CN103873479B - A kind of panel data transmission algorithm assessed based on cross-layer - Google Patents

Abstract

The invention proposes a parallel data transmission method based on cross-layer evaluation, which solves the problem of out-of-sequence and reordering of data packets caused by simple data scheduling in the multi-path parallel transmission process of the flow control transmission protocol. In order to obtain accurate path quality assessment, the heartbeat mechanism is used to obtain the RTT time and determine the path activity, and the path capacity is obtained according to the effective signal-to-noise ratio obtained from the data link layer and the estimated rate and bandwidth obtained at the transport layer. jointly determine the path quality. Priorities are determined based on path quality, and data distribution and scheduling are intelligently performed. Use the path capacity to determine the cause of packet loss, and take different measures for retransmission. This method can perceive the dual information of the data link layer and the transport layer, overcome the information incompleteness of the traditional data distribution method, obtain a more accurate and comprehensive path quality assessment, perform efficient data distribution, and reduce packet loss and reordering. At the same time, distinguish the cause of packet loss to avoid unnecessary reduction of the congestion window. Through the parallel data transmission method based on cross-layer evaluation, it can perceive and adapt to the dynamics of wireless networks, provide high-quality data transmission services, and improve user experience.

Description

A Parallel Data Transfer Algorithm Based on Cross-Layer Evaluation

technical field

The invention relates to the technical field of communication and communication data transmission technology. Specifically, it involves evaluating the path quality through a cross-layer method on the flow control transmission protocol and the realization of multi-path parallel transmission based on the path quality.

Background technique

With the development of the Internet and related technologies, mobile devices tend to have multiple network interfaces, and at the same time have stronger computing capabilities, storage capabilities, and more advanced communication functions. In the future, the network will support a variety of network access anytime and anywhere. Multihoming technology is generally considered to be able to apply these access devices to provide network diversity. This makes CMT (Concurrent Multipath Transfer) the best way to improve wireless communication performance. CMT can use multiple network interfaces to transmit data in parallel. Compared with single path, multipath has greater bandwidth, and errors on one path can be recovered on another path. More importantly, the sender can perform data scheduling and congestion balancing among multiple paths to maximize the quality of service (QoS) of network services and applications. Generally speaking, in the parallel distributed system of wireless network, CMT has better bandwidth aggregation ability, fault tolerance and load balancing ability.

SCTP (Stream Control Transmission Protocol) is a new transport layer protocol developed by IETF that supports multi-homing features. It can support multiple IP addresses on the same association at the same time to achieve multi-path transmission. The inherent multi-homing feature of SCTP makes it very promising for CMT implementation. So far, IETF has made some valuable progress on SCTPCMT, but there are still many important and challenging work to be done. In SCTP CMT, the key consideration is how to deal with data reordering and packet loss. Due to the large differences and dynamics of each path parameter, the traditional round-robin scheduling will cause buffer blocking, that is, serious out-of-order reception and reordering of data packets in the overloaded buffer. A great deal of work has been done to decide how to distribute data across multiple paths and reduce reordering, but most of the work is based on data exchange, monitoring send-receive interactions, or simply relying on connection parameters at the transport layer. Since the time and frequency variation characteristics of the wireless link are not directly reflected in the transport layer, the information obtained by these mechanisms on the wireless connection is incomplete and may not be accurate.

In addition to this, because the wireless channel is very unreliable, packet loss often occurs and is confused with packet reordering. In a wireless environment, packet loss is often caused by wireless errors, and it is inappropriate to treat it as congestion according to the original design of SCTP/TCP. If the window halving algorithm is sent repeatedly by mistake, the performance of the CMT will be degraded thereby. Therefore, it is very important to determine whether packet loss is caused by congestion or wireless error, and then take different corresponding measures for packet loss. Obviously, due to ignorance of the physical layer and data link layer, the transport layer cannot accurately determine the cause of packet loss.

SUMMARY OF THE INVENTION - CLAIMS SECTION

In view of this, the present invention proposes a parallel data transmission method based on cross-layer evaluation. During the multi-channel parallel transmission process of the flow control transmission protocol, the path quality is obtained according to the information of the transmission layer and the data link layer, and according to the path quality Priority intelligently distributes and transmits data to provide high-performance data transmission services. The present invention designs an effective signal-to-noise ratio calculation module and a rate & bandwidth estimation module to obtain the information of the data link layer and the transmission layer respectively, adopts a path quality evaluation model based on cross-layer cooperation to judge the path quality, and designs a A parallel data distribution algorithm based on path quality is used for data distribution, and finally a perceptual retransmission strategy is proposed for retransmission processing. The invention can effectively evaluate the path quality, intelligently carry out data distribution and scheduling, adapt to the dynamic nature of the wireless network, and provide effective multimedia transmission for mobile users.

