JP6145190B1 - Relay device, relay method, and relay program - Google Patents

Abstract

Congestion control using a slow start threshold that is calculated based on a band estimation value in relaying communication between a plurality of networks, controlling communication for each network, and suppressing an increase in processing amount, Provided are a relay device, a relay method, and a relay program that are efficiently executed. A relay unit that relays communication between a plurality of networks and controls communication for each network, and uses a minimum round-trip delay time in a relay device that performs communication between networks via the control unit. At least one control unit 1 that estimates a band estimation value and performs congestion control using a slow start threshold value calculated based on the band estimation value is provided. [Selection] Figure 5

Description

  The present invention relates to a technique for relaying communication between a plurality of networks and controlling the communication for each network, and more particularly to a technique for performing congestion control using a slow start threshold value calculated based on a band estimation value.

  Currently, TCP (Transmission Control Protocol) is widely used in networks such as the Internet. In TCP communication, congestion control is performed to avoid a situation in which user access concentrates on a specific destination in the network and communication that should normally be performed cannot be performed. Regarding congestion control, various congestion control algorithms (TCP Tahoe, TCP New Reno, TCP Vegas, etc.) have been proposed in addition to TCP Reno defined in RFC (Request for Comments) 2581, and have been proposed by the OS (Operating System). Furthermore, a congestion control algorithm that differs depending on the version of the OS is installed.

  In the congestion control, control is performed using a parameter called a congestion window, and the transmitting terminal transmits the confirmation response from the receiving terminal up to the size of the congestion window (hereinafter, sometimes simply referred to as “congestion window”). The segment can be transmitted without waiting for the arrival of the segment (hereinafter referred to as “acknowledge segment”) (hereinafter, the segment transmitted by the transmitting terminal is referred to as “data segment”). The congestion window is increased while it is determined that the network is not in a congestion state, but is decreased at once when it is determined that the network is in a congestion state. The amount of decrease depends on the congestion control algorithm, but in any case, the congestion window control in such congestion control at the time of segment loss depends on the network throughput (effective transfer amount per unit time of network or communication line). Influence. In particular, the influence is large in a long-distance high-speed communication network that requires time for data transfer and a wireless network in which segment loss easily occurs.

  Compared to a wired network, a wireless network is more likely to cause segment loss due to various causes such as radio wave interference and bit errors regardless of communication congestion, and is more likely to be determined as a congestion state. Therefore, if congestion control is performed each time it is judged as a congestion state, the entire network on the communication path will be affected, such as a decrease in throughput, network bandwidth pressure due to segment retransmission, and an increase in retransmission processing at the transmitting terminal. As a result, congestion control is being improved.

  TCP Westwood and TCP Westwood + are algorithms proposed for such wireless networks as targets. Although a wireless network is assumed, some of the implemented technologies can be applied to long-distance high-speed communication networks. Here, the congestion control by TCP Westwood + will be described.

When sending a segment using TCP Westwood +, the control at each stage of congestion control, such as slow start, congestion avoidance, and congestion detection, is the same control as TCP Reno etc., but the congestion window after congestion detection is extremely reduced. In order to avoid this, the slow start threshold after congestion detection is obtained using the band estimation value calculated when the confirmation response segment is received, and the congestion window is also changed accordingly. Specifically, the estimated bandwidth value SBWE _k for the _kth acknowledgment segment is calculated from the following equation 1, and the available bandwidth BWE _k for the _kth acknowledgment segment is calculated from the following equation 2.

Here, A _S is the number of acknowledgments included in the acknowledgment segment, if the acknowledgment segment and data segment is transmitted and received in a one-to-1, A _S but is 1, for a plurality of data segments If one acknowledgment segment is transmitted and received, the number of transmitted and received data segment is a _S. SS is the segment size. Therefore, A _S × SS is the total data size of the data segment that is the basis of the acknowledgment segment. RTT _k is the round trip delay time for the kth acknowledgment segment. The round trip delay time is the time from when the transmitting terminal transmits the data segment until the data segment reaches the receiving terminal and the acknowledgment segment from the receiving terminal reaches the transmitting terminal. α is a constant, and 0.9 is a recommended value. When the same acknowledgment segment, which is the case of early retransmission, is received three times (hereinafter referred to as “triple response”) among the conditions for detecting congestion, the slow start threshold ssthresh and the congestion as shown in Equation 3 below. When the window CWND is changed and the retransmission timer times out (hereinafter referred to as “retransmission timeout”), the window CWND is changed as shown in the following equation 4.

Here, RTTmin is the minimum value among all round-trip delay times until RTT _k is measured.

  When trying to improve congestion control in a wireless network by using the TCP West + that performs the control as described above, the improvement is possible in an environment where the entire communication path in which the transmitting terminal and the receiving terminal communicate only with the wireless network can be observed. I can plan. However, there are currently many environments in which wireless networks and wired networks coexist (hereinafter referred to as “mixed network environments”). For example, in data communication using a mobile phone or a smartphone, data communication is performed in an environment in which a wireless network established by a mobile phone company and the Internet are mixed, and a public wireless LAN (Local In the area network, data communication is performed via a wireless network and the Internet. In such a mixed network environment, there is a possibility that appropriate control cannot be performed because conventional TCP, such as TCP Westwood +, is executed by the transmission / reception terminal and the communication path cannot be specified. As one of measures against this problem, Split TCP is proposed in RFC2757. According to Split TCP, improvement of congestion control can be expected by dividing a network section according to network characteristics such as wireless or wired and applying an appropriate TCP according to the network characteristics of each section. As a method for realizing Split TCP, there is a method of installing a relay device at a point where a network section is divided.

