CN1327677C - ECN based congestion control method with prediction verification - Google Patents

Abstract

The present invention relates to a congestion control method with prediction verification on the basis of ECN, which belongs to the field of network technology. The present invention is characterized in that prediction parameters of the current packet is calculated firstly through the difference of the adjacent packet, the corresponding prediction length of the queue is calculated according to the prediction parameters, and the common prediction threshold value of all the packets is corrected according to the congestion conditions of nodes; the prediction value of the average queue length of all the packets is calculated by using all the packets as a prediction group by using twice of a first packet delay as the signal transmission time, and a receiving end compares the difference between the prediction value of the average queue length of all the packets and the prediction threshold value with the prediction floating value, so that the receiving end can transmit an acknowledgement packet for increasing or decreasing or keeping the length of the congestion window to the resource end, which is convenient for the resource end to regulate the packet transmission speed. The receiving end driving method has the advantage that the receiving end driving method can reduce loss rate and packet average delay, has stable and high utilization rate of the network, and can detect the forthcoming congestion condition of the network.

Description

Jamming control method based on ECN and band prediction checking

Technical field

Jamming control method based on ECN and band prediction checking belongs to congested control technology field, the Internet.

Background technology

Algorithm among the present invention be a kind of use in the Internet congested control and avoid algorithm, belong to source algorithm based on " pipe " model.Congested control is a major issue in the present the Internet.According to the realization position of algorithm, congestion avoidance algorithm can be divided into two big classes: link algorithm and source algorithm.The link algorithm is carried out in the network equipment (as router and switch), and effect is the generation that detects network congestion, produces congestion feedback information; The source algorithm is carried out in main frame and network edge device, and effect is to adjust transmission rate according to feedback information.

Use generally has following several based on measuring classical congestion avoidance algorithm: Wand and Crowcroft proposes the increase that the DUAL algorithm is based on the RTT time (signal dealing transmission time) and responds in paper Eliminating Periodic Packet Losses in 4.3-Tahoe BSD TCP Congestion Control Algorithm; Congestion window is the same with the Reno algorithm to be increased, but per two RTT are after the time, and algorithm will check that the current RTT time is whether greater than the mean value of minimum and maximum RTT, if then congestion window is reduced 1/8.

R.Jain. CARD (the Congestion Avoidance using Round-tripDelay) method that proposes in paper A Delay-Based Approach for Congestion Avoidance in InterconnectedHeterogeneous Computer Networks is based on and derives optimum window size.This judgement is based on current RTT and window size.Per two RTT postpone to carry out acting, (current window size-parent window size) * (the former RTT of current RTT-), if the result for just, window reduces 1/8, if the result for bearing or 0, then increases the block length of a maximum.

Another jamming control method is smooth transmission rate.Each RTT window will increase the throughput that a block length relatively increases the window front and back simultaneously in the Tri Scheme algorithm that Wand and Crowcroft ' s proposes in paper A New CongestionControl Scheme:Slow Start and Search (Tri-S).If difference during less than a transmitted in packets throughput that obtains 1/2, then window is reduced a block length.How many bytes RTT of Tri-S calculate throughput use has sent in the time determines.

The TCP Vegas algorithm that proposes in paper TCP Vegas:End to End Congestion Avoidance on a Global Internet and Analysisof the Increase and Decrease Algorithms for Congestion Avoidance in Computer Networks is a kind of implementation method of better performances, this method is very similar with Tri-S, only is concerned about the variation of throughput rate or transmission rate.But this algorithm is different with Tri-S when calculate throughput, and it is not to use the variation of the slope of throughput, but relatively desired value of throughput and the difference between the measured value.

Vegas makes and measures and control link in this way and upload defeated " additionally " data.Excessive data increases the grouping that sends after being meant that the bandwidth of those links uses equals available bandwidth.Clearly, sent too many additional packet, will cause congested if connect.The Vegas jamming control method is based on control excessive data total amount, and is not only to abandon grouping.The judgement of Vegas algorithm is carried out at the source end, like this judgement of measuring based on RTT can not correctly reflect link congestion sending direction still be at answer party to middle appearance, and the present invention can reflect that link congestion occurs at sending direction.

