset gate is used to alleviate the information attenuation of the input vector in the deep neural structure; the update gate is used to control the extent to which the state information of the previous moment is brought into the current state, that is, to help the model decide how much past information to pass to the future, thereby alleviating the gradient disappearance phenomenon in the model.

[0147] In the convolution operation module, the input sequence x t ={x i} t Each vector x in i After performing the convolution operation according to the method shown in formula (5), the reset gate obtains the input sequence x t ={x i} t The vector x in i , and with the vector x i-1 The output h of the corresponding convolution operation module i-1 , and reset the gate parameter r according to the method shown in formula (6) i The calculation of is as follows:

[0148] r i =σ(W r h i-1 +V r x i +b r ) (6)

[0149] In formula (6), r i is the reset gate parameter, σ is the Sigmoid function, W r and V r is the weight parameter, b r is the bias parameter.

[0150] At the same time, the update gate is updated according to the method shown in formula (7) to update the gate parameter u i The calculation of is as follows:

[0151] u i =σ(W u h i-1 +V u x i +b u ) (7)

[0152] In formula (7), u i is the update gate parameter, σ is the Sigmoid function, W u and V u is the weight parameter, bu is the bias parameter.

[0153] in, is a sequence of d real numbers, which is the model variable trained by the machine learning system. The initial real number is 0, which will change during the learning process without external intervention and is determined by the deep learning algorithm. For example, d = 15, is a sequence of 15 real numbers.

[0154] The gate control module determines the reset gate parameter r i and update the gate parameter u i Afterwards, the calculation is performed according to the method shown in formula (8) to obtain the local information vector after each vector processing, and output it as the output result of the gating module. Formula (8) is as follows:

[0155]

[0156] In formula (8), is the vector x i The processed local information vector, h i is the vector x i The output of the corresponding convolution operation module, Represents element-wise multiplication.

[0157] S14. Perform a linear operation on the local information vector to obtain a fitted local information vector.

[0158] For example, in order to enable the fairness prediction model to better mine the features of the input sequence, the training data is fitted. The linear transformation module in the local feature extraction layer performs a linear operation on the output of the gating module, that is, the local information vector after each vector processing, according to the method shown in formula (9), and obtains the fitted local information vector e i , formula (9) is as follows:

[0159]

[0160] In formula (9), e i is the local information vector after fitting, W e is the weight parameter, b e is the bias parameter.

[0161] S15, adding the time position code representing the transmission time information to the fitted local information vector to generate the final local feature vector of each vector, that is, the local feature vector of the multipath transmission at the current moment under each preset parameter.

[0162] Exemplarily, the temporal position encoding module in the local feature extraction layer is used to output the sequence e of the linear transformation module. i The position coding representing the transmission time information is added to the , so that the fairness prediction sub-device can improve the ability of learning sequences through relative positions. The time position coding module represents the time position coding according to the sine and cosine functions of different frequencies shown in formula (10), which is as follows:

[0163]

[0164] In formula (10), PE i For the input sequence x t ={x i} t The position encoding of each vector in Represents dimension. Each dimension of the time position encoding corresponds to a sine curve, whose wavelength forms a geometric series from 2π to 10000·2π. Specifically, if s is an even number, the sin formula is used, and if it is an odd number, the cos formula is used.

[0165] Then, the final local feature vector f of each vector is generated according to formula (11): i , which is the output of the local feature extraction layer, that is, the local feature vector f of multipath transmission at the current moment under each preset parameter i , formula (11) is as follows:

[0166] f i =e i +PE i (11)

[0167] At this point, the extraction of the local feature vector is completed. There are multiple groups of preset parameters, and the local feature vector of the multipath transmission corresponding to each group of preset parameters at the current moment needs to be extracted accordingly.

[0168] S2032. Represent the local feature vector as a local feature matrix, perform dimension conversion on the local feature matrix, and perform normalization calculation and multi-head parallel attention calculation on the conversion result in sequence to determine the prediction target; perform linear transformation on the prediction target to obtain the linear transformation result.