  When communication between the relay device and the terminal is performed in a wireless network on the assumption that the relay device is used, if TCP Westwood + is adopted as the TCP used in the relay device, the method of determining the slow start threshold based on the estimated bandwidth value Inconvenience may occur. For example, when a wireless network that is short-distance communication and a wired network of the Internet that is long-distance communication are connected by a relay device, such as a public wireless LAN, the round-trip delay time in communication in the wireless network is the communication in the wired network. It is very small compared with the round trip delay time. However, for example, when communication is performed using a CSMA / CA (Carrier Sense Multiple Access with Collision Avoidance) method adopted in a wireless LAN, as shown in FIG. When transmitting data segment), if terminal 1A is transmitting before terminal 1B transmits, a signal is detected from SIFS (Short Interframe Space), which is the shortest waiting time in the frame transmission interval, and a busy channel. Through DIFS (DCF (Distributed Coordination Function) Interframe Space), which is the time until it is determined that the idle state has shifted to the idle state, and backoff, which is a random time to wait for frame transmission to avoid frame collision The terminal 1B transmits data. Therefore, when the terminals 1A and 1B transmit and receive data, delay jitter may occur in the wireless network as compared with the wired network due to the above time. For example, as shown in FIG. 2, when the server 3 connected to the wireless access point 2 via a wired network transmits data to the terminal 1A via the wireless access point 2, in FIG. A delay jitter may occur between data 1 and data 2 between data 2 and data 3 in FIG. In addition, if the standby data on the wireless access point increases rapidly due to a large number of terminals transmitting a large amount of data at the same time, if the standby data is discarded due to a buffer overflow, an early In order to take retransmission processing at the time-out of retransmission instead of retransmission, retransmission is waited until timeout, and the more waiting data, the more retransmission processing of discarded data occurs. Therefore, the round-trip delay time is likely to change due to other factors other than the amount of communication data (communication volume), and the estimated bandwidth value calculated using the round-trip delay time when detecting congestion, and the slow start threshold also changes due to other factors. It will be easy to do. Also, usually in the slow start phase, the congestion window is rapidly increased to the slow start threshold, and the transmission rate also increases rapidly, which increases the probability of segment loss in a wireless network. Therefore, the probability of occurrence of retransmission processing due to segment loss also increases.

  As described above, the congestion control using the slow start threshold obtained based on the estimated bandwidth value is an effective means as long as the entire communication path can be observed. When applied to a network such as a network in which the round-trip delay time changes regardless of the amount of communication data, improvement is necessary.

  As a method for improving communication by a relay device, for example, there is a method disclosed in Japanese Patent No. 5059976 (Patent Document 1). In the apparatus of Patent Document 1, the communication band is measured for each session, the measured band is compared with a preset target band, and the packet is transmitted in the target band when the measured band exceeds the target band. If the measured bandwidth is lower than the target bandwidth, communication is stabilized by promoting the bandwidth by proxy retransmission of the buffered packet. Bandwidth suppression and promotion are performed without changing the packet header, and are independent of existing TCP / IP processing.

Japanese Patent No. 5059976

  However, since the processing in the device of Patent Document 1 is processing that does not depend on existing TCP / IP processing, processing different from normal TCP processing is required, and a large amount of packets (segments) are to be transmitted. In such a case, the processing may cause a delay.

  The present invention has been made under the circumstances described above, and an object of the present invention is to relay communication between a plurality of networks, control the communication for each network, and is calculated based on a band estimation value in the control. It is an object of the present invention to provide a relay device, a relay method, and a relay program that efficiently execute congestion control using a slow start threshold that suppresses an increase in processing amount.

The present invention relates to a relay device that relays communication between a plurality of networks and includes a control unit that controls communication for each network, and performs communication between the networks via the control unit. The control unit estimates a bandwidth estimation value using a predetermined size according to a segment size and a minimum value of the round trip delay time, and uses a slow start threshold calculated based on the bandwidth estimation value to perform congestion control. This is achieved by providing at least one control unit 1 that performs the following.

The object of the present invention is that the control unit 1 controls communication in a network in which the round-trip delay time varies greatly due to factors other than communication volume, or the control unit 1 performs communication in a wireless network. Each of the control units further includes a storage unit that stores the segment received from the transmission terminal by controlling, and the input / output of the segment between the control units is performed via the storage unit. When the control unit can independently control, or when each control unit confirms that the segment received from the transmission terminal is received by the reception terminal, the confirmed segment is stored in the storage unit. When deleting and retransmitting the segment received from the transmitting terminal to the receiving terminal, the segment stored in the storage unit is retransmitted. And a, or the control unit 1, by performing the estimation of the bandwidth estimate by dividing the segment size with a minimum value of the round trip time, or the control unit 1, the control unit 1 is communication If at least one of the number of receiving terminals and the number of communication sessions in the network that controls the network is greater than or equal to a predetermined value, the band estimation value is estimated using the minimum value of the round trip delay time. Is achieved.

