JP3989197B2 - Packet discard device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an apparatus for guaranteeing the minimum bandwidth of each connection accommodated in a communication network.
[0002]
[Prior art]
In recent years, data traffic such as IP traffic has increased exponentially due to the rapid spread of the Internet and LAN. Along with this, the frequency of occurrence of congestion is increasing on the network, and degradation of service quality for users is a problem. For example, in the conventional Internet, the best effort service that does not guarantee the transfer quality has been mainstream, but the number of cases where sufficient throughput cannot be obtained only by the best effort service is increasing, and between ISPs (Internet Service Providers) and between companies. In the case of connecting with a high-speed line, it is considered that the demand for services that guarantee the minimum bandwidth and delay quality for each line will increase in the future.
[0003]
As a service for guaranteeing the minimum bandwidth contracted for each user, there is a GFR (Guaranteed Frame Rate) service. At the time of non-congestion, each user can share and use an available bandwidth (residual bandwidth).
[0004]
Many proposals have been made regarding a method for realizing the GFR service, and one of them is a WRR (Weighted Round Robin) method. FIG. 16 is a diagram showing an outline of the WRR system. The WRR method has a queue for each connection, and a weight is assigned to each connection, and a bandwidth is distributed among the connections by performing read control according to the weight on each queue.
[0005]
By accommodating all connections that share a bandwidth in the WRR, the bandwidth can be distributed according to the weight of the accommodated connection. With the WRR scheme, it is possible to distribute the surplus bandwidth fairly between connections according to a certain rule while guaranteeing the minimum bandwidth.
[0006]
However, the complexity of the WRR hardware is proportional to the number of connections accommodated, and becomes a bottleneck for speeding up. In addition, since it is necessary to have a queue for each connection, the hardware amount of the buffer unit also becomes a problem. That is, the WRR method becomes high speed, and if the number of connections accommodated increases, it is difficult to economically realize the GFR service. Therefore, there is a demand for a method for realizing the GFR service with a simple hardware configuration.
[0007]
As a method for realizing the GFR service with a simple hardware configuration, a FIFO-Tagging method is known. In this method, it is not necessary to have a buffer for each connection, and only one buffer is required for each line. In the FIFO-Tagging method, the input rate to the network is observed at the entrance of the network for each connection, and if the measured rate is equal to or less than the MCR (Minimum Cell Rate), the packet of that connection passes as it is and exceeds the MCR. Then, Tag is added to the header part of the packet. Here, the MCR is a band for which the network guarantees transfer for a connection.
[0008]
The FIFO buffer is provided with a threshold, and it is always observed whether or not the queue length exceeds the threshold. If the queue length exceeds the set threshold value, the packet with Tag in the header is discarded before entering the FIFO buffer, and the packet without Tag in the FIFO buffer, that is, to the network. Only packets with connections whose input rate is less than or equal to MCR pass.
[0009]
In the FIFO-Tagging method, the MCR can be guaranteed at the time of congestion by setting the output speed of the FIFO buffer to be equal to or higher than the total MCR of the connections accommodating the FIFO buffer. In addition, when there is a part of the bandwidth exceeding the sum of the MCRs of each connection and a surplus bandwidth at the time of non-congestion, the bandwidth is shared by a plurality of connections.
[0010]
[Problems to be solved by the invention]
However, in this method, a buffer is shared by a plurality of connections, so the bandwidth and surplus bandwidth that exceed the sum of the MCRs of each connection are distributed among the connections in proportion to the input rate to the FIFO buffer. Therefore, there was a problem in terms of “fairness”. In the FIFO-Tagging method, when there is a surplus bandwidth, there is no provision on how to distribute the bandwidth among the connections, so in an extreme case an unfair situation where one connection occupies the surplus bandwidth There was a problem that occurred.
