CN101645765A - Reliable transmission acceleration method facing networks with high error rate and long delay characteristics - Google Patents

Abstract

The invention relates to a reliable transmission acceleration method facing networks with high error rate and long delay characteristics, belonging to a reliable end-to-end transmission method of internet. In the invention, reasons that the network loses packages are distinguished according to the following method on the basis of a FAST TCP protocol: when a sending end receives 3 repeated ACK packages, the difference value of the transmission speed of a previous RTT and the transmission speed of a present RTT can be determined, when the transmission speed of the present RTT is larger than thetransmission speed of the previous RTT, the network does not have congestion, and the reason for losing package is link error; and when the transmission speed of the present RTT is greatly smaller than the transmission speed of the previous RTT, then the network can have congestion, and the package loss can be determined to be caused by network congestion. Calculated RTT can accurately and timelyreflect the current network status through the invention, thereby the reflection sensitivity of a congestion window of the sending end to the network status can be enhanced.

Description

Reliable transmission acceleration method towards high bit-error, long delay characteristics network

Technical field

The present invention relates to the method for the end-to-end reliable transmission in the Internet, particularly towards the monolateral reliable transmission acceleration method of high bit-error, long delay network.

Background technology

Along with the development of radio network technique and deepening continuously of application, all kinds wireless network (as satellite network, aeronautical communications network, mobile radio network etc.) is day by day obvious with the convergence that traditional ground is fixed with spider lines.The evolution of network causes the structure of network complicated more, and the isomerism of network is strengthened.From application point of view, this class network have the higher bit error rate (Bit Error Rate, BER), characteristic such as long delay, the bandwidth time delay of network long-pending (Bandwidth Delay Product, BDP) bigger.(Transmission Control Protocol in the time of TCP), can cause transmission performance to go down significantly to use traditional reliable transport protocol on this class network.Originally this be to design for wired network because of traditional Transmission Control Protocol, error rate is quite low in wired network, it is generally acknowledged that packet loss is all caused by network congestion, but in the network environment of some high bit-errors, long delay, the reason of packet loss no longer is unique, it may be to be caused by congested, also may be to be caused by link error; Simultaneously because the characteristic of long delay makes the speed of network recovery slow down.Therefore in the network of high bit-error, long delay characteristics, how effectively to distinguish the packet loss reason and take relevant transmission method (promptly when congestion packet loss, to call congestion avoidance algorithm, reduce data package transmission velocity; When link error caused packet loss, explicit re-transmission kept original transmission rate) become the transmission acceleration, improve the key of reliable transmission performance.

At the problems referred to above, many improving one's methods proposed, wherein having of current main-stream shows congestion notification mechanism method (Explicit Congestion Notification, ECN), confirm (Acknowledgement, ACK) the TCP-Westwood method of effective bandwidth measuring technique based on the TCP-Peach method of NIL packet detection with based on receiving terminal.Wherein the ECN method needs intermediate router to do special support, and is not strong with the interoperability of existing network configuration; The TCP-Peach method then is packet loss of link to be judged that need intermediate router to support priority policy, probe messages has also taken the network bandwidth simultaneously, has increased the burden of network by sending probe messages; The TCP-Westwood method is to measure the ACK speed of on average returning by the source end in a TCP connection to finish effective bandwidth estimation end to end, can utilize amount of bandwidth running into when congested congestion window and start the link that threshold value is adjusted into estimation slowly at every turn, but test shows that TCP-Westwood scheme performance in high bit-error, long delay network is still very low.

Summary of the invention

The purpose of this invention is to provide a kind of reliable transmission acceleration method (New Fast Tcp) towards high bit-error, long delay characteristics network, to judge that current network packet loss reason is that network congestion or link error cause, utilize judged result to adjust the transmission strategy of transmitting terminal in the reliable transport protocol again.

