KR100913897B1 - Transmission Control Protocol Congestion Control Method to Reduce the Number of Retransmission Timeouts - Google Patents

Abstract

The present invention provides a congestion control method for a communication system using a transmission control protocol, the method comprising: determining whether a data segment response (ACK) received from a receiver is in a fast recovery state; Retransmitting the corresponding data segment if the received data is duplicated by a preset number of duplicate responses, and storing the sequence number of the next data segment to be transmitted after the retransmission in a score board; When receiving a duplicate response for the retransmitted data segment, determining whether the retransmitted data segment is lost by using buffer status information of the receiver received through the duplicate response, and if the retransmitted data segment is lost, Resend Data Seg It characterized in that it comprises the step of retransmitting the tree again.

TCP SACK, TCP SACKPlus, Retransmission Data Segment, Scoreboard, nxt_snd.

Description

Transmission control protocol congestion control method to reduce the number of retransmission timeouts {Method for controlling congestion of TCP for reducing the number of retransmission timeout}             

1 is a block diagram showing the configuration of a packet data transmission system to which the present invention can be applied;

2 is a block diagram showing the configuration of the communication terminal shown in FIG. 1;

3 is a state diagram showing an operation for congestion control of a transmission control protocol;

4 is a flowchart illustrating a congestion control operation of a general transmission control protocol selection response;

5 is a flowchart illustrating a congestion control operation of a transmission control protocol selection response according to an embodiment of the present invention.

The present invention relates to a communication system using a transmission control protocol, and more particularly, to a method for preventing performance degradation due to retransmission timeout during congestion control of a transmission control protocol.

As a protocol for transmitting data over the Internet, Request For Comments (RFC) 793 issued by the Internet Engineering Task Force (IETF), an Internet standardization organization, defines the Transmission Control Protocol (hereinafter referred to as TCP). Doing. Transmission control protocols are used to transfer data in the form of packets between nodes over the Internet. Such a transmission control protocol is usually referred to as TCP / IP because it operates as a transport layer protocol above the Internet protocol of the network layer (hereinafter referred to as IP).

Transmission Control Protocol (TCP) uses the Internet Protocol (IP) as a network layer like the User Datagram Protocol (UDP), but provides connection-oriented services and reliable byte stream services that are basically different from UDP. do. TCP is a protocol that adjusts the transmission rate in consideration of network conditions, and is being considered as a transport layer in wired / wireless converged networks in the future.

Versions of TCP, such as Tahoe, Reno, Newreno, are used, and in TCP communication, when a receiver receives a data segment from a sender, the receiver transmits an acknowledgment (ACK) having the same sequence number as the data segment. The sender simultaneously transmits a predetermined number of data segments simultaneously without an ACK from the receiver, that is, a congestion window (hereinafter referred to as CWND), and receives ACKs having the same sequence number, that is, duplicate ACKs. Immediately resend the data segment with the sequence number. Otherwise, if no ACK is received for a certain timeout after transmitting the data segment, the data segment is retransmitted.

In TCP Newreno's loss recovery according to the prior art, when a retransmitted data segment is lost again, the sender continues to receive only duplicate ACKs for the segment lost during retransmission, not an ACK indicating that the retransmitted segment was successfully received. Inevitably, a retransmission timeout occurs. As the probability of segment loss increases, the number of segments lost in one window increases and the probability of loss increases in the process of retransmitting the lost segments. Thus, the loss recovery of TCP Newreno according to the prior art is performed in a high loss rate environment. It can significantly reduce transmission performance.

In addition, TCP Newreno retransmits the segment by duplicate ACKs, and then transmits new segments after every duplicate ACK is received.When a new segment is lost, it is timed out or by receiving duplicate ACKs for the newly transmitted segment. The newly transmitted segment is retransmitted. However, in this case, the size of the congestion window is reduced by half by retransmitting the previous segment, so that it may not receive enough duplicate ACKs to retransmit the newly lost segment, resulting in a timeout.                         

Since the retransmission timeout uses a timer having a relatively long unit resolution of, for example, 500 ms, the time for transmitting the data segment and waiting for the ACK is made in the time unit of the timer. The operation of such a timer can take no action for a long time on a lost segment, resulting in wasting transmission bandwidth. Therefore, there is a need for a method for reducing the retransmission timeout probability in the transmission control protocol.

