JP2013121028A - Data packet transmission method and system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To restrict a reduction in transmission rate by congestion control in a transmission line in which packets do not disappear due to congestion.SOLUTION: In a data packet transmission system of the invention herein applied for, a counter device 20 on the transmitting side comprises: a plurality of TCP connection units 21 which, when a congestion window exceeds a prescribed value, notify to the effect that TCP connections are in a standby state; a management unit 22 which manages the usage state of the TCP connection units; and a transmitting unit 23 which acquires sockets for TCP connections in a standby state from the management unit 22, divides the original data to be transmitted to which a single TCP connection belongs, and packetizes the divided data and the beginning positions of the divided data and then assigns them to a plurality of sockets in a standby state.

Description

  The present invention relates to a data packet transmission method and system, and more particularly to a TCP (Transmission) in a system in which a transmission path between a data transmission side and a data reception side uses the entire physical band or is not congested in an upper limit band in which bandwidth is guaranteed. The present invention relates to a data packet transmission method and system for performing high-speed and stable data transmission using a parallel connection of (Control Protocol).

  As shown in Non-Patent Document 1, in data transmission by TCP, the data transmission rate is determined based on the size of the congestion window uniquely owned by the transmission side, and this congestion window size is a very small value at the time of starting data transmission. Thus, every time an acknowledgment packet from the receiving side is returned, it is increased slightly (slow start). Therefore, a throughput close to the theoretical value cannot be obtained unless the number of packets from the start of data transmission exceeds a certain number. It is also recommended to set the congestion window size to the initial value even when data transmission is not performed for a certain period.

  As shown in [HTTP1.1 Keep Alive Function], instead of establishing / terminating a TCP connection for each request from an application, the session should be reused without being immediately terminated after the first session is established. Thus, a device has been devised to save time for establishing / ending a TCP connection. [HTTP1.1 Keep Alive function] results in the congestion window size being maintained when there is a data transfer request at a relatively short interval, but immediately after establishing a TCP connection or when the interval between data transfer requests is long. TCP has a slow start function.

M.M. Allman, V.M. Paxon and W.M. Stevens, “TCP congestion control”, RFC2581, April 1999, http: // www. ietf. org / rfc / rfc2581. txt M.M. Mathis, et al. , “TCP Extended Statistics MIB.”, Network Working Group, Request for Comments: 4898, Category: Standards Track, May 2007, http: // www. ietf. org / rfc / rfc4898. txt

  When packet loss due to congestion does not occur in a transmission path where the data transmission side and data reception side are directly connected or the bandwidth is strictly controlled, congestion control is not necessary, but data transmission is performed at the start of transmission or for a certain period of time. If not, the slow start function works. Even if a packet is lost due to a transmission error, congestion control works and the bandwidth is limited to half due to a decrease in the congestion window size.

  An object of the present invention is to suppress a decrease in transmission rate due to congestion control in a transmission path in which packet loss due to congestion does not occur.

  In order to achieve the above object, a data packet transmission system according to the present invention is a data packet transmission system that performs data transmission between opposing devices using a TCP connection on one or more physical links, wherein A plurality of TCP connection units that establish a TCP connection and notify that the TCP connection unit is in a waiting state; and a management unit that manages a usage state of the TCP connection unit by receiving a notification from the TCP connection unit The socket of the TCP connection in the standby state is acquired from the management unit, the original data to be transmitted is divided, the divided data and the start position of the divided data are packetized, and the plurality of sockets in the standby state A transmitting unit for allocating the divided data to the receiving side, the opposite device on the receiving side establishes a TCP connection and A plurality of TCP connection units to be received and a reception that writes to the reception buffer from a position corresponding to a pointer indicating the start position of the divided data, and outputs the original data after completion of writing all the divided data to the reception buffer A part.

  In the data packet transmission system according to the present invention, the transmission-side counter device includes a plurality of TCP connection units, a management unit, and a transmission unit, and the reception-side counter device includes a TCP connection unit and a reception unit. Packets including the original data can be allocated to a plurality of TCP connections, and can be transmitted in parallel in the TCP layer, and returned to the original data after transmission. Therefore, the data packet transmission system of the present invention can suppress a decrease in transmission rate due to congestion control in a transmission path where packet loss due to congestion does not occur.