The present invention also relates to a relay method having a control step of relaying communication between a plurality of networks and controlling communication for each network, and performing processing for communication between the networks in the control step. The object of the present invention is to estimate a band estimation value using a predetermined size corresponding to a segment size and a minimum round-trip delay time as the control step, and to calculate a slow start threshold value calculated based on the band estimation value This is achieved by having at least one control step 1 for performing congestion control using.

The above object of the present invention further includes a storage step for storing the segment received from the transmission terminal in the storage unit for each control step, and after the execution of the storage step, the segment received from the transmission terminal is received by the reception terminal. If it is confirmed in the control step that is, the control step of the segment the confirmed and deleted from the storage unit, after the execution of the storage step, the retransmission of the segments received from the transmitting terminal to the receiving terminal It carried out if, by retransmitting the segment stored in the storage unit, or in the control step 1, to perform the estimation of the bandwidth estimate by dividing the segment size with a minimum value of the round trip time in Or in the control step 1, a network that controls communication in the control step 1 If at least one of the number of receiving terminals or the number of communication sessions in the network is greater than or equal to a predetermined value, it is more effective by estimating the band estimation value using the minimum value of the round trip delay time. Achieved.

  Furthermore, the above object of the present invention is achieved by a relay program for executing the above relay method.

  According to the relay device, the relay method, and the relay program of the present invention, in congestion control while relaying communication between a plurality of networks, the bandwidth estimation value is estimated using the minimum value of the round-trip delay time. By using the slow start threshold value calculated based on the estimated value, the segment loss due to a sudden increase in traffic is suppressed and the time required for retransmission due to congestion detection is shortened within the framework of conventional congestion control. Efficient control can be performed.

It is the schematic of the time chart which shows the timing of the data transmission in wireless LAN. It is a figure which shows the image of the data transmission in the network environment where the wireless network and the wired network are connected. It is a figure which shows the structural example (1st Embodiment) of the whole network environment containing the relay apparatus which concerns on this invention. It is an image figure which shows the flow of the segment in 1st Embodiment. It is a block diagram which shows the structural example of the relay apparatus of 1st Embodiment. 3 is a block diagram illustrating a configuration example of a control unit 10. FIG. 3 is a block diagram illustrating a configuration example of a control unit 30. FIG. 3 is a flowchart showing a part of an operation example of a control unit 30. 3 is a flowchart showing a part of an operation example of a control unit 30. It is a figure which shows the example of the change of the congestion window by congestion control. (A) is a figure in the case of the conventional TCP Westwood +, (B) is a figure in the case of 1st Embodiment. It is a block diagram which shows the structural example of the relay apparatus of 2nd Embodiment. 3 is a block diagram illustrating a configuration example of a control unit 50. FIG. 4 is a flowchart showing a part of an operation example of a control unit 50.

In the present invention, in a relay device that relays communication between a plurality of networks, congestion control is performed on a specific network by a method that improves the calculation method of the slow start threshold sthresh in the TCP West +. Specifically, the bandwidth estimation value SBWE _{k for} the _kth acknowledgment segment is calculated using the following equation 5 instead of equation 1.

In the conventional TCP Westwood +, the observed round-trip delay time RTT _k is used. Instead, RTTmin which is the minimum value among all round-trip delay times until RTT _k is measured is used. The band estimation value SBWE _k is calculated by assuming that the RTTmin is a time required to transmit one segment. After the bandwidth estimation value SBWE _{k is} calculated, the slow start threshold value ssthresh is calculated by the same calculation as in the conventional TCP Westwood + (Equations 2 to 4), and congestion control is performed. By using Equation 5 to calculate the bandwidth estimates SBWE _k, variation of bandwidth estimate SBWE _k is suppressed, and can be estimated smaller bandwidth estimates SBWE _k. For example, the segment size SS is 1460 bytes, first, in the second and third number of acknowledgments included in the acknowledgment segment _{A S} respectively 5,11 and 2, the round trip time is 26, 30 and respectively 24 In the case of seconds, the bandwidth estimation value SBWE _k (k = 1, 2, 3) in the conventional TCP Westwood + is expressed by the following formula 6, and the bandwidth estimation value SBWE _k in the present invention is expressed by the following formula 7 (the fractional part is rounded down). In the present invention, it can be seen that the fluctuation of the band estimated value SBWE _k is suppressed, and the band estimated value SBWE _k becomes small.

When a network in which such congestion control is executed (hereinafter referred to as an “applicable network”), such as a wireless network, where the round-trip delay time is likely to change due to another factor other than the traffic, is adopted. Therefore, the fluctuation of the estimated bandwidth value SBWE _k is suppressed, and the estimated bandwidth value SBWE _k is reduced, so that the slow start threshold value sthresh reset at the time of congestion detection can be reduced, and after the reset, the congestion window CWND It is possible to shift to an early stage of congestion avoidance in which the rate of increase is suppressed from the slow start stage. Therefore, a rapid increase in transmission rate can be suppressed, stable transmission can be performed, and segment loss can be suppressed.

  For networks other than the application network (hereinafter referred to as “non-application network”), congestion control by standard TCP such as TCP CUBIC, which has been conventionally used, is performed.

In Equation 5, RTTmin is considered to be a time required for transmitting one segment, but it is necessary for transmitting a number of segments other than 1 (not necessarily an integer). The band estimation value SBWE _k may be calculated in consideration of time.

