JP4599554B2 - TCP congestion control method for broadband, high-delay wireless networks - Google Patents

Description

The present invention relates to a technique for speeding up TCP communication in a broadband and high-delay wireless network.

  In the conventional TCP congestion control method that is currently implemented as standard in many OSs, the congestion window size is increased based on two algorithms (slow start algorithm and congestion avoidance algorithm).

The slow start algorithm is executed from the start of communication, and is executed when the congestion window size is smaller than a value called a slow start threshold. In the slow start algorithm, the initial value of the congestion window size is set to one segment size, and the congestion window size is increased by one segment size every time one ACK for the transmitted segment is returned. On the other hand, the congestion avoidance algorithm is an algorithm that is executed when the congestion window size is greater than or equal to the slow start threshold. In the congestion avoidance algorithm, the congestion window size is increased by one segment size each time an ACK for a segment corresponding to the congestion window size is returned. Accordingly, the congestion window size is exponentially increased by the slow start algorithm until the congestion window size reaches the slow start threshold, and thereafter, the congestion window size is linearly increased by the congestion avoidance algorithm.

  However, the conventional congestion window size increasing algorithm has the following problems. In a network having a small propagation delay time, since the time (round trip time) required from when data is transmitted until ACK returns is small, the congestion window size increases rapidly. However, when the propagation delay time is increased, the round trip time is also increased, so that the increase rate of the congestion window size is very slow. In particular, when a line having a very long propagation delay time such as a satellite line exists in the network, the decrease in the increase speed becomes remarkable. Furthermore, if the available bandwidth is very large (several hundred Mbps to several Gbps), it will take a very long time to increase to the congestion window size where maximum performance is obtained.

  As a solution to this problem, an improved method has been proposed in which a network segment is measured using a special segment called a dummy segment, and the congestion window size is increased using the measurement bandwidth immediately after the start of transmission. (For example, refer nonpatent literature 1.) The dummy segment proposed by this improved system is a duplication of a transmission data segment, and has a low priority and is not retransmitted. Accordingly, when congestion occurs in the router, the dummy segment is preferentially discarded by the router. However, in the improved method, it is necessary to introduce control for handling dummy segments not only on the transmitting side but also on all routers existing between the receiving side and the receiving side. Accordingly, considering that the number of routers existing on the Internet is enormous, there is a problem that it is very difficult to realize the improved method.

I. F. Akyildiz, G. Morabito, S. Palazzo: “TCP-Peach: A New Congestion Control Scheme for Satellite IP Newtorks,” IEEE / ACM Transactions of Networking, 9, 3, pp.307-321, June 2001.

  Next, in the conventional TCP congestion control method, when data loss occurs, it is determined that congestion has occurred. In this case, however, the data loss is determined by receiving three or more duplicate ACKs. Then, the TCP sets the congestion window size to a value that is half of the value at the time of data loss in order not to cause further congestion by compressing the network bandwidth. However, data loss is caused not only by congestion but also by bit errors occurring in the transmission path. In particular, in a wireless network, bit errors occur more frequently than in a wired network due to rainfall or obstacles along the route. When data loss due to bit errors occurs, congestion does not occur, so there is no need to halve the congestion window size so as not to compress the network bandwidth. Therefore, when conventional TCP is applied to a wireless network, data loss due to bit errors occurs more frequently than in a wired network, and there is a problem that performance is greatly degraded.

  In order to prevent performance degradation when data loss occurs due to this bit error, an improved method has been proposed in which the network bandwidth is measured and the congestion window size after data loss is determined based on the measured bandwidth value. (For example, refer to Non-Patent Document 2.) However, the network bandwidth measurement method proposed in the improved method can be used only in a specific network, and the bandwidth tracking performance when used in a network in which various networks are mixed. Decreases and the accuracy of the measurement band decreases.

Tokuhiko Sato, Mitsunobu Kuniji, Fumio Teraoka: “TCP-J: Transport protocol suitable for wireless networks,” IPSJ Journal, 43, 12, pp.38483858, Dec. 2002.

  In the conventional TCP congestion control method, a timer called a retransmission timer is held. TCP measures the round trip time during communication, and sets the value of the retransmission timer based on this measurement time. The retransmission timer is set when data is transmitted, and performs an operation of retransmitting data when the retransmission timer expires (hereinafter referred to as retransmission timeout). When a retransmission timeout occurs, TCP determines that heavy congestion has occurred in the network, sets the congestion window size to one segment size, and resumes data transmission using the slow start algorithm.

