Optimization method for mixed parallel transmission based on multi-path network TCP-UDP

Abstract

The application relates to the technical field of data transmission, in particular to a mixed parallel transmission optimization method based on a multi-path network TCP-UDP. In the application, the transmission rate, the sending queue length and the round trip time of the multi-path connection are periodically monitored, a dynamic network portrait containing real-time states and historical trends is established, a weight coefficient is allocated for the rate value of each time point according to the portrait, the connection throughput of a future time point is predicted, the estimated transmission delay is calculated in combination with the queue length and the round trip time, a decision space is constructed based on the two core indexes of the throughput prediction and the delay estimation, positive and negative ideal solutions are introduced for multi-objective optimization, the optimal transmission path with high throughput and low delay is selected, the predicted performance of the optimal path and the application demand are dynamically judged to adopt the TCP or UDP protocol, and the TCP fragmentation size is adjusted when necessary, so that the fast self-adaptation to network changes is realized.

Description

Technical Field

[0001] This invention relates to the field of data transmission technology, and in particular to an optimization method for hybrid parallel transmission based on multi-channel network TCP-UDP. Background Technology

[0002] The field of data transmission technology primarily studies efficient and secure data transmission methods and technologies in different network environments, including data encoding, transmission control, error detection and correction, and network protocol optimization. Data transmission technology is widely used in communication, the Internet, mobile devices, and various smart devices for data exchange, ensuring reliable data transmission under different temporal and spatial conditions.

[0003] Among them, the hybrid parallel transmission optimization method based on multi-path network TCP-UDP refers to combining the two transmission protocols, TCP and UDP, to achieve high efficiency and reliability in data transmission over the network. This method primarily addresses the issue that, in practical applications, while TCP provides reliability guarantees, it can suffer from high latency due to congestion control mechanisms, while UDP, although offering faster transmission speeds, lacks reliable control.

[0004] The shortcomings of existing technologies lie in their reliance on a single and fixed protocol selection strategy. Before transmission begins, a decision is made once based on the application type or static network assessment to use either TCP or UDP. This approach cannot adapt to real-time dynamic changes in network conditions. When network congestion or jitter intensifies, the originally selected UDP path may experience a severe decline in service quality due to increased packet loss, while the originally selected TCP path may experience a sharp increase in latency due to the intervention of congestion control mechanisms, failing to meet low latency requirements. Due to the lack of continuous monitoring and dynamic prediction capabilities for network conditions, existing technologies struggle to consistently guarantee the efficiency and stability of data transmission in the face of complex and ever-changing network environments, easily falling into the trap of local optima rather than global optima. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a hybrid parallel transmission optimization method based on multi-path network TCP-UDP.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a hybrid parallel transmission optimization method based on multi-path network TCP-UDP, comprising the following steps: S1: Periodically monitor multiple TCP and UDP connections in the transmission network, obtain the transmission rate time series set of each TCP and UDP connection within a period, and the real-time sending queue length set of the number of data packets to be transmitted in parallel in each TCP and UDP connection. At the same time, send preset probe data packets to each TCP and UDP connection and record the real-time round-trip time set of the probe data packets. S2: Assign weighting coefficients to the transmission rate values ​​at each period time point in the transmission rate time series set, predict the throughput of TCP and UDP connections at future time points, and construct a predicted throughput set; S3: Based on the predicted throughput set, the real-time transmission queue length set, and the real-time round-trip time set, calculate the estimated transmission delay for each TCP and UDP connection and construct the estimated transmission delay set. S4: Construct an initial decision data table based on the estimated transmission delay set and the predicted throughput set and map it to the decision space; select the optimal transmission path identifier for TCP and UDP connections in the decision space; and classify the selected protocol type. S5: Based on the selected protocol type, schedule the data stream to be transmitted in parallel to the TCP or UDP connection corresponding to the optimal transmission path identifier for transmission, and obtain the transmission optimization result.

[0007] As a further embodiment of the present invention, the transmission rate time series set specifically includes connection rate profiles and sampling time series; the real-time transmission queue length set includes the amount of data to be transmitted and queue status snapshots; the real-time round-trip time set includes path round-trip delay and network delay samples; the predicted throughput set specifically includes connection bandwidth prediction values ​​and performance trend assessments; the estimated transmission delay set includes data packet queuing delay and one-way propagation delay; the selected protocol type specifically includes transmission mode decision and protocol selection criteria; and the transmission optimization result specifically includes path scheduling instructions, protocol application instructions, and fragment size configuration.

[0008] As a further aspect of the present invention, the steps for obtaining the transmission rate timing set, the real-time transmission queue length set, and the real-time round-trip time set are specifically as follows: S111: Periodically monitor multiple TCP and UDP connections in the transmission network, count the total amount of data successfully transmitted by each TCP and UDP connection within each preset time interval, calculate the periodic transmission rate of each TCP and UDP connection based on the total amount of data and the length of the corresponding time interval, and establish a time series set of transmission rates. S112: Within the same monitoring period, directly obtain the number of data packets to be transmitted in parallel in each TCP and UDP connection and record them as the real-time sending queue length, and integrate them into a set of real-time sending queue lengths; S113: Within the same monitoring period, actively construct and send preset probe data packets to each TCP and UDP connection, record the time elapsed from the time of sending each probe data packet to the time of receiving the acknowledgment response as the real-time round-trip time, and obtain the set of real-time round-trip times.