The object of the present invention is achieved like this: a kind of reliable transmission acceleration method towards high bit-error, long delay characteristics network, on the basis of FAST Transmission Control Protocol, distinguish the reason of network generation packet loss according to following method: when transmitting terminal is received 3 repeat ACK bags, can judge the transmission rate of previous RTT and the difference of current RTT transmission rate, if the transmission rate of current RTT is during greater than the transmission rate of previous RTT, also do not have congested generation in the expression network, the packet loss reason is a link error; If the transmission rate of current RTT then represents to have congested generation in the network much smaller than the transmission rate of previous RTT, can think that packet loss is owing to network congestion causes;

Transmission rate (the TP of current RTT _{Current_rtt}) computing formula as follows:

${\mathrm{TP}}_{\mathrm{Current}\_\mathrm{rtt}}=\frac{\mathrm{snd}\_\mathrm{nxt}-\mathrm{seg}\_\mathrm{ack}}{\mathrm{avgRTT}\×\mathrm{snd}\_\mathrm{mss}}---\left(1\right)$

Wherein, snd_nxt is the sequence number of the section that is being sent out, and seg_ack is the sequence number of the section confirmed of the current ACK bag of receiving, both differences are the data volumes that sends in the current RTT, snd_mss is the maximum segment size of transport layer, and avgRTT is current average RTT;

diff＝TP _{Last_rtt}-TP _{Current_rtt} (2)

Wherein, diff is the transmission rate (TP of previous RTT _{Last_rtt}) with the transmission rate (TP of current RTT _{Current_rtt}) poor; The method of judging the packet loss reason according to threshold parameter α is as follows:

(a) if diff＞α thinks that then packet loss is caused by network congestion;

(b) if diff≤α thinks that then packet loss is caused by link error; Threshold parameter α value 0.99～1.

Transmitting terminal is taked following transmission strategy after distinguishing the packet loss reason:

(a) if packet loss is that then transmitting terminal calls congestion avoidance algorithm owing to network congestion, the bag of retransmission of lost enters the fast quick-recovery stage then fast, reduces data package transmission velocity;

(b) if because the packet loss that causes of link error, then explicit re-transmission enters fully and recovers the stage, and being about to congestion window increases by 3, keeps original transmission rate.

According to current average RTT and current variable δ of RTT value structure that measures, its value can obtain by following formula:

$\mathrm{\δ}=\frac{|T_i\left(k\right)-{\stackrel{\‾}{T}}_i\left(k\right)|}{{\stackrel{\‾}{T}}_i\left(k\right)}---\left(5\right)$

T wherein _i(k) be the RTT value that k is measured, T _i(k) be k RTT value that measures; Construct two groups of threshold values again, be respectively { δ ₁, δ ₂..., δ _i..., δ _nAnd { λ ₁, λ ₂..., λ _i..., λ _n, λ _N+1, 0≤δ wherein ₁＜δ ₂... δ _i＜δ _I+1...＜δ _n, 0≤λ ₁＜λ ₂... λ _i＜λ _I+1... λ _n＜λ _N+1＜1, the computing formula of improving the smoothing factor η (t) in the computational methods of calculating average RTT in the back is:

$\mathrm{\&eta;}\left(t\right)=\left\{\begin{array}{cc}{\mathrm{\&lambda;}}_1,& 0\&le;\mathrm{\&delta;}\&le;{\mathrm{\&delta;}}_1\\ {\mathrm{\&lambda;}}_i,& {\mathrm{\&delta;}}_{i-1}<\mathrm{\&delta;}\&le;{\mathrm{\&delta;}}_i,1<i\&le;n\\ {\mathrm{\&lambda;}}_{n+1},& \mathrm{\&delta;}>{\mathrm{\&delta;}}_n\end{array}\right.---\left(6\right)$

Can utilize formula (5) and formula (6), upgrade, can reflect the average RTT value of the available bandwidth situation of network exactly according to formula (3);

T _i(k+1)＝(1-η(t _k))T _i(k)+η(t _k)T _i(k) (3)。

The method of distinguishing the packet loss reason is as follows:

The basic principle of this method is: because the transmission rate of adjacent R TT can reflect the situation of change of the available bandwidth of network, if the transmission rate of current RTT is during greater than the transmission rate of previous RTT, the expression network does not also have saturated, does not have congested generation in the network; If it is saturated that the transmission rate of current RTT, represents then that network has begun much smaller than the transmission rate of previous RTT, has congested generation in the network.Therefore, when transmitting terminal is received 3 repeat ACK bags, can judge the transmission rate of previous RTT and the difference of current RTT transmission rate, if the transmission rate of current RTT is during greater than the transmission rate of previous RTT, also do not have congested generation in the expression network, the packet loss reason is a link error; If the transmission rate of current RTT then represents to have congested generation in the network much smaller than the transmission rate of previous RTT, can think that packet loss is owing to network congestion causes.Transmission rate (the TP of current RTT wherein _{Current_rtt}) computing formula as follows:

${\mathrm{TP}}_{\mathrm{Current}\_\mathrm{rtt}}=\frac{\mathrm{snd}\_\mathrm{nxt}-\mathrm{seg}\_\mathrm{ack}}{\mathrm{avgRTT}\&times;\mathrm{snd}\_\mathrm{mss}}---\left(1\right)$

Wherein, snd_nxt is the sequence number of the section that is being sent out, and seg_ack is the sequence number of the section confirmed of the current ACK bag of receiving, both differences are the data volumes that sends in the current RTT, snd_mss is the maximum segment size of transport layer, and avgRTT is current average RTT.

diff＝TP _{Last_rtt}-TP _{Current_rtt} (2)

Wherein diff is the transmission rate (TP of previous RTT _{Last_rtt}) with the transmission rate (TP of current RTT _{Current_rtt}) poor, it can reflect the available bandwidth situation of network.We introduce a threshold parameter α, judge that the method for packet loss reason is as follows:

(a) if diff＞α thinks that then packet loss is caused by network congestion;

(b) if diff≤α thinks that then packet loss is caused by link error.

Through a large amount of analysis of experiments, effect is better when obtaining threshold parameter α and getting the value of 0.99～1 scope.

Transmitting terminal is taked following transmission strategy after distinguishing the packet loss reason:

(a) if packet loss is that then transmitting terminal calls congestion avoidance algorithm owing to network congestion, the bag of retransmission of lost enters the fast quick-recovery stage (the fast quick-recovery of TCP-NewReno) then fast, reduces data package transmission velocity;

(b) if because the packet loss that causes of link error, then explicit re-transmission enters fully and recovers the stage, and being about to congestion window increases by 3, keeps original transmission rate;

The computational methods of the average RTT of New Fast TCP are as follows:

The algorithm that calculates average RTT in the FAST Transmission Control Protocol is:

T _i(k+1)＝(1-η(t _k))T _i(k)+η(t _k)T _i(k) (3)

η(t)＝min(3/w _i(t)，1/4) (4)

T wherein _i(k+1) be (to be T to the average RTT of k _i(k)) renewal, T _i(k) be k RTT value that measures, t _kBe to measure T _i(k) the moment, η (t) and w _i(t) be smoothing factor and window value respectively at moment t.

From formula (4) value used value 1/8 much smaller than traditional Transmission Control Protocol of η (t) as can be seen, so the RTT after level and smooth can not reflect the available bandwidth situation of network exactly, make that the renewal of congestion window is also insensitive, can not utilize the bandwidth of link well, so a kind of improvement project of calculating average RTT has been proposed this patent.It improves one's methods as follows:

Set a variable δ, its value can obtain by following formula:

$\mathrm{\&delta;}=\frac{|T_i\left(k\right)-{\stackrel{\&OverBar;}{T}}_i\left(k\right)|}{{\stackrel{\&OverBar;}{T}}_i\left(k\right)}---\left(5\right)$

Set two groups of threshold values, be respectively { δ ₁, δ ₂..., δ _i..., δ _nAnd { λ ₁, λ ₂..., λ _i..., λ _n, λ _N+1, 0≤δ wherein ₁＜δ ₂... δ _i＜δ _I+1...＜δ _nAnd 0≤λ ₁＜λ ₂... λ _i＜λ _I+1... λ _n＜λ _N+1＜1, the computing formula of improving the smoothing factor η (t) in the computational methods of calculating average RTT in the back is:

$\mathrm{\&eta;}\left(t\right)=\left\{\begin{array}{cc}{\mathrm{\&lambda;}}_1,& 0\&le;\mathrm{\&delta;}\&le;{\mathrm{\&delta;}}_1\\ {\mathrm{\&lambda;}}_i,& {\mathrm{\&delta;}}_{i-1}<\mathrm{\&delta;}\&le;{\mathrm{\&delta;}}_i,1<i\&le;n\\ {\mathrm{\&lambda;}}_{n+1},& \mathrm{\&delta;}>{\mathrm{\&delta;}}_n\end{array}\right.---\left(6\right)$

Can select the size of corresponding threshold value number and threshold value according to actual conditions, select suitable value can improve the utilance and the reliable transmission performance of link.