[0169] Exemplarily, the encoding layer in the network fairness prediction sub-device is used to capture the information between the output sequences of the local feature extraction layer through the attention mechanism, that is, the sequence information characteristics between each moment. The encoding layer can be composed of 6 identical sub-layers, each of which can include two modules: a multi-head self-attention module and a feedforward neural network module. The multi-head self-attention module is used to perform deep learning on the input sequence and output the prediction target. Its purpose is to focus on the state information that is more critical to predicting the fairness of the network among a large amount of input information, reduce the attention to other information, solve the problem of information overload, and improve the efficiency and accuracy of task processing. The feedforward neural network module is used to perform data fitting on the prediction target output by the multi-head self-attention module, so that the fairness prediction model can better mine the characteristics of the input sequence and avoid the insufficient fitting degree of the attention mechanism to the complex process.

[0170] For example, the multi-head self-attention module transmits the local feature vector f at the current moment under each set of preset parameters output by the local feature extraction layer through multiple paths. i , the local feature matrix f is represented by the method shown in formula (12) to calculate f t +1 Sequence and f t+2 Sequence and f t The sequence itself relative to f t The importance weight of the sequence, and then the sequence f t 、f t+1 and f t+2 According to the weighted softmax addition, update f with the result t .

[0171] Wherein, formula (12) is as follows:

[0172]

[0173] The multi-head self-attention module first converts f t Convert to d k The query vector Q t , represents f t The vector K of correlation with other information t , and the vector V representing the queried information t , formula (13) is as follows:

[0174]

[0175] In formula (13), is a trainable parameter.

[0176] Then, according to the method shown in formula (14), using Q t , K t , Vt Perform softmax normalization calculation, formula (14) is as follows:

[0177]

[0178] In formula (14), d k = d / H, d represents the output dimension of each sub-layer, which can be set, H represents the number of heads of multi-head parallel attention, that is, the number of input matrices f, Z t The remaining state parameters representing the state parameter W and the importance weights of the state parameter itself relative to the predicted target.

[0179] Exemplarily, in the embodiment of the present application, d is 12 in the training phase, d is 15 in the testing phase, and H=3.

[0180] In addition, in order to prevent data overfitting and improve computational efficiency, the method shown in formula (15) is further used to perform multi-head parallel attention calculation and output the predicted target Z t , formula (15) is as follows:

[0181] MutiHead(Q,K,V)=Concat({Z t} H )W A (15)

[0182] In formula (15), MutiHead(Q, K, V) represents the attention mechanism of all sub-stream state information in the input sequence.

[0183] in, It is set according to the dimensions of the weight parameters and bias parameters. represents a one-dimensional array of real numbers, represents a two-dimensional array of real numbers, Represents a three-dimensional array of real numbers.

[0184] When the multi-head self-attention module transmits the local feature vector f at the current moment under each set of preset parameters output by the local feature extraction layer through multiple paths i It is represented as a local feature matrix f, and its dimension conversion, normalization calculation and multi-head parallel attention calculation are completed to determine the predicted target Z t After that, the feedforward neural network module predicts the target Z according to formula (16). t Perform linear transformation and obtain the linear transformation result g t , where the first layer is activated by the ReLU function and the second layer is a linear function. Formula (16) is as follows:

[0185] g t =FFN(Z t)=W2·ReLU(W1Z t +b1)+b2 (16)

[0186] In formula (16), W1, W2 are weight parameters, b1, b2 are bias parameters.

[0187] At this point, the encoding layer outputs the linear transformation result g corresponding to each set of preset parameters t .

[0188] S2033. Normalize the linear transformation result and output the network fairness prediction value.

[0189] For example, the linear regression layer in the network fairness prediction sub-device uses formula (17) to transform the linear transformation result g output by the encoding layer: t Normalization is performed to eliminate the adverse effects caused by too large or too small sample data. Formula (17) is as follows:

[0190] y t =σ(w0g t +b0) (17)

[0191] In formula (17), w0 is the weight parameter, b0 is the bias parameter, and σ is the Sigmoid function.