[0011]
The present invention is made in such a background, and by installing it in the communication network, the minimum bandwidth contracted even during congestion is guaranteed for each connection accommodated in the communication network. However, it is an object of the present invention to provide a packet discarding device that can distribute the surplus bandwidth fairly between connections during normal times (non-congestion) and enable more efficient use of network resources. . It is an object of the present invention to provide a packet discard apparatus that can be realized economically even on a high-speed line. It is another object of the present invention to provide a packet discarding apparatus that can be installed in a node such as a packet switching switch to realize a simple hardware configuration and efficiently avoid congestion in the node.
[0012]
[Means for Solving the Problems]
The present invention is a packet discard apparatus comprising means for determining whether or not an incoming packet can be accepted, and means for discarding a packet that has been rejected due to this determination.
[0013]
Here, the feature of the present invention is that Acceptability The means to determine When the arrival rate or queue length of incoming packets exceeds the admission permission rate or admission permission queue length and the ratio exceeds the specified value, At least two consecutive packets are rejected.
[0014]
In addition, Acceptability The means for determining may comprise means for prohibiting continuous acceptance refusal. In this case, the means for switching between the means for refusing acceptance of the two consecutive packets and the means for prohibiting continuous acceptance refusal are selected. It is also possible to provide means for
[0015]
For example, when the queue length Xj suddenly exceeds the permission queue length AQj, it is necessary to perform rapid traffic control in order to avoid congestion. When such a situation occurs, it can be dealt with by continuously discarding packets. On the other hand, when the queue length Xj gently exceeds the permitted queue length AQj, it is possible to realize a stable traffic control by performing a gentle traffic control. It is desirable to use such conflicting control modes by switching according to the situation. Such switching control is desirably provided as an additional function of the means for determining. The packet can be described similarly even if it is a cell that is a fixed-length packet.
[0016]
By installing it in the communication network, it guarantees the minimum bandwidth contracted for each connection accommodated in the communication network even at the time of congestion, while surplus at normal time (when there is no congestion) A bandwidth can be evenly distributed among connections, network resources can be used more efficiently, and a packet discard device that can be realized economically even on a high-speed line can be realized. Furthermore, by installing in a node such as a packet switching switch, a simple hardware configuration can be realized, and efficient congestion avoidance in the node can be realized.
[0017]
DETAILED DESCRIPTION OF THE INVENTION
(First Example)
A multiplexing apparatus according to the first embodiment of the present invention to which the packet discarding apparatus of the present invention is applied will be described with reference to FIGS. FIG. 1 is a block diagram of a multiplexing apparatus according to the first embodiment of the present invention. FIG. 12 is a diagram for explaining queue length detection using a monitoring packet.
[0018]
As shown in FIG. 1, a multiplexing apparatus to which the packet discarding apparatus 5 of the present invention is applied includes a connection identifying unit 1 for identifying connections 1 to k to which an arriving packet belongs, and a packet for determining whether or not to accept the packet. The discard device 5, the FIFO type buffer 6 that multiplexes packets according to the determination result of the packet discard device 5, and the queue length of the buffer 6 for each connection i from among the packets of multiple connections accumulated in the buffer 6 The queue length monitoring unit 7 and the virtual queue management unit 3 that detect Xi, and the connection information storage unit 2 that holds the value of the weight Wi predetermined for each connection i are provided. The queue length Qmax is set, and the queue length becomes almost zero when the maximum queue length Qmax is exceeded. The continuous function of ΣXi that is the sum of i is α (ΣXi), the sum of the weights of the connection i in which one or more packets are stored in the buffer 6 is Wact, and the permission queue length calculation unit 4 AQi
AQi = α (ΣXi) ・ Wi / Wact
The packet discarding device 5 determines that the packet that belongs to the connection whose queue length Xi for each connection i is equal to or smaller than the permitted queue length AQi is accepted.
[0019]
In the present embodiment, the queue length monitoring unit 7 and the virtual queue management unit 3 buffer each connection i with respect to the actual queue length of the buffer 6 in a situation where a plurality of connections are mixed and accumulated in one buffer 6. What detected the queue length Xi of 6 is called a virtual queue length. In the following, for simplicity of explanation, the virtual queue length is simply referred to as queue length Xi or Xj.