Summary of the invention

The object of the present invention is to provide a kind of jamming control method that makes the source end can control the transmission rate of packet to be sent based on the result of prediction checking based on ECN and band prediction checking according to receiving terminal.

The invention is characterized in: it is a kind of source side method based on explicit congested announcement ECN and band prediction checking FAV, promptly in receiving terminal host PC and network edge device, carry out the prediction checking, source end host PC is then adjusted transmission rate according to feedback information, and described jamming control method contains following steps successively:

At the OPTION territory of IP of packet head enrollment time stamp, timestamp was made up of 4 bytes when (1) the source end sent packet;

(2) all support the node of explicit congested announcement ECN mark to transmit its congestion situation in the path, after promptly queue length surpasses keeper's preset threshold in node, the CE position set in the COS TOS territory;

(3) calculating each transmission packets at receiving terminal postpones: after receiving terminal receives packet, obtain the time receive accordingly by system call, be the time of this machine system, deduct the value of the timestamp in the data packet header, obtain the link transmission time of delay of this grouping with this value;

The predicted value of queue length when (4) inferring that by anticipation formula in following group this n grouping enters congested node is simultaneously according to receiving whether this n grouping is labeled CE and adjusts the anticipation threshold value , this steps in sequence is carried out in the following order:

(4.1) use the difference of the link transmission delay of the adjacent packets that arrives first to calculate Prediction Parameters K _n, it is a kind of asynchronous for delay measurements brings the parameter of influence owing to clock in order to eliminate source end and receiving terminal,

$K_n=\frac{{\mathrm{Td}\text{'}}_n-{\mathrm{Td}\text{'}}_{n-1}}{{\mathrm{Td}\text{'}}_{n-1}-{\mathrm{Td}\text{'}}_{n-2}}$

Wherein, Td ' _nThe link transmission that is the n grouping postpones,

Td ' _N-1The link transmission that is the n-1 grouping postpones,

Td ' _N-2The link transmission that is the n-2 grouping postpones,

(4.2) use following group of interior anticipation formula, the predicted value λ of queue length when the current grouping n of recursion acquisition enters congested node _n:

λ _n＝(K _n+1)×λ _n-1-K _n×λ _n-2

Wherein, λ _N-1Be the queue length predicted value of n-1 grouping,

λ _N-2It is the queue length predicted value of n-2 grouping

(4.3) use current group n queue length predicted value λ _nVerify the accuracy of anticipation with the ce mark of this grouping n, thereby revise the anticipation threshold value of setting

If $\stackrel{-}{X}<{\mathrm{\&lambda;}}_n$ And the grouping of receiving unmarked CE is

Then $\stackrel{-}{X}={\mathrm{\&lambda;}}_n+1;$

If $\stackrel{-}{X}>{\mathrm{\&lambda;}}_n$ And the grouping of receiving mark CE is P _CE,

Then $\stackrel{-}{X}={\mathrm{\&lambda;}}_n-1;$

(5) by pre-judging method between following group the estimation result in current each group is added up, and the anticipation variation of data traffic in the network in the future, sending the affirmation grouping with corresponding meaning according to statistics in each group to the source end simultaneously is special marking ACK:

(5.1) at receiving terminal all packets according to signal link propagation delay time RTT as the condition of dividing the anticipation group, the RTT time calculates according to the twice of first block transmission delay, promptly a RTT in the time receiving terminal send the affirmation grouping ACK of a Controlling Source end transmission rate to the source end, announcement source end

(5.2) find the solution interior average queue prediction length of RTT time:

${E\left[\mathrm{\&lambda;}\right]}_n=\frac{\underset{k=1}{\overset{S}{\mathrm{\&Sigma;}}}\mathrm{\&lambda;k}}{S}$

Wherein, k=1,2,3 ... S is the subscript of the packet in this anticipation group;