1. A method for parallel data transmission based on cross-layer evaluation, the steps of which include:

a) First, the path activity is judged at the transport layer, the heartbeat message mechanism HEARTBEAT is used to measure the path activity, the path activity is judged, and the operation is performed on the active path;

b) Then, for the active path, calculate the effective signal-to-noise ratio ESNR at the data link layer, count the frames that need to be retransmitted to obtain the frame error rate FER, and calculate the bit error rate BER by FER, according to BER and signal-to-noise ratio The relationship between SNR obtains the effective signal-to-noise ratio ESNR, and periodically reports to the transport layer to realize the cross-layer evaluation of path quality;

c) Then, rate & bandwidth estimation is performed on the active path at the transport layer, the occupancy of path resources is obtained through time average rate value estimation, and the real-time change of the path is tracked through the real-time bandwidth estimation value;

d) Next, design a parallel data distribution strategy based on path quality at the transport layer, integrate the ESNR information periodically reported by the data link layer in b and the rate & bandwidth parameters detected by this layer in c, to achieve dual cognition of the path state, Solve the problem of inaccurate control caused by the ignorance of the transport layer to the data link layer information; by establishing a path quality evaluation model based on cross-layer cooperation, using cross-layer parameters to obtain the remaining path capacity, and setting the path data distribution priority according to the path capacity, forming Path list SL; design a parallel data distribution algorithm based on path quality, according to the order of path list SL, distribute appropriate amount of data to each transport layer path according to the remaining capacity of each path, and the total amount of data is limited by the window;

e) Finally, design a perceptual retransmission strategy at the transport layer, analyze the cause of packet loss according to the remaining capacity of the path, and adopt different retransmission algorithms for packet loss caused by wireless bit errors and congestion.

2. A method for parallel data transmission based on cross-layer evaluation as described in 1, wherein the path activity judgment includes:

a) Set dynamic estimation interval u for each path, compare u and RTO time to judge path activity;

b) Send two HEARTBEAT chunks to detect the path. If retransmission occurs and exceeds the maximum number of retransmissions, set the path as inactive; if two ACKs are successfully received, set the path as active; retransmission occurs but two ACKs are received ACK to allow data allocation even if the path is inactive.

3. A method for parallel data transmission based on cross-layer evaluation as described in 1, wherein the effective signal-to-noise ratio (ESNR) calculation includes:

a) On the data link layer, mark the frame that needs to be retransmitted as error_frame, count the error_frame, and obtain FER by recording the wrong frame and all sent frames in the measurement interval; use the independent bit model to obtain BER from FER, Finally, ESNR is obtained according to the relationship of BER-SNR.

4, a kind of parallel data transmission method based on cross-layer evaluation as described in 1, it is characterized in that, rate & bandwidth estimation comprises:

a) Utilize the amount of data successfully sent in the estimated interval and the total time these data occupy the cache to obtain the average sending rate R _e of the path in an estimated interval; the available bandwidth is the ratio of the average packet length to the average entry-exit time Indicates that the time moving average filter is used to obtain bandwidth samples, and then the Kalman filter is used for prediction-verification processing to obtain the estimated bandwidth value BW _e .

5. A method for parallel data transmission based on cross-layer evaluation as described in 1, characterized in that the parallel data distribution strategy based on path quality includes:

a) Path quality evaluation model based on cross-layer cooperation _{: Calculate the remaining available capacity of the path from the effective signal-to-noise ratio ESNR of cross-layer parameters, the average transmission rate Re and the estimated value of available bandwidth BW e ,} and set the data distribution priority according to the path capacity, Form a path list SL;

b) Parallel data distribution algorithm based on path quality: according to the order of the path list SL, an appropriate amount of data is distributed to the path according to the remaining capacity of each path in turn. When the remaining capacity of the path R _c is positive, the amount of distributed data can be On the basis of the congestion window, an appropriate amount will be increased; otherwise, the data will be conservatively distributed according to the window value; the total amount of sent data is limited by the window.