[0020]
As a feature of the packet discard device 5 of the present invention, the ratio that the queue length Xj of the connection j that has been rejected is calculated exceeds the permission queue length AQj, and when the calculated ratio exceeds a predetermined value It is also possible to reject the acceptance of packets of the connection j in succession at least two, and prohibit the continuous acceptance rejection of the packets of the connection j whose acceptance has been rejected when the calculated ratio is not more than a predetermined value. Which acceptance determination policy is adopted can be switched and selected by a user operation.
[0021]
Further, the packet discarding device 5 can determine that the packet to be determined to be unacceptable is permitted to be accepted when the packet of the connection j that should be determined to be unacceptable does not exist in the buffer 6.
[0022]
A packet is attached to a packet that exceeds the minimum guaranteed bandwidth and arrives at the network, and the packet discarding device 5 determines that the packet is to be rejected when the tag is attached to the packet that is determined to be rejected. Is determined to be accepted.
[0023]
The queue length monitoring unit 7 detects the queue length Xi for a change cycle that is equal to or less than a predetermined change cycle of the queue length Xi. Alternatively, the queue length monitoring unit 7 outputs the maximum value of the queue length X within the unit time as the detection result of the queue length Xi for each unit time.
[0024]
Further, as shown in FIG. 12, a monitoring packet insertion unit 18 for sending a monitoring packet to the buffer 6 is provided, and the queue length monitoring unit 7 receives the monitoring packet sent from the monitoring packet insertion unit 18 as input to the buffer 6. The queue length Xi can also be detected according to the time difference between the received time and the time when the monitoring packet is output from the buffer 6.
[0025]
A plurality of service classes are provided, and the connection identifying unit 1 simultaneously identifies the service class to which the packet belongs together with the connection of the arriving packet, and the virtual queue management unit 3 performs the service of the buffer 6 for each connection i. The queue length Xi for each class is detected, and the connection information storage unit 2 holds the value of the weight Wi for each service class that is predetermined for each connection i, and the packet discard device 5 determines the permission for each service class. Queue length AQi
AQi = α (ΣXi) ・ Wi / Wact
And a packet belonging to a connection whose queue length Xi for each service class per connection i is equal to or smaller than the permissible queue length AQi for each service class can be determined to be accepted.
[0026]
Hereinafter, the first embodiment of the present invention will be described in more detail.
[0027]
FIG. 14 is a block diagram of a network that applies the present invention and provides a minimum bandwidth guarantee service. The calling user sends a packet over the network to the called user. The network guarantees the transfer rate up to the minimum bandwidth contracted with the user. A plurality of users are accommodated in each link of the network, and the surplus bandwidth is distributed among users according to a weight determined based on a fee, a minimum guaranteed bandwidth, and the like.
[0028]
As described above, the multiplexing device including the packet discarding device of the present invention can be used as a bandwidth control device (UPC: Usage Parameter Control) by being provided in the network as shown in FIG. Further, by providing it in the packet switching switch, it can be used to avoid buffer congestion in the switch.
[0029]
First, when the originating user sends a packet to the network, the packet sending rate to the user's network is measured by a rate observation device located at the incoming edge of the network. If the packet transmission rate exceeds the minimum guaranteed bandwidth, Tag is attached to the packet header, and packets transmitted below the minimum guaranteed bandwidth are transmitted to the network as they are. A packet without a tag is transferred to the called user without being discarded in the network, thereby guaranteeing the minimum bandwidth. The multiplexing device of the present invention is located in the relay node and performs a process of distributing a link band shared by a plurality of users to each user according to the weight.
[0030]
A first embodiment of the present invention will be described with reference to FIGS. As shown in FIG. 1, the multiplexing apparatus includes a connection identification unit 1, a bandwidth control unit 10, and a buffer unit 11. The connection identifying unit 1 identifies the connection from the header of the packet that arrived. The bandwidth control unit 10 performs processing related to bandwidth control based on the connection information. The buffer unit 11 is a simple FIFO buffer 6 and is shared among all users accommodated in the link.