S is the quantity of packet in this anticipation group,

(5.3) calculate average queue prediction length E[λ in the RTT time] _nAnd threshold value

Poor Diff,

$\mathrm{Diff}={E\left[\mathrm{\&lambda;}\right]}_n-\stackrel{\&OverBar;}{X},$

(5.4) send the corresponding grouping ACK that confirms according to different difference range

If Diff＞α E[λ] _n, the parameter of α for setting,

Then send P _DEC

If Diff＜-α E[λ] _n),

Then send P _NOR

If-α E[λ] _n＜Diff＜α E[λ] _n,

Then send P _HOLD

Wherein, P _NORBe the affirmation grouping ACK of no special marking,

P _DECBe the affirmation grouping ACK of underlined position DEC,

P _HOLDBe the affirmation grouping ACK of underlined position HOLD;

DEC and HOLD mark can be arranged on replys in a certain position of confirming packet header TOS territory.According to the explanation of RFC2780, the 6th, 7 of TOS is used as the ECN agreement and uses, this method in ACK with the 6th position 1 as P _DECFlag bit, and with the 7th position 1 as P _HOLDFlag bit;

(5.5) after confirming that grouping ACK arrives source end PC, source end PC adjusts transmission rate with known transmission rate algorithm, promptly

(5.5.1) the source termination is received P _DECAfter, the congestion window of source end reduces 1/8,

(5.5.2) the source termination is received P _NORAfter, the congestion window of source end increases by 1,

(5.5.3) the source termination is received P _HOLDAfter, the congestion window of source end remains unchanged.

Experimental results show that: and existing method is relatively, the inventive method effectively reduces offered load, has also reduced to postpone end to end when reducing Loss Rate; And the increase of shared congestion link bandwidth can obtain higher and stable link utilization.

Description of drawings:

Fig. 1. institute of the present invention method realizes schematic diagram.

Fig. 2. the program flow chart of the method for the invention.

Fig. 3. the topological structure of using during experimental simulation of the present invention.

Fig. 4. the analog result when the congestion link bandwidth is 8Mbps.

Fig. 5. the analog result when the congestion link bandwidth is 16Mbps.

Fig. 6. the analog result when the congestion link bandwidth is 32Mbps.

Fig. 7. the analysis experimental result of parameter alpha.

Fig. 8. fairness is analyzed experimental result.

Embodiment

In Fig. 1, use the FAV algorithm to realize effective congested control.Wherein, source end PC is as server, and operation realServer8.01 provide streaming media service, purpose PC is as client computer, files in stream media on the access originator end PC, and use RealPlayer8.0 Plus to observe image transmission result, thus relatively congestion avoidance algorithm is to the influence of transmission performance.Employed equipment disposition is as follows in realization:

The PC configuration of source end and destination end:

Mainboard intel D845

Internal memory 256DDR

Video card Geforce2MX400

The prompt 80G in hard disk west

Network interface card realtek 8139

Router: can be all models of supporting ECN, that select for use in the experiment be Cisco 2600, and the router in the network can carry out interconnected according to different topologys.

The method of the invention specifically is made of successively following steps:

OPTION territory enrollment time at the IP of packet head when (1) the source end sends packet stabs.Timestamp is made up of four bytes.

The Version of version number (4bit)	The long IHL of header (4bit)	COS Type of Service (8bit)	Datagram length Total Length (16bit)
					Sign Identification (16bit)			Sign Flags (3bit)	Sheet displacement Fragment Offset (13bit)
Life span Time to Live (8bit)		Transport layer protocol Protocol (8bit)	Header check and Header Checksum (16bit)
					Source IP address Source Address (32bit)
Target ip address Source Address (32bit)
					OPTION territory Time-Stamp (32bit)

Table 1

(2) all routers are supported congestion situation in explicit congested announcement ECN (the Explicit Congestion Notification) flag node in the forward-path.Just queue length surpasses threshold value in node

After, will be the CE position 1 in the TOS territory.

The TOS territory:

Table 2

(3) calculating transmission packets at receiving terminal postpones.