[0009] As a further aspect of the present invention, the step of obtaining the predicted throughput set specifically includes: S211: Assign weight coefficients to TCP and UDP connections according to the time order of the periodic transmission rate of each TCP and UDP connection in the transmission rate time sequence set, and combine the multiple weight coefficients assigned to each sequence in time order to establish a rate weight coefficient sequence set. S212: Perform weighted calculation on the periodic transmission rate of each TCP and UDP connection in the transmission rate time series set and the corresponding weight coefficient in the rate weight coefficient sequence set to obtain the weighted transmission rate value at each time point and generate a connection weighted rate sequence. S213: Perform cumulative calculation on each weighted and corrected transmission rate value in the connection weighted rate sequence to determine the predicted throughput of TCP and UDP connections at future time points and construct a predicted throughput set.

[0010] As a further aspect of the present invention, the step of obtaining the estimated transmission delay set specifically includes: S311: For each TCP and UDP connection, call the queue length in the real-time sending queue length set, and at the same time call the preset average size of the data packets of the data stream to be transmitted in parallel. Calculate the queue data volume based on the queue length value and the preset average size of the data packets, and establish a connection queue data volume sequence. S312: Based on the connection queue data volume sequence and the corresponding predicted throughput calculated from the corrected transmission rate in the predicted throughput set, calculate the queuing time of each TCP and UDP connection to obtain the connection queuing time set; S313: Calculate the estimated transmission delay of the corresponding TCP and UDP connections based on the queuing time in the connection queuing time set and half of the round-trip time in the real-time round-trip time set, and construct the estimated transmission delay set.

[0011] As a further aspect of the present invention, the step of obtaining the selected protocol type specifically comprises: S411: Based on the estimated transmission delay set and the predicted throughput set, construct an initial decision data table for all TCP and UDP connections, and perform vector normalization processing on the data in the initial decision data table to map it to a standardized decision space. Then, determine the positive ideal solution composed of the optimal normalization index value and the negative ideal solution composed of the worst normalization index value in the decision space, and establish a positive and negative ideal solution vector. S412: Based on the positive and negative ideal solution vectors and the spatial coordinates of each TCP and UDP connection in the decision space, calculate the Euclidean distance between the positive and negative ideal solutions in the positive and negative ideal solution vectors respectively, and calculate the relative proximity of each TCP and UDP connection based on the Euclidean distance. S413: Select the largest relative proximity value from the relative proximity values ​​of each TCP and UDP connection, use the corresponding TCP and UDP connection as the optimal transmission path identifier, call the throughput value in the predicted throughput set corresponding to the optimal transmission path identifier, and compare it with the preset throughput threshold. If the throughput value is greater than the throughput threshold, record the UDP protocol as the selected protocol type; otherwise, record the TCP protocol as the selected protocol type.

[0012] As a further aspect of the present invention, the step of obtaining the transmission optimization result specifically includes: S511: Schedule the data stream to be transmitted in parallel to the TCP or UDP connection corresponding to the optimal transmission path identifier for transmission, and at the same time determine the category of the selected protocol type, and establish the scheduling path and protocol determination result; S512: Based on the scheduling path and protocol determination result, if the determined protocol is TCP, the queue length value of the corresponding connection in the real-time sending queue length set is called and compared with the preset queue threshold. Based on the comparison result, it is determined whether the data packet fragment size needs to be adjusted, and the TCP fragmentation adjustment criterion is obtained. S513: Based on the transmission path determined in the scheduling path and protocol determination result, according to the TCP fragmentation adjustment criterion, when it is determined that the data packet fragment size needs to be adjusted, the data packet fragment size of the TCP connection is synchronously reduced to obtain the transmission optimization result.

[0013] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, a dynamic network profile containing real-time status and historical trends is established by periodically monitoring the transmission rate, sending queue length, and round-trip time of multiple connections. Based on this profile, a weighting coefficient is assigned to the rate value at each time point to predict the connection throughput at future time points. Simultaneously, the estimated transmission latency is calculated by combining queue length and round-trip time. A decision space is constructed based on the two core indicators of throughput prediction and latency prediction, and positive and negative ideal solutions are introduced for multi-objective optimization to select the optimal transmission path that balances high throughput and low latency. Finally, the optimal path's predictive performance and application requirements are dynamically determined to use either TCP or UDP protocols, and the TCP fragment size is adjusted when necessary. This achieves rapid adaptation to network changes and effectively solves the problem of decreased transmission efficiency and stability caused by static decision-making in existing technologies due to network dynamism. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the main steps of the present invention; Figure 2 This is a flowchart of step S1 of the present invention; Figure 3 This is a flowchart of step S2 of the present invention; Figure 4 This is a flowchart of step S3 of the present invention; Figure 5 This is a flowchart of step S4 of the present invention; Figure 6 This is a flowchart of step S5 of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0016] Please see Figure 1 This invention provides a technical solution, a method for optimizing hybrid parallel transmission based on multi-path network TCP-UDP, comprising the following steps: S1: Periodically monitor multiple TCP and UDP connections in the transmission network, obtain the transmission rate time series set of each TCP and UDP connection within a period, and the real-time sending queue length set of the number of data packets to be transmitted in parallel in each TCP and UDP connection. At the same time, send preset probe data packets to each TCP and UDP connection and record the real-time round-trip time set of the probe data packets. S2: Assign weight coefficients to the transmission rate values ​​at each period time point in the transmission rate time series set, predict the throughput of TCP and UDP connections at future time points, and construct a predicted throughput set; S3: Based on the predicted throughput set, the real-time sending queue length set, and the real-time round-trip time set, calculate the estimated transmission delay for each TCP and UDP connection and construct the estimated transmission delay set. S4: Construct an initial decision data table based on the estimated transmission delay set and the predicted throughput set and map it to the decision space. Select the optimal transmission path identifier for TCP and UDP connections in the decision space and classify the selected protocol types. S5: Based on the selected protocol type, schedule the data stream to be transmitted in parallel to the TCP or UDP connection corresponding to the optimal transmission path identifier for transmission, and obtain the transmission optimization result.