[0192] At this point, the linear regression layer outputs the network fairness prediction value y corresponding to each set of preset parameters t .

[0193] S204: Calculate the network fairness prediction result of the multipath transmission under each preset parameter according to the network fairness prediction value and the network fairness value.

[0194] For example, the performance analysis layer in the network fairness prediction sub-device calculates the network fairness prediction value y corresponding to each set of preset parameters output by the linear regression layer. t , and the network fairness value F calculated by the data acquisition and processing sub-device, the root mean square error (RMSE) corresponding to the multipath transmission and each set of preset parameters is calculated according to the method shown in formula (18), that is, the network fairness prediction result of the multipath transmission under each set of preset parameters, formula (18) is as follows:

[0195]

[0196] In formula (18), D is the data set input by the data collection and processing sub-device into the fairness prediction sub-device.

[0197] The smaller the root mean square error is, the higher the network fairness is under the prediction parameters. The root mean square error can be used to more intuitively determine the accuracy of the fairness prediction sub-device in predicting fairness.

[0198] S205: Determine the congestion window size of the transmission sub-flow at the next moment according to the network fairness prediction result.

[0199] Exemplarily, the congestion control sub-device in the multipath transmission congestion control device compares the network fairness prediction results of the multipath transmission under each preset parameter, selects the network fairness prediction result with the smallest value, and uses the corresponding preset parameters to determine the congestion window size of the transmission sub-flow at the next moment, thereby controlling the congestion window size of the transmission sub-flow in the multipath transmission at the next moment.

[0200] In addition, in some embodiments, before the real-time status information and the preset congestion window size are introduced into the pre-trained fairness prediction model and the network fairness prediction value corresponding to the multipath transmission under each preset congestion window size is output, the method further includes: training the fairness prediction model. The training of the fairness prediction model includes a training phase and a testing phase. The training phase mainly determines the weight parameters and bias parameters in the fairness prediction model, and the testing phase mainly tests the accuracy of the output result of the fairness prediction model.

[0201] The training set and test set are generally constructed according to the 7:3 principle. For example, if the training set includes 1400 groups, 600 test sets need to be constructed in the same way. The data set consisting of the calculated network fairness value F of the multipath transmission at the current moment is a training set.

[0202] Different from the testing phase, in the training phase, the input sequence of the convolution operation module includes only the real-time status information of each transmission sub-stream. For example, taking multipath transmission including three transmission sub-streams, and real-time status information including data throughput, transmission rate, average round-trip delay, and congestion window size as an example, the input sequence of the convolution operation module in the training phase can be expressed as:

[0203]

[0204] Then, convolution operation is performed on each vector in the input sequence, the output of the gating module is calculated, the linear transformation result is determined, and the time position encoding is added to obtain the local feature vector f of each vector in the input sequence. i , and further obtain the local feature matrix f. The local feature matrix F in the training stage can be expressed as a 3×12 matrix as follows:

[0205]

[0206] Then, as in the test phase, the local feature matrix f is transformed in dimension, and the transformation results are normalized and multi-head parallel attention is calculated in turn, the prediction target is determined, and the prediction target is linearly transformed to obtain the linear transformation result, and the linear transformation result is normalized, and the network fairness prediction value is output, and then the network fairness prediction result of multi-path transmission is calculated based on the network fairness prediction value and the network fairness value. In this way, after repeated training, the various weight parameters and bias parameters in the fairness prediction model are determined. Among them, the initial values ​​of all weight parameters and bias parameters are set using the normal distribution W, b~U(0,1), that is, the initial values ​​are randomly obtained in (0,1), and then the Adam optimizer is used to calculate the loss function every time the input sequence is deep learned (i.e. repeated training), and the parameters are updated to the most appropriate value.

