JPH11187031A - ATM switch - Google Patents

Abstract

(57) [Summary] [PROBLEMS] To avoid the same connection continuous discard, retransmission segment discard, segment discard immediately after communication, discard concentration, etc. caused by indiscriminate cell discard of UBR service of ATM exchange, and to improve throughput and usefulness. Realize an improved ATM switch. Instead of conventional indiscriminate cell discarding at the time of congestion, selective discarding based on window size, stepwise discarding promotion according to congestion, retransmission cell discarding, and avoidance of discarding discarded connections are performed. For this purpose, an inflow data unit flow rate measuring means, a discard class assigning means according to the flow rate, and a storage means of a predetermined discard class are provided, and the cells are selectively discarded from cells having a discard class larger than the predetermined discard class.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an ATM exchange, and a non-guaranteed service (UBR service: Unsp.
The present invention relates to an ATM switch having excellent congestion control of cells transmitted by an enhanced bit-rate service.

[0002]

2. Description of the Related Art In an ATM network in which information communication services such as voice, video, data, etc. are handled in an integrated manner, it is important to accommodate each traffic in the network with high efficiency for efficient operation as a whole. 2. Description of the Related Art In recent years, the following two types of services have been introduced as services for data systems in which the amount of communication is remarkably increasing.

[0003] One of them is that the network notifies the terminal of the resource usage status every moment to avoid discarding the data unit caused by the network congestion due to the traffic exceeding the limit determined by the network resource status. A adaptive service (ABR service: Available) that performs rate control, regulates outgoing traffic autonomously, and avoids unnecessary congestion to achieve comprehensive efficient operation.
e Bit-Rate service). "

[0004] Second, the terminal itself does not perform rate control according to the notification, but sends out traffic to the network at a predetermined speed of the terminal. If discarding due to network congestion occurs, the terminal detects loss and performs end-to-end. "Non-guaranteed service (UBR service: Unspecify)
ed Bit-Rate service).

[0005] These two types of services have the following drawbacks, respectively, and are obstacles to the spread of the services. That is, ABR
For the service, a buffer for overflow data at the time of congestion is indispensable for both the terminal and the network, and the number of facilities in the entire network becomes enormous. On the other hand, the UBR service is simpler and cheaper than the ABR service in network and terminal equipment, but the communication quality is not guaranteed, so the reliability is a hindrance factor, and the throughput is reduced due to excessive regulation of retransmission control and flow control. I do.

[0006] In consideration of the advantages and disadvantages, the services in these two modes are considered to be introduced in accordance with the trend of service demand and the evolution of the network generation. It is desirable to improve efficiency and throughput.

[0007] In the UBR service, the retransmission control and the flow control are performed in the terminal and the ATM exchange depending on a protocol higher than the ATM. Hereinafter, TCP / IP, which is generally used in a data service accommodated in the UBR service of the ATM network for this purpose, will be described.
(Transmission Control Pro
The retransmission control and the window flow control method of the Internet Protocol (tocol / Internet Protocol) will be described.

First, FIG. 2 is an explanatory diagram of a TCP / IP protocol stack. Data of the application layer shown in the upper part of FIG. 2 is exchanged between each application and a TCP unit which is located in the TCP layer and controls transmission control, and is transmitted to and received from the IP layer which is located in the IP layer and controls communication control.
Section divided between the TCP section and the TCP section for communication control.
It is exchanged in units called segments.

Further, the TCP segment is exchanged as an IP packet with the next ATM / AAL5 unit via the IP unit. Next, the IP packet is AAL5 / A
At the TM section, control information such as a trailer is added, and the ATM
It is converted into a data unit of the AAL5 (connection-type data communication service class) of the fifth category in the service category 5.

Next, the AAL5 data unit is 4
The cell is divided into 8 octets, and a cell header of 5 octets is added for each division, so that the ATM
Transferred in the network.

On the other hand, when network congestion occurs, the AAL5
The data unit is discarded. However, even if discarding is performed on an ATM cell basis, the lost data unit obtained by reorganizing the ATM cell group is received and detected as an error frame. Will be.