6. A parallel data transmission method based on cross-layer evaluation as described in 1, wherein the perceptual retransmission strategy includes:

a) When fast retransmission detects packet loss, the sender judges the cause of packet loss according to the remaining capacity of the path _Rc . If the value is positive, it is judged to be packet loss caused by a wireless error, and only the threshold value is adjusted; otherwise, it is judged as Packet loss due to congestion, reducing the congestion window while adjusting the threshold;

b) When overtime retransmission detects packet loss, the standard SCTP retransmission processing method is used to adjust the congestion window and threshold value.

The present invention has following technical effect:

1. In the present invention, the data link layer periodically reports ESNR information to the transport layer. Obtaining ESNR from FER measurement is simple and convenient, and solves the inaccuracy caused by the inability to deal with co-channel interference, signal multi-path influence, frequency selective fading and different frame sizes in actual direct SNR measurement. The ESNR integrates various factors in wireless transmission, simplifies complex wireless communication, and can well reflect the path status of wireless communication to the upper layer.

2. In the present invention, the transport layer performs dual recognition of the path state. One is rate estimation, which provides a relatively long-term path average transmission rate value, which reflects the resource occupation of the path; the other is bandwidth estimation, which shows the real-time transmission rate of each path, and reflects the real-time changes of the path. Dual cognition can obtain a clear cognition of the path status and provide effective information for the decision-making of the transport layer.

3. In the present invention, the comprehensive information of the path is obtained by using the cross-layer parameters effective signal-to-noise ratio ESNR, the average transmission rate _Re and the estimated value of available bandwidth BW _e to evaluate the path quality. The cross-layer cooperation method can obtain complete wireless connection information, provide comprehensive path quality assessment for parallel data distribution algorithms, enable efficient and reasonable application of paths, and reduce data rearrangement and packet loss caused by improper data distribution algorithms.

Description of drawings

Fig. 1 is the overall framework of the parallel data transmission algorithm of the present invention;

Fig. 2 is a flowchart of the path activity judgment algorithm;

Fig. 3 is a flow chart of the path bandwidth estimation algorithm;

Figure 4 is a flow chart of the data distribution algorithm;

FIG. 5 is a flow chart of a perceptual retransmission strategy algorithm;

detailed description

In order to facilitate the understanding of the technical means, creative features, objectives and effects of the present invention, the present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

1. The overall framework of the system

In order to solve the data reordering and packet loss problems caused by the diversity of paths in SCTP CMT, the present invention proposes a parallel data transmission algorithm based on cross-layer evaluation. As shown in FIG. 1 , it is the overall framework of the system of the present invention, including the sending end, the receiving end and the multipath on the wireless network. The SCTP receiving end receives the data sent from the sending end, and then feeds back the receiving status. The SCTP sender collects feedback, obtains the path quality through the status of the transport layer and the information of the data link layer, and distributes the transmission data according to the path quality.

The cross-layer evaluation model uses the effective signal-to-noise ratio ESNR, estimated rate value, and estimated bandwidth value to judge the path capacity, and sends HEARTBEAT to judge the path activity, and the two constitute the path quality. Parallel data distribution scheduling uses the parameters obtained from the cross-layer evaluation model to distribute data sequentially and by volume. Packet loss is handled by a perceptual retransmission policy.

2. Path Activity Judgment

In cross-layer cooperation, the first question is how often ESNR needs to be calculated and path quality updated. A dynamic estimation interval u is set for each path. When packet loss occurs, use the confidence interval to obtain an interval sample x _k .

in is the average value, σ is the standard deviation, and Z _1-ε/2 is a function of ε. When the previous estimation interval is completed, select the lower bound to update the value of u.

The estimated interval is related to path activity: after u is updated, if a path has u≤RTO, that is, packet loss often occurs, the path is set as inactive; otherwise, the path is set as active. These two states are as defined in standard SCTP.

However, this algorithm has shortcomings. Because u is a statistical result based on historical samples, the sender can only know the long-term performance of the path. If the path fluctuates suddenly, there is no guarantee that such path activity will change the data distribution rapidly. Furthermore, SCTP CMT allows replying SACK on another path, and the traditional RTT measurement method may bring wrong delay estimation, so it cannot be used for path quality evaluation. Here, we do not use the RTT measurement method of DATA chunks. Since HEARTBEAT_ACK must be replied on the same path as HEARTBEAT, we extract the RTT time of HEATBEAT chunks as the reference RTT ie RTT _ref .