In the data packet transmission system according to the present invention, the TCP connection unit on the transmission side detects a decrease in a congestion window to a predetermined value or a link abnormality, and the management unit detects a decrease in the congestion window or a link abnormality. The transmission unit notifies the transmission unit that a socket abnormality has occurred in the TCP connection unit on the transmission side, and the transmission unit assigns the divided data allocated to the socket in which the socket abnormality has occurred to the standby socket. Allocate.
According to the present invention, when a congestion window falls below a predetermined value or a link abnormality occurs in any of a plurality of TCP connections, a TCP connection in a standby state that is set in advance but is not used for actual data transmission is set. Can be switched. As a result, it is possible to quickly cope with a decrease in the congestion window to a predetermined value or lower or a link abnormality, and it is possible to prevent occurrence of untransferred data.

In the data packet transmission system according to the present invention, the TCP connection unit on the transmission side enters the waiting state by sending a dummy data packet until the congestion window exceeds a set value.
According to the present invention, it is possible to increase the congestion window size on the transmission side of a plurality of TCP connections to which divided original data is allocated.

  In order to achieve the above object, a data packet transmission method of the present invention is a data packet transmission method for performing data transmission between opposing devices using a TCP connection on one or more physical links, and a TCP connection on the transmission side The management unit manages the usage state of the TCP connection unit on the transmission side by notifying the management unit that the congestion window has exceeded the predetermined value, and the transmission unit The original data to be transmitted is divided, the divided data and the head position of the divided data are packetized and allocated to the plurality of standby sockets, and the TCP connection unit on the transmission side follows the instruction from the transmission unit. A divided data transmission procedure for transmitting the divided data, and a TCP connection unit on the receiving side establishes a TCP connection to receive the divided data, and a receiving unit A divided data reception procedure for writing the divided data into a reception buffer from a position corresponding to a pointer indicating a head position of the divided data, and a receiving unit after the writing of all the divided data into the reception buffer is completed An original data reading procedure for reading data in order.

  Since the data packet transmission method of the present invention has a use state management procedure, a divided data transmission procedure, a divided data reception procedure, and an original data reading procedure in order, a packet including divided original data to be transmitted Can be allocated to a plurality of TCP connections, parallel transmission can be performed in the TCP layer, and the original data can be restored after transmission. Therefore, the data packet transmission method of the present invention can suppress a decrease in transmission rate due to congestion control in a transmission path where packet loss due to congestion does not occur.

In the data packet transmission method of the present invention, the TCP connection unit on the transmission side detects a decrease in the congestion window to a predetermined value or lower or a link abnormality detection procedure, and the transmission unit is a socket in which a socket abnormality has occurred. A socket replacement procedure for allocating the divided data allocated to the socket in the waiting state between the divided data transmission procedure and the divided data reception procedure.
According to the present invention, when a congestion window decreases or a link abnormality occurs in any of a plurality of TCP connections, it is possible to switch to another TCP connection that is set in advance but is not used for actual data transmission. As a result, packet loss can be dealt with quickly, and occurrence of untransferred data can be prevented.

In the data packet transmission method of the present invention, in the use state management procedure, the TCP connection unit on the transmission side transmits the dummy data packet until the congestion window exceeds a set value, thereby setting the standby state.
According to the present invention, it is possible to increase the congestion window size on the transmission side of a plurality of TCP connections to which divided original data is allocated.

  The above inventions can be combined as much as possible.

  According to the present invention, it is possible to reduce the influence of a decrease in throughput due to TCP congestion control even in a situation where a congestion window decreases due to packet loss due to a transmission error or the like in a transmission path in which congestion does not occur.

1 shows an example of a data packet transmission system according to a first embodiment of the present invention. It is a block diagram which shows the structure of the outline of 2nd embodiment of this invention. It is explanatory drawing which shows the operation | movement of the outline of 3rd embodiment of this invention. It is explanatory drawing which shows the operation | movement of the outline of 4th embodiment of this invention.