Next, the ATM cell is composed of 53 octets of frame data shown in FIG. The header part includes a GFC (4 bits) for flow control of the terminal, a virtual path identifier VPI (8 bits), a virtual channel identifier VC
I (16 bits), PTI (3 bits) indicating the cell format, cell discard priority information CLP (1 bit), and header error control information HEC (8 bits).

When a call is generated by a connection request from a terminal or the like, a virtual path identifier VPI and a virtual channel identifier VCI corresponding to the call are given, a connection (logical communication path) is set, and the connection information (logical communication path information) is set. ) Is held until the call is released.

As described above, the AAL5 data unit is composed of a plurality of continuous ATM cells. A specific pattern indicating discontinuity of the cell is displayed in the PTI portion of the last cell header, and the last cell is identified by the header monitoring. It can be identified.

In TCP / IP, at the time of communication control, data of an application layer is divided into units for controlling a communication procedure called the above-mentioned TCP segment, and AT
Data transmission / reception with other ATM exchanges and destination terminals via the M network is performed, and retransmission control for recovering lost data and window flow control for maintaining efficiency are performed in this unit.

First, retransmission control is performed as follows. That is, if there is no acknowledgment (ACK) of the destination within a predetermined time after the transmission of the segment, it is regarded that the segment is lost and the transmission of the segment is repeated. The ACK monitoring time (RTO) at this time is increased to about twice the monitoring time at the time of immediately preceding transmission, which is generally considered appropriate for recovery from accidental abnormalities, to achieve stable recovery. Therefore, when retransmission is repeated, the ACK monitoring time (RTO) increases, and the recovery standby time rapidly increases. FIG. 4 is an explanatory diagram of the back-off at the time of retransmission showing this pattern, and shows a state where the ACK monitoring time (RTO) is set to double at the time of two retransmissions, that is, at the time of re-retransmission.

Next, the window flow control is performed as follows. In TCP / IP, usually, transmission is performed well, and a plurality of segments are collectively transmitted without waiting for the acknowledgment (ACK) to increase efficiency and increase throughput. This collective amount of allowance is called a congestion window (CWND).

The transmission source transmits a plurality of segments collectively within a congestion window (CWND) set at that time. Also, the congestion window (CWND)
Is reduced at the start of communication or at the time of retransmission, and is increased according to a predetermined rule during normal transmission and reception, so that the arrival at the maximum congestion window size is accelerated.

This control is performed, for example, as follows. The initial value of the congestion window (CWND) is set to the maximum size (MSS) of the TCP segment.
ND), apart from the slow start threshold (ssthres
A threshold for acceleration determination called h) is defined, and its initial value is also set as the maximum size (MSS) of the TCP segment.

Next, every time transmission and reception is normally completed, the congestion window (CWND) is set to a slow start threshold (ssthres).
h) While below, the congestion window (CWND) is set to T
CP segment maximum size (MSS) is increased immediately, and the size is set to a slow start threshold (ssthre).
After sh), the increase unit is reduced by narrowing the increment unit to the ratio of the maximum segment size (MSS) to the current congestion window (CWND).

If congestion occurs due to an increase in burstiness during transmission / reception, the congestion window (CWND) at the start of retransmission is not the congestion window size that has grown up to that point, but the TCP segment that is the first set value. As the maximum size (MSS) and the slow start threshold (ssthresh), a half value of the congestion window (CWND) at that time is set.

With the above control, the congestion window (CWND) increases rapidly up to half the size of the point at which congestion occurs, and thereafter the growth decreases sharply. That is, a congestion window of CWND ≦ ssthresh (acceleration period) CWND ′ CWND ′ = CWND + MSS A congestion window of CWND> ssthresh (suppression period) CWND
/ 2 CWND '= MSS On the other hand, the initial value and the increase value of the congestion window size are closely related to the line efficiency, and the larger the initial value, the larger the amount of data transmitted and received once from the beginning and the larger the increase value. About
Since the maximum window size is reached quickly, the transmission efficiency is increased and the throughput is increased. However, conversely, the burstiness of the data is increased, and the concentration and discreteness become extreme, and congestion due to local concentration tends to occur.