[0017] The transmission rate time series set specifically includes connection rate profiles and sampling time series; the real-time transmission queue length set includes the amount of data to be transmitted and queue status snapshots; the real-time round-trip time set includes path round-trip delay and network delay samples; the predicted throughput set specifically includes predicted connection bandwidth values ​​and performance trend assessments; the estimated transmission delay set includes packet queuing delay and one-way propagation delay; the selected protocol type specifically includes transmission mode decisions and protocol selection criteria; and the transmission optimization results specifically include path scheduling instructions, protocol application instructions, and fragment size configurations.

[0018] Please see Figure 2 The specific steps for obtaining the transmission rate time series set, the real-time transmission queue length set, and the real-time round-trip time set are as follows: S111: Periodically monitor multiple TCP and UDP connections in the transmission network, count the total amount of data successfully transmitted by each TCP and UDP connection within each preset time interval, calculate the periodic transmission rate of each TCP and UDP connection based on the total amount of data and the length of the corresponding time interval, and establish a time series set of transmission rates. First, probes targeting multiple TCP and UDP connections are deployed in the transmission network. These probes perform monitoring at regular time intervals. The interval is determined by autocorrelation analysis of historical network speed data, selecting the time delay before the autocorrelation function value first drops below a low level (e.g., 0.5) as the monitoring interval to ensure the independence and timeliness of each monitoring session. Here, this interval is set to 1 second. Within each 1-second monitoring period, the probe continuously records the size of each successfully transmitted and acknowledged data packet through a specific TCP connection (e.g., TCP_Conn1) and a specific UDP connection (e.g., UDP_Conn2), and these sizes are accumulated. For example, in the first monitoring period (time T1), the total amount of data successfully transmitted by the TCP_Conn1 connection is 5 megabytes (MB), while the total amount of data successfully transmitted by the UDP_Conn2 connection is 10 megabytes. In the subsequent second monitoring period (time T2), the total amount of data successfully transmitted by TCP_Conn1 is 6 megabytes, and by UDP_Conn2 it is 9 megabytes. In the third monitoring period (time T3), the total data volume of TCP_Conn1 was 5.5 megabytes, and that of UDP_Conn2 was 11 megabytes. After obtaining the total data volume in each time interval, the periodic transmission rate was calculated. For the TCP_Conn1 rate at time T1, the total data volume of 5 megabytes was converted to 40 megabits per second (Mbps), and then divided by the 1-second time interval length to obtain a periodic transmission rate of 40 megabits per second (Mbps). Similarly, the TCP_Conn1 rate at time T2 was calculated to be 48 Mbps, and the rate at time T3 was calculated to be 44 Mbps. The same calculation was performed on UDP_Conn2, yielding periodic transmission rates of 80 Mbps, 72 Mbps, and 88 Mbps at times T1, T2, and T3, respectively. These rate values ​​calculated at different time points are organized according to connection identifiers and time order to form two independent rate time series: one belongs to TCP_Conn1 with the content [40, 48, 44], and the other belongs to UDP_Conn2 with the content [80, 72, 88]. Finally, the rate time series of all connections are integrated to establish the transmission rate time series set.

[0019] S112: Within the same monitoring period, directly obtain the number of data packets to be transmitted in parallel in each TCP and UDP connection and record them as the real-time sending queue length, and integrate them into a set of real-time sending queue lengths; Within the same 1-second period of transmission rate monitoring, the length of the send queue for each TCP and UDP connection is monitored in parallel. This process is achieved by directly querying the send buffers allocated to each connection (e.g., TCP_Conn1 and UDP_Conn2) in the operating system kernel network stack. At time T1, the query accesses the send buffer of TCP_Conn1 and counts 100 packets waiting to be sent; simultaneously, it accesses the send buffer of UDP_Conn2 and counts 50 packets waiting to be sent. These values ​​are recorded. At the following time T2, this query is repeated, and the number of packets waiting to be sent for TCP_Conn1 becomes 120, and for UDP_Conn2 becomes 40. At time T3, the query is executed again, and the queue length for TCP_Conn1 is 110 packets, and for UDP_Conn2 it is 60 packets. Thus, at each monitoring time point (T1, T2, T3), a real-time send queue length value for each connection is obtained. Arrange these values ​​in connection and time order. The real-time transmission queue length sequence for TCP_Conn1 is [100, 120, 110], and the sequence for UDP_Conn2 is [50, 40, 60]. Collecting the queue length values ​​of all monitored TCP and UDP connections at each time point constitutes the real-time transmission queue length set.