  Embodiments of the present invention will be described with reference to the accompanying drawings. The embodiments described below are examples of the present invention, and the present invention is not limited to the following embodiments. In the present specification and drawings, the same reference numerals denote the same components.

(First embodiment)
FIG. 1 shows an example of a data packet transmission system according to the first embodiment of the present invention. In the data packet transmission system according to the present embodiment, an arbitrary application 10 transmits data using the opposite device 20, and an arbitrary application 40 receives data using the opposite device 30. The transmission-side counter device 20 includes a plurality of TCP connection units 21, a management unit 22, and a transmission unit 23. The reception-side counter device 30 includes a plurality of TCP connection units 31 and a reception unit 32. The data transmission method according to the present embodiment includes a use state management procedure, a divided data transmission procedure, a divided data reception procedure, and an original data reading procedure in this order.

  In the usage state management procedure, the TCP connection unit 21 establishes a TCP connection and transmits a packet. When the congestion window exceeds a predetermined value, the TCP connection unit 21 notifies the standby state. Then, the management unit 22 manages the usage state of the TCP connection unit 21 by receiving a notification from the TCP connection unit 21. Here, the TCP connection unit 21 enters a standby state by sending dummy data packets until the congestion window exceeds the set value.

  In the divided data transmission procedure, the transmission unit 23 receives a transmission request from an application on the transmission side, acquires a socket of a TCP connection in a standby state from the management unit 22, divides original data to be transmitted, The divided data and the start position of the divided data are packetized and allocated to a plurality of standby sockets. In the present embodiment, an example in which the transmission unit 23 is provided in the transmission-side counter device 20 will be described, but an application may have the function of the transmission unit 23.

  In the divided data reception procedure, the receiving unit 32 receives the divided data via the TCP connection unit 31 and writes it into the reception buffer 321 from the position corresponding to the pointer indicating the head position of the divided data. After the writing of all the divided data to the reception buffer 321 is completed, the original data reading procedure is executed.

  In the original data reading procedure, the receiving unit 32 outputs the original data to the receiving-side application 40.

  By executing the above procedure in order, the socket can be passed to the transmission unit 23 in a state where the congestion window size is large, so that the influence of the TCP slow start can be eliminated. As a result, even in a situation where a congestion window is lowered due to packet loss due to a transmission error or the like in a transmission path in which congestion does not occur, it is possible to reduce the influence of throughput reduction due to TCP congestion control.

  Here, a socket abnormality detection procedure and a socket replacement procedure may be further provided between the divided data transmission procedure and the divided data reception procedure.

In the socket abnormality detection procedure, the TCP connection unit 21 detects a decrease of the congestion window to a predetermined value or a link abnormality. The management unit 22 notifies the transmission unit 23 that a socket abnormality has occurred in the TCP connection unit that has detected a decrease in the congestion window or a link abnormality.
In the socket replacement procedure, the transmission unit 23 allocates the divided data allocated to the socket in which the socket abnormality has occurred to another socket in the standby state.

Here, when the TCP connection is switched in the event of packet loss, it is necessary to transmit the divided data that has not been transferred in the connection before the switching. In this case, one of the following operations needs to be performed.
(A) The entire divided data that has not been transferred is transferred.
(B) The receiving side notifies the transmitting side of the area where the divided data has been received or has not been received, and transmits the divided data that has not been completed.
By performing the operation (A) or (B), generation of untransferred data can be prevented.

  By having the socket abnormality detection procedure and the socket replacement procedure described above, when a congestion window is lowered or a link abnormality occurs, the socket can be replaced with another waiting socket. This makes it possible to eliminate the influence of halving the congestion window size due to packet loss.

(Second embodiment)
FIG. 2 shows a schematic configuration of the second embodiment of the present invention.
In connection between devices, each device generates a plurality of connection threads 100 and management threads 110 based on a connection start request. Each connection thread establishes a TCP connection with the opposite apparatus as a connection start 101.

  Here, the connection thread 100 functions as the TCP connection unit 21, and the management thread 110 functions as the management unit 22. The connection thread 100 includes a connection start unit 101, a warm-up state unit 102, a standby state unit 103, a busy state unit 104, an abnormality detection state unit 105, and a connection end state unit 106. The management thread 110 includes a standby connection list 111 and a busy connection list 112.