In order to get accurate path activity and RTT _ref , we change SCTP's fixed HEARTBEAT interval to dynamic u. At the beginning of each estimation interval, the sender sends two HEARTBEAT chunks to probe the path. If the path replies with corresponding two HEARTBEAT_ACKs within an RTO time, the error counter is cleared, and the path is set to allocate data even if u≤RTO at this time. Otherwise, the error counter counts and HEARTBEAT is resent. When the counter reaches Path.Max.Retrans, the path is made inactive. When the sender successfully receives two HEARTBEAT_ACKs, select the shorter RTT as the RTT _ref . Finally, we can get n available active paths, all inactive paths must not participate in the following process.

Figure 2 is a flowchart of the algorithm, and its steps are as follows:

1) Judging the path activity according to u: if u≤RTO, set the path as inactive, otherwise set it as active.

2) Send two HEARTBEATS, start the timers of the two HEARTBEATS respectively, set the path as non-distributable data, and set the counter HBcount that records the number of times of successfully receiving HEARTBEAAT_ACK to zero.

3) If the value of the counter HBcount is less than 2, go to step 4, otherwise, go to step 6.

4) For each HEARTBEATS sent, if HEARTBEAT_ACK is successfully received, the RTT time is recorded, the counter HBcount is incremented by 1, and the corresponding timer is stopped; if the timer expires, HEARTBEAT is retransmitted, the timer restarts, and the timeout counter error is incremented by 1 .

5) If the value of the counter error is greater than the maximum number of retransmissions of the path, set the path as inactive and go to step 6. Otherwise go back to step 3.

6) If the value of the counter HBcount is 2, that is, two HEARTBEAT_ACKs are successfully received, then proceed to step 7, otherwise the algorithm ends.

7) If the path state is inactive at this time and the value of the error counter is 0, that is, there is no HEARTBEAT overtime, then the path state is changed to active.

8) Set the path as distributable data, clear the timeout counter error, and select a shorter RTT time as RTT _ref . Algorithm ends.

3. Calculation of effective signal-to-noise ratio (ESNR)

Wireless channels are unpredictable and unreliable due to different types of noise, fading and interference. In general, SNR can provide an overall assessment of wireless communications. However, in actual SNR measurement, existing devices only measure the preamble and header of the PLCP at the physical layer, and do not consider data frames with a higher rate. Furthermore, this direct-reported SNR cannot handle co-channel interference, signal multipath effects, frequency-selective fading, and different frame sizes. So we adopt the concept of ESNR to solve the deficiency of directly reporting SNR.

In a specific wireless communication, there is a known and definite one-to-one relationship between BER and SNR. Assuming that in the current environment, BER=F(SNR), ESNR is defined as:

ESNR＝F ^-1 (BER) (3)

Calculating SNR can be converted to measuring BER. The direct measurement of BER will bring a lot of overhead in the physical layer, so choose to use the FER of the data link layer to calculate BER. It is very convenient to record transmitted frames and FER over an estimated interval u. On the data link layer, once a data frame requires retransmission, this frame is marked as error_frame. By recording error_frame and all frames sent, FER can be obtained:

Here, error_frame_num is the number of error_frames in an estimation interval, and total_frame_num is the total number of frames sent in an estimation interval. If there is a path with FER=0, set the FER of the path to a reasonable minimum value. A frame is successfully received if and only when all bits of a frame are successfully decoded, so the BER can be obtained by the following formula:

Here, L is the number of bytes in a data frame. Then by formula (3), the ESNR can be calculated.

4. Rate & Bandwidth Estimation

During the processing of SCTP data exchange, some data is already being transmitted and occupying path resources. We employ a time-averaged rate estimate to inform the system about the occupancy of path resources. Define R _e as the average transmission rate of the path in an estimation interval:

Here, sendsize is the amount of data that has been successfully sent in the estimated interval, T _e is the time when the first data enters the path buffer, and T _l is the time when the last data leaves the buffer.

Change the view of bandwidth estimation to send duration, that is: the available bandwidth is equal to the ratio of the average packet length to the average entry-exit time. In order to eliminate fluctuations caused by random transmission behavior, bandwidth samples must first be smoothed through a time-averaging filter. Then use the Kalman filter to get the bandwidth estimate. The Kalman filter is a discrete-time recursive filter. Firstly, the time update is performed to predict the current bandwidth state through the previous results; and then the measurement update is performed to bring the new sample into the prior estimate to obtain the posterior estimate.