After receiving terminal receives packet, obtain this machine system time by system call, the value that deducts timestamp in the data packet header with this value can obtain this transmission packets delay.As the condition of dividing the anticipation group, calculate according to the twice of first block transmission delay by the RTT time according to signal link propagation delay time RTT for receiving terminal.

(4) by the following group of interior queue length of anticipation algorithm (The Ingroup Forecast Algorithm) when inferring that this grouping enters the congested node formation, simultaneously according to receiving whether grouping is labeled CE and adjusts the anticipation threshold value Comprising following committed step:

(a) use the link delay difference of the adjacent packets of going ahead of the rest to calculate Prediction Parameters K _nThereby, eliminate the asynchronous influence that brings for delay measurements of clock. $K_n=\frac{{\mathrm{Td}\text{'}}_n-{\mathrm{Td}\text{'}}_{n-1}}{{\mathrm{Td}\text{'}}_{n-1}-{\mathrm{Td}\text{'}}_{n-2}}$

(b) use " anticipation formula in the group ", the predicted value of queue length, that is: λ when recursion acquisition current group enters congested node _n=(K _n+ 1) * λ _N-1-K _n* λ _N-2

(c) use the predicted value of current group queue length and the ce mark of grouping to verify the accuracy of anticipation, thereby revise the anticipation threshold value

$\mathrm{If}(\stackrel{-}{X}<l_n\mathrm{andP}\stackrel{\&OverBar;}{\mathrm{CE}})$

$\stackrel{-}{X}={\mathrm{\&lambda;}}_n+1;$

$\mathrm{If}(\stackrel{-}{X}>{\mathrm{\&lambda;}}_n\mathrm{and}{P}_{\mathrm{CE}})$

$\stackrel{\&OverBar;}{X}={\mathrm{\&lambda;}}_n-1;$

Wherein, use

The grouping that unmarked CE is received in expression, and P _CEThe grouping that mark CE is received in expression.

(5) by anticipation algorithm (The Intergroup Forecast Algorithm) between group estimation result in the current anticipation group is added up, and the anticipation variation of data traffic in the network in the future, send ACK to the source end according to statistics in the anticipation group simultaneously with special marking.Comprising following committed step:

(a) all packets are divided into the anticipation group according to RTT, the RTT time can use 2 * first packetization delay to calculate.That is to say that receiving terminal is to the ACK of a Controlling Source end transmission rate of source end transmission in the RTT time.

(b) find the solution interior average queue prediction length of RTT time $E[\mathrm{\&lambda;}{]}_n=\frac{\underset{k=1}{\overset{S}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_k}{S};$

(c) calculate average queue prediction length and threshold value

Difference $\mathrm{Diff}=E[\mathrm{\&lambda;}{]}_n-\stackrel{\&OverBar;}{X};$

(d) send corresponding ACK according to the difference result.

If (Diff＞α E[λ] _n) α is in order to adjust E[λ] _nScope,

sendACK(P _DEC)；

if(Diff＜-α·E[λ] _n)

sendACK(P _NOR)；

if(-α·E[λ] _n＜Diff＜α·E[λ] _n)

sendACK(P _HOLD)；

Wherein, ACK comprises following three kinds: (1) P _NOR, this is the ACK of no special marking; (2) P _DEC, mark DEC position among the ACK; (3) P _HOLD, mark HOLD position among the ACK.DEC and HOLD mark can be arranged in a certain position in acknowledgment packet head TOS territory.

Wherein, according to the explanation of RFC2780, the 6th, 7 of TOS is used as the ECN agreement and uses, during experiment in ACK with the 6th position 1 as P _DECFlag bit, and with the 7th position 1 as P _HOLDFlag bit, as shown below.

The TOS territory:

0 1 2 3 4 5 6 7

DEC

HOLD

Table 3

(6) after ACK arrived source end PC, source end PC carried out the transmission rate adjustment, just uses transmission rate adjustment algorithm (The Sending Rate Adjust Algorithm), and its step is as follows:

(a) the source termination is received P _DECAfter, the congestion window of source end reduces 1/8;

(b) the source termination is received P _NORAfter, the congestion window of source end increases by 1;

(c) the source termination is received P _HOLDAfter, the congestion window of source end remains unchanged.