[0031]
A queue length monitoring unit 7 is connected to the buffer 6, and the queue length of the buffer 6 and information on input / output packets are transmitted to the virtual queue management unit 3 at the previous stage when packets are input / output. Here, the information relating to the input / output packet includes the packet length of the packet input to or output from the buffer 6 and the identification number of the connection to which the packet belongs.
[0032]
When a packet arrives from the input line side, the connection identifying unit 1 identifies the connection of the packet, and the packet is sent to the subsequent bandwidth control unit 10 to determine whether or not to input to the buffer 6 and is determined to be discarded. Packets are discarded on the spot, and packets that are not determined to be discarded are input to the buffer unit 11 at the subsequent stage and output to the output line in accordance with the FIFO rules.
[0033]
In the present invention, bandwidth control based on packet discard in the bandwidth controller 10 plays an important role. Therefore, the processing of the bandwidth control unit 10 will be described in detail. The bandwidth control unit 10 includes a virtual queue management unit 3, a permission queue length calculation unit 4, a connection information storage unit 2, and a packet discard device 5.
[0034]
The virtual queue management unit 3 calculates the queue length of the buffer 6 and the queue length for each connection based on information from the queue length monitoring unit 7. The virtual queue management unit 3 is connected to the connection information storage unit 2, and the calculated virtual queue length for each connection is recorded in the connection information storage unit 2.
[0035]
A method for calculating each queue length will be described. First, regarding the calculation of the queue length of the buffer 6, the queue length information transmitted from the queue length monitoring unit 7 is simply held in the virtual queue management unit 3 as it is. Next, a virtual queue length calculation method for each connection will be described. Since the buffer 6 is empty in the initial state, the virtual queue length of each connection is zero. When a packet is input to the buffer 6, the packet length and connection identification number of the packet are transmitted through the queue length monitoring unit 7. Based on the information, the virtual queue management unit 3 accesses the connection identification unit 1 and performs a process of adding the packet length sent to the virtual queue length of the connection. On the contrary, at the time of packet output, a process of subtracting the packet length transmitted from the virtual queue length is performed.
[0036]
FIG. 2 shows an example of the connection information storage unit 2 in which information on the virtual queue length for each connection calculated by the virtual queue management unit 3 and information on connection weights are stored.
[0037]
Next, the permission queue length calculation unit 4 receives information on the queue length of the buffer 6 from the virtual queue management unit 3 and the connection information storage unit 2, and information on the virtual queue length and weight for each connection, and based on them Has a function to calculate the authorization queue length. Here, the permitted queue length of the connection represents the maximum queue length that is allowed to be input to the buffer 6.
[0038]
A method for calculating the permission queue length AQi of the connection i will be specifically described. Connection weight is Wi, queue length of current buffer 6 Sum of The Σ Let Xi be
AQi = α (ΣXi) ・ Wi / Wact
Calculated by α (ΣXi) is a continuous function, and the graphs of FIGS. 3 to 9 are examples of the queue length function α (ΣXi). Queue length on the horizontal axis Sum of ΣXi is taken, and the value of the function α (ΣXi) is taken on the vertical axis. Wact is the sum of the weights of active connections. Here, the active connection is a connection having a virtual queue length larger than zero, that is, a connection in which at least one packet is stored in the buffer 6.
[0039]
The function α (ΣXi) becomes zero when the queue length exceeds a certain threshold value (Qmax in the figure). That is, when the queue length exceeds a certain threshold value, it is determined that the queue is congested, and the permission queue length of each connection becomes zero.
[0040]
For example, in the example of FIG. 3, the function α (ΣXi) takes a constant value within a range not exceeding the threshold value Qmax, and the function α (ΣXi) becomes zero when the threshold value Qmax is exceeded. That is, the permission queue length AQi is distributed in proportion to the connection weight within a range not exceeding the threshold value Qmax.
