JPH10150435A - Method for controlling data transfer error - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide the method for controlling a data transfer error by which a high throughput is obtained even when a band width delay product is high without the need for a buffer of a large capacity. SOLUTION: Transfer object data from a transmitter side are all packet- processed as a main transfer phase (102), a sequence number is provided to the data and the resulting packet is transferred to a receiver side (103), the receiver side receives the packet and stores the packets sequentially through random access and stores the sequence number of the packet normally received and informs it to the transmitter side for each prescribed period during the reception of the packets. Then the transmitter side stores the sequence number informed from the receiver side during packet transfer to a packet group information(PGI) table (112), and reads the packet with the sequence number stored in the PGI table through random access in response to the end of the main transfer phase and transfers the packet again to the receiver side in the re-transfer phase.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data transfer error control method, and more particularly to a data transfer error control method suitable for high-speed transfer of large-scale data via a predetermined communication network.

[0002]

2. Description of the Related Art In general, in a large-scale data transfer application requiring highly reliable data transfer, a sequence control procedure for performing error control based on a sequence number sequentially assigned to each data to be transferred, for example, to each packet, is described. The data transfer is performed by using the data transfer.
In such a data transfer error control method, usually, when a transfer error such as packet loss occurs, the data transfer is interrupted by a sequence control procedure. For example, T
In the CP, the transfer of the packet after the lost packet is interrupted, the lost packet is retransmitted, and the transfer of the subsequent packet is restarted in accordance with the success of the transfer.

On the other hand, in XTP, a selective retransmission means is used, and another protocol such as RAP (Random A) is used.
ccess Protocol), a selective re-transfer unit using a random access unit of a terminal device is used. However, in the RAP, the transmission side is interrupted during the original packet transfer in response to the occurrence of an error, so that the overall throughput is reduced. Also, recently proposed Netwa
rp (NTT) also targets high throughput, but its error control part is still unclear.

[0004]

[0005] Large-scale data transfer applications that require highly reliable data transfer are:
When the communication is performed between terminal devices capable of randomly accessing transfer data, no sequence control procedure is required. However, in the conventional general-purpose data transfer error control method, the transfer is stopped when the packet is damaged or lost, and the transfer is restarted after the damaged or lost packet is recovered. For this reason,
There is no problem if the distance between the terminal devices performing communication is short and the data transfer speed is low, but if the distance between the terminal devices is long and the data transfer speed is high, that is, the data transfer delay and the transmission bandwidth used for data transfer (Hereinafter referred to as bandwidth delay product) is extremely large, there is a problem that the throughput is reduced.

[0005] In particular, in a method of periodically performing confirmation notification between transmission and reception during data transfer as used in TCP or the like, when large-scale data is transferred over a long distance, the throughput becomes extremely poor, and furthermore, prediction is not possible. When there is a large fluctuation in the round trip time, the transmitting side often stops. Also, in theory, in order to increase the throughput, the conventional data transfer error control method needs to provide an extremely large buffer in the terminal device. For this reason, there is a problem that the memory is excessively used in the terminal device, and it is difficult to simultaneously set a plurality of communication connections. The present invention has been made to solve such a problem, and it is an object of the present invention to provide a data transfer error control method capable of obtaining a high throughput even when a bandwidth delay product is large without requiring a large-capacity buffer. The purpose is.

[0006]

To achieve the above object, a data transfer error control method according to the present invention comprises:
On the transmission side, the transfer target data stored in the transmission side data storage means is sequentially read and divided, sequentially stored in a series of packets with sequence numbers, and transferred to the reception side. , And the data stored in the normally received packet is randomized.
The access side accumulates the data in the data storage unit on the receiving side, holds the sequence number of the packet that could not be received normally and notifies the transmitting side of the sequence number at predetermined intervals during packet reception. Hold the sequence number notified from the receiving side in the table,
If a sequence number is stored in the table after the end of a series of packet transfer, data corresponding to the sequence number is read out by random access from transfer target data stored in the transmission side data storage means,
The packet is stored in the packet to which the sequence number is added, and is retransmitted to the receiving side. Therefore, all data to be transferred is transmitted from the transmitting side to the receiving side in the main transfer phase, and only packets for which retransmission is requested from the receiving side during that time are retransmitted from the transmitting side to the receiving side in the retransmitting phase after the main transfer phase ends. Is done.

[0007]

Next, the present invention will be described with reference to the drawings. FIG. 1 is a flowchart showing a transmitting side control procedure according to a data transfer error control method according to one embodiment of the present invention, and FIG. 2 shows a receiving side control procedure according to a data transfer error control method according to one embodiment of the present invention. It is a flowchart. The gist of the present invention is to transmit all data to be transferred from a transmitting side to a receiving side in a main transfer phase,
During this time, only the packet requested to be retransmitted by the receiving side is transferred in the retransmission phase after the end of the main transfer phase.

[0008] In particular, the transmitting side and the receiving side have random access means capable of reading and writing only an arbitrary part of transfer target data and received data stored in data storage means such as a hard disk device and a memory. I have.
This random access means is common as in the prior art, and a detailed description is omitted. The random access means of the present invention reads and writes an arbitrary part of the transfer target data and the received data in units of the data length of one packet or a plurality of packets.

The data transfer rate when the data transfer error control method of the present invention is used is determined by a communication line speed (hereinafter referred to as a communication line speed) determined by a negotiation performed between a communication party, that is, a transmitting terminal device and a receiving terminal device, and a communication network. , Negotiated transfer speed). It should be noted that whether rate control is provided by a transport protocol or directly by a lower layer mechanism is not directly related to the function according to the present invention.

First, the transmitting side establishes a communication connection with the receiving side via the communication network and determines a negotiated transfer speed with the communication network (step 101). Subsequently, the transfer target data is sequentially read from the transmission side data storage means, and packetization is started (step 102). As shown in FIG. 3, each packet has its own sequence number and error check information. FIG. 3 is an explanatory diagram showing the structure of a packet used in the data transfer error control method according to one embodiment of the present invention.

In FIG. 3, error check information 31 is information for checking the packet itself for damage, data 32 is information in which transfer target data is divided and stored, and a sequence number 33 is in accordance with the division order of transfer target data. Is a number assigned to each packet. In this manner, the data to be transferred is sequentially divided and packetized, transferred to the receiving side (step 103), and repeatedly performed until the transfer of the last packet is completed (step 1).
04: NO). This corresponds to the main transfer phase described above.

As means for identifying the last packet,
For example, the last bit of the sequence number 33 may be used as a “last packet indication” flag, or a higher-level file transfer application may notify how much data is to be transmitted. . This can also be used as a means for notifying the receiving side of the last packet from the transmitting side to the receiving side.

On the other hand, in the main transfer phase shown in FIG. 2A, the receiving side first establishes a communication connection with the transmitting side via the communication network in the same manner as the transmitting side, and establishes a connection with the communication network. A negotiated transfer rate is determined between the two (step 120). The receiving side is provided with a virtual buffer called PG (packet group) that does not physically exist. This PG is composed of n packets, and the n packets are processed as error control units. Note that n is determined by parameters such as the size of the transfer target data and the negotiated transfer speed.

On the receiving side, a PG information (PGI) table is provided for the PG. In the PGI table, among the packets contained in the same PG, data damage is performed by the data inspection information 31 as PG information. Are stored, and the sequence number of the packet whose packet loss has been confirmed by the sequence number 33 is stored. Then, the contents of the PGI table are notified from the receiving side to the transmitting side by PGI packets every predetermined period set in the timer. Note that the PGI packet itself has a series of sequence numbers.

This period, that is, the timer value t _out is calculated based on the following formula (1) based on the negotiated transfer rate. t _out = (packet size × PG size) / negotiation transfer speed Therefore, the receiving side sets the timer value t _out calculated in this way and starts the timer (step 12).
1). Thereafter, until the timer times out, the packets sequentially transferred from the transmitting side are sequentially received, the original data to be transferred is partially restored, and among the received data of the receiving side data storage means by the random access means, For example, it is stored at a position based on the sequence number (step 12
3). If data damage or packet loss is detected, the sequence number of the packet is stored in the PGI table (step 124).

If the timer has timed out in step 122 (step 122: YES), the PG
The PG information in the I table is transferred to the transmitting side by a PGI packet. As a result, the sequence number of the packet in which data damage or packet loss has occurred is notified to the transmission side. When all packets included in the same PG are normally received, blank PG information is transferred to the transmission side. On the other hand, on the transmitting side, packet transfer (step 10)
In parallel with 3), PG information from the receiving side is sequentially received (step 111), and the PGI table is updated (step 112).

In this way, the transfer target data is sequentially packetized and transferred to the receiving side in the main transfer phase. Thereafter, the transmitting side checks the PGI table to determine whether there is a packet requested to be retransmitted by the receiving side (step 105). If there is a packet requested to be retransmitted (step 105: YES). Then, the process shifts to the retransmission phase to retransmit these packets (step 106).

In the retransfer phase, first, of the data to be transferred by the transmission-side data storage means, desired data is read out from the position corresponding to the sequence number of the retransmission-requested packet by the random access means and packetized (step S1). 102), these are sequentially transferred to the receiving side (steps 103 and 104), and the PGI table is updated. On the other hand, the receiving side responds to the notification of the end of the main transfer phase from the transmitting side (step 126: YE
S), proceed to the retransfer phase (step 127).
FIG. 2B shows the retransmission phase on the receiving side. First, the receiving side calculates the size SPG of the _{PG based} on the following equation (2) (step 130).

S _PG (i) = R _p -1 if R _p <S _PG (i-1) S _PG (i) = S _PG (i-1) -1 if R _p ≧ S _PG (i-1) Here, i is a natural number, S _PG (i) is the PG size at the i-th retransmission, and R _P is the number of packets to be retransmitted. Here, if the previous PG size is larger than the number of packets to be retransmitted, a number smaller by one packet than the number of packets to be retransmitted is set as a new PG size. If the previous PG size is smaller than the number of packets to be retransmitted, a new PG size that is smaller by one packet than the previous PG size is set.

Subsequently, the receiving side calculates a timer value based on the PG size in the same manner as in the above-described main transfer phase and starts the timer (step 131). Thereafter, the receiving side sequentially receives the packets in steps 132 to 134, updates the PGI table, and transfers the PG information in the PGI table to the transmitting side in response to a timeout (step 132: YES) (step 135). ). If there is a packet for which a retransmission request is present in the PGI table (step 136: YES), the flow returns to step 130, where the retransmission phase is executed again. (Step 1)
37).

Incidentally, forward error correction may be used in the reverse direction in order to facilitate error control and ensure proper transfer of PG information. Because the size of the PGI packet is very small, the overhead imposed by forward error correction is minimal. Also, PG
If there are enough packets, the variation in the delay of individual packets in the same PG will be canceled out by the statistical averaging property, and the possibility of overlapping PG information will be reduced.

If all the packets corresponding to the new PG are normal, the receiving side sets the P related to the previous PG.
The G information is transferred to the transmitting side. Therefore, the PGI table maintained on the transmission side is not only cleared in the forward direction, that is, in response to the end of the packet transfer from the transmission side, but also updated every time in the reverse direction. This avoids duplicate transfer in the retransfer phase. Also, all PG
Even if the information is lost during the transfer, the transmitting side can transfer the PG information properly by requesting the transfer of the PG information after transferring all the data to be transferred.

When transferring the transfer target data stored in a predetermined file from the transmission side to the reception side, that is, in the case of file transfer, the reception side receives a correct packet and deletes the original transfer target data. Partially generated sequentially,
The data is written to a predetermined storage medium such as a disk device as a file. From this, the random
The order of the received packets is not important, since the access means allows any part of the data to be transferred to be stored.

[0024]

An embodiment of the present invention will be described with reference to FIG. FIG. 4 is a sequence diagram showing a data transfer error control method as one embodiment of the present invention. First, the packet # to which the sequence number “1” is assigned from the transmitting side
The data is transferred to the receiving side in order from 1. Here, the negotiated transfer speed is 100 Mb / sec, the PG size is 1000,
If the size is 1500 bytes, the timer value t
_out is 1500 × 8 × 1000) / 100,000,000,0
It is calculated as 00 = 0.12 seconds.

The receiving side extracts data from packets that have arrived normally and stores them sequentially. At the same time, if data damage or packet loss is detected, the sequence number corresponding to the packet is stored in the PGI table. In this case, the sequence numbers “5” and “3”
Packet # 5, packet # 34 to which "42" is assigned
Two have been lost. When the timer times out for the first time, the receiving side notifies that a packet # 5 and a packet # 342 have not been received by a PGI packet. This PGI
The packet has its own sequence number, in this case "1". Thereafter, the receiving side restarts the timer, performs reception processing of the subsequent packet, and transfers a new PGI packet according to the timeout.

The transmitting side receives this PGI packet during packet transfer and updates the PGI table. In addition,
As long as there is data to be transferred, no new retransmission phase starts. Each time a PGI packet is sent from the receiving side, it is updated in both the backward and forward directions. In other words, each packet previously reported as lost / errored is reported again if it has not yet arrived.

For example, when a packet requested to be retransmitted in the previous PGI packet arrives after a while (for example, in the next period), this packet is deleted from the PGI table held by the receiving side. Subsequently, since the PG information in this table is notified to the transmitting side, the transmitting side can implicitly know from the PG information that the packet has been normally transferred.

At the end of the main transfer phase, the sender has a list of packets in the PGI table that must be retransmitted. Therefore, the process immediately shifts to the retransmission phase. The receiving side recalculates the timer value according to the equation (2), and the transmitting side calculates the packet # 5 and the packet # 342.
To start transferring packets numbered packet #z. This eliminates the need for the transmitting side to wait for a retransmission request notification from the receiving side after the main transfer phase is completed, and can reduce an idle period of 2 × RTT (round trip time) required for the retransmission request notification, thereby improving the overall throughput. I do. After confirming that there are no more items remaining in the PGI table, the receiving side starts the disconnection process.

As described above, first, in the main transfer phase, all data to be transferred is packetized from the transmitting side, a series of sequence numbers are assigned and transferred to the receiving side, and the receiving side receives these packets and transfers them by random access. The target data is sequentially stored, the sequence number of the packet that has not been received normally is retained as PG information, and the packet is notified to the transmitting side at predetermined intervals during packet reception. Then, the transmitting side receives the PG information from the receiving side during the packet transfer, stores the PG information in the PGI table, and reads out the packet of the sequence number stored in the PGI table by random access at the end of the main transfer phase. In the retransmission phase, the data is retransmitted to the receiving side.

Therefore, as compared with the conventional method in which the transfer is stopped when the packet is damaged or lost, and the transfer is restarted after the damaged or lost packet is recovered, the packet is damaged or lost. Even when such an error occurs, data transfer is not interrupted, and data transfer can be performed efficiently without requiring a large-capacity buffer. This eliminates the time spent on retransmission request notification performed between transmission and reception in response to the occurrence of an error, the distance between terminal devices is long, and when the data transfer speed is fast, that is, when the bandwidth delay product is extremely large, The overall throughput is dramatically improved as compared to the prior art.

[0031]

As described above, according to the present invention, on the transmitting side, data to be transferred is sequentially stored in a packet with a series of sequence numbers and transferred to the receiving side.
The data stored in the packet normally received from the transmitting side is stored in the data storing means of the receiving side by random access, and the sequence number of the packet which could not be received normally is retained, and the packet is sequenced at predetermined intervals during packet reception. The number is notified to the transmitting side, and the transmitting side holds the sequence number notified from the receiving side during the packet transfer in a table, and when the sequence number is stored in the table after a series of packet transfer ends, The data corresponding to the sequence number is read from the transfer target data by random access, stored in a packet to which the sequence number is assigned, and retransmitted to the receiving side. Therefore, compared to the conventional method of stopping transmission when a packet is damaged or lost and restarting transmission after the damaged or lost packet is recovered, errors such as packet damage or loss are compared. Even when the error occurs, data transfer is not interrupted, and data transfer can be performed efficiently without requiring a large-capacity buffer. Furthermore, the time spent for retransmission request notification performed between transmission and reception in response to the occurrence of an error is eliminated, and when the distance between terminal devices is long and the data transfer rate is high, that is, when the bandwidth delay product is extremely large, the conventional method is used. The overall throughput is dramatically improved as compared to.

[Brief description of the drawings]

FIG. 1 is a flowchart illustrating a transmission side control procedure of a data transfer error control method according to an embodiment of the present invention.

FIG. 2 is a flowchart showing a receiving side control procedure of the data transfer error control method according to one embodiment of the present invention.

FIG. 3 is an explanatory diagram showing a configuration of a packet.

FIG. 4 is a sequence diagram illustrating an example of data transfer error control.

[Explanation of symbols]

101 to 112: transmission side control procedure, 120 to 127 ... reception side control procedure (main transfer phase), 130 to 137 ... reception side control procedure (retransmission phase), 31 ... data inspection information, 32 ... data, 33 ... sequence number.

Claims (1)

[Claims] In a data transfer error control method for transferring data to be transferred from a transmission side to a reception side via a communication network, the transmission side stores the transfer target data stored in the transmission side data storage means. The packet is sequentially read and divided, sequentially stored in a series of packets with sequence numbers, and transferred to the receiving side. The receiving side sequentially receives the packets from the transmitting side, and randomizes the data stored in the normally received packet. -While being stored in the data storage means on the receiving side by access,
Holding the sequence number of the packet that could not be received normally, and notifying the sequence number to the transmitting side at predetermined intervals during packet reception, the transmitting side stores the sequence number notified from the receiving side during packet transfer in a table. If a sequence number is stored in the table after the end of a series of packet transfer, the data corresponding to the sequence number is randomly accessed from the transfer target data stored in the transmission side data storage means. A data transfer error control method, wherein the data is read out, stored in a packet provided with the sequence number, and retransmitted to a receiving side.