The congestion avoidance algorithm that prediction is verified based on the ECN band is not very high to the requirement of hardware, can be applied in diverse network.This jamming control method has following four advantages:

The transmission rate that anticipation comes control data between the group of use anticipation group has effectively reduced packet loss;

Reduced the average retardation of grouping when reducing Loss Rate;

In forecasting process, use anticipation in the group of anticipation group and, make algorithm can adapt to the variation of flow in the network, thereby keep stable higher network utilization based on the authentication mechanism of ECN;

Algorithm uses the recipient to drive the method for (receiver-drive), whether can clearly judge send taken place on the link congested.

The FAV algorithm is by measuring the variation tendency that packetization delay comes formation in the predict congestion node at destination, thereby be about to the congestion situation of appearance in the supposition network, the CE that uses ECN to provide in forecasting process indicates the accuracy of verifying supposition, thereby has avoided congested generation preferably.The method that this prediction adds checking is a key point of the present invention.

The model that adopts during simulation as shown in Figure 3.

Wherein, S1 is the source end of data to Sn, and R1 and Rn are the destination of data, and sets up from Si to Ri that (the TCP Reno of 1≤i≤n) connects, and one of these link sharing are the congestion links from Route1 to Route2.Experiment is to simulating under FAV and classical source congestion avoidance algorithm TCP Reno, the environment that TCP Vegas at gateway is DropTail and comparing.

The parameter of using during simulation is as follows: α=1/8.Queue length 150 in Router1, and all sources start constantly 0, and stop behind 60s.

Experiment 1, congestion link bandwidth 8Mbps measures Loss Rate (Loss Rate), the utilance (Utilization) of congestion link and postpones (Delay), result such as Fig. 4.

As seen from Figure 4, in the congestion link bandwidth is 8Mbps, and fluxion was less than 60 o'clock, TCP Vegas obtains minimum Loss Rate, but when fluxion in the network continued to increase, the Loss Rate of TCP Vegas and TCP Reno algorithm rose very soon, and the FAV algorithm has still kept less Loss Rate, be 1/2 of TCP Vegas, but TCP Reno 1/4; The FAV algorithm is because level and smooth data traffic in the network so network utilization is low slightly than TCPVegas and TCP Reno, but generally is no more than 6%; After fluxion in the network surpasses 30, FAV will obtain minimum end-to-end delay.Statistics such as table 4.

Mean value	FAV	Vegas	Reno	FAV/Vegas	FAV/Reno
						Loss Rate (%)	3.5	6.3	14.3	55％	24％
Utilance (%)	86.5	92.2	90.8	94％	95％
						Postpone (ms)	68	85	114	80％	60％

Table 4 experiment 1 statistics

Experiment 2, congestion link bandwidth 16Mbps measures Loss Rate (Loss Rate), the utilance (Utilization) of congestion link and postpones (Delay), result such as Fig. 5.

As seen from Figure 5, be 16Mbps in the congestion link bandwidth, and fluxion was less than 60 o'clock, TCP Vegas still obtains minimum Loss Rate, and when fluxion in the network continues to increase, the FAV algorithm will still keep less Loss Rate simultaneously, be 1/2 of TCP Vegas, but TCP Reno 1/5; After the increase of congestion link bandwidth, the FAV algorithm is keeping more stable link utilization, near the performance of TCP Vegas and TCP Reno; Simultaneously, FAV still has minimum end-to-end delay.Statistics such as table 5.

Mean value	Index (Average)	FAV	Vegas	Reno	FAV/Vegas	FAV/Reno
							Loss Rate (%)	Loss Rate (％)	2.8	5.5	13.1	51％	21％
Utilance (%)	Utilization (％)	87.1	84.6	90.0	103％	97％
							Postpone (ms)	Delay (ms)	43	47	63	91％	68％

Table 5 experiment 2 statisticss

Experiment 3, congestion link bandwidth 32Mbps measures Loss Rate (Loss Rate), the utilance (Utilization) of congestion link and postpones (Delay), result such as Fig. 6.