Accordingly, an object of the present invention is to provide a congestion control method of a transmission control protocol for retransmitting a lost segment without retransmission timeout when the retransmitted segment is lost.

Another object of the present invention is to determine whether the retransmitted segment is lost by using the buffer status information (SACK block information) sent by the receiver through the ACK when the retransmitted segment is lost, and retransmits it again so that a retransmission timeout occurs A congestion control method of a transmission control protocol for lowering

The method for achieving the object of the present invention, in the congestion control method for a communication system using a transmission control protocol, the process of determining whether the data segment response (ACK) received from the receiver is in a fast recovery state, Retransmitting the corresponding data segment when the received data segment response is not a fast recovery state and is duplicated by a preset number of duplicate responses, and storing the sequence number of the next data segment to be transmitted after the retransmission in a score board; When the received data segment response is in a fast recovery state and receives a duplicate response for the retransmitted data segment, checking whether the retransmitted data segment is lost using buffer status information of the receiver received through the duplicate response; The resend data segment If the bit is lost, it characterized in that it comprises the step of retransmission, the retransmitted data segment.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description of the present invention, if it is determined that a detailed description of a related known function or configuration may unnecessarily obscure the subject matter of the present invention, the detailed description thereof will be omitted.

In the following description of the configuration and operation principle according to the present invention in detail will use a packet data transmission system including at least partly a wireless network.

1 is a block diagram showing the configuration of a packet data transmission system to which the present invention can be applied.

Referring to FIG. 1, a packet data transmission system includes a base transceiver station (BTS) 12 serving a mobile terminal 10 and a base station controller controlling the base station transceiver 12 through a wireless channel. Global Asynchronous Transfer Mode (GAN) that connects a Station Controller (BSC) 14 and the base station controller 14 to a packet network 30 through a Packet Data Service Node (PDSN) 18. Network) exchanger (16). Through this configuration, the mobile terminal 10 exchanges data with the computer 20 connected to the packet network 30.

The mobile terminal 10 may be a laptop phone connected to a cellular phone or a personal communications service (PCS) phone, a personal digital assistant (PDA) phone or an International Mobile Telecommunication (IMT) -2000 terminal capable of Internet communication. The computer 20 is a web server, a file transfer protocol (FTP) server, or the like. For example, FTP server 20 loads the file requested by the user in data segments and provides it to mobile terminal 10. Where the data segments contain sequence numbers for ordering. In this case, the FTP server becomes a sender and the mobile terminal 10 becomes a receiver.

2 is a block diagram showing the configuration of the communication terminal shown in FIG. This configuration is applied to the mobile terminal 10 and the FTP server 20 shown in FIG.

Referring to FIG. 2, the communication node includes a transmitter 40, a receiver 50, a network interface 70, and a bus 60 interconnecting them. The receiver 50 receives the data segments transmitted from the transmitting side via a communication line, and analyzes the packet. The transmitter 40 converts the data to be transmitted into data segments of a predetermined format and transmits the data to a receiving node through a communication line.

Although FIG. 1 discloses a plurality of devices connecting a sender as a mobile terminal and a receiver as a web server or an FTP server and a sender and a receiver, the present invention is not limited to this configuration. That is, the present invention can be applied to other data transmission systems having a similar technical background with a slight modification without departing from the scope of the present invention, which can be determined by those skilled in the art. will be.

In the above configuration, the sender transmits data segments generated by dividing files to be transmitted to the receiver. Communication in the packet network 30 based on the Internet Protocol (IP) is usually performed using a Transmission Control Protocol (TCP) (hereinafter referred to as "TCP / IP").

3 is a state diagram illustrating an operation for congestion control of a transmission control protocol. As shown in FIG. 3, congestion control in TCP is basically performed in a slow start mode 80 and a congestion avoidance mode 82. Consists of.

The slow start 80 is to prevent the segments from being discarded due to the limitation of the router buffer capacity when a large number of segments arrive to the router located on the network. Even if the transmission bandwidth of the network can afford to accommodate all the segments to be delivered, it may happen that routers cannot transmit segments due to the bursty nature of the traffic. The slow start 80 uses a congestion window (hereinafter referred to as CWND) on the sender side as an algorithm for avoiding such a phenomenon.                     