[0020] S113: Within the same monitoring period, actively construct and send preset probe data packets to each TCP and UDP connection, record the time elapsed from the time of sending each probe data packet to the time of receiving the acknowledgment response as the real-time round-trip time, and obtain the set of real-time round-trip times. Within the same 1-second monitoring period, active probing is performed to obtain the real-time round-trip time. This process constructs a 32-byte probe packet for each monitored TCP and UDP connection. This size is chosen to be small enough that its impact on network load is negligible, yet larger than the minimum packet size typically handled by the network protocol stack, effectively preventing it from being discarded as an abnormal packet by network devices, and accommodating a unique probe identifier. Taking the probe of TCP_Conn1 at time T1 as an example, the current high-precision timestamp is recorded as the sending time, and then this 32-byte probe packet is injected into the TCP_Conn1 sending queue. Upon receiving this probe packet, the remote server immediately returns an acknowledgment response packet. Upon receiving this acknowledgment response packet, the local machine again records the current high-precision timestamp as the receiving time. The time difference between the receiving timestamp and the sending timestamp is calculated; this difference is the real-time round-trip time of TCP_Conn1 at that moment. For example, at time T1, the round-trip time (RTT) measured for TCP_Conn1 is 50 milliseconds, and the RTT measured for UDP_Conn2 is 20 milliseconds. At time T2, the above process is repeated, and the RTT measured for TCP_Conn1 is 60 milliseconds, and for UDP_Conn2 is 22 milliseconds. At time T3, the RTT measured for TCP_Conn1 is 55 milliseconds, and for UDP_Conn2 is 21 milliseconds. By performing this probing process on all connections in each monitoring period, a set of real-time RTT data sorted by connection and time can be obtained. For example, the RTT sequence for TCP_Conn1 is [50, 60, 55], and the sequence for UDP_Conn2 is [20, 22, 21]. These sequences together constitute the real-time RTT set.

[0021] Please see Figure 3 The specific steps for obtaining the predicted throughput set are as follows: S211: Assign weight coefficients to TCP and UDP connections based on the time order of the periodic transmission rates of each TCP and UDP connection in the transmission rate time sequence set, and combine the multiple weight coefficients assigned to each sequence in chronological order to establish a rate weight coefficient sequence set. First, each connection's rate time series is checked for missing data points due to probe malfunction or network interruption. If missing data is found, linear interpolation is used to fill in the gaps. This method is chosen because, within short monitoring periods of seconds, network rate changes typically exhibit strong local linearity, and linear interpolation can obtain reasonable estimates with low computational cost. For example, suppose the rate sequence of TCP_Conn1 becomes [40, missing, 44] due to missing data at time T2. The preprocessing operation takes the two known data points 40 and 44 before and after it, calculates their arithmetic mean, i.e., (40+44)÷2=42. This calculated result, 42, is then filled into the missing position, resulting in a repaired sequence of [40, 42, 44]. This check and repair process is performed on the rate time series of all connections, ultimately generating a complete rate time series set without missing data.

[0022] S212: Perform weighted calculations on the periodic transmission rates of each TCP and UDP connection in the transmission rate time series set and the corresponding weight coefficients in the rate weight coefficient sequence set to obtain the weighted transmission rate value at each time point and generate a connection weighted rate sequence. A weighted calculation is performed on the periodic transmission rate of each connection in the complete rate time series, using a double exponential smoothing model. This model consists of two core equations, used to update the baseline level and trend of the sequence, respectively, and is executed iteratively time-by-time. The weighted and corrected transmission rate value at each time point is obtained through the following calculations: , The calculated connection-weighted rate, i.e. the predicted rate for the next time point, is derived from... Confirmed. Among them, : The sequence number obtained from the time point sequence of the complete rate time series set, for example, T1 corresponds to . : Obtained from the complete rate time series set, at the 1st The periodic transmission rate of the connection monitored at each time point is measured in Mbps. The result calculated in this step, at the [number]th [time] The smoothed rate values ​​at each time point are expressed in Mbps. The result calculated in the previous iteration of this step. The rate smoothing value at each time point, measured in Mbps. For the initial value... Set as the first observation of the sequence . The result calculated in this step, at the [number]th [time] The trend smoothing value at each time point is measured in megabits per second per monitoring period. The result calculated in the previous iteration of this step. The trend smoothing value at each time point, measured in megabits per second per monitoring period. For the initial value... Setting it to 0 indicates that there is no initial trend. : Data smoothing coefficient. Its value is determined through backtesting on historical rate datasets. It takes values ​​in the interval [0, 1] with a fixed step size (e.g., 0.1), and for each value, its mean squared error of prediction on the historical dataset is calculated. The value that minimizes the mean squared error is selected. Value. For example, if The mean square error is 5.5. The mean square error is 4.8. If the mean square error is 5.1, then choose... The larger this value, the greater the weight of the current observation. Trend smoothing coefficient. Its value is determined using the same method as... Consistent, but applies to the trend portion. For example, selected after testing. The larger this value, the greater the weight given to recent trend changes.

[0023] Taking TCP_Conn1 as an example, its rate sequence is as follows: Mbps (assuming no missing data). The calculation process is as follows: Initialization: Mbps, Mbps / monitoring period. When hour: Mbps. Mbps / monitoring period. When hour: Mbps. Mbps / monitoring period. When hour: Mbps. Mbps / monitoring period. Ultimately, the connection-weighted rate of TCP_Conn1 is... Mbps. The calculation results for all connections are used to generate a connection-weighted rate sequence.

[0024] S213: Perform cumulative calculations on each weighted and corrected transmission rate value in the connection weighted rate sequence to determine the predicted throughput of TCP and UDP connections at future time points and construct a predicted throughput set; The process of determining the predicted throughput of TCP and UDP connections at future points in time is as follows: Obtain the weighted and corrected transmission rate value for each TCP and UDP connection at the last time point in the connection weighted rate sequence; Obtain the length of the time interval set in the periodic monitoring operation; Based on the obtained weighted and corrected transmission rate values, the predicted throughput of TCP and UDP connections is estimated for a duration equal to the time interval. The time duration used for cumulative calculation is consistent with the monitoring period set in S111, which is 1 second here, to ensure that the prediction timescale matches the data acquisition frequency. Taking TCP_Conn1 as an example, its connection weighted rate is 45.56 Mbps, and the connection weighted rate of UDP_Conn2 obtained through the same calculation is 84.52 Mbps. The predicted throughput of TCP_Conn1 is 45.56 Mbps × 1 second, resulting in 45.56 Mbps. The predicted throughput of UDP_Conn2 is 84.52 Mbps × 1 second, resulting in 84.52 Mbps. This calculation is performed on the weighted rate values ​​of all connections, and the combined values ​​construct the predicted throughput set.