Conversely, the smaller the initial value or increase value, the more
Since the amount of data transmitted / received at one time is small or the window size grows slowly, the transmission efficiency decreases and the throughput decreases, but the burstiness is weak and congestion does not easily occur.

In addition, by the above-mentioned size increase acceleration control,
The segment with the larger congestion window size,
Since the slow start threshold is large, the increase in the congestion window size is quickly recovered by retransmission when congestion occurs, and the influence of discarding is reduced.

Therefore, when discarding a frame during congestion, discarding a frame of a connection under communication with a large congestion window size results in a faster recovery rise and less decrease in throughput.

A conceptual description of this TCP / IP retransmission control and window flow control is shown in the TCP window flow control explanatory diagram of FIG. In FIG. 5, each time segment transmission is performed, the congestion window increases, the transmission unit amount increases by the maximum segment size, and after several segment transmissions, a timeout of the ACK monitoring time (RTO) occurs at the X mark. In this example, the data is retransmitted and the recovery is quicker.

[0027]

As described above, when a conventional TCP / IP retransmission control and window flow control method are used as a UBR service congestion control method in an ATM network, There were the following problems.

First, when data units are discarded due to congestion immediately after the start of communication or at the time of retransmission, the window size of the subsequent retransmission is small, so that the efficiency is reduced, and the start-up and recovery of communication are delayed, and window reduction is performed. There was a problem of reduced efficiency at the time.

Further, there is a problem of excessive degeneration of throughput related to window control, such as a repetition of the trigger congestion, and a cumulative increase of ACK monitoring time, which significantly increases standby time.

In spite of the potential problem, data unit discarding or ATM cell discarding is performed indiscriminately regardless of the size of the window as discarding processing when network congestion occurs. Therefore, a segment immediately after the start of communication with a small window size or a retransmission segment is discarded, and the excessive degeneration is induced to cause a remarkable recovery delay or a rise delay, a recovery delay of a specific connection due to continuous discarding of segments belonging to the same connection, and May lead to centralized disposal due to continued congestion,
There is a throughput degradation problem based on indiscriminate cell discard.

On the other hand, when normal transmission / reception continues, the congestion window size monotonically increases, so that the burstiness of data increases with the increase in the window size.
As a result, intermittent congestion is caused, and consequently, there is a problem of congestion due to window enlargement, which causes the excessive degeneration and induces continuous discarding.

According to the present invention, when performing retransmission and flow control by the TCP / IP protocol in the UBR service in the ATM network, these problems caused by the conventional indiscriminate cell discarding method can be prevented. It is an object of the present invention to provide an ATM switch which can effectively avoid these problems by adopting a method and improve the throughput and availability of the UBR service.

[0033]

The present invention employs the following means to solve the above-mentioned problems. First, the ATM switch of the present invention always measures the flow rate of the AAL5 data unit of each incoming connection to estimate the congestion window size in order to solve the throughput degeneration caused by indiscriminate cell discarding. Instead of discarding, a segment having a size larger than a predetermined congestion window size is identified and the cell is discarded.

For this reason, the interval of the data unit as a flow rate element is measured for each connection, and the priority information corresponding to the interval becomes smaller as the interval becomes smaller and becomes larger as the interval becomes larger. Is provided as a discard class.

The smaller the discard class, the larger the flow rate, that is, the larger the congestion window size, and the smaller the effect of cell discard. Therefore, at the time of congestion, priority is given to other connections to which a larger discard class is assigned. Indicates that it should be discarded.

On the other hand, there is provided means for setting a threshold value of a predetermined discarding class to be discarded when congestion is detected as a selected discarding level, and increasing or decreasing the threshold value. The selective discard level indicates a threshold at which the incoming cells of the connection having the discard class equal to or lower than the level should be discarded at the time of congestion. When the maximum value is set, all cells of all connections are discarded at the time of congestion, and the minimum value is set. Is set, only the connection cell having the lowest minimum discard class is discarded.

The level is sequentially updated within the maximum and minimum ranges, triggered by congestion. Further, when a congestion is detected when a segment arrives, a selective discarding means is provided for discriminating a segment of the connection to which the segment belongs and whose discard class is equal to or lower than the selective discard level and discarding the cell.