[0041]
In the example of FIG. 4, the queue length Sum of Between ΣXi and a predetermined value less than the threshold value Qmax, the function α (ΣXi) gradually decreases until halfway, rapidly decreases as it approaches the threshold value Qmax, and becomes zero at the threshold value Qmax. That is, the permission queue length AQi is the queue length Sum of When ΣXi is small, it is distributed over the weight of connection i, but the queue length Sum of When ΣXi exceeds a predetermined value, the distribution ratio of the permission queue length AQi sharply decreases.
[0042]
In the example of FIG. 5, the queue length Sum of ΣXi and the value of the function α (ΣXi) are linearly inversely proportional. That is, the permission queue length AQi is the queue length Sum of Linearly inversely proportional to ΣXi. This is the simplest and basic control example.
[0043]
In the example of FIG. 6, the queue length Sum of ΣXi and the value of the function α (ΣXi) are inversely proportional according to a quadratic function. That is, the permission queue length AQi is the queue length Sum of Large when ΣXi is small, but queue length Sum of As ΣXi increases, it begins to decrease suddenly, and the queue length Sum of As ΣXi approaches the threshold value Qmax, the decrease becomes gradual. This allows the queue length Sum of When ΣXi starts to increase, traffic increase can be strongly suppressed.
[0044]
In the example of FIG. 7, the queue length Sum of The value of ΣXi and the function α (ΣXi) are inversely proportional to the quadratic function reversed from the quadratic function shown in FIG. That is, the permission queue length AQi is the queue length Sum of Large when ΣXi is small, but queue length Sum of As ΣXi increases, it begins to decrease gradually, and the queue length Sum of The decrease becomes steeper as ΣXi approaches the threshold value Qmax. This allows the queue length Sum of As ΣXi approaches the threshold value Qmax, an increase in traffic can be strongly suppressed.
[0045]
In the example of FIG. 8, the queue length Sum of ΣXi and the value of the function α (ΣXi) are inversely proportional in steps. That is, the permission queue length AQi is the queue length Sum of Large when ΣXi is small, but queue length Sum of As ΣXi increases, it begins to decrease gradually, and the specified queue length Sum of The queue length starts to decrease suddenly from ΣXi. Sum of As ΣXi approaches the threshold value Qmax, the decrease becomes gentle again. This allows the queue length Sum of When ΣXi is near the middle between zero and the threshold value Qmax, an increase in traffic can be strongly suppressed.
[0046]
In the example of FIG. 9, the queue length Sum of ΣXi and the value of the function α (ΣXi) are reversed from the example of FIG. That is, the permission queue length AQi is the queue length Sum of Large when ΣXi is small, but queue length Sum of As ΣXi increases, it begins to decrease suddenly, starts to decrease gradually from the predetermined queue length ΣXi, and queue length Sum of As ΣXi approaches the threshold value Qmax, it suddenly starts decreasing again. This allows the queue length Sum of The increase in traffic can be strongly suppressed between the time when ΣXi starts to increase from zero and the time when ΣXi approaches the threshold value Qmax.
[0047]
Also queue length Sum of The current queue length (the instantaneous value of the queue length when the packet arrives) is used as ΣXi. Sum of You may employ | adopt the average value of the queue length within a fixed time as (SIGMA) Xi.
[0048]
The packet discard device 5 determines whether a packet arriving from the permission queue length and the virtual queue length for each connection is input to the buffer 6 or discarded. FIG. 10 is a flowchart showing a processing flow at the time of arrival of a packet for performing deterministic discarding processing. As shown in FIG. Long X If i is less than or equal to the permission queue length AQi, the packet is input to the buffer 6. If the queue exceeds the permission queue length AQi, the packet with Tag is discarded, and the packet without Tag is input to the buffer 6. Thus, control is performed so that the occupation ratio of the buffer 6 becomes fair while always guaranteeing the minimum bandwidth of each connection.