[0025] Please see Figure 4 The specific steps for obtaining the estimated transmission delay set are as follows: S311: For each TCP and UDP connection, call the queue length in the real-time sending queue length set, and at the same time call the preset average size of the data packets to be transmitted in parallel. Calculate the queue data volume based on the queue length value and the preset average size of the data packets, and establish a connection queue data volume sequence. For each TCP and UDP connection, the queue length recorded in the real-time send queue length set at the latest monitoring time (T3) is retrieved. For TCP_Conn1, the queue length is 110 packets; for UDP_Conn2, it is 60 packets. Simultaneously, an average packet size is calculated. This value is not fixed but dynamically calculated based on recent historical transmission data of a specific application stream (e.g., high-definition video stream), specifically the arithmetic mean of the sizes of the most recently transmitted batch of packets (e.g., 1000 packets). For example, by analyzing recent packet samples, the average size is calculated to be 1500 bytes. Next, the queue data volume is calculated based on the queue length and average packet size. For TCP_Conn1, this is calculated as 110 × 1500 bytes, resulting in 165,000 bytes. For UDP_Conn2, this is calculated as 60 × 1500 bytes, resulting in 90,000 bytes. The calculation results for all connections together establish the connection queue data volume sequence.

[0026] S312: Based on the connection queue data volume sequence and the corresponding predicted throughput calculated from the corrected transmission rate in the predicted throughput set, calculate the queuing time of each TCP and UDP connection to obtain the connection queuing time set. Before calculation, the data units are standardized. The data volume in the connection queue data volume sequence is converted from bytes to bits, with 1 byte equaling 8 bits. The queue data volume of TCP_Conn1 is 165,000 bytes, which is converted to 1,320,000 bits. The queue data volume of UDP_Conn2 is 90,000 bytes, which is converted to 720,000 bits. Then, the corresponding values ​​in the predicted throughput set are called. The predicted throughput of TCP_Conn1 is 45.56 megabits per second (i.e., 45,560,000 bits per second), and the predicted throughput of UDP_Conn2 is 84.52 megabits per second (i.e., 84,520,000 bits per second). The queuing time is calculated by dividing the queue data volume (bits) by the predicted throughput (bits per second). For TCP_Conn1, the calculation is 1,320,000 bits ÷ 45,560,000 bits per second, resulting in a queuing time of approximately 0.0290 seconds, or 29.0 milliseconds. For UDP_Conn2, the calculation is 720,000 bits ÷ 84,520,000 bits per second, resulting in a queuing time of approximately 0.0085 seconds, or 8.5 milliseconds. The calculation results for all connections together form the set of connection queuing times.

[0027] S313: Calculate the estimated transmission delay of the corresponding TCP and UDP connections based on the queuing time in the connection queuing time set and half of the round-trip time in the real-time round-trip time set, and construct the estimated transmission delay set. The estimated transmission delay is calculated based on the queuing time in the connection queuing time set and the round-trip time in the real-time round-trip time set. The round-trip time used in the calculation is the value of the corresponding connection at the latest monitoring time point (time T3). The round-trip time of TCP_Conn1 at time T3 (55 milliseconds) and the round-trip time of UDP_Conn2 at time T3 (21 milliseconds) are retrieved from the real-time round-trip time set. The estimated transmission delay is calculated by adding half of the queuing time to the round-trip time. For TCP_Conn1, half of its round-trip time, 55 milliseconds ÷ 2, is calculated, resulting in 27.5 milliseconds. This is then added to the queuing time of TCP_Conn1 (29.0 milliseconds) to obtain an estimated transmission delay of 56.5 milliseconds. For UDP_Conn2, half of its round-trip time, 21 milliseconds ÷ 2, is calculated, resulting in 10.5 milliseconds. This is then added to the queuing time of UDP_Conn2 (8.5 milliseconds) to obtain an estimated transmission delay of 19.0 milliseconds. The calculation results for all connections together construct the estimated transmission delay set.

[0028] Please see Figure 5 The specific steps for obtaining the selected protocol type are as follows: S411: Based on the estimated transmission delay set and the predicted throughput set, construct an initial decision data table for all TCP and UDP connections, and perform vector normalization processing on the data in the initial decision data table to map it to a standardized decision space. Then, in the decision space, determine the positive ideal solution composed of the optimal normalization index value and the negative ideal solution composed of the worst normalization index value, and establish a vector of positive and negative ideal solutions. The process of determining the positive ideal solution composed of the optimal normalized index value and the negative ideal solution composed of the worst normalized index value is as follows: For each column of data in the initial decision data table that represents the predicted throughput and the estimated transmission delay, calculate the arithmetic square root of the sum of the squares of all the values ​​in the data column. Divide each value in the data column by the arithmetic square root calculated for the same data column to obtain the normalized value; The maximum value of the normalized throughput and the minimum value of the normalized latency in all TCP and UDP connections together constitute the positive ideal solution. The minimum normalized throughput value and the maximum normalized latency value of all TCP and UDP connections together constitute the negative ideal solution. Based on the data from the estimated transmission delay set and the predicted throughput set, an initial decision data table is constructed for all TCP and UDP connections. In this table, each row represents a connection, and each column represents a decision metric. The metric value for TCP_Conn1 is [45.56Mbps, 56.5ms], and the metric value for UDP_Conn2 is [84.52Mbps, 19.0ms]. Vector normalization is then performed on the data in this table. For the predicted throughput column, the square root of the sum of the squares of the values ​​in the column is calculated. Then, divide each value in this column by 96.02 to obtain a normalized throughput of 0.474 for TCP_Conn1 and 0.880 for UDP_Conn2. Perform the same operation on the estimated transmission delay column to calculate... The normalized latency of TCP_Conn1 is 0.948, and that of UDP_Conn2 is 0.319. After processing, a standardized decision space is obtained. Within this space, positive and negative ideal solutions are determined. The positive ideal solution consists of the optimal normalized values ​​of all indicators (maximum values ​​for benefit-type indicators and minimum values ​​for cost-type indicators), i.e., [0.880, 0.319]. The negative ideal solution consists of the worst normalized values ​​of all indicators, i.e., [0.474, 0.948]. These two solutions together establish the positive and negative ideal solution vectors.