  When the connection start unit 101 establishes the TCP connection, the operation is shifted to the warm-up state by the operation of the warm-up state unit 102, and dummy data is exchanged with the opposite device at a high frequency to expand the congestion window. When the congestion window is enlarged beyond the set value, the process shifts to the waiting state unit 103 and notifies the management thread 110. The management thread 110 registers the connection thread in the waiting connection list 111. In the connection thread 100 in the waiting state 103, the waiting state unit 103 exchanges dummy data at a low frequency so that the congestion window is not reduced. As a result, the congestion window size on the transmission side of the TCP connection can be increased.

  When the management thread 110 receives a socket request from the application on the transmission side, the management thread 110 passes the necessary number of connection information (sockets, etc.) from the waiting connection list 111 to the application, and notifies the corresponding connection thread 100 and receives information from the waiting connection list 111. Registration in the deleted / in-use connection list 112 is performed. The application on the transmission side has a function as the transmission unit 23. The connection thread 100 that has received the notification from the management thread 110 shifts to the busy state unit 104.

  When the management thread 110 receives a socket release from the application, the management thread 110 notifies the connection thread 100 and registers it in the deletion / waiting connection list 111 from the in-use connection list 112. The connection thread 100 that has received the notification shifts to the busy state unit 103. When each connection thread 100 detects a communication abnormality, the process proceeds to the abnormality detection state unit 105.

  When an abnormality is detected in the in-use state unit 103, the management thread 110 is notified, and the management thread 110 notifies the application of a socket abnormality. The abnormality detection state unit 105 shifts to the warm-up state unit 102 when the connection can be recovered, and shifts to the connection end state unit 106 when the connection cannot be recovered. At the end of the connection between the opposing devices, all the connection threads 100 used shift to the connection end state unit 106, and then all the threads are ended.

(Third embodiment)
FIG. 3 shows the schematic operation of the third embodiment of the present invention.
In this example, apparatus A (201) and apparatus B (221) are connected by physical links LA1 (202) to LB1 (222) and LA2 (203) to LB2 (223), and TCP connections ( 204, 224) CA1 to CB1, CA2 to CB2, CA3 to CB3, and LA2 to LB2 are stretched with CA4 to CB4 and CA5 to CB5. The vertical axis shows the progress of time.

  In the initial state in this example, CA1 to CB1, CA2 to CB2 are in use (205, 225), and CA3 to CB3, CA4 to CB4, CA5 to CB5 are in a standby state from the warm-up state (206, 226) ( 207, 227). When a decrease in the congestion window (208) is detected in the TCP connections CA2 to CB2, the connections are preferentially selected from the physical links LA2 to LB2 that are not the physical links LA1 to LB1 of the CA2 to CB2 among the waiting connections ( Here, CA4 to CB4 are selected). This takes into account that packet loss is likely to occur due to an increase in the transmission error rate as a precursor to a physical link failure. In the case of a link failure (209), all the connections in use are transferred to connections in the standby state of another link, and are set in use (210, 230). Immediately after the failure of the link is recovered, the state shifts to the warm-up state (211, 231).

(Fourth embodiment)
FIG. 4 shows the schematic operation of the fourth embodiment of the present invention.
In this example, the transmission buffer (310) of the application transmits data to the reception buffer (330) of the application of the opposite apparatus via the TCP connection (320).

  The application appends the data to be transferred to the transmission buffer empty area (314) from the end of the written data (313). When written data (313) is present in the area next to transmitted data (311), the data is divided and transferred to a TCP connection in use (320a) to be transmitted data (312). The divided data (312a to 312c) passed to each connection is packetized with a pointer indicating the start position of the data (322a to 322c) and transferred to the receiving side.

  When a decrease in congestion window (324) is detected during transfer, the TCP connection in the waiting state (320b) is assigned as an alternative, and the data packet is retransmitted (323b).

  In the application of the opposite apparatus, the divided data is written into the reception buffer from the position corresponding to the pointer indicating the start position of the divided data included in the packet, and is set as received data (331).