[0049]
Further, when the virtual queue length exceeds the permitted queue length, it is possible to discard packets probabilistically instead of deterministically discarding packets as described above. FIG. 11 is a flowchart showing a processing flow at the time of arrival of a packet for performing probabilistic discard processing. As shown in FIG. 11, the permitted queue length is AQi, and the virtual queue length. Sum of Is assumed to be ΣXi, the discard probability P is
P = (ΣXi−AQi) / ΣXi
Given in. In other words, when the virtual queue length exceeds the permission queue length, the excess queue length is dropped. By controlling bandwidth by probabilistic discarding, when focusing on one connection, even if the virtual queue length exceeds the permitted queue length, the phenomenon that packets are discarded continuously is deterministic. Therefore, it can be said that the compatibility with TCP rate control is increased. The packet input to the buffer 6 is output to the output line according to the FIFO standard.
[0050]
The generation of the pseudo random number Rand in FIG. 11 is performed every time the packet arrives, and by executing the discard based on the result of comparing the discard probability P and the pseudo random number Rand, it is possible to actually realize the discard with the discard probability P. it can. For example, when the discard probability P is 0.5, the probability that the pseudo random number Rand takes a value larger than 0.5 is also 0.5, and discarding is performed when the pseudo random number Rand is larger than 0.5. By executing, discard according to the discard probability P is performed.
[0051]
The multiplexing apparatus to which the packet discarding apparatus 5 of the present invention is applied estimates the network congestion state based on the queue length, and if it is determined to be congested, the permitted queue length becomes equal to zero and is below the minimum guaranteed bandwidth. The network transfers only packets that are sent to the network at a rate of. When the queue length is less than or equal to a certain threshold value, it is determined that there is an excess bandwidth, and each connection is equally distributed between connections by occupying a buffer proportional to the weight.
[0052]
In addition, while realizing a GFR service with a FIFO buffer with a simple hardware configuration, the surplus bandwidth is fairly shared between connections as in the method of performing packet read control with a buffer individually for each connection, such as the WRR method. Can be distributed. In addition, since one buffer is shared by many connections, there is an advantage that a necessary buffer amount can be reduced by the statistical multiplexing effect, compared with a method in which a buffer is individually provided for each connection.
[0053]
Next, as a feature of the packet discard device 5 in the first embodiment of the present invention, the ratio that the queue length Xj of the connection j rejected by the packet discard device 5 exceeds the permitted queue length AQj is calculated. When the ratio exceeds the predetermined value, at least two consecutive connection j packets are rejected. When the calculated ratio is equal to or less than the predetermined value, the connection j packets rejected are continuously received. Prohibit refusal.
[0054]
That is, when the queue length Xj suddenly exceeds the permission queue length AQj, it is necessary to perform rapid traffic control in order to avoid congestion. When such a situation occurs, it can be dealt with by continuously discarding packets. On the other hand, when the queue length Xj gently exceeds the permitted queue length AQj, it is possible to realize a stable traffic control by performing a gentle traffic control.
[0055]
Further, as a function of the queue length monitoring unit 7, the queue length Xi is detected for a change period equal to or less than a predetermined change period of the queue length Xi. Thereby, the instantaneous change in the queue length Xi can be absorbed, and the malfunction of traffic control due to disturbance can be avoided.
[0056]
Alternatively, as a function of the queue length monitoring unit 7, the maximum value of the queue length Xi within the unit time is output as the detection result of the queue length Xi for each unit time. Thereby, the instantaneous change in the queue length Xi can be absorbed, and the malfunction of traffic control due to disturbance can be avoided.
[0057]
As shown in FIG. 12, the buffer 6 includes a monitoring packet insertion unit 18 for sending a monitoring packet, and the queue length monitoring unit 7 is the time when the monitoring packet sent from the monitoring packet insertion unit 18 is input to the buffer 6. And the queue length X is detected according to the time difference between the time when the monitoring packet is output from the buffer 6.
[0058]
As a result, the effective queue length Xi including parameters such as the actual delay time can be detected as compared with the detection of the queue length Xi simply based on the number of accumulated packets.
[0059]
Also, packets belonging to one connection can be further classified into two service classes and handled. One service class is a real-time (R) class for packets that place importance on real-time characteristics, and another service class is a non-real-time (NR) class for packets that do not place importance on real-time characteristics.