  In the relay apparatus of the present invention, in order to perform such congestion control, a control unit is provided for each network to be controlled, but a storage unit may be further provided for each network. The data segment transmitted from the transmission terminal to the relay device is stored in the storage unit, and the relay device transmits the data segment stored in the storage unit to the reception terminal. When retransmission processing is necessary, the data segment stored in the storage unit is used for retransmission to the receiving terminal. Since processing using this storage unit is executed regardless of whether it is an applied network or a non-applied network, each control unit can execute processing independently of segment transmission / reception in other control units. In addition, since retransmission processing is performed between the relay device and the receiving terminal, not between the transmission terminal and the receiving terminal, the time required for the retransmission processing can be shortened, and the amount of communication for retransmission between the transmission terminal and the relay device can be reduced. The increase can be suppressed. Instead of providing a storage unit for each network, only one storage unit may be provided, and a storage area in the storage unit may be divided and used for each network.

  Embodiments of the present invention will be described below with reference to the drawings.

  FIG. 3 shows a configuration example (first embodiment) of the entire network environment including the relay device according to the present invention. In this configuration example, the wireless network is the application network, and the wired network on the Internet is the non-application network. In the wireless network, segments are transmitted and received between the receiving terminal 4 such as a personal computer or a tablet and the wireless access point 2, and the relay device 5 is directly connected to the wireless access point 2. In a wired network on the Internet, a server 7 is connected to the relay device 5 via the Internet 6. Normally, a router or the like is interposed between the relay device 5 and the server 7, but this is omitted in this configuration example.

  In the first embodiment, as shown in FIG. 4, a data segment (indicated as “segment” in FIG. 4) is transmitted from the server 7 connected to the non-applicable network to the relay device 5. When the data segment is received, an acknowledgment segment (indicated as “response” in FIG. 4) is transmitted to the server 7 and the data segment is stored in the storage unit 40. Then, the data segment stored in the storage unit 40 is transmitted to the receiving terminal 4A connected to the application network. When receiving the data segment, the receiving terminal 4A transmits an acknowledgment segment to the relay device 5. The storage unit 40 is provided with an area for storing a data segment for each communication session (hereinafter referred to as “session-specific storage buffer”). When a data segment is received from the server 7, it is stored in the corresponding session-specific storage buffer. Remember.

  The retransmission processing in the segment transmission / reception from the server 7 to the receiving terminal 4A is also performed via the relay device 5. That is, when a part of the data segment transmitted from the server 7 to the relay device 5 does not reach the relay device 5 correctly due to segment loss or the like, the relay device 5 uses the same method as that of the conventional TCP to confirm the confirmation response segment. The server 7 is requested to retransmit the data segment using the presence / absence and the sequence number in the acknowledgment segment. When a part of the data segment transmitted from the relay device 5 to the receiving terminal 4A does not reach the receiving terminal 4A correctly, the relay device 5 determines whether there is an acknowledgment segment transmitted from the receiving terminal 4A or the confirmation response segment. This is determined from the sequence number of the data segment, and the data segment is retransmitted using the data segment stored in the storage unit 40.

  An example of the configuration of the relay device 5 of the first embodiment is shown in FIG. The relay device 5 includes a control unit 10, a segment management unit 20, a control unit 30, and a storage unit 40. FIG. 5 shows only components necessary for explaining the congestion control when a data segment is transmitted from the non-application network to the application network in FIG. This also applies to FIGS. 6 and 7 showing examples of configurations of the control unit 10 and the control unit 30 described later.

  The control unit 10 receives the data segment Dseg transmitted from the server 7, outputs the confirmation response segment AckS transmitted to the server 7, and outputs the data segment Dseg to the segment management unit 20. The control unit 10 performs congestion control using conventional TCP, for example, TCP CUBIC. Note that TCP is located in the transport layer in the protocol structure. Normally, data transmitted from the server 7 is input to the transport layer through a layer lower than the transport layer, but is input by the control unit 10. It is assumed that the data segment Dseg has passed through the lower layers. Similarly, it is assumed that the confirmation response segment AckS output from the control unit 10 is transmitted to the server 7 via the lower layer.

  A configuration example of the control unit 10 is shown in FIG. The control unit 10 includes a segment reception unit 110, an acknowledgment generation unit 120, and a segment transmission unit 130. The segment receiving unit 110 receives the data segment Dseg, performs data input processing such as error confirmation normally performed in TCP, and then outputs the data segment Dseg to the confirmation response generation unit 120 and the segment management unit 20 Output to. The confirmation response generation unit 120 generates the confirmation response segment AckS from the information of the data segment Dseg by the same procedure as that of the conventional TCP, and outputs it to the segment transmission unit 130. The segment transmission unit 130 performs the process for outputting the confirmation response segment AckS to the lower layer, and then outputs it.

  The segment management unit 20 receives the data segment Dseg output from the control unit 10 and stores it in the storage unit 40. As described above, since the storage unit 40 is provided with a session-specific storage buffer for each communication session, the communication session is specified based on the destination port number in the data segment Dseg, and the corresponding session-specific storage unit buffer. Stores the data segment Dseg.

  The data segment Dseg stored in the storage unit 40 is input to the control unit 30 via the segment management unit 20 in accordance with a segment request Rs from the control unit 30 described later. The segment management unit 20 reads the data segment Dseg corresponding to the requested communication session and sequence number from the storage unit 40 based on the segment request Rs and outputs the data segment Dseg to the control unit 30. However, the data segment Dseg that is initially stored in the session-specific storage buffer is output to the control unit 30 when it is stored in the storage unit 40.