In the conventional TCP congestion control method, it is assumed that the cause of retransmission timeout is all data loss due to congestion. However, a retransmission timeout occurs even when the ACK is lost in bursts or when the arrival of the ACK is temporarily delayed more than before. However, in this case, no data loss has occurred and no congestion has occurred. ACK burst loss is often caused by bit errors in wireless networks. Also, it is known that timeout due to ACK delay is likely to occur in a network such as a satellite network that has a high propagation delay time and is very stable. The conventional TCP congestion control method does not consider such ACK loss or delay. Further, the conventional TCP congestion control method does not have a function of searching for the cause of retransmission timeout on the initiative of the transmission side. Moreover, there is no function for notifying the transmission side that an ACK loss or delay has occurred from the reception side. Therefore, when a loss or delay of ACK that occurs frequently in a wireless network with high-speed and stable delay occurs, there is no way to detect the loss or delay of ACK even though the cause of retransmission timeout is not congestion. Therefore, the slow start algorithm is executed as in the conventional TCP congestion control method to lower the transfer rate. Furthermore, in this case, only when an ACK loss or delay occurs, it is erroneously determined that the data has been lost although the data has been received, and the data is erroneously retransmitted. As a result, significant performance degradation occurs and becomes a problem.

  As a solution to this, an improved method for detecting the ACK delay by examining the relationship between the round trip time of the retransmitted data and the round trip time until the retransmission timeout occurs after the retransmission timeout occurs and the data is retransmitted is proposed. Has been. (For example, refer to Non-Patent Document 3.) However, this improved method has a problem that it can not cope with a loss of ACK only by considering the delay of ACK. In addition, when the ACK delay is detected in the improved method, the congestion window size after the retransmission timeout is set to one segment size, so that even if the ACK delay can be detected, a decrease in throughput cannot be avoided.

Takeshi Hasegawa, Masayuki Murata, Hideo Miyahara: “Improvement of TCP congestion control / error recovery method considering erroneous retransmission,” IEICE Transactions B, J82-B, 11, pp.2074-2084. Nov. 1999.

The present invention is a problem when TCP is used in a broadband, high-delay wireless network such as a satellite Internet using a satellite line, a decrease in the increase rate of the congestion window due to a large propagation delay, and a congestion at the time of data loss due to a bit error. TCP congestion control that solves all the problems of unnecessarily reducing the window, the erroneous retransmission of data after the retransmission timeout caused by ACK loss and delay, and the unnecessary reduction of the transmission rate, and speeding up TCP communication The main challenge is to provide a method.

In order to solve the above problems, the present invention first measures the available bandwidth based on the arrival interval of ACK and the amount of received data notified by ACK during communication, and uses this measurement bandwidth to maximize the bandwidth. The maximum performance can be obtained from the start of transmission by deriving the window size (measurement window size) that can be used for the transmission, and adding the derived measurement window size to the congestion window size based on the conventional TCP slow start algorithm and congestion avoidance algorithm This paper proposes a TCP congestion control method that enables communication with a large congestion window size.

  Further, in the present invention, the setting of the congestion window size and the slow start threshold after occurrence of data loss determined by receiving three or more duplicate ACKs is set to the measurement window size obtained from the available bandwidth measured at the time of receiving the duplicate ACK. We propose a TCP congestion control method characterized by using and setting.

  Also, in the present invention, when measuring the network bandwidth during communication, the bandwidth smoothing bandwidth can be adapted to various networks by dynamically changing the coefficient for smoothing the bandwidth according to the network. A TCP congestion control method characterized by having a measurement method is proposed.

  Further, in the present invention, after a retransmission timeout has occurred, in order to determine whether the cause of the retransmission timeout is a data loss or an ACK loss or delay, Data having a sequence number is transmitted as a probe packet, and the maximum sequence number of data continuously received from the receiving side is notified by ACK. TCP congestion characterized in that after receiving an ACK for a probe packet, it uses the sequence number of the probe packet and the sequence number notified by the ACK for the probe packet to determine whether data loss, ACK loss or delay has occurred A control method is proposed.