  The application on the reception side reads the received data (331) from the reception buffer (330) and sets it as a reception buffer free area (334).

  As described above in the first embodiment to the fourth embodiment, according to the first aspect of the present invention, the parallel connection of TCP connected by one or more physical links that do not cause congestion between the opposing devices. In the data transmission according to the above, each TCP connection is established before the transmission of user data is started, and the management thread is set to a waiting state. When there is a socket request from the application to the management thread, the TCP connection in the waiting state is selected and the socket is passed to the application. When a congestion window drop or link abnormality is detected, the management thread notifies the application of the socket abnormality, the application releases the corresponding socket, passes the alternative socket to the application, and transmits unsent data.

  The acquisition / monitoring of the congestion window can be directly performed by any OS / product corresponding to [TCP Extended Statistics MIB] in Non-Patent Document 2. Indirectly, it is possible to know the packet loss or the decrease in the congestion window by setting a timeout on the reading side of the TCP based on the communication delay time (RTT). In addition, the timer equivalent to the timer used in TCP flow control such as RTO (Retransmission Time Out) and delayed ACK, duplicate ACK is used, or the time required for transferring a predetermined number of bytes on the receiving side is measured and effective. It is possible to know a decrease in the congestion window by converting to a proper throughput and notifying the transmission side.

  The connection in which the abnormality is detected starts transmission of dummy data packets due to a failure reason and enters a waiting state, or terminates the connection and disables the connection. The connection in the waiting state transmits dummy data packets at a low frequency in order to maintain the congestion window.

  In this way, the influence of the TCP slow start can be eliminated by passing the socket to the application in a state where the congestion window size is large. In addition, when packet loss occurs, it is possible to eliminate the influence of the congestion window size halving due to packet loss by exchanging with another waiting socket.

  Since the present invention is independent of the TCP congestion control method, when the TCP congestion control method can be selected from a plurality of methods, the method having the largest congestion window size in the actual use environment is adopted. The average throughput is expected to be improved by selecting the TCP congestion control method. In the present invention, by setting the congestion window lowering detection threshold value to the average throughput, it is possible to guarantee the minimum bandwidth of the average throughput even in an environment where the congestion window size varies.

  According to the second aspect of the present invention, when two or more physical links are used, the socket of the TCP connection on another physical link is given priority when notifying the socket abnormality in addition to the first aspect. Select and pass to the application as an alternative. In this way, by selecting a TCP connection of another physical link, it is possible to avoid the congestion window from being lowered in the alternative socket when there is a cause of failure in the physical link itself.

  According to the third aspect of the present invention, the transmission thread and the reception thread are arranged between the management thread and the application in the first aspect. In the transmission thread, a transmission buffer is prepared, and when there is a transmission request from the application, the data is written in the empty area of the transmission buffer as untransmitted data. On the other hand, the necessary number of sockets are received from the management thread, and when there is unsent data in the transmission buffer, the data is divided, and the start position of the divided data and the divided data are packetized and transmitted through the sockets. . When a transmission abnormality such as a decrease in the congestion window is detected, the socket is released, an alternative socket is received from the management thread, and the corresponding divided data is retransmitted. In the reception thread, a reception buffer is prepared, and when a packet is received via a socket, the divided data received from the reception buffer address corresponding to the head position of the divided data included in the packet is written and used as received data. When there is a reception request from the application, the received data is passed and the corresponding reception buffer is released. As described above, by using the reception thread and the transmission thread, it is possible to configure an application without being aware of the parallel connection. Further, by directly using the socket, it becomes possible to detect a decrease in the congestion window earlier.

  According to the fourth aspect of the present invention, when a transmission abnormality is detected in the third aspect, the TCP connection of another physical link is selected as an alternative when the transmission abnormality is detected. If there is a cause of the problem, the alternative socket can also avoid the decrease in the congestion window.

  The present invention can be applied to the information communication industry.