  The segment management unit 20 also performs processing in accordance with an erase command Cd from the control unit 30 described later. Based on the erase command Cd, the communication session and sequence number to be erased are confirmed, and the corresponding data segment Dseg is erased from the storage unit 40.

  The control unit 30 receives the data segment Dseg output from the segment management unit 20 and transmits it to the receiving terminal 4A. Also, the confirmation response segment AckR transmitted from the receiving terminal 4A is input, the segment request Rs and the erasure command Cd are generated based on the information of the confirmation response segment AckR, and output to the segment management unit 20. The control unit 30 performs congestion control based on Equation 5. As with the control unit 10, the acknowledgment segment AckR input to the control unit 30 is assumed to pass through the lower layer, and the output data segment Dseg is assumed to be sent via the lower layer.

  A configuration example of the control unit 30 is shown in FIG. The control unit 30 includes a segment reception unit 310, a congestion control unit 320, and a segment transmission unit 330.

  The segment receiving unit 310 receives the confirmation response segment AckR, and performs the data input processing such as error confirmation normally performed in TCP on the confirmation response segment AckR, as in the case of the segment receiving unit 110, and then the congestion. Output to the controller 320.

  The congestion control unit 320 performs congestion control based on Equation 5. Congestion control includes stages of slow start, congestion avoidance, and congestion detection, and congestion detection includes congestion detection based on a triple overlap response and congestion detection based on a retransmission timeout. In the congestion detection by the triple overlap response, if the confirmation response segment AckR having the same sequence number is received from the receiving terminal 4A three times in succession, it is determined that the congestion is detected. In congestion detection due to retransmission timeout, if the acknowledgment segment AckR for the data segment Dseg is not received even after a predetermined time has elapsed after the data segment Dseg is transmitted to the receiving terminal 4A, it is determined that congestion is detected. The timing other than the congestion detection timing is the slow start or congestion avoidance stage, and the distinction between the slow start and the congestion avoidance is performed by comparing the congestion window CWND and the slow start threshold ssthresh. That is, when the congestion window CWND is smaller than the slow start threshold ssthresh, it is a slow start stage, and when it is not, it is a congestion avoidance stage.

  In the slow start, the congestion window CWND is updated according to the following formula 8, and in the congestion avoidance, the congestion window CWND is updated according to the following formula 9.

After the update, the communication session and sequence number of the data segment Dseg to be transmitted to the receiving terminal 4A are specified from the confirmation response segment AckR, and output to the segment management unit 20 as the segment request Rs. Note that one segment request Rs may be output in response to requests for a plurality of data segments Dseg.

In the case of congestion detection due to the triple overlap response, the slow start threshold ssthresh and the congestion window CWND are changed according to the above equation 3, and in the case of congestion detection due to retransmission timeout, the slow start threshold ssthresh and the congestion window CWND are determined according to the above equation 4. change. The bandwidth estimation value SBWE _k and the available bandwidth BWE _k used at this time are set to the latest values according to the equations 5 and 2 each time the confirmation response segment AckR is input. The round-trip delay time minimum value RTTmin used for calculating the band estimation value SBWE _k is calculated as the round-trip delay time from the time when the control unit 30 transmits the data segment Dseg to the reception of the acknowledgment segment AckR. Then, the calculated round-trip delay time is updated by comparing it with the minimum value at that time. The constant α used to calculate the available bandwidth BWE _k is 0.9, which is the recommended value for TCP Westwood +. After changing the slow start threshold ssthresh and the congestion window CWND, the communication session and sequence number of the data segment Dseg to be retransmitted to the receiving terminal 4A are output to the segment management unit 20 as the segment request Rs.

  When the data segment Dseg is normally transmitted to the receiving terminal 4A, the data segment Dseg stored in the storage unit 40 becomes unnecessary, and therefore the congestion control unit 320 outputs an erasure command Cd to the segment management unit 20. The data segment Dseg is erased from the storage unit 40. Whether the data segment Dseg has been transmitted normally is determined by using the sequence number in the acknowledgment segment AckR at the slow start and congestion avoidance stage, as in the case of the conventional TCP, the input acknowledgment segment AckR is the next one. It is determined whether it is an acknowledgment segment that requests transmission of a data segment. The communication session and sequence number of the data segment Dseg to be erased are output as the erase command Cd, and the segment management unit 20 identifies the data segment Dseg to be erased from the erase command Cd and erases it from the storage unit 40. Note that one erasure command Cd may be output for erasing a plurality of data segments Dseg.

  When the congestion control unit 320 inputs the data segment Dseg output from the segment management unit 20 in response to the segment request Rs, the data segment Dseg is output to the segment transmission unit 330. However, since the segment request Rs based on the acknowledgment segment AckR cannot be generated for the data segment Dseg that is initially stored in the session-specific storage buffer, the segment when the data segment Dseg is stored in the storage unit 40 cannot be generated. The data segment Dseg is input from the management unit 20 to the congestion control unit 320 and is output to the segment transmission unit 330.

  Similar to the segment transmission unit 130, the segment transmission unit 330 outputs the data segment Dseg after performing processing for output to the lower layer.

  Note that when transmitting a data segment to the receiving terminal 4A, flow control based on a window size set based on a buffer provided in the receiving terminal 4A is also executed. However, in this embodiment, the window size is always set as compared to the congestion window. Largely, it is assumed that data segment transmission is performed only depending on congestion control.