As seen from Figure 6, in the congestion link bandwidth is 32Mbps, and fluxion was less than 80 o'clock, the Loss Rate of TCP Vegas is minimum, after but fluxion increased in the network, the Loss Rate of TCP Vegas and TCP Reno algorithm rose very soon, and the growth of the Loss Rate of FAV algorithm is very slow, average Loss Rate is 3/4 of TCPVegas, 1/4 of TCP Reno; It is stable that the network utilization of FAV algorithm still keeps, high slightly than TCP Vegas and TCP Reno; At this moment, the end-to-end delay of FAV algorithm is still minimum.Statistics such as table 6.

Mean value	Index (Average)	FAV	Vegas	Reno	FAV/Vegas	FAV/Reno
							Loss Rate (%)	Loss Rate (％)	2.3	3.1	10.3	74％	22％
Utilance (%)	Utilization (％)	87.0	85.2	86.4	102％	101％
							Postpone (ms)	Delay (ms)	31	32	42	97％	74％

Table 6 experiment 3 statisticss

By above experiment as can be known and TCP Vegas, TCP Reno algorithm relatively, the Loss Rate of FAV algorithm has tangible reduction, has reduced the retransmission packet in the network, thereby has effectively reduced offered load; Reduced end-to-end delay when reducing Loss Rate.Along with the increase of congestion link bandwidth, the FAV algorithm can obtain the link utilization of higher stable.

Experiment 4, the parameter alpha analysis is established α and is respectively 1/10,1/8,1/6 o'clock is that the FAV algorithm under the situation of 16Mbps experimentizes to congestion link, measure Loss Rate (Loss Rate), the utilance (Utilization) of congestion link and postpone (Delay), result such as Fig. 7.

Experiment 5, the fairness analysis.

Use among the paper Analysis of the Increase and Decrease Algorithms for Congestion Avoidance inComputer Networks fairness computing formula about Fairness Index:

$F\left(x\right)=\frac{{\left(\mathrm{\&Sigma;xi}\right)}^2}{n\left(\mathrm{\&Sigma;}{\mathrm{xi}}^2\right)}$

When the congestion link bandwidth is respectively 8Mbps, 16Mbps, during 32Mbps, experimental result such as Fig. 8.

Under three kinds of situations, the fairness parameter of FAV algorithm has all surpassed 0.92, and under most of situation all more than 0.95, especially fluxion surpasses after 80sly in the network, the fairness parameter arrives 0.99, the fairness of visible FAV algorithm is better.

Claims (1)

1, the jamming control method that prediction is verified based on the ECN band, it is characterized in that: it is a kind of source side method based on explicit congested announcement ECN and band prediction checking, promptly in receiving terminal host PC and network edge device, carry out the prediction checking, source end host PC is then adjusted transmission rate according to feedback information, and described jamming control method contains following steps successively:

At the OPTION territory of IP of packet head enrollment time stamp, timestamp was made up of 4 bytes when (1) the source end sent packet;

(2) all support the node of explicit congested announcement ECN mark to transmit its congestion situation in the path, after promptly queue length surpasses keeper's preset threshold in node, the CE position set in the COS TOS territory;

(3) calculating each transmission packets at receiving terminal postpones: after receiving terminal receives packet, obtain the time receive accordingly by system call, be the time of this machine system, deduct the value of the timestamp in the data packet header, obtain the link transmission time of delay of this grouping with this value;

The predicted value of queue length when (4) inferring that by anticipation formula in following group this n grouping enters congested node is simultaneously according to receiving whether this n grouping is labeled CE and adjusts the anticipation threshold value

, this steps in sequence is carried out in the following order:

(4.1) use the difference of the link transmission delay of the adjacent packets that arrives first to calculate Prediction Parameters K _n, it is a kind of asynchronous for delay measurements brings the parameter of influence owing to clock in order to eliminate source end and receiving terminal,