The slow start 80 is used for the first TCP connection and when a timeout occurs, and the value of CWND becomes larger than the Slow Start Threshold (SSTHRESH) or a predetermined number (usually three) of duplicate ACKs. If it detects a loss of a segment transmitted through reception, it terminates. The duplicate ACK indicates an ACK generated by the receiver using the same sequence number when the received segments are not in the correct order. Here, the ACK contains the sequence number of the immediately next segment that is normally received. For example, 2,3, (), 3,3,3,3 are sent for 1,2,3 (loss), 4,5,6,7 that is sent, so 4,5,6,7 You will receive a duplicate ACK.

The value of CWND is increased by the number of ACKs received during each round trip time (hereinafter referred to as RTT). Where RTT represents the round trip delay measured from sender to receiver. In practice, however, the sender compares the sender's CWND with the receiver's current remaining buffer size (hereinafter referred to as AWND) when transmitting the data segment and uses the smaller of the two as the CWND. That is, the sender transmits the segment using Min (CWND, AWDN). Here, the value of AWND is reported from the receiver to the sender. The maximum value is 64Kbyte and is used for the receiver's flow control. As such, slow start exponentially increases CWND to quickly access the available bandwidth of the network.

Congestion avoidance 82 linearly increases the CWND value by 1 for every RTT after the slow start 80 terminates or recovers the loss of the segment due to congestion. In congestion avoidance mode, CWND is increased by 1 / CWND for every ACK received. While the increase in CWND at slow start 80 is exponential, the increase in CWND at congestion avoidance 82 is linear. That is, slow start 80 increases CWND by the number of ACKs received during one RTT, while congestion avoidance 82 increases CWND by 1 during one RTT.

Slow start 80 and congestion avoidance 82 are separate algorithms but implemented together in TCP. That is, when the value of CWND becomes larger than the slow start threshold value SSTHRESH in the slow start 80, the congestion avoidance 82 is used.

Meanwhile, segments lost in TCP are recovered by retransmission timeout or fast retransmit 84.

Each time a data segment is transmitted in slow start mode 80 or congestion avoidance mode 82, a preset retransmission timer (e.g. set to 500 ms) is started and no response is received from the receiver until the retransmission timer expires. Otherwise a timeout occurs. If a timeout occurs, CWND is set to 1 and slow start mode 80 is restarted.

Fast retransmission 84 is used when a certain number of duplicate ACKs are received. In fast retransmission 84, if a predetermined number of duplicate ACKs are received, the segment having the corresponding sequence number is regarded as lost, and retransmission of the segment expected to be lost prevents timeout. In other words, when receiving consecutive duplicate ACKs above the Retransmit Threshold, the sender immediately retransmits the segment regardless of whether the retransmission timer expires. Normally, the retransmission threshold is three.

Fast Recovery 86 starts after the segment expected to be lost in fast retransmission 84 has been retransmitted. In fast recovery mode 86, the CWND is increased for each received redundant ACK, the increased CWND is increased, and a new segment is transmitted based on the increased CWND. And when an ACK is received for the segment expected to be lost, the fast recovery mode 86 ends and the congestion avoidance mode 82 begins. This fast recovery mode 86 is particularly effective when the bandwidth delay (hereinafter referred to as BDP) is large. If fast recovery mode 86 is not used, congestion avoidance mode 82 begins after fast retransmission 84.

The congestion control method in which the mixed control operation of the transmission control protocol is applied to the transmission control protocol selection response (TCP Selective Acknowledgment, hereinafter called TCP SACK) version among the various versions of the transmission control protocol will be described with reference to the drawings. In this case, the TCP SACK is for the receiver to accurately inform the sender of which segments have been lost, so that the sender can retransmit the lost segment more quickly and can prevent unnecessary retransmissions.

4 is a flowchart illustrating a congestion control operation according to a selection response (TCP SACK) of the transmission control protocol. The flow shown here is done by the sender.