  With the above configuration, an example of the operation of the control unit 30 after inputting the confirmation response segment AckR will be described with reference to the flowcharts of FIGS.

  When the control unit 30 inputs the confirmation response segment AckR (step S10), the confirmation response segment AckR is input to the segment reception unit 310. The segment receiving unit 310 performs data input processing on the confirmation response segment AckR (step S20) and outputs the result to the congestion control unit 320.

The congestion control unit 320 calculates the round-trip delay time based on the information of the confirmation response segment AckR. If the calculated round-trip delay time is smaller than the minimum round-trip delay time value RTTmin, the calculated round-trip delay time is set as RTTmin. Otherwise, RTTmin is left as it is (step S30). Then, using the minimum value RTTmin of the round-trip delay time, the band estimation value SBWE _k and the usable band BWE _k are calculated from Formula 5 and Formula 2 (Step S40).

  Next, the congestion control unit 320 confirms the sequence number in the confirmation response segment AckR, confirms whether the response is a triple response with the same sequence number three times in succession (step S50), and the confirmation result is not a triple response. In this case, the congestion window CWND is compared with the slow start threshold value ssthresh (step S60). If the congestion window CWND is smaller than the slow start threshold ssthresh, it is determined as a slow start stage, and the congestion window CWND is updated by Equation 8 (step S70). If the congestion window CWND is greater than or equal to the slow start threshold ssthresh, congestion avoidance is performed. The stage is determined, and the congestion window CWND is updated by Equation 9 (step S80). Then, it is confirmed whether the acknowledgment segment AckR is an acknowledgment segment that requests transmission of the next data segment (step S90), and in the case of the next data segment transmission request, the communication session and sequence number of the data segment Dseg to be transmitted Is identified from the confirmation response segment AckR, and is output to the segment management unit 20 as the segment request Rs (step S100). Further, in order to erase the data segment Dseg confirmed that the acknowledgment segment AckR is normally transmitted by the next data segment transmission request from the storage unit 40, the communication session and sequence number of the data segment Dseg Is output to the segment management unit 20 as an erasure command Cd (step S110). If the confirmation response segment AckR is not a next data segment transmission request, a return is returned. Note that the order of step S100 and step S110 may be reversed.

If the confirmation result is a triple response in step S50, it is determined that congestion is detected, and the slow start threshold ssthresh and the congestion window are calculated from Equation 3 using the estimated bandwidth value SBWE _k and the available bandwidth BWE _k calculated in step S40. CWND is changed (step S120). Then, the data segment Dseg to be retransmitted is specified from the information of the acknowledgment segment AckR, and the communication session and sequence number of the data segment Dseg are output to the segment management unit 20 as the segment request Rs (step S130).

  In response to the segment request Rs, when the data segment Dseg is output from the segment management unit 20 and the congestion control unit 320 inputs the data segment Dseg (step S140), the data segment Dseg is output to the segment transmission unit 330.

  The segment transmission unit 330 performs processing for outputting the input data segment Dseg to the lower layer and outputs the processed data segment Dseg (step S150). At the time of output, a retransmission timer for the output data segment Dseg is started (step S160).

  The above operation is executed every time an acknowledgment segment AckR is input.

Retransmission processing in congestion detection due to retransmission timeout is executed by interruption from the retransmission timer. In other words, when the retransmission timer times out beyond the set time, an interrupt is generated, and the bandwidth estimation value SBWE _k and the available bandwidth BWE _k calculated in step S40 are used as congestion detection due to retransmission timeout, The slow start threshold ssthresh and the congestion window CWND are changed from Equation 4 (step S200). Thereafter, as in the case of the congestion detection by the triple overlap response, the data segment Dseg to be retransmitted is specified from the information of the confirmation response segment AckR, and the communication session and sequence number of the data segment Dseg are set as the segment request Rs to the segment management unit 20 (step S210). Then, by the same operation as steps S140 to S160, the data segment Dseg is output and the retransmission timer is started (steps S220 to S240).

  Note that the retransmission processing in congestion detection due to retransmission timeout may be executed not by processing by interruption but by processing by appropriately checking the retransmission timer. Further, steps S220 to S240 in the retransmission process may not be executed in the retransmission process, and may be used in the execution of steps S140 to S160.

  Although the above description is about the configuration and operation when a data segment is transmitted from the non-application network to the application network, transmission of the data segment from the application network to the non-application network is also possible in this embodiment. It is. In this case as well, the data segment is transmitted via the relay device 5. However, in the data segment transmission from the receiving terminal 4A to the relay device 5 and the data segment transmission from the relay device 5 to the server 7, congestion control by conventional TCP is used. The control unit 30 performs the same operation as that of the control unit 10. In this case, the control unit 30 may be provided with components equivalent to the segment reception unit 110, the confirmation response generation unit 120, and the segment transmission unit 130, or the segment reception unit 310, the congestion control unit 320, and the segment transmission unit 330, respectively. However, they may also function as the segment receiver 110, the confirmation response generator 120, and the segment transmitter 130, respectively. The control unit 10 performs the same operation as the control unit 30, but congestion control is performed by conventional TCP. Also in this case, the control unit 10 may be provided with components equivalent to the segment reception unit 310, the congestion control unit 320 (however, the conventional congestion control by TCP) and the segment transmission unit 330 may be provided, and the segment reception may be performed. The unit 110, the confirmation response generation unit 120, and the segment transmission unit 130 may function as the segment reception unit 310, the congestion control unit 320, and the segment transmission unit 330, respectively. In addition, about a memory | storage part, the data segment to memorize | store is distinguished by providing a memory | storage part different from the memory | storage part 40, or providing another area | region in the memory | storage part 40. FIG.