By means of the above, at the time of congestion, at a given predetermined selective discard level, which has a smaller discard class, which indicates that the data unit flow rate is larger, the congestion window size is larger. ,
Segments that are less affected by cell discard are selectively discarded.

Next, according to the present invention, in order to cope with the congestion concentration due to the enlargement of the congestion window size, when updating the selective discard level at the time of the congestion, the discard promoting side is linked with the level according to the degree of congestion. Alternatively, a step-by-step updating means of the selective discard level for increasing or decreasing the discard suppression side is provided.

When the congestion state continues for a predetermined period of time or more, the updating means detects that the congestion is in a concentrated state, and performs the selection so as to discard even the discard class having the next higher discard priority. Raise the disposal level to the disposal promotion side. While the continuation state continues, the promotion process is repeated up to a predetermined maximum value.

On the other hand, when the congested state is resolved, the selective discarding level is reduced, and the discarding is repeated. As a result, in response to intermittent congestion due to an increase in burstiness based on the enlargement of the congestion window size, stable discarding is performed, and the above-described excessive response is suppressed.

Further, according to the present invention, in order to avoid continuous discard in the same connection, when the cell is discarded during the congestion,
Means is provided for raising the discard class of the connection once discarded and lowering the discard priority. As a result, when the next data unit of the connection arrives, the discard class and the discard condition of the collation process of the selected discard level do not match, thereby avoiding continuous discard in the connection.

On the other hand, in the present invention, in order to avoid discarding a retransmission segment, a retransmission segment identification unit and a unit for changing a discard class of a connection to which the segment belongs are provided.

When the flow rate of the inflowing data unit is a small amount below a certain level, for example, the monitoring time RT of the response ACK
If the arrival of a data unit at an interval exceeding O is detected, there is no transfer data of the connection or A
Since the retransmission is performed due to the timeout of the CK reception and the transfer interval of the segment is increased, the retransmission segment is identified by this.

The discard class of the connection at this time is
By giving a specific large value, the discard condition at the time of matching processing between the discard class and the selected discard level is not satisfied, thereby avoiding cell discard of the retransmission segment at the time of congestion. The singular value is held for a certain period of time after the retransmission recovery until the record of the discard class is updated by a normal ATM cell.

[0046]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention is shown in FIG. 1, and the configuration and operation of each section will be described below. In addition, FIG.
FIG. 2 mainly shows a transfer unit of a data unit of the ATM exchange of the present invention, and other parts generally required when configuring the ATM exchange, that is, a central control unit, a switch, and a line connection unit. Are not shown.

They are to be provided separately based on commonly known techniques in the art. In FIG. 1, 11
Is a data unit flow monitoring unit, 12 is a discard class table, 13 is a data unit discard unit, 14 is a cell buffer unit, 15 is a queue length monitoring unit, 16 is a congestion continuation monitoring timer unit, and 17 is a discard start class recording / instruction unit. It is.

First, the data unit flow rate monitoring unit 11
Receives an ATM cell flowing in from the outside of its own switch such as a line or another ATM switch, and receives the PT of the header of the ATM cell.
From the non-continuous display of the I section, a data unit break is identified.

On the other hand, the flow monitoring unit 11 responds to the connection indicated by the VPI unit and the VCI unit of the ATM cell header.
The time interval of the data unit is measured, a discard class is assigned, and the incoming ATM cell is sent to the data unit discarding unit 13 at the next stage. Further, a data unit having a long interval longer than a predetermined time is detected, and a retransmission frame is identified.

When congestion is detected when the ATM cell arrives, the data unit discarding unit 13 refers to the discard class of the connection to which the cell belongs stored in the discard class table 12 and records the discard start class record. / Discards the data unit to which the incoming ATM cell belongs by discriminating that it is equal to or lower than the selected discard level notified from the instruction unit 17. Further, the data unit discarding unit 13 includes means for changing the discard class in the discard class table 12 by one stage every time discarding is performed.