[0029] S412: Based on the positive and negative ideal solution vectors and the spatial coordinates of each TCP and UDP connection in the decision space, calculate the Euclidean distance between the positive and negative ideal solutions in the positive and negative ideal solution vectors, and calculate the relative proximity of each TCP and UDP connection based on the Euclidean distance. Based on the positive and negative ideal solution vectors and the spatial coordinates corresponding to each connection in the decision space, the VIKOR model is used to calculate the overall performance of each connection. This model ranks the alternatives by calculating group utility, individual regret, and making trade-offs. Its execution process is as follows: first, it calculates the performance of each connection... and The values ​​are then normalized and synthesized into the final ranking index. The specific calculations are as follows: , , ,in, : Join index obtained from the decision data table, for example Represents TCP_Conn1. Indexes of decision indicators, such as Corresponding to normalized throughput. : The total number of decision indicators, which is 2 here. : Accessed from the S411 standardized decision space, connecting In terms of indicators The normalized value on. and The index obtained from the positive and negative ideal solution vectors established by S411 The positive and negative ideal solutions. :index The weights are determined based on the Analytic Hierarchy Process (AHP). By constructing a pairwise comparison matrix between indicators, the relative importance of different indicators is quantified, and then the normalized principal eigenvectors of the matrix are calculated to obtain the weights. For example, in real-time video conferencing, AHP analysis determines that latency is more important than throughput, thus yielding the throughput weight. Delay weight . : The calculated connection The group utility value. : The calculated connection Individual regret value. and All connections calculated in this step The minimum and maximum values. and All connections calculated in this step The minimum and maximum values. : Decision-making mechanism coefficient. Its setting reflects the decision-making strategy; a value of 0.5 indicates a focus on "maximizing group interests" (as determined by...). (representative) and "minimization of individual regret" (by) Giving equal importance to representatives is a compromise strategy. : The calculated connection The final ranking metric. When the best and worst values ​​of a metric are equal, that is... If the value is zero, this term contributes nothing to the calculation.

[0030] The calculation process is as follows: For TCP_Conn1( ), its value is . . For UDP_Conn2( ), its value is . . .Sure .calculate : . The relative proximity of each TCP and UDP connection is determined by... OK, TCP_Conn1 is 0, UDP_Conn2 is 1.

[0031] S413: Select the largest relative proximity value from the relative proximity of each TCP and UDP connection, use the corresponding TCP and UDP connection as the optimal transmission path identifier, call the throughput value in the predicted throughput set corresponding to the optimal transmission path identifier, and compare it with the preset throughput threshold. If the throughput value is greater than the throughput threshold, record the UDP protocol as the selected protocol type; otherwise, record the TCP protocol as the selected protocol type. From all the relative proximity values ​​of TCP and UDP connections, the largest relative proximity value of 1 is selected, and its corresponding connection UDP_Conn2 is used as the optimal transmission path identifier. Next, the throughput value of 84.52 Mbps is retrieved from the predicted throughput set corresponding to the optimal transmission path identifier UDP_Conn2. This throughput value is compared with a throughput threshold. This threshold is determined based on the application's quality of service requirements. By analyzing the correspondence between throughput and objective quality of service scores (such as scores based on stuttering rate and clarity) in historical data, the 10th percentile of the throughput distribution that consistently yields an "acceptable" quality of service score is used as the threshold. For example, analysis shows that the acceptable session throughput for a quality of service score is above 58 Mbps, and its 10th percentile is 60 Mbps; therefore, the threshold is set to 60 Mbps. The comparison is performed: 84.52 Mbps is greater than 60 Mbps, meeting the condition; therefore, the UDP protocol is recorded as the selected protocol type.

[0032] Please see Figure 6 The specific steps for obtaining the transmission optimization results are as follows: S511: Schedule the data stream to be transmitted in parallel to the TCP or UDP connection corresponding to the optimal transmission path identifier for transmission, and at the same time determine the category of the selected protocol type and establish the scheduling path and protocol determination result. Based on the optimal transmission path identifier UDP_Conn2 calculated in S412 and selected in S413, the send handle of the data stream to be transmitted in parallel is bound to the UDP_Conn2 connection. This operation means that all subsequent data packets belonging to this data stream will be sent through the UDP_Conn2 transmission channel, thus achieving dynamic path selection. Simultaneously, the selected protocol type determination result obtained from S413 is UDP. Finally, the scheduling path (UDP_Conn2) and the protocol determination result (UDP) are integrated into a specific execution instruction, which specifies the exit path of the data stream and the transport layer protocol used.