10: Transmission-side application 20: Transmission-side counter device 21: Connection thread 22: Management thread 23: Transmission thread 24: Transmission buffer 31: Connection thread 32: Reception thread 33: Reception buffer 40: Reception-side application 100: Connection Thread 101: Connection start state portion 102: Warm-up state portion 103: Waiting state portion 104: In-use state portion 105: Abnormality detection state portion 106: Connection end state portion 110: Management thread 111: Waiting connection list 112: Use Medium connection list 120: Application 201: Device A
202: Physical link LA1
203: Physical link LA2
204: TCP connection (CA1 to CA5)
205: State in use (CA1 to CA2)
206: Warm-up state (CA3 to CA5)
207: Waiting state (CA3 to CA5)
208: Packet loss, 209 link failure 210: Busy state (CA4 to CA5)
211: Warm-up state (CA1 to CA3)
221: Device B
222: Physical link LB1
223: Physical link LB2
224: TCP connection (CB1 to CB5)
225: State in use (CB1 to CB2)
226: Warm-up state (CB3 to CB5)
227: Waiting state (CB3 to CB5)
230: State in use (CB4 to CB5)
231: Warm-up state (CB1 to CB3)
310: transmission buffer 311: transmitted data 312 (312a to 312c): transmitting data 313: written data 314: transmission data empty area 320: TCP connection 321: transmitted packet 322 (322a to 322c): transmitting packet 323b: Retransmitting packet 324: Transmission failure 330: Receive buffer 331: Received data 332: Received data (332a to 332c)
334: Receive buffer free area

Claims (6)

A data packet transmission system that performs data transmission between opposing devices using a TCP (Transmission Control Protocol) connection on one or more physical links,
The opposite device on the transmission side is
A plurality of TCP connections for establishing a TCP connection and notifying that it is in a standby state;
A management unit that manages a usage state of the TCP connection unit by receiving a notification from the TCP connection unit;
Obtains a TCP connection socket in the standby state from the management unit, divides the original data to be transmitted, packetizes the divided data and the start position of the divided data, and sends it to the plurality of sockets in the standby state. A transmitter to allocate;
With
The opposite device on the receiving side is
A plurality of TCP connection units for establishing a TCP connection and receiving the divided data;
A receiving unit that writes to the reception buffer from a position corresponding to a pointer indicating the start position of the divided data, and outputs the original data after completion of writing all the divided data to the reception buffer,
Data packet transmission system. The TCP connection unit on the transmission side detects a decrease in the congestion window to a predetermined value or less or a link abnormality,
The management unit notifies the transmission unit that a socket abnormality has occurred in the TCP connection unit on the transmission side that has detected a decrease in congestion window or link abnormality,
The data packet transmission system according to claim 1, wherein the transmission unit allocates the divided data allocated to the socket in which the socket abnormality has occurred to the standby socket.   3. The data packet transmission system according to claim 1, wherein the TCP connection unit on the transmission side enters the standby state by sending a dummy data packet until a congestion window exceeds a set value. 4. A data packet transmission method for performing data transmission between opposing devices using a TCP connection on one or more physical links,
Usage state management in which the management unit manages the usage state of the TCP connection unit on the transmission side by notifying the management unit that the TCP connection unit on the transmission side is in a standby state in which the congestion window exceeds a predetermined value Procedure and
The transmission unit divides the original data to be transmitted, packetizes the divided data and the start position of the divided data, and allocates the packets to the plurality of standby sockets. A divided data transmission procedure for transmitting the divided data according to an instruction;
A divided data reception procedure in which a TCP connection unit on the receiving side establishes a TCP connection to receive the divided data, and the receiving unit writes the divided data into a reception buffer from a position corresponding to a pointer indicating the head position of the divided data When,
An original data reading procedure in which the receiving unit reads the original data after completing the writing of all the divided data to the reception buffer;
A data packet transmission method comprising: A socket abnormality detection procedure in which the TCP connection unit on the transmission side detects a decrease in the congestion window below a predetermined value or a link abnormality;
A socket replacement procedure for allocating the divided data allocated to the socket in which the socket abnormality has occurred to the socket in the waiting state;
The data packet transmission method according to claim 4, further comprising: between the divided data transmission procedure and the divided data reception procedure.   6. The data according to claim 4, wherein, in the use state management procedure, the TCP connection unit on the transmission side enters the waiting state by sending a dummy data packet until a congestion window exceeds a set value. Packet transmission method.

Images