The cell buffer unit 14 is a buffer for temporarily storing the incoming ATM cells and transferring the ATM cells to the outside of the next line or another ATM switch. The queue length monitoring unit 15 stores the AT stored in the cell buffer unit 14.
The number of M cells is monitored, and when it reaches a predetermined level, it is detected as occurrence of congestion, and the result is notified to the congestion continuation monitoring timer unit 16.

The congestion continuation monitoring timer unit 16 monitors the continuation of the congestion occurrence by the queue length monitoring unit 15. If it has continued for a predetermined time or longer, the discard start class recording / instruction unit 17 is notified to promote discarding at the selected discarding level, and to detect discarding at the discarding level when detecting congestion resolution.

The discard start class recording / instruction unit 17 stores a predetermined selected discard level and instructs the data unit discard unit 13 of the identification discard level. Further, the update of the selective discard level based on the notification from the congestion continuation monitoring timer unit 16 is performed.

The operation of the entire apparatus of the present invention shown in the above embodiment is performed as follows. First, as an operation for avoiding throughput degradation due to selective discarding, the ATM cells flowing from the network to the local exchange are connected to the data unit flow monitor 1.
1 and the queue length monitoring unit 15 of the ATM cell detects congestion in the data unit flow monitoring unit 11 and the discard class table 12 after the above-described discard class assignment processing for each connection. If not, the data is sent to the cell buffer 14 via the data unit discarding unit 13 and accumulated.

On the other hand, when the queue length monitoring unit 15 detects congestion, the predetermined selection discard level stored in the discard start class recording / instruction unit 17 and the corresponding connection The ATM cell is discarded by the data unit discarding unit 13 when the discard class is compared with the discard class.

Next, the update processing operation of the selective discard level is as follows.
As described above, this is achieved by sequentially changing the selective discard level stored in the start class recording / instruction unit 17 according to the congestion continuation state.

On the other hand, the operation for avoiding continuous discarding of the same connection is performed by means of raising the discarding class of the connection by one level and lowering the discarding priority at the time of cell discarding by the data unit discarding unit 13 at the time of congestion, by the following means. This is achieved by setting the discard condition when the AAL5 data unit of the connection arrives to be inconsistent.

The operation for avoiding retransmission segment discard is performed as described above. In other words, when the data unit flow monitoring unit 11 detects a frame at an interval equal to or longer than a predetermined interval, the discard class of the corresponding connection is specifically increased.

The singular value keeps the singular value larger than the selected discard level for a certain period of time until an ATM cell having a normal data interval arrives at the corresponding connection and the record of the discard class is updated. Since the discard condition does not match in the collation process between the selected discard level and the discard class, the retransmission segment is prevented from being discarded during the congestion for the predetermined time.

Next, specific implementation of the present invention and details of each unit will be described. First, the data unit flow monitor 1
Regarding the measurement of the data unit time interval and the assignment of the discard class of 1, the flow monitoring counter operation explanatory diagram of FIG.
A description will be given focusing on the connection.

The above-mentioned discard class table 12 is a table for storing control information required for the measurement. As shown in FIG. 7, for each connection, the flow rate measurement counter CM, the judgment processing execution time t0, the discard class Nc section.

On the other hand, the flow rate measuring counter section CM is a counter that repeats the subtraction every time a data unit is detected by the non-continuous display of the PTI section of the header of the incoming ATM cell. An initial value is set for each tnow (t0) point at which the value becomes 0 by subtraction during measurement.

Next, when the value of the CM becomes 0 at the time tnow (t0) in FIG. 7 during the repetition of the subtraction every time the data unit passes, the previous judgment time recorded in the t0 section of the discard class table 12 With reference to the current time,
The connection discard class Nc is determined and assigned.
That is, the discard class Nc is as follows: Nc = min ｛Int [(tnow−t0) / Δt],
Nmax｝ Here, min ｛a, b｝: Take the smaller of a and b. Int []: Take an integer value (round down) tnow: Processing time t0: Previous determination processing time Δt: Unit time for discarding class assignment Nmax: The predetermined maximum value of Nc This equation is set to 1 every time Δt elapses according to the elapsed time (tnow−t0) from the time t0 when the previous determination was performed.
This indicates that the priority class Nc is added, and the value of the Int [] is the time distance from the previous determination time t0 to the processing time tnow.