If the ACK arrives from the receiver in step 100, the sender checks whether the current state is a fast recovery state in step 110. As a result of the check, if the current state is not a fast recovery state, in step 120, if the received ACK is a new ACK, and if it is a new ACK, the sender initializes a score board in step 121, and then CWND is allowed in step 122. Send a new segment within the range and go back to step 100. On the other hand, if it is determined in step 120 that the new ACK is not updated in step 130 scoreboard. The score board is updated based on buffer status information (SACK block information) received through the received ACK from the receiver, and stores a sequence of data segments to be retransmitted.

In step 140, the sender checks whether the number of received duplicate ACKs is equal to a threshold value, for example, three. As a result, if the number of received duplicate ACK is 3, the sender retransmits the lost segment through fast retransmission in step 141, and newly sets nxt_snd, STHRESH, pipe, and CWND values in step 142. SSTHRESH is set to 1/2 * CWND, CWND is cut in half, and pipe value to be used instead of CWND is set to the value obtained by subtracting the number of duplicate ACKs received from the current CWND. The sender then transitions to fast recovery mode and proceeds to step 100 again. On the other hand, if the number of received duplicate ACK is not 3, the process proceeds to step 100 again.

On the other hand, if the current state is confirmed as a fast recovery state in step 110, in step 150 the sender decreases the pipe value by 1. In this case, if the partial ACK is received, the pipe value is reduced by 2 in step 150. Then, in step 160, the fast recovery state is terminated. As a result of the check, in step 171, the score board is initialized, and in step 172, after transmitting a new segment, the process proceeds to step 100 again. In this case, the recovery end point in the TCP SACK is when the ACK arrives to respond to all segments transmitted before the fast recovery starts.

On the other hand, if the fast recovery state is maintained in step 170, the sender updates the score board and increases the pipe value by 1 in step 180. In step 181, the sender retransmits the lost segment or transmits a new segment. In this case, in case of retransmission, the transmitter newly sets nxt_snd. The sender then proceeds back to step 100 to listen for the next ACK. In this case, the pipe value can be additionally transmitted within the limit smaller than CWND and increases with each transmission.

As such, the conventional congestion control method uses TCP's buffer status information (SACK block information), and it is possible to know exactly which segments are lost and need to be retransmitted even if multiple segments are lost in one window. Retransmit several segments in the RTT. This reduces the chance of timeouts by retransmitting lost segments faster than other versions of TCP.

However, the conventional congestion control method (TCP SACK) recovers several segment losses in one window. That is, when a retransmitted segment is lost again, a retransmission timeout always occurs in SACK regardless of the window size. When TCP SACK loses a retransmitted segment during the loss recovery process, the sender will not receive an acknowledgment that the retransmitted segment has been successfully received, but only a duplicate ACK for the segment lost during retransmission. This is impossible and inevitably causes a retransmission timeout. In addition, as the probability of segment loss increases, the number of segments lost in one window increases, and the probability of loss again in the process of retransmitting the segments increases, so that the performance of TCP SACK may be degraded in a high loss rate environment. . In order to improve this, the present invention performs a congestion control operation by applying TCP SACKPlus (hereinafter, referred to as TCP SACKPulse), which is an improvement of the conventional congestion control method.

The present invention utilizes buffer status information (SACK block information) sent by the receiver through ACK to check whether the retransmitted segment is lost and retransmits it to lower the retransmission timeout probability. That is, when the transmitted segment is lost, the lost segment is retransmitted without retransmission timeout. What happens when a segment that is retransmitted in a SACK is lost is a duplicate ACK to send a retransmitted segment and eventually a timeout. To do this, the sequence of the next segment to be sent after the retransmission is sent to the scoreboard. Save it. The SACK block information included in the received duplicated ACK is continuously checked. In this case, when the retransmitted segment is not received at the receiving end and the segment after retransmission stored in the score board arrives at the receiving end normally, it can be seen that the retransmitted segment is lost. This congestion control operation will be described with reference to the drawings.

5 is a flowchart illustrating a congestion control operation of an improved transmission control protocol selection response according to an embodiment of the present invention.

In step 200, when the ACK is received from the receiver, in step 210, the sender checks whether the current state is a fast recovery state. As a result of the check, if the current state is not a fast recovery state, in step 220, the received ACK is checked whether it is a new ACK. Then, in step 222, the sender transmits a new segment to the receiver within the range allowed by CWND and proceeds to step 200 again.