  Further, according to the present invention, if it is known that the cause of retransmission timeout is data loss in the above TCP communication method, the congestion window size is set to one data segment size after data retransmission as in the conventional TCP operation. After setting the slow start threshold to the measurement window size, restart the transmission with the slow start algorithm, and if the loss or delay of the ACK is the cause of the retransmission timeout, the congestion window size and the slow start threshold are set to the retransmission timeout. We propose a TCP congestion control method that is characterized by setting the previous value and transmitting new data.

According to the present invention, even in a wireless network having a wide band and a large propagation delay time, the available bandwidth of the network can be used quickly and to the maximum, and high-speed TCP communication is possible. In addition, high-speed TCP communication is possible without reducing the data transmission rate unnecessarily at the time of data loss due to bit errors that frequently occur in a wireless network rather than congestion. In addition, ACK delays that occur in networks where the delay is stable due to wide bandwidth, and unnecessary data retransmission and transmission rate reduction caused by ACK loss due to bit errors that are likely to occur in wireless networks. High-speed TCP communication is possible not only when data is lost, but also when ACK loss or delay occurs.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is an overall configuration diagram of a TCP congestion control method according to the present invention. The TCP congestion control method according to the present invention has a configuration in which four mechanisms (11, 12, 13, 14) are added to the TCP congestion control mechanism 10 that is generally implemented at present. However, the currently widely used version of TCP is TCP Reno. In FIG. 1, reference numeral 10 denotes a congestion control mechanism of TCP Reno, which is a conventional TCP version. Reference numeral 11 denotes a network bandwidth measuring mechanism for measuring a network bandwidth. 12 is an LWC (Lift Window Control) mechanism that accelerates the increase speed of the congestion window size, and 13 is a CWS (Congestion Window Setting) that sets the congestion window size and slow start threshold after data loss that is revealed by receiving duplicate ACKs. Each mechanism is shown. The 12 LWC mechanisms and the 13 CWS mechanisms operate using the available bandwidth of the network obtained from the 11 mechanisms. Reference numeral 14 denotes an AEN (Acknowledgement Error) which detects whether the cause of timeout is due to data loss, ACK loss or delay after the occurrence of retransmission timeout, and sets the congestion window size and slow start threshold after retransmission timeout. Notification) mechanism.

  The TCP (Original) congestion control mechanism 10 of FIG. 1 in the method of the present invention indicates a TCP Reno congestion control mechanism that is a generally installed version of TCP. Further, the system of the present invention assumes a case where a selective acknowledgment (Selective ACKnowledgement, SACK) that can notify the loss of a plurality of data on the receiving side is used. Here, a normal ACK that does not use SACK can be notified of only one data loss. Therefore, in the present invention, the congestion control mechanism 10 of TCP (Original) includes four algorithms: a slow start algorithm, a congestion avoidance algorithm, an early retransmission algorithm, and an early recovery algorithm.

  The network bandwidth measuring mechanism 11 measures the available bandwidth of the network using ACK transmitted from the receiving side. First, let t (k) be the time at which the sending host received the ACK, t (k-1) the time at which the previous ACK was received, and d (k) the amount of transmission data. Using these pieces of information, the band sample value BWsample (k) is calculated using the formula of FIG. 2 every time an ACK is received. Next, the available band BWk is derived by using the calculated band sample value and the past band value and smoothing with the equation of FIG. In FIG. 3, the available bandwidth BWk at a certain time t (k) includes the past available bandwidth BWk-1 and the bandwidth sample value BWsample (k) acquired at time t (k) and the bandwidth sample value BWsample of the previous measurement. It is obtained from the average value of (k-1). However, in FIG. 3, a indicates a smoothing coefficient (0 <a <1), and it is necessary to set an optimum value for each network that performs communication. Further, when the smoothing coefficient a is made to react sensitively to changes in the network, the value of the smoothing coefficient is reduced (closer to 0). Conversely, when it is not desired to react very sensitively to a drastic change in the network, the smoothing coefficient is increased (closer to 1).

  Further, the smoothing coefficient a in FIG. 3 uses a fixed value suitable for each network when communication is performed using only a specific network such as a satellite network using only a satellite line. When passing through a variety of networks such as the Internet, if the smoothing coefficient a is dynamically set in accordance with the network conditions, the followability is improved and a usable bandwidth with higher accuracy can be derived. Also, the usable bandwidth becomes sensitive to changes in the smoothing coefficient when the smoothing coefficient is close to the limit value (maximum value 1, minimum value 0). Therefore, as a method of dynamically setting the smoothing coefficient a, it is conceivable to use a mathematical expression that draws a curve that gradually increases and decreases as it approaches the limit value, and linearly increases and decreases in other parts. However, when TCP communication is performed on the Internet via various networks, it is possible to obtain a sufficiently accurate band value even if a fixed value is used as the smoothing coefficient.