  Here, the effect of the first embodiment will be described by comparing with the congestion control by the conventional TCP Westwood +.

FIG. 10 is a diagram illustrating a change in the congestion window CWND in the congestion control according to the conventional TCP Westwood + and a change in the congestion window CWND in the congestion control according to the first embodiment. FIG. 10A shows the case of the conventional TCP Westwood +, FIG. 10B shows the case of the first embodiment, the vertical axis is the congestion window CWND, and the horizontal axis is time. A solid line represents a change in the congestion window CWND, a thick dotted line represents a change in the slow start threshold ssthresh, and a value of BWE _k × RTTmin is indicated by a broken line. In FIG. 10, it is assumed that the value of the congestion window CWND that causes retransmission timeout is the same from the time point when the congestion control is executed by the triple overlap response (time point t ₁ ). and it became the retransmission timeout in ₂ and t _3. As can be seen from FIG. 10, in the case of the first embodiment shown in FIG. 10B, stable communication is possible with a small value of BWE _k × RTTmin and little fluctuation, and the time until a retransmission timeout is relative. becomes long, the transmission amount from the time point t ₁ where the congestion control is performed up to the point of retransmission timeout (the triangular area has a gray) has also become relatively large, it is possible to reduce the number of the congestion detecting It has become. Also, it can be seen that the transition time from the retransmission timeout to the congestion avoidance stage (the stage where the congestion window CWND becomes equal to or higher than the slow start threshold ssthresh) is shorter in FIG. As a result, a sudden increase in transmission rate can be suppressed and segment loss can be suppressed.

  Another embodiment of the present invention will be described.

In communication via a wireless access point as in the first embodiment, random delay jitter occurs regardless of the amount of communication data, as shown in FIG. 2, so the minimum value RTTmin of the round trip delay time as shown in Equation 5 Is effective since it can suppress the influence of random delay jitter when there are many receiving terminals. However, when the number of receiving terminals is extremely small, such as one, the conventional congestion control using TCP Westwood + works effectively. Therefore, the number of receiving terminals or the number of communication sessions is detected at the time of starting relay processing in the relay device, and the number of detected receiving terminals (hereinafter referred to as “the number of detected terminals”) or the number of communication sessions (hereinafter, Only when the “number of detected sessions” is equal to or greater than a predetermined value (threshold), the congestion control using the number 5 is performed. That is, when the number of detected terminals is equal to or greater than the threshold for receiving terminals (hereinafter referred to as “terminal threshold”) or the number of detected sessions is equal to or greater than the threshold for communication sessions (hereinafter referred to as “session threshold”), Formula 5 is used. calculating a bandwidth estimates SBWE _k Te, otherwise (speed detection terminals rather smaller than terminal a threshold, and if the number of detection session is less than the session threshold), the number 1 being used in conventional TCP Westwood + The band estimation value SBWE _k is calculated by using this.

  FIG. 11 shows a configuration example of the relay device of the present embodiment (second embodiment). Compared to the configuration example of the first embodiment shown in FIG. 5, the control unit 30 is a control unit 50. Other constituent elements are the same as those in the first embodiment, and the description thereof will be omitted.

  A configuration example of the control unit 50 is shown in FIG. Compared to the configuration example of the control unit 30 in the first embodiment shown in FIG. 7, a threshold setting unit 540 and a target detection unit 550 are added, and the congestion control unit 320 is a congestion control unit 520. The other components are the same as the configuration example of the control unit 30 in the first embodiment, and thus the description thereof is omitted.

  The threshold setting unit 540 outputs the terminal threshold Tht and the session threshold Ths set based on the wireless access point to be used, the purpose of use, the installation environment, and the like to the congestion control unit 520.

  The target detection unit 550 detects the number of reception terminals and the number of communication sessions in the applied network (in this embodiment, a wireless network) at the time when the control unit 50 starts relay processing, and detects the number of detection terminals Nt and the number of detection sessions Ns. Is output to the congestion control unit 520.

The congestion control unit 520 has the same basic function as that of the congestion control unit 320, but uses a calculation formula for the estimated bandwidth value SBWE _k by comparing the number of detected terminals Nt and the terminal threshold Tht and the number of detected sessions Ns and the session threshold. Change based on comparison with Ths. Specifically, when the number of detected terminals Nt is equal to or greater than the terminal threshold Tht, or when the number of detected sessions Ns is equal to or greater than the session threshold Ths, the band estimation value SBWE _k is calculated using Formula 5, otherwise, the number 1 is used to calculate the band estimation value SBWE _k .

  The operation of the control unit 50 after receiving the acknowledgment segment AckR in the second embodiment is the “bandwidth estimation value” in step S40 in the operation example in the first embodiment shown in FIGS. The operation of “available bandwidth calculation” is different, and the other operations are the same as the operation example in the first embodiment. An example of the operation of the different portions will be described with reference to the flowchart of FIG. It is assumed that the terminal threshold Tht and the session threshold Ths are already set by the threshold setting unit 540, and the detected terminal number Nt and the detected session number Ns are already detected by the target detecting unit 550.