$K_n=\frac{T{d^{\text{'}}}_n-T{d^{\text{'}}}_{n-1}}{T{d^{\text{'}}}_{n-1}-T{d^{\text{'}}}_{n-2}}$

Wherein, Td ' _nThe link transmission that is the n grouping postpones,

Td ' _N-1The link transmission that is the n-1 grouping postpones,

Td ' _N-2The link transmission that is the n-2 grouping postpones,

(4.2) use following group of interior anticipation formula, the predicted value λ of queue length when the current grouping n of recursion acquisition enters congested node _n:

λ _n＝(K _n+1)×λ _n-1-K _n×λ _n-2

Wherein, λ _N-1Be the queue length predicted value of n-1 grouping,

λ _N-2It is the queue length predicted value of n-2 grouping

(4.3) use current group n queue length predicted value λ _nVerify the accuracy of anticipation with the ce mark of this grouping n, thereby revise the anticipation threshold value of setting

If $\stackrel{\&OverBar;}{X}<{\mathrm{\&lambda;}}_n$ And the grouping of receiving unmarked CE is

Then $\stackrel{\&OverBar;}{X}={\mathrm{\&lambda;}}_n+1;$

If $\stackrel{\&OverBar;}{X}>{\mathrm{\&lambda;}}_n$ And the grouping of receiving mark CE is P _CE,

Then $\stackrel{\&OverBar;}{X}={\mathrm{\&lambda;}}_n-1;$

(5) by pre-judging method between following group the estimation result in current each group is added up, and the anticipation variation of data traffic in the network in the future, sending the affirmation grouping with corresponding meaning according to statistics in each group to the source end simultaneously is special marking ACK:

(5.1) at receiving terminal all packets according to signal link propagation delay time RTT as the condition of dividing the anticipation group, the RTT time calculates according to the twice of first block transmission delay, promptly a RTT in the time receiving terminal send the affirmation grouping ACK of a Controlling Source end transmission rate to the source end, announcement source end

(5.2) find the solution interior average queue prediction length of RTT time:

$E{\left[\mathrm{\&lambda;}\right]}_n=\frac{\underset{k=1}{\overset{S}{\mathrm{\&Sigma;}}}{\mathrm{\&lambda;}}_k}{S}$

Wherein, k=1,2,3 ... S is the subscript of the packet in this anticipation group;

S is the quantity of packet in this anticipation group,

(5.3) calculate average queue prediction length E[λ in the RTT time] _nAnd threshold value Poor Diff,

${\mathrm{Diff}=E\left[\mathrm{\&lambda;}\right]}_n-\stackrel{\&OverBar;}{X},$

(5.4) send the corresponding grouping ACK that confirms according to different difference range

If Diff＞α E[λ] _n, the parameter of α for setting,

Then send P _DEC

If Diff＜-α E[λ] _n,

Then send P _NOR

If-α E[λ] _n＜Diff＜α E[λ] _n,

Then send P _HOLD

Wherein, P _NORBe the affirmation grouping ACK of no special marking,

P _DECBe the affirmation grouping ACK of underlined position DEC,

P _HOLDBe the affirmation grouping ACK of underlined position HOLD;

DEC and HOLD mark can be arranged on replys in a certain position of confirming packet header TOS territory, and according to the explanation of RFC2780, the 6th, 7 of TOS is used as the ECN agreement and uses, this method in ACK with the 6th position 1 as P _DECFlag bit, and with the 7th position 1 as P _HOLDFlag bit;

(5.5) after confirming that grouping ACK arrives source end PC, source end PC adjusts transmission rate with known transmission rate algorithm, promptly

(5.5.1) the source termination is received P _DECAfter, the congestion window of source end reduces 1/8,

(5.5.2) the source termination is received P _NORAfter, the congestion window of source end increases by 1,

(5.5.3) the source termination is received P _HOLDAfter, the congestion window of source end remains unchanged.

Images