  The LWC mechanism 12 increases the congestion window size using the available bandwidth obtained by the network bandwidth measuring mechanism 11 described above. First, the LWC mechanism derives a measurement window size cwnd_abe that can use the bandwidth to the maximum by the equation of FIG. In FIG. 4, BW is an available bandwidth obtained by the equation of FIG. 3 (where BWk = BW), MSS is the maximum segment size, and RTTmin is the minimum value of the round trip time RTT. Here, the transmission side records the RTT obtained from the transmission time of certain data and the reception time of the ACK for the data from the start of transmission, and always holds the minimum value as RTTmin. Next, the congestion window size cwnd is obtained by adding the measurement window size cwnd_abe to the congestion window size cwnd_reno obtained by the slow start algorithm and the congestion avoidance algorithm of the TCP (Original) congestion control mechanism (10 in FIG. 1). Derived by the formula shown in.

  In the slow start algorithm and congestion avoidance algorithm of the TCP (Original) congestion control mechanism (10 in FIG. 1) used in the LWC mechanism 12, the congestion window size cwnd_reno is set by the equation shown in FIG. In FIG. 6, the slow start algorithm is executed at the start of data transmission and when a retransmission timeout occurs due to data loss, and increases the congestion window size cwnd_reno by one segment size every time one ACK is received. The congestion avoidance algorithm is executed when the congestion window size cwnd_reno becomes larger than the slow start threshold ssthresh, and increases the congestion window size cwnd_reno by one segment size every time an ACK corresponding to the congestion window size cwnd_reno is received.

  The CWS mechanism 13 uses the measurement window size cwnd_abe calculated from the available bandwidth derived by the network bandwidth measurement mechanism 11, and the congestion window size cwnd after occurrence of data loss that is found by receiving three or more duplicate ACKs. Set the slow start threshold ssthresh. FIG. 7 shows formulas for setting the congestion window size and the slow start threshold in the CWS mechanism. As shown in FIG. 7, when cwnd is smaller than ssthresh, ssthresh is set to the measurement window size cwnd_abe, and cwnd holds the previous value. If cwnd is greater than or equal to ssthresh, ssthresh is set to the measurement window size cwnd_abe, and cwnd is set to ssthresh (ie, the measurement window size). In addition, after data loss occurs and cwnd is set by the above method, the congestion window size cwnd_reno held by the congestion control mechanism of TCP (Original) (10 in FIG. 1) follows the normal operation of TCP Reno. , Half the value.

  When a retransmission timeout occurs, the AEN mechanism 14 investigates whether the cause of the timeout is data loss or ACK loss / delay. As a result, when the cause of the retransmission timeout is known, the AEN mechanism performs different operations for data loss and ACK loss or delay.

First, an outline of AEN operation will be described. FIG. 8A shows an example of AEN operation when a retransmission timeout occurs due to data loss, and FIG. 8B shows an AEN operation example when a retransmission timeout occurs due to ACK loss or delay. In FIG. 8, RTO indicates a retransmission timeout time, and RTT indicates a round trip time. First, when the data transmitted at time t0 is lost or the ACK for the transmitted data is lost / delayed and a retransmission timeout occurs at time t1, a probe packet is transmitted to investigate the cause of the retransmission timeout. Here, as the probe packet (hereinafter referred to as AEN-Prove), data having the maximum sequence number among the data transmitted by the transmission side is used. That is, the transmitted data having the latest sequence number is retransmitted. It should be noted that a plurality of probe packets can be sent out for reliable transmission. The receiving side that has received AEN-Prove returns an ACK (hereinafter referred to as AEN-ACK) indicating the maximum sequence number of the data that has been received so far to the transmitting side. The transmitting side that has received the AEN-ACK at time t2 uses the sequence number Pseg of AEN-Prove and the maximum sequence number Pack of data that has been received continuously until then, as shown in FIG. In this way, it is determined whether the cause of the retransmission timeout is a data loss or an ACK loss or delay. In FIG. 9, Pseg indicates the sequence number of AEN-Probe, and Pack indicates the sequence number notified by AEN-ACK. From FIG. 9, if Pseg is larger than Pack (Pseg> Pack), it is judged that data loss has occurred. If Pseg is less than Pack (Pseg <= Pack), ACK loss or delay occurs. Judge that you are doing.