Compared with prior art, the invention has the beneficial effects as follows:

The present invention by the improvement to its congestion control mechanism, fault recovering mechanism, has greatly improved the reliable transmission performance on the basis of FAST Transmission Control Protocol.The present invention has mainly solved two key issues, that is: (1) distinguishes the packet loss reason, avoids congestion window to reduce by half blindly, has strengthened the transmission performance of reliable transport protocol; (2) propose a kind of average RTT account form based on the sectionally smooth factor, it is sensitiveer to make that congestion window upgrades ground according to network condition, thereby has improved link bandwidth utilization rate and transmission performance.

Description of drawings

Fig. 1 is the test topological diagram of reliable transmission acceleration method (the New FAST TCP) performance of checking high bit-error, long delay characteristics network;

Fig. 2 is that the error rate at bottleneck link is 5 * 10 ^-6, when propagation delay was 50ms, the throughput ratio of multiple method for reliable transmission;

Fig. 3 is that the error rate at bottleneck link is 10 ^-5, when propagation delay was 50ms, the throughput ratio of multiple method for reliable transmission;

Fig. 4 is that the error rate at bottleneck link is 5 * 10 ^-6, when propagation delay was 125ms, the throughput ratio of multiple method for reliable transmission;

Fig. 5 is that the error rate at bottleneck link is 10 ^-5, when propagation delay was 125ms, the throughput ratio of multiple method for reliable transmission.

Embodiment

The present invention can be applicable in the network of high bit-error, long delay.Below in conjunction with accompanying drawing enforcement of the present invention is described further:

With reference to shown in Figure 1, the present invention is verified the topology diagram of testing for us, l-G simulation test has been simulated 10 pairs of business and has been carried out the ftp file transmission, and every pair of business all is the file of client from ftp server download 60M, and their share same section bottleneck link that bandwidth is 3Mbps.The bandwidth and the propagation delay of some links have wherein also been marked among the figure.According to topological structure shown in Figure 1, we have verified under different link error rates and the different link propagation delays, the transmission performance of each agreement.

With reference to shown in Figure 2, for the error rate of bottleneck link is 5 * 10 ^-6, when propagation delay is 50ms, the contrast of the throughput of the method that TCP-Reno, TCP-NewReno, FAST TCP and this patent are proposed (New FAST TCP).Because the error rate is bigger, the packet loss that a lot of link errors is arranged and cause, traditional Transmission Control Protocol and FAST TCP method are not distinguished packet loss reason mechanism, so cause a large amount of unnecessary congestion windows to reduce by half, transmission performance seriously reduces, the method that this patent proposed then can effectively be distinguished the packet loss reason, thereby adopts the corresponding strategy that sends, and has improved the utilance and the transmission performance of link.As can be seen from the figure the throughput performance of this patent has a distinct increment than the performance of other several agreements.

With reference to shown in Figure 3, for the error rate of bottleneck link is 10 ^-5, when propagation delay is 50ms, the contrast of the throughput of four kinds of method for reliable transmission, a kind of situation in its result and front is similar, because this moment, the error rate was bigger, the throughput of New FAST TCP and other agreement differs bigger.

With reference to figure 4 with reference to shown in Figure 5, for the propagation delay at bottleneck link is 125ms, the error rate is respectively 5 * 10 ^-6With 10 ^-5The time, the throughput contrast of four kinds of method for reliable transmission.As can be seen from the figure the two kinds of situations in result of the test and front are similar, and under high bit-error, long delay environment, the transmission performance of New FAST TCP is higher than other several agreements far away.

(RTT refers to round-trip delay to table 1 in the table 1, Round Trip Time, RTT) be that four kinds of method for reliable transmission are under the various error rates and propagation delay, the time that a pair of service downloading 10M file is consumed, by the data in the table 1 as can be seen institute's time spent of New FAST TCP much smaller than other several method for reliable transmission, under the certain situation, its time-consuming is about 20% of other method.