[0033] S512: Based on the scheduling path and protocol determination result, if the determined protocol is TCP, the queue length value of the corresponding connection in the real-time sending queue length set is called and compared with the preset queue threshold. Based on the comparison result, it is determined whether the data packet fragment size needs to be adjusted and the TCP fragmentation adjustment criterion is obtained. Based on the established scheduling path and protocol determination results, the protocol type is first determined. In this embodiment, the determined protocol is UDP. Since subsequent adjustment measures are for the TCP protocol, the subsequent operations in this step will be skipped. For complete explanation, the process when the determined protocol is TCP is described: Suppose the comparison result of S413 is that the throughput is less than the threshold, the TCP protocol is selected, and the optimal path is TCP_Conn1. At this time, the queue length value of TCP_Conn1 at time T3, which is 110 packets, will be called from the real-time sending queue length set. This value is compared with a queue threshold. The threshold is set based on the analysis of the relationship between sending queue length and network packet loss rate in historical data, and the inflection point of queue length where the packet loss rate begins to rise sharply is used as the threshold. For example, the analysis curve shows that when the queue length exceeds 80 packets, the packet loss rate rises rapidly from less than 1% to more than 5%, so the threshold is set to 80. The comparison is performed: 110 is greater than 80, and the comparison result is that adjustment is needed. This result is recorded, and the TCP fragmentation adjustment criterion is obtained, with the content "adjustment is needed".

[0034] S513: Based on the transmission path determined in the scheduling path and protocol determination results, according to the TCP fragmentation adjustment criteria, when it is determined that the data packet fragment size needs to be adjusted, the data packet fragment size of the TCP connection is reduced synchronously to obtain the transmission optimization result; Based on the UDP_Conn2 transmission path determined in the scheduling path and protocol judgment results, and the S512's decision to skip adjustment due to the UDP protocol type, the final transmission operation does not adjust the packet fragment size and directly uses UDP_Conn2 for transmission. If the scheduling path is TCP_Conn1, the protocol is TCP, and the criterion is "adjustment required," then for the TCP_Conn1 connection, the fragment size of subsequent packets to be sent will be synchronously reduced. Specifically, the maximum segment size parameter associated with this connection in the TCP protocol stack is adjusted, for example, from the default 1460 bytes to 1024 bytes, to reduce the risk of a single large packet causing congestion in the network.

[0035] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A hybrid parallel transmission optimization method based on multi-path network TCP-UDP, characterized in that, Includes the following steps: S1: Periodically monitor multiple TCP and UDP connections in the transmission network, obtain the transmission rate time series set of each TCP and UDP connection within a period, and the real-time sending queue length set of the number of data packets to be transmitted in parallel in each TCP and UDP connection. At the same time, send preset probe data packets to each TCP and UDP connection and record the real-time round-trip time set of the probe data packets. S2: Based on the order of the periodic transmission rates of each TCP and UDP connection in the transmission rate time series set, assign weight coefficients to the transmission rate values ​​at each periodic time point in the transmission rate time series set, use a double exponential smoothing model to perform weighted calculations on the periodic transmission rates of each TCP and UDP connection, obtain the weighted transmission rate values ​​at each time point, generate a connection weighted rate sequence, predict the throughput of TCP and UDP connections at future time points, and construct a predicted throughput set; S3: Based on the predicted throughput set, the real-time transmission queue length set, and the real-time round-trip time set, calculate the estimated transmission delay for each TCP and UDP connection and construct the estimated transmission delay set. S4: Construct an initial decision data table based on the estimated transmission delay set and the predicted throughput set and map it to the decision space; select the optimal transmission path identifier for TCP and UDP connections in the decision space; and classify the selected protocol type. S5: Based on the selected protocol type, schedule the data stream to be transmitted in parallel to the TCP or UDP connection corresponding to the optimal transmission path identifier for transmission, and obtain the transmission optimization result.

2. The method for optimizing hybrid parallel transmission based on multi-path network TCP-UDP according to claim 1, characterized in that, The transmission rate time series set specifically includes connection rate profiles and sampling time series; the real-time transmission queue length set includes the amount of data to be transmitted and queue status snapshots; the real-time round-trip time set includes path round-trip delay and network delay samples; the predicted throughput set specifically includes predicted connection bandwidth and performance trend evaluation; the estimated transmission delay set includes packet queuing delay and one-way propagation delay; the selected protocol type specifically includes transmission mode decision and protocol selection criteria; and the transmission optimization result specifically includes path scheduling instructions, protocol application instructions, and fragment size configuration.

3. The hybrid parallel transmission optimization method based on multi-path network TCP-UDP according to claim 1, characterized in that, The steps for obtaining the transmission rate time series set, the real-time transmission queue length set, and the real-time round-trip time set are as follows: S111: Periodically monitor multiple TCP and UDP connections in the transmission network, count the total amount of data successfully transmitted by each TCP and UDP connection within each preset time interval, calculate the periodic transmission rate of each TCP and UDP connection based on the total amount of data and the length of the corresponding time interval, and establish a time series set of transmission rates. S112: Within the same monitoring period, directly obtain the number of data packets to be transmitted in parallel in each TCP and UDP connection and record them as the real-time sending queue length, and integrate them into a set of real-time sending queue lengths; S113: Within the same monitoring period, actively construct and send preset probe data packets to each TCP and UDP connection, record the time elapsed from the time of sending each probe data packet to the time of receiving the acknowledgment response as the real-time round-trip time, and obtain the set of real-time round-trip times.