  As a result of the determination using the sequence number, if the cause of retransmission timeout is data loss, the lost data indicated by the sequence number notified by AEN-ACK is retransmitted (time t3 in FIG. 8 (a)), and congestion occurs. The window size cwnd is set to one segment size, and transmission is restarted with the slow start algorithm. However, the slow start threshold value is set to the measurement window size derived from the bandwidth obtained from the network bandwidth measurement mechanism. On the other hand, when the cause of the retransmission timeout is the loss or delay of ACK, the congestion window size and the slow start threshold value remain unchanged, and new continuous data transmission is started (time t3 in FIG. 8B).

  In the above determination, in the congestion control mechanism of TCP (Original) (10 in FIG. 1), when the cause of retransmission timeout is data loss, the congestion window size cwnd_reno is set to one segment size. When the cause of retransmission timeout is loss or delay of ACK, the value of cwnd_reno is not changed.

  Next, the detailed operation of AEN will be described using the state transition diagram of FIG. In FIG. 10, conditions when transitioning from one state to another state and operations performed at that time are indicated by numbers from (1) to (14). In the items from (1) to (14) in FIG. 10, the upper stage shows the conditions for transitioning from one state to another state, and the lower stage shows the operations performed when transitioning from one state to another state. In FIG. 10, Pseg indicates the sequence number of AEN-Probe, Pack indicates the sequence number notified by AEN-ACK, and Nack indicates the sequence number of normal ACK that is not AEN-ACK.

Here, in order to indicate that the probe data (AEN-Probe) used in AEN and the ACK (AEN-ACK) in response to the probe are different from the normal data and ACK, a reserved field (6 in the TCP header) is used. Consider using one bit). That is, if 1 is set in 1 bit of the reserved field of the TCP header, this indicates that this packet is a packet (AEN_Probe or AEN_ACK) used in AEN. In normal TCP communication, the values of the reserved fields in the TCP header are all 0.

In FIG. 10, the communication state when retransmission timeout has not occurred is NORMAL_STATE 20, the state after transmitting AEN-Probe is AEN-ACK_WAIT 21, and the state after receiving a duplicate ACK in the state of AEN-ACK_WAIT 21 is DUP-ACK_RCVD22. Respectively. However, in FIG. 10, NORMAL_STATE 20 is written in two places for the sake of easy understanding of the drawing, but the same thing is shown. First, a case where the state transitions from the state of NORMAL_STATE 20 to the state of AEN-ACK_WAIT 21 is shown. When a retransmission timeout occurs in the state of NORMAL_STATE 20, an AEN-Probe is transmitted and a transition is made to the AEN-ACK_WAIT 21 state (1). Next, a case where no transition is made from the state of AEN-ACK_WAIT21 is shown. When a normal ACK with Pseg> Nack is received in the state of AEN-ACK_WAIT21, nothing is done (ACK is ignored), and the state does not change (2). Next, a case where the state returns from the state of AEN-ACK_WAIT21 to the state of NORMAL_STATE20 is shown. When an AEN-ACK with Pseg <= Pack is received in the state of AEN-ACK_WAIT21, it is determined that an ACK loss or delay has occurred, transmission of new data is started, and transition is made to NORMAL_STATE20 (3). When AEN-ACK with Pseg> Pack is received in the state of AEN-ACK_WAIT21, it is determined that the data has been lost, the slow data indicated by AEN-ACK is retransmitted, and at the same time, the slow start algorithm is started and transition to NORMAL_STATE20 (4). When the AEN-Probe causes a retransmission timeout in the state of AEN-ACK_WAIT21, the slow start algorithm is started simultaneously with retransmitting the unacknowledged data, and transits to NORMAL_STATE20 (5). When a normal ACK with Pseg <= Nack is received in the state of AEN-ACK_WAIT21, it is determined that the data that has timed out has already been received, and transmission of new data is started immediately, and transition to NORMAL_STATE20 (6). Next, a case will be described in which a transition is made from the state of AEN-ACK_WAIT 21 to the state of DUP-ACK_RCVD 22. If a duplicate ACK is received for data other than the data that has caused retransmission timeout in the state of AEN-ACK_WAIT21, nothing is done (ACK is ignored), and the process transits to DUP-ACK_RCVD22 (7). Next, a case where there is no transition from the state of DUP-ACK_RCVD 22 is shown. When a normal ACK with Pseg> Nack is received in the state of DUP-ACK_RCVD 22, nothing is done (the ACK is ignored), and the state does not change (8). In the state of DUP-ACK_RCVD22, when no more than three duplicate ACKs are received for data other than the data that has caused retransmission timeout, nothing is done (ACK is ignored), and the state does not change (9). In the state of DUP-ACK_RCVD22, when three duplicate ACKs are received for data other than the data that has caused retransmission timeout, the lost data indicated by the duplicate ACK is immediately retransmitted and the state does not change (10). Finally, a case where a transition from the state of DUP-ACK_RCVD 22 to the state of NORMAL_STATE 20 is shown. If an AEN-ACK with Pseg <= Pack is received in the state of DUP-ACK_RCVD 22, it is determined that an ACK loss or delay has occurred, transmission of new data is started, and a transition is made to NORMAL_STATE 20 (11). When AEN-ACK with Pseg> Pack is received in the state of DUP-ACK_RCVD22, it is determined that the data has been lost, the lost data indicated by AEN-ACK is retransmitted, and at the same time, the slow start algorithm is started and transition to NORMAL_STATE20 (12). If the AEN-Probe causes a retransmission timeout in the state of DUP-ACK_RCVD 22, the slow start algorithm is started simultaneously with the retransmission of unreachable data, and transition is made to NORMAL_STATE 20 (13). When a normal ACK is received with Pseg <= Nack in the state of DUP-ACK_RCVD22, it is determined that the data that has timed out has already been received, starts transmission of new data, and transitions to NORMAL_STATE20 (14).