As shown in Figure 3, the algorithm flow for path bandwidth estimation:

1) Obtain data length samples and time samples.

2) Send data. Record the time when the last data was sent.

3) Update the average data length and the average time length, and calculate the bandwidth samples.

4) Perform time update to obtain prior estimation values of bandwidth and error variance.

5) Perform measurement update, and obtain a posteriori estimated values of bandwidth and error variance according to the bandwidth samples.

5. Parallel data distribution strategy based on path quality

a) Path quality assessment model based on cross-layer cooperation

Choose to distribute data in another dynamic cycle. In this way, cross-layer evaluation and data distribution can be combined simultaneously but at different paces. The distribution period Pd is _dynamically related to the processing capability of each path. The processing time for one path is:

In order to fully utilize all active paths, P _d chooses the maximum processing time:

After that, the data will be distributed to all candidate paths simultaneously within _Pd time. In a fixed environment, the distribution frequency is:

The path capacity C can be obtained in a cross-layer manner:

C＝BW _e ·log ₂ (1+ESNR) (9)

Since a path has been occupied by R _e , the remaining capacity R _c of the path is:

R _c =CR _e (10)

Next, arrange all candidate paths in descending order according to R _c to form a new and classified path list SL. This makes paths with larger R _c have higher priority for data distribution.

In the distribution interval P _d , the maximum amount of data that can be distributed D _max is limited by the following formula:

b) Parallel data distribution algorithm based on path quality

According to the queue order of SL, the right amount of data is distributed to the path according to the _Rc of each path in turn. If R _c is a positive value, it means that the path has potential processing capacity to increase its throughput, and the amount of data distributed is the current cwnd plus some R _c . Otherwise, just conservatively distribute the data as the difference between cwnd and outstanding. In this period P _d , as long as cwnd allows, the amount of distributed data D will gradually approach D _max . If the data distribution amount has reached D _max , the data distribution task in this period is completed and the data distribution is stopped immediately. Wait for the end of the period, and then recalculate T _handle and the next P _d . Update C and R _c , and repeat the algorithm at the next P _d to continue data distribution.

As shown in Figure 4, the flow of the data distribution algorithm:

1) For the path i in the list SL, if _Rc is greater than zero, set the amount of distributed data D according to formula (11); otherwise, set the amount of distributed data according to formula (12):

2) Calculate the distributed data D _total =D _total +iD.

3) If D _total ≤ D _max , distribute data iD on path i when cwnd permits; if D _total >D _max , stop data distribution and wait for the next data distribution time.

4) For the next path, repeat steps 1-3 until all paths are allocated.

6. Perceptual retransmission strategy

If packet loss occurs during data exchange, the cause of packet loss can be analyzed according to R _c . And take different steps to recover lost packets.

When fast retransmission detects packet loss, the sender can judge the cause of packet loss: if R _c > 0, it means that the path is not congested, attribute the packet loss to wireless bit error, and adjust ssthresh but not cwnd:

ssthresh=max(R _e RTT _ref ,cwnd/2,4 MTU) (13)

Where _Re is the result of rate estimation, and RTT _ref refers to the reference RTT mentioned above. We use periodic rate estimates because wireless bit errors are random and bursty, and their adjustments need to be brought back to the average. On the contrary, if R _c <0, the path is overutilized, attribute packet loss to congestion, and adjust cwnd while adjusting ssthresh:

ssthresh=max(BW _e RTT _ref ,cwnd/2,4 MTU) (14)

cwnd=min(cwnd,ssthresh) (15)

where BW _e is the estimated value of the bandwidth. We use real-time bandwidth estimates because congestion represents a change in the nature of the connection.

For timeout retransmission, because the RTO is a relatively long time, the timeout of the T3-rtx timer indicates serious congestion or wireless errors, and we only adopt the standard SCTP processing method conservatively.

After adjustment, the path with the largest cwnd is chosen and lost packets are retransmitted as quickly as possible. Fig. 5 is the flow of the retransmission algorithm.