[0207] The multipath transmission congestion control method provided by the embodiment of the present application obtains the real-time status information of the transmission subflow in the multipath transmission, performs sum operation processing on the data throughput of the transmission subflow in the multipath transmission at the current moment, obtains the sum result information, performs logarithmic operation processing on the sum result information, and obtains the network fairness value of the multipath transmission at the current moment; then determines the preset congestion window size of the transmission subflow under each preset parameter; then brings the real-time status information and the preset congestion window size into a pre-trained fairness prediction model, outputs the network fairness prediction value corresponding to the multipath transmission under each preset congestion window size; and calculates the network fairness prediction result of the multipath transmission under each preset parameter according to the network fairness prediction value and the network fairness value; and then determines the congestion window size of the transmission subflow at the next moment according to the network fairness prediction result. The method of the present application compares the network fairness prediction results of multi-path transmission under different preset parameters and selects the optimal control scheme for congestion control. The method has more selectivity, stronger adaptability, and more versatility, and can be better applied to rapidly changing network scenarios. At the same time, it also effectively ensures the network fairness when the transmission sub-flows in multi-path transmission compete with the traditional TCP flow for the bottleneck link bandwidth, and optimizes the congestion control mechanism.

[0208] The following is an embodiment of the device of the present application, which can be used to execute the embodiment of the method of the present application. For details not disclosed in the embodiment of the device of the present application, please refer to the embodiment of the method of the present application.

[0209] Figure 5 A schematic diagram of the structure of a multi-path transmission congestion control device provided in an embodiment of the present application. Figure 5 As shown, the multipath transmission congestion control device 50 provided in this embodiment includes: a data unit 501 , a fairness prediction unit 502 and a control unit 503 .

[0210] The data unit 501 is used to obtain the real-time status information of the transmission sub-flow in the multi-path transmission, and calculate the network fairness value of the multi-path transmission at the current moment according to the data throughput in the real-time status information, wherein the real-time status information represents the status characteristics of the transmission sub-flow when performing data transmission;

[0211] The fairness prediction unit 502 is used to generate a network fairness prediction result of multipath transmission under each preset parameter according to the real-time status information and the network fairness value;

[0212] The control unit 503 is used to determine the congestion window size of the transmission sub-flow at the next moment according to the network fairness prediction result.

[0213] Figure 6 A schematic diagram of the structure of another multi-path transmission congestion control device provided in an embodiment of the present application. Figure 6 As shown, the multipath transmission congestion control device 60 provided in this embodiment includes: a data unit 601 , a fairness prediction unit 602 and a control unit 603 .

[0214] The data unit 601 is used to obtain the real-time status information of the transmission sub-flow in the multi-path transmission, and calculate the network fairness value of the multi-path transmission at the current moment according to the data throughput in the real-time status information, wherein the real-time status information represents the status characteristics of the transmission sub-flow when performing data transmission;

[0215] The fairness prediction unit 602 is used to generate a network fairness prediction result of multipath transmission under each preset parameter according to the real-time status information and the network fairness value;

[0216] The control unit 603 is used to determine the congestion window size of the transmission sub-flow at the next moment according to the network fairness prediction result.

[0217] In an example, the data unit 601 includes a first processing module 6011 and a second processing module 6012 .

[0218] The first processing module 6011 is used to perform a summation operation on the data throughput of the transmission sub-flows at the current moment in the multi-path transmission to obtain summation result information.

[0219] The second processing module 6012 is used to perform logarithmic operation on the summation result information to obtain the network fairness value of the multi-path transmission at the current moment.

[0220] In one example, the fairness prediction unit 602 includes a determination module 6021 , a prediction module 6022 , and a result module 6023 .

[0221] The determination module 6021 is used to determine the preset congestion window size of the transmission sub-flow under each preset parameter.

[0222] The prediction module 6022 is used to bring the real-time status information and the preset congestion window size into the pre-trained fairness prediction model, and output the network fairness prediction value corresponding to the multi-path transmission under each preset congestion window size.