Table 1

Claims (3)

1, a kind of reliable transmission acceleration method towards high bit-error, long delay characteristics network, it is characterized in that, on the basis of FAST Transmission Control Protocol, distinguish the reason of network generation packet loss according to following method: when transmitting terminal is received 3 repeat ACK bags, can judge transmission rate and current round trip time delay (the RoundTrip Time of previous RTT, RTT) difference of transmission rate, if the transmission rate of current RTT is during greater than the transmission rate of previous RTT, also do not have congested generation in the expression network, the packet loss reason is a link error; If the transmission rate of current RTT then represents to have congested generation in the network much smaller than the transmission rate of previous RTT, can think that packet loss is owing to network congestion causes;

Transmission rate (the TP of current RTT _{Current_rtt}) computing formula as follows:

${\mathrm{TP}}_{\mathrm{Current}\_\mathrm{rtt}}=\frac{\mathrm{snd}\_\mathrm{nxt}-\mathrm{seg}\_\mathrm{ack}}{\mathrm{avgRTT}\&times;\mathrm{snd}\_\mathrm{mss}}---\left(1\right)$

Wherein, snd_nxt is the sequence number of the section that is being sent out, and seg_ack is the sequence number of the section confirmed of the current ACK bag of receiving, both differences are the data volumes that sends in the current RTT, snd_mss is the maximum segment size of transport layer, and avgRTT is current average RTT;

diff＝TP _{Last_rtt}-TP _{Current_rtt} (2)

Wherein, diff is the transmission rate (TP of previous RTT _{Last_rtt}) with the transmission rate (TP of current RTT _{Current_rtt}) poor; The method of judging the packet loss reason according to threshold parameter α is as follows:

(a) if diff＞α thinks that then packet loss is caused by network congestion;

(b) if diff≤α thinks that then packet loss is caused by link error; Threshold parameter α value 0.99～1.

2, according to the described a kind of reliable transmission acceleration method of claim 1, it is characterized in that transmitting terminal is taked following transmission strategy after distinguishing the packet loss reason towards high bit-error, long delay characteristics network:

(a) if packet loss is that then transmitting terminal calls congestion avoidance algorithm owing to network congestion, the bag of retransmission of lost enters the fast quick-recovery stage then fast, reduces data package transmission velocity;

(b) if because the packet loss that causes of link error, then explicit re-transmission enters fully and recovers the stage, and being about to congestion window increases by 3, keeps original transmission rate.

According to the described a kind of reliable transmission acceleration method of claim 1, it is characterized in that 3, according to current average RTT and current variable δ of RTT value structure that measures, its value can obtain by following formula towards high bit-error, long delay characteristics network:

$\mathrm{\&delta;}=\frac{|T_i\left(k\right)-{\stackrel{\&OverBar;}{T}}_i\left(k\right)|}{{\stackrel{\&OverBar;}{T}}_i\left(k\right)}---\left(5\right)$

T wherein _i(k) be the RTT value that k is measured, T _i(k) be k RTT value that measures; Construct two groups of threshold values again, be respectively { δ ₁, δ ₂..., δ _i..., δ _nAnd { λ ₁, λ ₂..., λ _i..., λ _n, λ _N+1, 0≤δ wherein ₁＜δ ₂... δ _i＜δ _I+1...＜δ _n, 0≤λ ₁＜λ ₂... λ _i＜λ _I+1... λ _n＜λ _N+1＜1, the computing formula of improving the smoothing factor η (t) in the computational methods of calculating average RTT in the back is:

$\mathrm{\&eta;}\left(t\right)=\left\{\begin{array}{cc}{\mathrm{\&lambda;}}_1,& 0\&le;\mathrm{\&delta;}\&le;{\mathrm{\&delta;}}_1\\ {\mathrm{\&lambda;}}_i,& {\mathrm{\&delta;}}_{i-1}<\mathrm{\&delta;}\&le;{\mathrm{\&delta;}}_i,1\&le;i\&le;n\\ {\mathrm{\&lambda;}}_{n+1},& \mathrm{\&delta;}>{\mathrm{\&delta;}}_n\end{array}\right.---\left(6\right)$

Can utilize formula (5) and formula (6), upgrade, can reflect the average RTT value of the available bandwidth situation of network exactly according to formula (3);

T _i(k+1)＝(1-η(t _k))T _i(k)+η(t _k)T _i(k) (3)。

Images