Congestion control unit 520, after the minimum value RTTmin calculating the round trip time (step S30 in FIG. 8), check the number _{A S} acknowledgments contained in the acknowledgment segment AckR (step S410). Next, the terminal threshold Tht and the session threshold Ths are input from the threshold setting unit 540 (step S420), and the number of detected terminals Nt and the number of detected sessions Ns are input from the target detection unit 550 (step S430). Then, the number of detected terminals Nt is compared with the terminal threshold Tht and the number of detected sessions Ns is compared with the session threshold Ths (step S440). The detected terminal number Nt is equal to or greater than the terminal threshold Tht or the detected session number Ns is the session threshold Ths. for more than a few 5 than to calculate the bandwidth estimates SBWE _k (step S450), otherwise, with the number _{a S} acknowledgment confirmed in round trip time RTT _k and step S410 calculated in step S30 The band estimation value SBWE _k is calculated from Equation 1 (step S460). After calculating the estimated bandwidth value SBWE _k , the available bandwidth BWE _k is calculated from Equation 2 (step S470).

The condition for switching the calculation formula of the band estimation value SBWE _k may be a case _where the number of detected terminals Nt is equal to or greater than the terminal threshold Tht and the number of detected sessions Ns is equal to or greater than the session threshold Ths. Moreover, it is good also as switching using only the detection terminal number Nt and the terminal threshold value Tht, or switching using only the detection session number Ns and the session threshold value Ths.

  In the case where there is a possibility that communication with high delay may occur in the network environment in the above-described embodiments (first embodiment, second embodiment), in order to improve the basic performance of TCP, Thus, “large initial window” defined by RFC3390, “window scaling option” defined by RFC1323, “SACK (Selective Acknowledgment)” defined by RFC2018, etc. may be applied. Thereby, the performance improvement of a relay apparatus and the synergistic effect with this embodiment can be aimed at. In the above-described embodiment, one application network and one non-application network are assumed, but a plurality of each may exist. In that case, a control unit is prepared for each network.

  It is also possible to realize the processing of each component excluding the storage unit in the above-described embodiment by a program and realize the above-described embodiment by the configuration of a computer and a storage device.

  In addition, this invention is not limited to the said form, A various deformation | transformation is possible in the range which does not deviate from the summary of this invention.

1A, 1B terminal 2 Wireless access point 3, 7 Server 4, 4A Reception terminal 5 Relay device 10, 30, 50 Control unit 20 Segment management unit 40 Storage unit 110, 310 Segment reception unit 120 Confirmation response generation unit 130, 330 Segment transmission Units 320 and 520 Congestion control unit 540 Threshold setting unit 550 Target detection unit

Claims (12)

In a relay device that relays communication between a plurality of networks, includes a control unit that controls communication for each network, and performs communication between the networks via the control unit,
As the control unit, the band estimation value is estimated using a predetermined size corresponding to the segment size and the minimum value of the round trip delay time, and congestion control is performed using a slow start threshold calculated based on the band estimation value. A relay device comprising at least one control unit 1 for performing.   The relay apparatus according to claim 1, wherein the control unit 1 controls communication in a network in which the round-trip delay time varies greatly due to factors other than the traffic.   The relay device according to claim 2, wherein the control unit 1 controls communication in a wireless network. A storage unit for storing the segment received from the transmission terminal is further provided for each control unit,
4. The relay device according to claim 1, wherein the input / output of the segment between the control units is performed via the storage unit, whereby the control units can independently control. 5. Each of the control units is
When confirming that the segment received from the transmitting terminal is received by the receiving terminal, erase the confirmed segment from the storage unit,
The relay apparatus according to claim 4, wherein when the segment received from the transmission terminal is retransmitted to the reception terminal, the segment stored in the storage unit is retransmitted. The control unit 1, a relay device according to any one of claims 1 to 5 perform the estimation of the bandwidth estimate by dividing the segment size with a minimum value of the round trip time.   The control unit 1 uses the minimum value of the round trip delay time when at least one of the number of receiving terminals and the number of communication sessions in the network for which the control unit 1 controls communication is equal to or greater than a predetermined value. The relay apparatus according to claim 1, wherein a band estimation value is estimated. In a relay method for relaying communication between a plurality of networks, having a control step of controlling communication for each network, and performing processing for communication between the networks in the control step,
As the control step, a band estimation value is estimated using a predetermined size according to the segment size and a minimum value of the round trip delay time, and congestion control is performed using a slow start threshold calculated based on the band estimation value. A relay method comprising at least one control step 1 to be performed. A storage step for storing the segment received from the transmission terminal in the storage unit is further provided for each control step,
When the control step confirms that the segment received from the transmitting terminal is received by the receiving terminal after execution of the storing step , the confirmed segment is erased from the storage unit,
The relay method according to claim 8 , wherein, after the storage step is performed, when the segment received from the transmission terminal is retransmitted in the control step, the segment stored in the storage unit is retransmitted. In the control step 1, the relay method according to claim 8 or 9 performs estimation of the bandwidth estimate by dividing the segment size with a minimum value of the round trip time.   In the control step 1, when at least one of the number of receiving terminals or the number of communication sessions in the network that controls communication in the control step 1 is a predetermined value or more, the minimum value of the round trip delay time is used. The relay method according to claim 8, wherein a band estimation value is estimated.   The relay program for performing the relay method in any one of Claims 8 thru | or 11.