[0223] The result module 6023 is used to calculate the network fairness prediction result of the multi-path transmission under each preset parameter according to the network fairness prediction value and the network fairness value.

[0224] In one example, the prediction module 6022 includes a local feature extraction layer 60221 , a coding layer 60222 , and a linear regression layer 60223 .

[0225] The local feature extraction layer 60221 is used to obtain the local feature vector of multipath transmission at the current moment under each preset parameter according to the real-time status information and the preset congestion window size.

[0226] The encoding layer 60222 is used to represent the local feature vector as a local feature matrix, perform dimension conversion on the local feature matrix, and perform normalization calculation and multi-head parallel attention calculation on the conversion result in sequence to determine the prediction target; perform linear transformation on the prediction target to obtain the linear transformation result.

[0227] The linear regression layer 60223 is used to normalize the linear transformation results and output the network fairness prediction value.

[0228] In one example, the local feature extraction layer 60221 includes a convolution operation module 1, a gating module 2, a linear change module 3, and a time position encoding module 4.

[0229] The convolution operation module 1 is used to map the real-time status information and the preset congestion window size to form an input sequence for multipath transmission under each preset parameter; and to perform a convolution operation on each vector in each input sequence to obtain local information of each vector convolution kernel size k, where k is a natural number.

[0230] The gate control module 2 is used to process each vector and the corresponding local information according to the reset gate parameter and the update gate parameter, and obtain the local information vector after each vector is processed.

[0231] The linear change module 3 is used to perform linear operation on the local information vector to obtain the fitted local information vector.

[0232] The time position coding module 4 is used to add the time position coding representing the transmission time information to the fitted local information vector to generate the final local feature vector of each vector, that is, the local feature vector of the multipath transmission at the current moment under each preset parameter.

[0233] In one example, the determination module 6021 includes a calculation module 60211.

[0234] The calculation module 60211 is used to calculate the preset congestion window size of the transmission substream under each preset parameter based on the congestion window size of the transmission substream at the current moment and the preset relationship, and the preset relationship represents the functional relationship between the congestion window size at the current moment, the preset parameters, and the preset congestion window size.

[0235] In one example, the real-time status information includes: data throughput, sending rate, average round-trip delay, and congestion window size.

[0236] It should be noted that it should be understood that the division of the various modules of the above device is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. And these modules can all be implemented in the form of software called by processing elements; they can also be all implemented in the form of hardware; some modules can also be implemented in the form of software called by processing elements, and some modules can be implemented in the form of hardware. In addition, it can also be stored in the memory of the above device in the form of program code, and the functions of the above data processing modules can be called and executed by a processing element of the above device. The implementation of other modules is similar. In addition, all or part of these modules can be integrated together, or they can be implemented independently. The processing element here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each module above can be completed by an integrated logic circuit of hardware in the processor element or instructions in the form of software.

[0237] Figure 7 This is a schematic diagram of the structure of the control device provided in the embodiment of the present application. Figure 7 As shown, the control device 70 includes: a processor 701, and a memory 702 communicatively connected to the processor.

[0238] The memory 702 stores computer-executable instructions; the processor 701 executes the computer-executable instructions stored in the memory 702 to implement any of the aforementioned methods.

[0239] In the specific implementation of the above control device, it should be understood that the processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The method disclosed in the embodiment of the present application can be directly embodied as a hardware processor for execution, or a combination of hardware and software modules in the processor for execution.

[0240] An embodiment of the present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement any of the methods described above.

[0241] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to computer instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0242] An embodiment of the present application also provides a computer program product, including a computer program, which is used to implement any of the aforementioned methods when executed by a processor.

[0243] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the following claims.

[0244] It should be understood that the present application is not limited to the precise 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 application is limited only by the appended claims.