Accordingly, in accordance with the data unit arrival interval, the smaller the interval, the smaller the value, and the larger the interval, the larger the Nc. FIG. 7 shows an example in which classes from class 0 at the start time to J at the current time are set.

A general semiconductor memory device or the like is used for the discard class table. In the case of the method based on the above Nc determination formula, the discard class is assigned, because it is necessary to grasp the processing time tnow. Therefore, although not specifically shown, a clock mechanism is provided in the data unit flow rate monitoring unit 11.

As another method of assigning a discard class, a method of providing a class counter and using the value of the class counter when the CM value in the discard class table becomes 0 can be exemplified.

The cell buffer unit uses a first-in first-out memory, but there is another configuration method using a random access type memory element in combination with a pointer register.
Hereinafter, the operation until the incoming ATM cell is transmitted to the exchange destination will be described in order.

First, the data unit flow monitor 1 shown in FIG.
When the ATM cell arrives at No. 1, it is confirmed in the flow chart of the flow monitoring section of FIG. 8 that the incoming cell is the last cell constituting the data unit (8-1). If the CM part of I is subtracted and the result is not negative, the cell buffer write processing is performed after the CM part of the discard class table is updated (8).
-3, 8-6). If the result of the CM section subtraction in the discard class table is negative, the CM section initial value and the addition time (T
0) and the discard class (Nc) according to the above-mentioned decision formula are set to prepare for the arrival of the next ATM cell (8-4, 8-5).

FIG. 9 is an explanatory diagram of a discarding unit processing flow of the data unit discarding unit 13. This processing flow includes a process of discarding a series of succeeding cells constituting the data unit after the data unit head cell arrives at the time of congestion.

First, when an ATM cell arrives, a congestion state is determined from information of the congestion continuation monitoring timer unit 16 (9-).
1), a process of a flag CS for storing and displaying the connection discarding or transfer status is performed (9-2, 9-3).

Next, by referring to the flag D, it is determined whether or not cell discarding in data units of the connection is ongoing (9-4). If discarding continues, discarding is continued up to the last cell of the data unit (9-15), and when the last cell arrives, the discard status flag D is cleared (9-16). If discarding is not being continued, it is determined whether the arriving cell is the head of the data unit (TOP flag, 9-5). If not, the arriving cell is determined to be the head of the next arriving cell. Is held (TOP flag, 9-11 / 9-12 / 9-13), and the arriving cell is stored in the cell buffer (9-14). If the arriving cell is the first cell, it is in the congestion state at this time, so it is determined whether the discard class of the connection in the discard class table is equal to or lower than the selected discard level (9-7). The discard class is changed by one stage and discarding is performed (9-9 to 10), or cell buffer accumulation processing is performed (9-11 to 14).

Here, the means for avoiding the continuous discard of the once discarded connection is implemented by the change of the discard class one stage and the discard execution processing (9-9 to 10). Next, the execution of the above-described update process of the selective discard level will be described.

FIG. 6 shows a discard start class recording / instruction unit 17.
FIG. 9 is an explanatory diagram of a selective discard level processing flow of FIG. The control unit determines the state of congestion for a predetermined period of time from the congestion continuation monitoring timer unit 16 or a notification of congestion elimination (6-1), and according to the result, the selection discard level is sequentially determined for a predetermined number of times. (6-4, 6-7), update to the disposal promotion or disposal suppression side (6-8, 6-9).

As another embodiment of the selective discard level updating process, a mode in which the congestion continuation monitoring timer unit 16 is not provided and the process is based on the selective discard level process flow can be exemplified. In that case, the processing based on the flow is configured to be started by the arrival of the cell of the connection under communication.

Next, a description will be given of an embodiment of the present invention in which the flow rate of the inflowing AAL5 data unit is small, that is, a retransmission frame or a frame immediately after the start of communication is identified and these are discarded during congestion.