  Next, setting of the congestion window size and the slow start threshold when performing the state transition from (1) to (14) in FIG. 10 will be described. First, in cases (4), (5), (12), and (13), the congestion window size is set to one data segment size, and the slow start threshold is set to the measurement window size obtained from the band value. In other cases, the congestion window size and the slow start threshold are not changed, and the values before the state transition are held.

  In addition, when the AEN congestion window size and slow start threshold are set, when an ACK loss or delay is found, the values before the occurrence of retransmission timeout are held. However, there is a possibility that the state of the network suddenly changes (bandwidth decreases / increases) while transmitting the probe packet. In such a case, the congestion window size after the occurrence of ACK loss or delay and the slow start threshold can be set by using the measurement window size obtained from the network bandwidth.

  In the AEN mechanism, AEN-Probe and AEN_ACK are distinguished from normal data and ACK by using the reserved field (6 bits) in the TCP header, but the method of the present invention can operate even when the reserved field is not used. It is. At this time, only (1), (3), (4), and (5) are used in FIG. In this case, the receiving side does not need to change the implementation for handling AEN, and only the transmitting side needs to be changed.

  The operation from the transmission start to the transmission end of each mechanism described above will be described with reference to FIG. First, in order to start communication, a connection is set between the transmission side and the reception side (S101). When the connection setting is completed (C101), data transmission is started (S102). When data transmission is started, every time an ACK is received, the network bandwidth measurement mechanism module 30 measures the available bandwidth of the network, and the measurement window size calculation processing of the LWC mechanism module 31 and the CWS mechanism module 32 (S103, S106) Notify the measurement band. The setting of the congestion window size used when performing the data transmission process (S102) is always performed by the LWC mechanism module 31 except when a data loss and a retransmission time-out determined by three or more duplicate ACKs occur. The LWC mechanism module 31 adds the measurement window size obtained from the measurement window size calculation process (S103) and the congestion window size based on the TCP Reno obtained from the TCP (Original) congestion window size setting process (S104) to transmit A window size is set (S105). The transmission side performs transmission processing using the congestion window size obtained in this processing (S105). If three duplicate ACKs are received and data loss is detected (C102), the CWS mechanism module 32 uses the measurement window size obtained from the measurement window size calculation process (S106), and the slow start threshold and the congestion window size. Is set (S107). When the process of the CWS mechanism is completed (C103), the data retransmission process is performed and the subsequent data transmission is resumed (S102). When a retransmission timeout occurs during data transmission (C104), the AEN mechanism module 33 first determines whether the cause of the retransmission timeout is data loss or ACK loss / delay (S108). After determining the cause of the retransmission timeout, the slow start threshold and the congestion window size are set (S109). When the processing of the AEN mechanism is completed (C105), if the data loss is the cause of the retransmission timeout, the data transmission is resumed after the lost data is retransmitted (S102). If the ACK loss / delay is the cause of the retransmission timeout, the new continuous data transmission is immediately resumed (S102). When all data transmission is completed (C106), the connection is disconnected (S110), and the communication is terminated.