On the other hand, if it is determined in step 220 that the new ACK is not updated in step 230 (Score Board). In step 240, the sender checks whether the number of received duplicate ACKs is equal to a threshold value, for example, three. As a result, if the number of duplicate ACKs received is not equal to the threshold value, the process proceeds to step 200 again. If the number of duplicate ACKs is equal to the threshold value, the sender retransmits the lost segment through fast retransmission in step 241, and then transmits the sequence of segments to be transmitted in step 242. Nxt_snd in the Score Board) is newly set. In this case, if only one loss occurs, nxt_snd becomes a segment sequence variable added by one from the maximum value (maxseq) of the sequence transmitted so far. Accordingly, the sender operates in fast recovery mode, sets the pipe value to be used instead of CWND to the value obtained by subtracting the number of duplicate ACKs received from the current CWND, and reduces the CWND by half. Then set SSTHRESH to 1/2 * CWND.

On the other hand, if the current state is a fast recovery state in step 210, in step 250 the sender decreases the pipe value by 1. At this time, if the partial ACK is received in step 250, the pipe value is reduced by 2. Then, in step 260, the fast recovery state is terminated. As a result of the check, when it is finished, the score board is initialized in step 261, the new segment is transmitted in step 262, and then the process proceeds to step 200 again. Thereafter, in step 280, the sender determines whether the retransmission segment is lost. That is, the SACK block information included in the received duplicate ACK is continuously checked. As a result of this check, if the retransmission data segment is lost, the sender increases the pipe value by 1 and retransmits the lost retransmission data segment in steps 290 to 291. In this case, if retransmission is performed, newly sets nxt_snd and then returns to step 200. Proceed. In this case, the pipe value can be additionally transmitted within the limit smaller than CWND and increases with each transmission.

On the other hand, if the retransmission segment is not lost in step 280, the sender increases the pipe value by 1 in step 281. In step 282, the sender retransmits the lost data segment or resends a new segment before retransmission. In this case, when retransmitting, the sender newly sets nxt_snd and then proceeds to step 200 again. In this case, the pipe value can be additionally transmitted within the limit smaller than CWND and increases with each transmission. If the retransmitted segment appears as not received at the receiving end and the next data segment to be transmitted is correctly received at the receiving end after retransmitting stored in the scoreboard, the sender determines that the resending data segment is lost. . At this time, the data segment expected to be lost is retransmitted to prevent a retransmission timeout.

Meanwhile, in the detailed description of the present invention, specific embodiments have been described, but various modifications are possible without departing from the scope of the present invention. Therefore, the scope of the present invention should not be limited to the described embodiments, but should be defined not only by the scope of the following claims, but also by the equivalents of the claims.

As described above, the present invention can reduce the probability of occurrence of the retransmission timeout by retransmitting the lost retransmission data segment once again by checking whether the retransmitted segment is lost based on the buffer state information of the receiver, and thus, the loss-prone environment is high. This has the effect of improving performance.

Claims (3)

In a congestion control method for a communication system using a transmission control protocol, Determining whether the data segment response (ACK) received from the receiver is in a fast recovery state; Retransmitting the corresponding data segment when the received data segment response is not a fast recovery state and is duplicated by a preset number of duplicate responses, and storing the sequence number of the next data segment to be transmitted after the retransmission in a score board; When the received data segment response is in a fast recovery state and receives a duplicate response for the retransmitted data segment, determining whether the retransmitted data segment is lost using buffer status information of the receiver received through the duplicate response; , And retransmitting the retransmitted data segment when the retransmitted data segment is lost. The method of claim 1, wherein the determining whether the retransmission data segment is lost comprises: Storing the sequence value of the next data segment of the retransmitted data segment as a variable to the receiver; Updating the scoreboard based on the buffer status information via the duplicate response to the retransmitted data segment; Determining whether the retransmission data segment is lost and updating the stored sequence value each time the retransmission data segment is retransmitted. The method of claim 2, And the buffer state information is information indicating a state in which the retransmission data segment is not received at the receiving end and a state in which the next data segment stored in the score board is not normally received at the receiving end.

Images