4. The hybrid parallel transmission optimization method based on multi-path network TCP-UDP according to claim 3, characterized in that, The specific steps for obtaining the predicted throughput set are as follows: S211: Assign weight coefficients to TCP and UDP connections according to the time order of the periodic transmission rate of each TCP and UDP connection in the transmission rate time sequence set, and combine the multiple weight coefficients assigned to each sequence in time order to establish a rate weight coefficient sequence set. S212: Perform weighted calculation on the periodic transmission rate of each TCP and UDP connection in the transmission rate time series set and the corresponding weight coefficient in the rate weight coefficient sequence set to obtain the weighted transmission rate value at each time point and generate a connection weighted rate sequence. S213: Perform cumulative calculation on each weighted and corrected transmission rate value in the connection weighted rate sequence to determine the predicted throughput of TCP and UDP connections at future time points and construct a predicted throughput set.

5. The hybrid parallel transmission optimization method based on multi-path network TCP-UDP according to claim 4, characterized in that, The process of determining the predicted throughput of TCP and UDP connections at future points in time is as follows: Obtain the weighted and corrected transmission rate value for each TCP and UDP connection at the last time point in the connection weighted rate sequence; Obtain the length of the time interval set in the periodic monitoring operation; Based on the obtained weighted and corrected transmission rate values, the predicted throughput of TCP and UDP connections is estimated for a duration equal to the time interval.

6. The hybrid parallel transmission optimization method based on multi-path network TCP-UDP according to claim 4, characterized in that, The specific steps for obtaining the estimated transmission delay set are as follows: S311: For each TCP and UDP connection, call the queue length in the real-time sending queue length set, and at the same time call the preset average size of the data packets of the data stream to be transmitted in parallel. Calculate the queue data volume based on the queue length value and the preset average size of the data packets, and establish a connection queue data volume sequence. S312: Based on the connection queue data volume sequence and the corresponding predicted throughput calculated from the corrected transmission rate in the predicted throughput set, calculate the queuing time of each TCP and UDP connection to obtain the connection queuing time set; S313: Calculate the estimated transmission delay of the corresponding TCP and UDP connections based on the queuing time in the connection queuing time set and half of the round-trip time in the real-time round-trip time set, and construct the estimated transmission delay set.

7. The hybrid parallel transmission optimization method based on multi-path network TCP-UDP according to claim 6, characterized in that, The specific steps for obtaining the selected protocol type are as follows: S411: Based on the estimated transmission delay set and the predicted throughput set, construct an initial decision data table for all TCP and UDP connections, and perform vector normalization processing on the data in the initial decision data table to map it to a standardized decision space. Then, determine the positive ideal solution composed of the optimal normalization index value and the negative ideal solution composed of the worst normalization index value in the decision space, and establish a positive and negative ideal solution vector. S412: Based on the positive and negative ideal solution vectors and the spatial coordinates of each TCP and UDP connection in the decision space, calculate the Euclidean distance between the positive and negative ideal solutions in the positive and negative ideal solution vectors respectively, and calculate the relative proximity of each TCP and UDP connection based on the Euclidean distance. S413: Select the largest relative proximity value from the relative proximity values ​​of each TCP and UDP connection, use the corresponding TCP and UDP connection as the optimal transmission path identifier, call the throughput value in the predicted throughput set corresponding to the optimal transmission path identifier, and compare it with the preset throughput threshold. If the throughput value is greater than the throughput threshold, record the UDP protocol as the selected protocol type; otherwise, record the TCP protocol as the selected protocol type.

8. The hybrid parallel transmission optimization method based on multi-path network TCP-UDP according to claim 7, characterized in that, The process of determining the positive ideal solution composed of the optimal normalized index value and the negative ideal solution composed of the worst normalized index value is as follows: For each column of data in the initial decision data table that represents the predicted throughput and the estimated transmission delay, calculate the arithmetic square root of the sum of the squares of all the values ​​in the data column. Divide each value in the data column by the arithmetic square root calculated for the same data column to obtain the normalized value; The maximum value of the normalized throughput and the minimum value of the normalized latency in all TCP and UDP connections together constitute the positive ideal solution. The minimum normalized throughput value and the maximum normalized latency value among all TCP and UDP connections together constitute the negative ideal solution.

9. The hybrid parallel transmission optimization method based on multi-path network TCP-UDP according to claim 7, characterized in that, The formula for calculating the relative proximity of each TCP and UDP connection is as follows: ； in, It is a connection The group utility value, It is a connection Individual regret value, and It is all connections The minimum and maximum values ​​in the value range. and It is all connections The minimum and maximum values ​​in the value range. It is the decision-making mechanism coefficient. It is a connection Ranking indicators; The relative proximity of each TCP and UDP connection is determined by Sure.

10. The hybrid parallel transmission optimization method based on multi-path network TCP-UDP according to claim 7, characterized in that, The specific steps for obtaining the transmission optimization results are as follows: S511: Schedule the data stream to be transmitted in parallel to the TCP or UDP connection corresponding to the optimal transmission path identifier for transmission, and at the same time determine the category of the selected protocol type, and establish the scheduling path and protocol determination result; S512: Based on the scheduling path and protocol determination result, if the determined protocol is TCP, the queue length value of the corresponding connection in the real-time sending queue length set is called and compared with the preset queue threshold. Based on the comparison result, it is determined whether the data packet fragment size needs to be adjusted, and the TCP fragmentation adjustment criterion is obtained. S513: Based on the transmission path determined in the scheduling path and protocol determination result, according to the TCP fragmentation adjustment criterion, when it is determined that the data packet fragment size needs to be adjusted, the data packet fragment size of the TCP connection is synchronously reduced to obtain the transmission optimization result.

Images