  As described above, according to the method of the present invention, it is possible to increase the speed of TCP communication in a broadband and high-delay wireless network. Furthermore, the system of the present invention can also be used in a wide-band and high-delay wired network (gigabit network or the like) having similar network characteristics.

Configuration diagram of TCP congestion control method of the present invention Sample bandwidth formula Available bandwidth formula Calculation window size formula Calculation formula of congestion window size by LWC Congestion window size setting based on slow start algorithm and congestion control algorithm Setting of congestion window size by CWS Detection of retransmission timeout cause by AEN and subsequent operation example Determination contents of cause of retransmission timeout in AEN State transition diagram of AEN operation Operation of the TCP congestion control method of the present invention

Explanation of symbols

10: TCP (Original) congestion control mechanism 11: Network bandwidth measurement mechanism 12: LWC (Lift Window Control) mechanism 13: CWS (Congestion Window Setting) mechanism 14: AEN (Acknowledgement Error Notification) mechanism 20: AEN is operating Normal communication state 21: Waiting for AEN-ACK after sending AEN-Probe 22: State after receiving duplicate ACK after sending AEN-Probe and waiting for AEN-ACK 30: Network bandwidth Measuring mechanism module 31: LWC mechanism module 32: CWS mechanism module 33: AEN mechanism module

Claims (6)

  A method of measuring the bandwidth of a communication network during communication and increasing the increase rate of the congestion window size using a window size (hereinafter referred to as a measurement window size) that can be obtained from the measured value to the maximum possible use of the network bandwidth; , A method to set the TCP congestion window size and slow start threshold after data loss found by duplicate ACK using the measurement window size, and detect the loss and delay of the ACK segment that is the acknowledgment of the data segment after retransmission timeout TCP congestion characterized by speeding up TCP communication in a wideband and high-delay wireless network, which prevents the retransmission of data by preventing erroneous retransmission of data and reducing the congestion window size control method.   The broadband window according to claim 1, wherein the measurement window size obtained from the measurement band during communication is added to the value of the congestion window size based on the slow start algorithm and the congestion avoidance algorithm, which is an algorithm for increasing the congestion window size of TCP. A TCP congestion control method characterized by accelerating the increase in congestion window from the start of communication in a high-latency network.   2. The network bandwidth according to claim 1, wherein the bandwidth is measured using an ACK arrival interval and a data reception amount notified by the ACK, and dynamically according to the measured bandwidth, the past bandwidth, and the network status. A TCP congestion control method characterized by being able to cope with various network environments because the current bandwidth value is derived using the smoothing coefficient that changes.   In Claim 1, the congestion window size after retransmission timeout and the slow start threshold value are set. After retransmission timeout occurs, a probe packet is transmitted, and from the response ACK, the cause of timeout is data loss or ACK. If the cause of the retransmission timeout is data loss, it is determined that the network is in a heavy congestion state, and the retransmission and congestion window size of the corresponding data is set to the initial value and thrown. If transmission is restarted after setting the start threshold to the measurement window size and the cause of retransmission timeout is ACK loss or delay, the corresponding data is not transmitted and the congestion window size and slow start threshold are set to the values before retransmission timeout. By setting this, retransmission time may be lost due to ACK loss or delay. TCP Congestion Control, characterized in that the speed of communication when the bets has occurred.   In claim 4, as a probe packet used for investigating the cause of retransmission timeout, data having the largest sequence number among data transmitted by the data transmission side before the occurrence of the timeout is used, and the data transmission side leads the data reception side. A TCP congestion control method that notifies the maximum sequence number of data received so far. 5. When determining the loss of data and the loss or delay of ACK in claim 4, the sequence number of the probe packet (hereinafter referred to as Pseg) and the latest data received by the data receiver notified by ACK for the probe packet. If the data sequence number (hereinafter referred to as Pack) is used and Pseg is greater than Pack (Pseg> Pack), it is determined that data loss has occurred and Pseg is less than Pack. If (Pseg <= Pack), it is determined that an ACK loss or delay has occurred.

Images