Claims

1. A multipath transmission congestion control method, characterized in that: The method comprises: Acquire real-time status information of transmission sub-streams in multi-path transmission, wherein the real-time status information includes data throughput at a current moment; Performing a summation operation on the data throughput of the transmission sub-flows at the current moment in the multi-path transmission to obtain summation result information; Performing logarithmic operation on the summation result information to obtain a network fairness value of multipath transmission at the current moment, wherein the real-time status information represents the status characteristics of the transmission substream when performing data transmission; Generate a network fairness prediction result of multipath transmission under each preset parameter according to the real-time status information and the network fairness value; Determining the congestion window size of the transmission subflow at the next moment according to the network fairness prediction result; The generating, according to the real-time status information and the network fairness value, a network fairness prediction result of multipath transmission under each preset parameter comprises: Determining a preset congestion window size of the transmission substream under each preset parameter; According to the real-time status information and the preset congestion window size, obtaining a local feature vector of multipath transmission at the current moment under each preset parameter; The local feature vector is expressed as a local feature matrix, the local feature matrix is ​​dimensionally transformed, and the transformation result is sequentially normalized and multi-head parallel attention calculated to determine the prediction target; the prediction target is linearly transformed to obtain a linear transformation result; Normalizing the linear transformation result and outputting the network fairness prediction value; According to the network fairness prediction value and the network fairness value, a network fairness prediction result of multipath transmission under each preset parameter is calculated.

2. The method according to claim 1, characterized in that The step of obtaining, according to the real-time status information and the preset congestion window size, a local feature vector of multipath transmission at the current moment under each preset parameter includes: Mapping the real-time status information and the preset congestion window size to form an input sequence of multipath transmission under each preset parameter; Perform a convolution operation on each vector in each of the input sequences to obtain local information of each vector convolution kernel size k, where k is a natural number; According to the reset gate parameter and the update gate parameter, each of the vectors and the corresponding local information are processed respectively to obtain the local information vector after each vector is processed; Performing a linear operation on the local information vector to obtain a fitted local information vector; A time position code representing transmission time information is added to the fitted local information vector to generate a final local feature vector of each vector, that is, a local feature vector of multipath transmission at the current moment under each preset parameter.

3. The method according to claim 1, characterized in that The determining of the preset congestion window size of the transmission substream under each preset parameter comprises: According to the congestion window size of the transmission substream at the current moment and the preset relationship, the preset congestion window size of the transmission substream under each preset parameter is calculated, and the preset relationship represents the functional relationship between the congestion window size at the current moment, the preset parameters, and the preset congestion window size.

4. The method according to any one of claims 1 to 3, characterized in that The real-time status information includes: data throughput, transmission rate, average round-trip delay, and congestion window size.

5. A multipath transmission congestion control device, characterized in that: The device comprises: A data unit is used to obtain real-time status information of a transmission substream in multipath transmission, and to perform a summation operation on the data throughput of the transmission substream in the multipath transmission at the current moment according to the data throughput in the real-time status information to obtain summation result information; and to perform a logarithmic operation on the summation result information to obtain a network fairness value of the multipath transmission at the current moment, wherein the real-time status information represents a status feature of the transmission substream when performing data transmission; A fairness prediction unit, configured to generate a network fairness prediction result of multipath transmission under each preset parameter according to the real-time status information and the network fairness value; A control unit, configured to determine a congestion window size of the transmission sub-flow at a next moment according to the network fairness prediction result; The fairness prediction unit is specifically used to determine the preset congestion window size of the transmission substream under each preset parameter; obtain the local feature vector of the multipath transmission at the current moment under each preset parameter according to the real-time status information and the preset congestion window size; represent the local feature vector as a local feature matrix, perform dimension conversion on the local feature matrix, and perform normalization calculation and multi-head parallel attention calculation on the conversion result in turn to determine the prediction target; perform linear transformation on the prediction target to obtain the linear transformation result; perform normalization processing on the linear transformation result and output the network fairness prediction value; calculate the network fairness prediction result of the multipath transmission under each preset parameter according to the network fairness prediction value and the network fairness value.

6. A control device, characterized in that: The control device includes: a processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 3.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 3 when executed by a processor.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.

Images