Claims (6)

1. A method for parallel data transfer based on cross-layer evaluation, the steps comprising: a) First, the path activity is judged at the transport layer, the heartbeat message mechanism HEARTBEAT is used to measure the path activity, the path activity is judged, and the operation is performed on the active path; b) Then, for the active path, calculate the effective signal-to-noise ratio ESNR at the data link layer, count the frames that need to be retransmitted to obtain the frame error rate FER, and calculate the bit error rate BER by FER, according to BER and signal-to-noise ratio The relationship between SNR obtains the effective signal-to-noise ratio ESNR, and periodically reports to the transport layer to realize the cross-layer evaluation of path quality; c) Then, rate & bandwidth estimation is performed on the active path at the transport layer, the occupancy of path resources is obtained through time average rate value estimation, and the real-time change of the path is tracked through the real-time bandwidth estimation value; d) Next, design a parallel data distribution strategy based on path quality at the transport layer, integrate the ESNR information periodically reported by the data link layer in b and the rate & bandwidth parameters detected by this layer in c, to achieve dual cognition of the path state, Solve the problem of inaccurate control caused by the ignorance of the transport layer to the data link layer information; by establishing a path quality evaluation model based on cross-layer cooperation, using cross-layer parameters to obtain the remaining path capacity, and setting the path data distribution priority according to the path capacity, forming Path list SL; design a parallel data distribution algorithm based on path quality, according to the order of path list SL, distribute appropriate amount of data to each transport layer path according to the remaining capacity of each path, and the total amount of data is limited by the window; e) Finally, design a perceptual retransmission strategy at the transport layer, analyze the cause of packet loss according to the remaining capacity of the path, and adopt different retransmission algorithms for packet loss caused by wireless bit errors and congestion. 2. A kind of parallel data transmission method based on cross-layer evaluation as claimed in claim 1, it is characterized in that, path activity judgment comprises: a) Set dynamic estimation interval u for each path, compare u and RTO time to judge path activity; b) Send two HEARTBEAT chunks to detect the path. If retransmission occurs and exceeds the maximum number of retransmissions, set the path as inactive; if two ACKs are successfully received, set the path as active; retransmission occurs but two ACKs are received ACK to allow data allocation even if the path is inactive. 3. a kind of parallel data transmission method based on cross-layer evaluation as claimed in claim 1, is characterized in that, effective signal-to-noise ratio (ESNR) calculates, comprises: a) On the data link layer, mark the frame that needs to be retransmitted as error_frame, count the error_frame, and obtain FER by recording the wrong frame and all sent frames in the measurement interval; use the independent bit model to obtain BER from FER, Finally, ESNR is obtained according to the relationship of BER-SNR. 4. A kind of parallel data transmission method based on cross-layer evaluation as claimed in claim 1, it is characterized in that, rate & bandwidth estimation comprises: a) Utilize the amount of data successfully sent in the estimated interval and the total time these data occupy the cache to obtain the average sending rate R _e of the path in an estimated interval; the available bandwidth is the ratio of the average packet length to the average entry-exit time Indicates that the time moving average filter is used to obtain bandwidth samples, and then the Kalman filter is used for prediction-verification processing to obtain the estimated bandwidth value BW _e . 5. A kind of parallel data transmission method based on cross-layer evaluation as claimed in claim 1, it is characterized in that, the parallel data distribution strategy based on path quality comprises: a) Path quality evaluation model based on cross-layer cooperation _{: Calculate the remaining available capacity of the path from the effective signal-to-noise ratio ESNR of cross-layer parameters, the average transmission rate Re and the estimated value of available bandwidth BW e ,} and set the data distribution priority according to the path capacity, Form a path list SL; b) Parallel data distribution algorithm based on path quality: according to the order of the path list SL, an appropriate amount of data is distributed to the path according to the remaining capacity of each path in turn. When the remaining capacity of the path R _c is positive, the amount of distributed data can be On the basis of the congestion window, an appropriate amount will be increased; otherwise, the data will be conservatively distributed according to the window value; the total amount of sent data is limited by the window. 6. A method for parallel data transmission based on cross-layer evaluation as claimed in claim 1, wherein the perceptual retransmission strategy comprises: a) When fast retransmission detects packet loss, the sender judges the cause of packet loss according to the remaining capacity of the path _Rc . If the value is positive, it is judged to be packet loss caused by a wireless error, and only the threshold value is adjusted; otherwise, it is judged as Packet loss due to congestion, reducing the congestion window while adjusting the threshold; b) When overtime retransmission detects packet loss, the standard SCTP retransmission processing method is used to adjust the congestion window and threshold value.