As means for retransmission frame identification, the data unit flow rate monitoring section 11 is provided with frame interval monitoring means (not shown) for monitoring the arrival interval of data units in addition to the data unit interval measuring means. As described above, the data unit interval is usually assumed to be the maximum RT
Detect the arrival of a data unit greater than O. When the arrival of the data unit at the long interval is detected, a singular value larger than the maximum value of the selected discard level is given as a discard class when the cell of the corresponding connection arrives.

As described above, the singular value is a value larger than the selected discard level until an ATM cell having a normal data interval of the corresponding connection arrives next and the record of the discard class is updated. Therefore, selective discard at the time of congestion can be avoided, and desired priority transfer can be performed.

One operation example of the configuration according to the present invention is shown in the operation example explanatory diagram of FIG. FIG. 10 focuses on the two connections (A, B), the congestion window (CWND) and the slow start threshold (ssthresres) at each transmitting terminal.
h) shows a transition example of the CM and the discard class Nc in the discard class table of the present invention.

When congestion occurs at time tc shown in the figure, connection A and connection B are discarded in classes 2 and 7, respectively, so connection A is preferentially discarded.

[0080]

According to the above-mentioned means, that is, in the ATM exchange which performs the UBR service, a discard class is assigned according to the data unit flow rate, identification discarding at the time of congestion by comparing a predetermined discard level with the discard class, discarded connection. Of the UBR service using the flow control method based on the conventional TCP / IP control method,
It is possible to avoid the throughput degeneration caused by the discarding process at the time of congestion, such as the continuous discarding of the same connection and the continuation of the congestion state, and to achieve an effect of realizing an ATM switch having a faster throughput recovery.

Further, the updating means of the predetermined discard level according to the congestion continuation time enables stable discard processing even at the time of congestion concentration due to the enlargement of the window size, and has an effect of preventing the discarding of connections between connections.

Further, the retransmission frame identification and the unique discard class assigning means have an effect of greatly reducing the throughput, immediately after the start of communication or of reducing the frequency of discarding the retransmission frame. Can be improved.

[Brief description of the drawings]

FIG. 1 shows an embodiment of the present invention;

FIG. 2 is an explanatory diagram of a TCP / IP protocol stack,

FIG. 3 is an explanatory diagram of a format of an ATM cell;

FIG. 4 is an explanatory diagram of back-off at the time of retransmission,

FIG. 5 is an explanatory diagram of TCP window flow control;

FIG. 6 is an explanatory diagram of a selective discard level processing flow,

FIG. 7 is a diagram for explaining the operation of a flow monitoring counter,

FIG. 8 is an explanatory diagram of a processing flow of a flow monitoring unit;

FIG. 9 is an explanatory diagram of a disposal unit processing flow,

FIG. 10 is an explanatory diagram of an operation example,

[Explanation of symbols]

 11 Data unit flow monitoring unit, 12 Discard class table, 13 Data unit discard unit, 14 Cell buffer unit, 15 Queue length monitoring unit, 16 Congestion continuation monitoring timer unit, 17 Discard start class recording / instruction unit.

Claims (4)

[Claims] An ATM exchange in which a plurality of connections are set via a local exchange and data units belonging to each connection flow at an arbitrary time interval, and wherein a frame flow rate corresponding to an inflow data unit within a predetermined continuation time is provided. Is always measured for each connection, and based on the measurement, a discard priority is set for each connection indicating that the higher the flow rate is, the higher priority is given to other connections and frame discard at the time of congestion should be performed. A comprising: priority setting means; and frame discarding means for discarding a frame belonging to a connection for which a discard priority higher than a predetermined priority is set when detecting network congestion.
TM exchange. 2. The ATM switch according to claim 1, further comprising a function of setting said predetermined priority to indicate a lower discard priority as the duration of congestion becomes longer. 3. The discard priority setting means, at the time of detecting network congestion, sets a discard priority of the connection for which a discard priority higher than the predetermined priority has been set and a frame has been discarded. 2. The ATM switch according to claim 1, further comprising a function of setting a lower priority. 4. The frame discarding means has a function of not discarding a frame for a predetermined time when network congestion is detected, for a connection in which it is detected that a time interval between incoming data units is equal to or longer than a predetermined time. The ATM switch according to claim 1, further comprising: