JP2000134279A - Flow control method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To avoid unnecessary cwnd reduction and to provide appropriate throughput by controlling a flow in accordance with the cause of the loss/ rejection of a packet. SOLUTION: When the occurrence of congestion is recognized in a transmission end node (step 15: YES), it is judged to be the loss of a packet owing to congestion occurred in a communication network and a congestion window value is reduced (step 33). When a congestion recognition means does not recognize the occurrence of congestion (step 15: N0), and an overlap Ack detection means recognizes overlap Ack reception (step 22; YES), it is judged to be the loss of the packet owing to the error of a line, which occurs in the communication network, and the lost data packet is retransmitted without reducing the congestion window value (step 23).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a flow control method, and more particularly, to a flow control method of a protocol layer 4 (OSI reference model / transport layer) in a node that executes a communication protocol process.

[0002]

2. Description of the Related Art At present, TCP (typical example: Reno-TCP) data flow control method widely used in a protocol layer 4 has been developed for a wire line, and the cause of a packet loss event is as follows. Treated as due to congestion. Therefore, in the flow control, the transmission amount is reduced by detecting a packet loss event using a window method. However, in order to control the transmission amount, a reception notification window (hereinafter, a
dwnd), a congestion window (hereinafter, referred to as cwnd) that estimates the transfer capability of the network is used.

The actual transmission amount is min (adwnd, c
wnd). However, min (x, y) indicates that the smaller value of x and y is used. When a packet loss event is detected, cwnd
Reduce. The flow control for adjusting the transmission amount is cw
nd, and a slow start threshold (ssthresh) is provided in addition to the above parameters.

In flow control, data packets are transmitted based on these parameters. Flow control has two phases called a slow start phase and a congestion avoidance phase, and the control method is different. Hereinafter, a conventional flow control method will be described with reference to FIG. Note that the unit of each parameter including the packet identification is the number of bytes in the implementation of TCP, but will be represented by the number of packets below for ease of explanation.

[0005] First, the transmission end node sets up a connection with the reception end node, and then sets up a ssthres
Initially, h = adwnd and cwnd = 1 are set (step 51). Then, enter the slow start phase, m
in (cwnd, adwnd), in this case cw
nd is adopted, one data packet (hereinafter, referred to as DT packet) is transmitted, and reception of an acknowledgment packet (hereinafter, referred to as Ack packet) returned from the receiving end node in response to the reception of the DT packet is performed. (Step 52).

Here, if an Ack packet is received within a predetermined period from the transmission of the DT packet (step 52: A
ck packet), this Ack packet is not a duplicate Ack described later (step 61: NO), and cwnd <s
sthresh (step 65: YE
S), +1 is added to cwnd (step 66), and step 5
Return to 2.

As a result, next, since cwnd = 2, two DT packets are transmitted, and thereafter, every time an Ack packet is received, the next transmittable amount, cwnd, is set.
Increases by +1 until the threshold value ssthresh is reached. If cwnd has reached ssthresh (step 65: NO), a congestion avoidance phase is entered, and cwnd is increased by 1 / cwnd every time an Ack packet is received (step 67). Therefore, compared to the slow start phase, the amount of packet transmission increases at a much slower level.

[0008] A packet loss event is detected by a method using a delivery confirmation timeout, and Fast Retra which detects and recovers early without waiting for the delivery confirmation timeout.
The nsmit method is used in combination. In the delivery confirmation timeout, when a predetermined period has elapsed from the transmission of the DT packet in step 52 (step 52: transmission timeout confirmation), the DT packet targeted for timeout is retransmitted (step 53), and ssthresh is set to min.
(Cwnd, adwnd) and the new cwnd
= 1 packet (step 54), and returns to step 52.

On the other hand, the Fast Retransmit method is also referred to as loss detection due to duplicate Ack. First, when an Ack packet is received in Step 52 (Step 52: Ack packet reception), if it is an Ack packet for the same data packet, that is, a duplicate Ack (Step 61: YES), a predetermined number (set in advance) It is a constant to be set, usually three). It is determined whether or not the received data is duplicated (step 62).

If a predetermined number of duplicate packets are received (step 62: YES), it is assumed that the packet next to the reception confirmation sequence number is lost, and the process enters the congestion avoidance phase. In this case, the lost packet is retransmitted (step 63), and the new cwnd is changed to min (old cwnd, ad).
wnd) / 2 and ssthresh = new c
As wnd (step 64), the process returns to step 52.

When the receiving end node detects a missing DT packet, it returns an Ack having the same acknowledgment sequence number for each subsequent received DT packet until the missing DT packet is received. The processing flow of FIG. 4 described above is described in, for example, the following references:
WRStevens, TCP Illustrated Vol.1, Chapter 21,
Addison Wesley, 1994 and the like.

[0012]

The mobile communication (wireless system) is expected to develop more and more in the future, but the line quality of wireless lines is worse than that of wired systems. Therefore, according to the above-described conventional TCP flow control method, cwnd is reduced even when a packet is lost / discarded due to line quality, so that there is a problem that throughput is inappropriately reduced.

Here, if a packet is lost / discarded due to a line error, not due to congestion, it is only necessary to retransmit the packet, and it is not necessary to reduce cwnd. Therefore, it is important to perform flow control according to the cause of packet loss / discard. The present invention has been made to solve such a problem, and a flow control method capable of avoiding unnecessary cwnd reduction and obtaining an appropriate throughput by performing flow control according to the cause of packet loss / discard. It is intended to provide.

[0014]

SUMMARY OF THE INVENTION In order to achieve the above object, a flow control method according to the present invention provides a transmitting end node with a congestion generated in a communication network connecting a transmitting end node and a receiving end node. Congestion confirmation means for confirming, and duplicate Ack reception for continuously receiving a predetermined number of Ack packets for the same data packet among Ack packets for acknowledgment returned from the receiving end node in response to the data packet is confirmed. And a duplicate Ack detecting means for detecting a packet loss event is provided. If the transmission end node receives the Ack packet and the congestion checking means confirms the occurrence of congestion, the packet is lost due to congestion in the communication network. It is determined that the congestion window value is reduced, and when the Ack packet is received,
The congestion confirmation means does not confirm the occurrence of congestion,
When the duplicate Ack reception is confirmed by the k detection means,
It is determined that the packet is lost due to a line error occurring in the communication network, and the lost data packet is retransmitted without reducing the congestion window value.
Therefore, when congestion occurring in the communication network is not confirmed, even when a duplicate Ack is detected and a packet loss event occurs, the data packet transmission amount is not reduced, and congestion occurring in the communication network is confirmed. Only in this case, the data packet transmission amount is reduced.

[0015]

Next, the present invention will be described with reference to the drawings. FIG. 1 is a flowchart illustrating a flow control method according to an embodiment of the present invention, and FIG. 2 is a block diagram illustrating an example of a data communication system to which the present invention is applied. In FIG. 2, a transmission end node 1 and a reception end node 2 are connected via a communication network 5 including a plurality of relay nodes 3 and 4 (for example, routers), and set a predetermined connection to the communication network 5. Then, data communication is performed by a packet via the connection.

In the transmitting end node 1 and the receiving end node 2, the application processing section 11 is a processing section for executing a higher-level application. The layer 4 processing unit 12 is a processing unit for data transmission / reception by packets based on connection setting release, flow control, and the like. The lower layer processing unit 13 is a layer 1 process for providing electrical consistency, a frame assembly / This is a processing unit that performs layer 2 processing for performing decomposition and the like and layer 3 processing for performing routing and the like.

The flow control of the protocol layer 4 executed by the layer 4 processing unit 12 is a window system as described above, and a reception notification window (adwnd) and a congestion window (cwnd) are used as parameters of the protocol layer 4 (TCP). ), Slow start threshold (ssth
refresh), the number of duplicate Ack by the Fast Retransmit method (default = 3).

In the following, as a method of confirming the occurrence of congestion in the communication network 5 by the transmission end node,
ECN (Explicit Connec- tion) that explicitly notifies the end node that each of the relay nodes 3 and 4 configuring the
(gestion Notification) control will be described. However, the method of confirming the occurrence of congestion in the communication network 5 is not limited to this, and other methods, for example, the transmission end node 1 using a predetermined control packet. May be notified.

In the ECN control, the relay node turns on the ECN indication in the packet header and notifies the end node positively when the relay node becomes busy before the actual packet discarding (that is, when the used amount of the transmission buffer exceeds a predetermined value). Things. That is, as shown in FIG. 3, when the relay node 3 relays a DT packet and the transmission buffer usage exceeds a predetermined threshold, the ECN display of the packet header is set to on and the DT packet is relayed. .

The receiving end node 2 copies the value of the ECN indication into the Ack packet header and returns it. The transmission end node 1 can determine that the relay node 3 is in the state of a radiant rod because the ECN display of the Ack packet is on.

For an overview of the ECN described above, for example, see Reference: S.floyd, "TCP and Explicit Congestio
n Notification ", ACM Compuler Communication Review,
Vol. 24, No. 5, 0ct. 1994, pp. 10-23. It should be noted that the current state of this ECN technology is that E
Setting of CN display and ECN display at sending end node 1
When receiving n packets, the amount of transmission is reduced in the same manner as when packet loss is detected. This is at a conceptual level, and this is used for the determination to avoid the reduction of cwnd as in the present invention. Not disclosed.

Since the ECN function is set by the relay nodes 3 and 4, the indication is provided in the header of the layer 3 (IP). At the end node, the information is passed from the IP processing module to the TCP processing module. . First, the transmission end node sets ssthresh = adwn when establishing a connection with the reception end node.
d, cwnd = 1, hi_seq = initial values of transmission sequence numbers are initialized (step 11).

The meaning of hi_scq is as follows. If the Ack packet that has been transmitted when the cwnd is reduced but the response is returned to the DT packet for which the reception has not been confirmed is ECN = on, c
Reducing wnd results in unnecessary over-reduction. Therefore, in order to prevent this, in the present invention, cwnd
Next transmission sequence number snd_ne when reduction of
xt is stored in hI_seq.

Then, a slow start phase is entered,
min (cwnd, adwnd), in this case c
wnd, and one data packet (hereinafter, DT)
Then, it waits for an acknowledgment packet (hereinafter, referred to as an Ack packet) returned from the receiving end node in response to receiving the DT packet (step 12).

The DT packet transmitted from the transmission end node 1 is provided with a series of transmission sequence numbers. On the other hand, the receiving end node 2 reads the transmission sequence number of the received DT packet to confirm the omission of the DT packet, and among the DT packets whose delivery has been confirmed by the receiving side at that time, the largest transmission sequence The number is stored as an acknowledgment sequence number in the Ack packet and returned.

As a result, when a missing DT packet occurs, an Ack packet corresponding to the subsequent DT packet is added to the last DACK that is continuously acknowledged by the receiving end node 2 until the DT packet is acknowledged.
The transmission sequence number of the T packet is stored repeatedly. Therefore, the transmission end node 1
If the same acknowledgment sequence number is confirmed a predetermined number of times as a packet acknowledgment sequence number,
That is, when the packet is a duplicate Ack, it can be determined that the packet next to the reception confirmation sequence number is lost (duplicate Ack confirmation means).

In step 12, if an Ack packet is received within a predetermined time after the transmission of the DT packet (step 12: Ack packet reception), the ECN display is checked (step 15 / congestion confirmation means). Where EC
If N = off (step 15: NO), duplicate Ack
(Step 21 / duplicate Ack confirmation means)
If it is a duplicate Ack (step 21: YES) and the number is 3 (step 22: YES /
Duplicate Ack confirmation means), resends the packet assuming that a packet loss has occurred (step 23), and returns to step 12.

As a result, even when the duplicate Ack is confirmed and packet loss is considered to have occurred, if the cause of congestion in the communication network 5 is not confirmed as a cause, it is determined that it is merely a line error and cwnd Is not reduced, unnecessary cwnd reduction can be avoided, and appropriate throughput can be obtained. Step 2
In step 1, although the received Ack packet is a duplicate Ack (step 21: YES), if the number has not reached 3 (step 22: NO), the process returns to step 12 and waits for the next Ack.

In step 21, the reception Ack
If the packet is not a duplicate Ack (step 21: N
O), the magnitude relationship between cwnd and the threshold ssthresh is compared (step 24), and if cwnd <ssthresh (step 24: YES), cwnd is incremented by 1 (step 25), and if cwnd ≧ ssthresh (step 24: NO). , Cwnd by 1 / cwnd (step 26). The process returns to step 12.

On the other hand, in step 15, the reception Ack
If the ECN of the packet is on (step 15: YE
S), it is determined that congestion has occurred in any of the relay nodes and the buffer is overloaded. In this case, first, the received A
The acknowledgment sequence number Ackscq of the ck packet is checked (step 31).

Here, if Ackseq> hi_seq (step 31: YES), it is determined that the DT packet is a lost event of a non-acknowledged DT packet newly transmitted after the time when cwnd was reduced immediately before, The next transmission sequence number snd_next is stored in hi_seq (step 32), and the flow control parameter is set to s
sthresh = min (cwnd, adwnd) /
2. Change to cwnd = 1 to reduce the amount of transmission (step 33).

If Ackseq ≦ hi_seq (step 31: NO), it is determined that the event is a loss event of the same DT packet as when cwnd was reduced immediately before.
cwnd is not reduced. As a result, unnecessary reduction of cwnd can be avoided, network resources can be effectively used, and appropriate throughput can be maintained.