[0060]
For this purpose, the connection identifying unit 1 shown in FIG. 1 identifies the connection of the arriving packet and identifies the service class of the packet. The virtual queue management unit 3 also acquires service class information together with packet connection information stored in the buffer 6. Thereby, the connection information storage unit 2 can be provided with a table as shown in FIG.
[0061]
For example, regarding connection # 1, X1 (R) belonging to the real-time class and X1 (NR) belonging to the non-real-time class are recorded as the virtual queue length. In addition, real-time class weight W1 (R) and non-real-time class weight W1 (NR) are recorded as weights. The weight W1 (R) is set to be heavier than W1 (NR), and packets belonging to the real-time class can be read with priority over packets belonging to the non-real-time class.
[0062]
Thereby, since the delay of the packet which attaches importance to the real time property can be reduced, the QoS for the user can be improved.
[0063]
(Second embodiment)
A multiplexing apparatus according to the second embodiment of the present invention to which the packet discarding apparatus of the present invention is applied will be described with reference to FIG. FIG. 15 is a block diagram of a multiplexing apparatus according to the second embodiment of the present invention.
[0064]
As shown in FIG. 15, the multiplexing apparatus includes a connection identification unit 1 that identifies connections 1 to k to which a cell that is a fixed-length packet arrives, a reception determination unit 15 that determines whether or not the cell is acceptable, A FIFO buffer 6 that temporarily accumulates cells according to the determination result of the acceptance determination unit 15, a rate detection unit 8 that detects the arrival rate Ri of the cell for each connection i, and a weight Wi that is predetermined for each connection i. A connection information storage unit 2 for holding a value, and a maximum rate serving as a threshold is set in advance for the cell arrival rate Ri, and the sum ΣRi of the cell arrival rates Ri that becomes almost zero when the maximum rate is exceeded. The continuous function is β (ΣRi), and the weight of the connection i in which one or more cells are stored in the buffer 6 of the buffer unit 11 Was used as a Wact, permitted the arrival rate calculation unit 14, the permission arrival rate ACRi
ACRI = β (ΣRi) · Wi / Wact
The reception determination unit 15 determines that this is a reception permission for cells belonging to connections whose cell arrival rate Ri for each connection i is equal to or lower than the permitted arrival rate ACRi.
[0065]
In the present embodiment, the queue length monitoring unit 7 and the virtual queue management unit 3 buffer each connection i with respect to the actual queue length of the buffer 6 in a situation where a plurality of connections are mixed and accumulated in one buffer 6. What detected the queue length Xi of 6 is called a virtual queue length.
[0066]
FIG. 14 is a block diagram of a network that applies the present invention and provides a minimum bandwidth guarantee service. The calling user sends a packet over the network to the called user. The network guarantees the transfer rate up to the minimum bandwidth contracted with the user. A plurality of users are accommodated in each link of the network, and the surplus bandwidth is distributed among users according to a weight determined based on a fee, a minimum guaranteed bandwidth, and the like.
[0067]
As described above, the multiplexing device including the packet discarding device of the present invention can be used as a bandwidth control device (UPC: Usage Parameter Control) by being provided in the network as shown in FIG. Further, by providing it in the packet switching switch, it can be used to avoid buffer congestion in the switch.
[0068]
First, when the originating user sends a packet to the network, the packet sending rate to the user's network is measured by a rate observation device located at the incoming edge of the network. If the packet transmission rate exceeds the minimum guaranteed bandwidth, Tag is attached to the packet header, and packets transmitted below the minimum guaranteed bandwidth are transmitted to the network as they are. A packet without a tag is transferred to the called user without being discarded in the network, thereby guaranteeing the minimum bandwidth. The multiplexing device of the present invention is located in the relay node and performs a process of distributing a link band shared by a plurality of users to each user according to the weight.
[0069]
As a feature of the reception determination unit 15 shown in FIG. 15 as the packet discarding apparatus in the second embodiment of the present invention, the ratio that the arrival rate Rj of the connection j rejected by the reception determination unit 15 exceeds the permitted arrival rate ACRj is calculated. When the calculated ratio exceeds a predetermined value, at least two consecutive packets of connection j are rejected. When the calculated ratio is equal to or lower than the predetermined value, the connection j of the rejected connection j is rejected. Prohibits continuous rejection of packets.
[0070]
That is, when the arrival rate Rj suddenly exceeds the permitted arrival rate ACRj, it is necessary to perform rapid traffic control in order to avoid congestion. When such a situation occurs, it can be dealt with by continuously discarding packets. On the other hand, when the arrival rate Rj gently exceeds the permitted arrival rate ACRj, it is possible to perform stable traffic control by performing gentle traffic control.
[0071]
【The invention's effect】
As described above, according to the present invention, by installing in the communication network, for each connection accommodated in the communication network, the guaranteed minimum bandwidth is guaranteed even during congestion, and at normal times. In the time of non-congestion, the surplus bandwidth can be evenly distributed among the connections, thereby enabling more efficient use of network resources. In addition, it is possible to realize a packet discard apparatus that can be realized economically even on a high-speed line. Furthermore, by installing in a node such as a packet switching switch, a simple hardware configuration can be realized, and efficient congestion avoidance in the node can be realized.
[Brief description of the drawings]
FIG. 1 is a block diagram of a multiplexing apparatus according to a first embodiment of the present invention.
FIG. 2 is a diagram showing a table example of a connection information storage unit.
FIG. 3 is a diagram illustrating an example of a function α (ΣXi).
FIG. 4 is a diagram illustrating an example of a function α (ΣXi).
FIG. 5 is a diagram illustrating an example of a function α (ΣXi).
FIG. 6 is a diagram illustrating an example of a function α (ΣXi).
FIG. 7 is a diagram illustrating an example of a function α (ΣXi).
FIG. 8 is a diagram illustrating an example of a function α (ΣXi).
FIG. 9 is a diagram illustrating an example of a function α (ΣXi).
FIG. 10 is a flowchart showing a processing flow when a packet arrives for deterministic discarding processing.
FIG. 11 is a flowchart showing a processing flow when a packet arrives to perform probabilistic discard processing.
FIG. 12 is a diagram for explaining queue length detection using a monitoring packet.
FIG. 13 is a diagram showing a table example of a connection information storage unit.
FIG. 14 is a diagram showing an example of a network configuration to which the present invention is applied.
FIG. 15 is a block diagram of a multiplexing apparatus according to the second embodiment of the present invention.
FIG. 16 is a diagram for explaining the outline of a conventional WRR method.
[Explanation of symbols]
1 Connection identifier
2 Connection information storage
3 Virtual queue manager
4 Allowed queue length calculator
5 Packet discard device
6 buffers
7 Queue length monitor
8 Rate detector
10 Band control part
11 Buffer section
14 Permitted arrival rate calculator
15 Reception determination part
18 Monitoring packet insertion part

Claims (2)

In a packet discarding device comprising means for judging whether or not to accept an incoming packet and means for discarding a packet rejected by this judgment,
The means for determining acceptability is:
Means for rejecting at least two consecutive packets when the queue length of the arriving packet exceeds the queue length of acceptance permission for each connection in the buffer and the ratio exceeds a predetermined value ;
Packet discard characterized in that it comprises means for prohibiting continuous acceptance refusal when the queue length of arriving packets exceeds the queue length of acceptance permission for each connection in the buffer and the ratio is below a predetermined value apparatus. In a packet discarding device comprising means for judging whether or not to accept an incoming packet and means for discarding a packet rejected by this judgment,
The means for determining acceptability is:
Means for rejecting at least two consecutive packets when the arrival rate of arriving packets exceeds the acceptance permission rate and the ratio exceeds a predetermined value;
Means for prohibiting continuous acceptance refusal when the arrival rate of arriving packets exceeds the acceptance permission rate and the ratio is below a predetermined value;
A packet discarding device.

Images