Subsequently, if the received Ack packet is a duplicate Ack
(Step 34 / duplicate Ack confirmation means)
If it is a duplicate Ack (step 34: YES) and the number is 3 (step 35: YES / duplicate A)
ck confirming means), resends the lost DT packet assuming that packet loss has occurred (step 36). Return to step 12.

In step 34, the reception Ack
If the packet is not a duplicate Ack (step 34: N
O), and if the number of duplicate Ack has not reached 3 in step 35 (step 35: NO), the process returns to step 12 and waits for the next Ack.

In step 12, when the Ack packet is not received within a predetermined period after the transmission of the DT packet and a delivery confirmation timeout occurs (step 1).
2: Check transmission timeout), determine that a packet loss has occurred, retransmit the lost packet (step 13), and set the flow control parameter to ssthres
h = min (cwnd, adwnd) / 2, cwnd =
1 to reduce the transmission amount (step 14), and return to step 12.

Thus, when it is determined that packet loss has occurred due to the confirmation of the duplicate Ack, if the occurrence of congestion in the communication network 5 has been confirmed as the cause,
Not just a line error but DT due to congestion in the communication network 5
It is determined that the packet is lost, and cwnd is reduced.
The congestion state of the communication network 5 is controlled to be reduced.

[0037]

As described above, according to the present invention, the transmission amount of data packets is controlled according to the cause of packet loss (congestion or line error), so that network resources can be used effectively. Appropriate throughput is obtained.

[Brief description of the drawings]

FIG. 1 is a flowchart illustrating a flow control method according to an embodiment of the present invention.

FIG. 2 is a block diagram showing an example of a data communication system to which the flow control method of the present invention is applied.

FIG. 3 is an explanatory diagram showing an example of ECN control.

FIG. 4 is a flowchart showing a conventional flow control method.

[Explanation of symbols]

1: transmission end node, 2: reception end node, 3, 4
.. Relay node, 5 communication network, 11 application processing unit, 12 layer 4 processing unit, 13 lower layer processing unit.

Continued on the front page (72) Inventor Eiichi Kondo 5-2-6 Toranomon, Minato-ku, Tokyo F-term in the ultra-high-speed network and computer technology laboratory (reference) 5K030 GA03 HA08 HB13 HC09 JL01 LA02 LC03 LC11 MB02 5K034 AA05 EE03 EE11 HH01 HH02 HH06 KK28 MM16 NN26

Claims (3)

[Claims] 1. A reception notification window (adwn) indicating the receiving capability of a receiving end node as a layer 4 protocol between transmitting and receiving end nodes connected via a communication network.
d) and when the data packet is transmitted from the transmitting end node to the receiving end node within a range of the number of packets indicated by a smaller value of the congestion window value (cwnd) which estimates the transfer capability of the network. A flow control method, comprising: a transmission end node; a congestion confirmation means for confirming congestion occurring in a communication network connecting the transmission end node and the reception end node; and an acknowledgment returned from the reception end node in response to a data packet. And a duplicate Ack detection means for detecting a packet loss event when it is confirmed that duplicate Ack packets for the same data packet are received consecutively for a predetermined number of Ack packets. When congestion occurrence means is confirmed by the congestion confirmation means at the point when It is determined that the packet has been lost due to congestion in the communication network, the congestion window value is reduced, and when the Ack packet is received, the occurrence of congestion is not confirmed by the congestion confirmation means, but the duplicate Ack is received by the duplicate Ack detection means. Is confirmed, it is determined that the packet is lost due to a line error occurring in the communication network, and the lost data packet is retransmitted without reducing the congestion window value. 2. The flow control method according to claim 1, wherein, among the relay nodes forming the communication network, the relay nodes that transfer the data packet from the transmitting end node to the receiving end node each transmit the data packet. When the outgoing route to be congested enters the congestion state, the congestion display information indicating the occurrence of congestion is set in the data packet transmitted to the outgoing route, and the receiving end node displays the congestion display information of the received data packet as the Acknowledgment of the data packet is set and returned, and the congestion confirmation means of the transmission end node confirms the occurrence of congestion in the communication network when the display information set in the Ack packet indicates the occurrence of congestion. Flow control method. 3. The flow control method according to claim 1, wherein, when the congestion confirming means confirms the occurrence of congestion at the time of receiving the Ack packet, the transmitting end node transmits the congestion window value at the end. Judge whether the received packet is for a non-acknowledged data packet transmitted after the point in time when the
A flow control method characterized in that a congestion window value is reduced only in the case of an Ack packet corresponding to a data packet whose acknowledgment has been transmitted after the reduction point.