JP6903436B2 - Communication device and communication method - Google Patents

Description

The present invention relates to a communication device and a method for detecting, predicting, and smoothing packet flow due to burst traffic or the like in a network.

As a technique for controlling the bandwidth of packet flow or guaranteeing the communication quality in a network, for example, there are techniques described in Patent Document 1 and Patent Document 2.
In Patent Document 1, transmission packets are classified into a plurality of queue groups to which individual bands are allocated according to the header information of each packet in the network, and a plurality of queues according to transmission priority are formed in each queue group. Packet read control that reads out transmission packets according to the transmission priority while guaranteeing the bandwidth allocated to the queue group from each queue group in the buffer memory and the transmission priority control device that queues to the buffer memory. The device is disclosed.

In Patent Document 2, in order to secure the communication quality of the flow corresponding to a large number of users and services while suppressing the capacity of the buffer used as a queue, the output is input with the transfer information regarding the transfer of communication packets as an input value. A distribution unit that distributes and stores received communication packets to one of a plurality of queues according to a value obtained by a predetermined function aggregated with respect to the queue, and a plurality of distribution units that control the bandwidth for each queue. A relay device including a band control unit that outputs communication packets stored in a queue for transmission is disclosed.

Further, in Patent Document 3, a virtual packet queue that manages only the queue length without actually inserting a packet is used, and an actual packet is converted into an actual packet according to the queue length and the discard condition held by the virtual packet queue. A technique for controlling whether to insert or discard a queue is disclosed. In this document, it is stated that the disposal conditions are determined based on the total queue length of a plurality of queues.

Japanese Unexamined Patent Publication No. 11-346246 Japanese Unexamined Patent Publication No. 2013-34164 Japanese Unexamined Patent Publication No. 2005-295524

Traffic from the server to the user is sent from a line close to the server with a high communication speed to the edge node accommodating the user line via the network. The edge node accommodating the user line has a queue for accumulating traffic to the user line, but since the user line is slower than the line in the network, the queue may overflow and packets may be discarded.
If communication packets are shaped for each user at the core node in the network, packet loss at the edge node can be reduced. However, for that purpose, it is necessary to provide a queue for all users including users who are not communicating, and to set the header condition of the communication packet for user identification and the bandwidth of the queue in the core node, and many users. It is difficult to accommodate users on the scale of, for example, 100,000.

In a method such as Patent Document 1 or Patent Document 2, in which a node in a network distributes communication packet header information and delivery information to one of a plurality of queues by table reference or a predetermined function, the number of users that can be accommodated is increased. However, since communication packets destined for a plurality of user lines are included in one queue, it is not possible to finely control each user, and it is difficult to manage fairness for each user.
The present invention has been made to solve the above problems, and in a network to which a large number of users are connected, all users even if the number of queues for shaping in the core node is smaller than the total number of accommodated users. It is an object of the present invention to provide a technique for avoiding packet overflow at an edge node.

In order to solve the above problems, in the present invention, as an example, a communication device for transmitting a packet addressed to a user delivered from a server to a network and transmitting a packet received via the network to a server, which is addressed to the user. When the packet is sent from the edge device located at the edge of the network to the user, the communication device extracts the packet-related information contained in the packet, and the packet-related information and the bandwidth information of the line between the edge device and the user. Based on, the virtual queue length, which is an estimated value of the queue length of the transmission queue addressed to the user in the edge device, is calculated and held for each user, and the bandwidth is controlled for each user based on the virtual queue length and predetermined conditions. Bandwidth control of packets addressed to the user is performed by assigning one of the first shaper queues, which is less than the total number of users, to each user who is judged to be necessary or not and which is judged to require bandwidth control. It was done.

According to the present invention, in a network to which a large number of users are connected, even if the number of queues for shaping in the core node is smaller than the total number of accommodated users, packet overflow at the edge node is avoided for all users. can do.

It is a figure which shows an example of the network configuration to which this invention applies, and the configuration of the core node in one Example of this invention. It is a figure which shows the structure of the virtual queue length calculation part in the core node in one Example of this invention. It is a figure which shows the example of the virtual queue length table in the virtual queue length update part. It is a figure which shows the model of the virtual queue which simulates a queue. It is a figure which shows the time change of a virtual queue length. It is a flowchart explaining the method of updating the virtual queue length table in the virtual queue length update part at the time of packet relay. It is a flowchart explaining the method of updating a virtual queue length table in a virtual queue length update part at the time of reading a maximum virtual queue length from a CPU part. It is a figure explaining the process in the core node when the virtual queue in the core node overflows in one Example of this invention. It is a flowchart explaining the replacement method of the user registered in a shaper queue. It is a flowchart explaining the process of registering a user in a shaper queue. It is a figure which shows the shaper queue search table and the queue of a shaper in a packet relay part. It is a figure which shows the example of the user identifier table in order of the virtual queue maximum occupancy rate. It is a figure which shows the example of the user information table. It is a figure which shows the example of the shaper queue management information table. It is a figure which shows that the discarding of an actual queue is avoided by exchanging an edge node with a device having a long queue length based on the monitoring result by this invention. It is a figure explaining the process in the core node when the virtual queue in the core node overflows in one Example of this invention.

Hereinafter, examples will be described with reference to the drawings.

FIG. 1 is a diagram showing an example of a network configuration to which the present invention is applied and a configuration of a core node in an embodiment of the present invention.
Reference numeral 1 denotes a core node for carrying out the present invention, in which distribution data from distribution servers 21 and 22 that distribute data such as moving images is distributed from the core node 1, the network 17 between the core node 1 and the edge nodes 31 and 32, and the edge nodes 31 and 32. Is relayed to user A 311 and user B 321 via the above. The network 17 represents a network including a relay device between the core node 1 and the edge nodes 31 and 32, but depending on the network configuration, there is no relay device and the core node 1 and the edge nodes 31 and 32 are directly connected. In some cases, it may go through one or more relay devices.

Since the distribution servers 21 and 22 are servers that provide services to a plurality of users, they generally have a higher processing capacity than the user A 311 and the user B 321 and are connected to the core node 1 from the distribution servers 21 and 22. , 221 generally uses a faster line than the line 3111 connecting the edge node 31 to the user A 311 and the line 3211 connecting the edge node 32 to the user B 321. For example, the lines 211 and 221 connected to the distribution servers 21 and 22 use fast lines of 10 Gbps, and the lines 3111 and 3211 connected to user A 311 and user B 321 use slow lines such as 100 Mbps and 1 Gbps, respectively.

The lines 211 and 221 connected to the distribution servers 21 and 22 are faster than the lines 3111 and 3211 connected to the user A 311 and the user B 321. Packets are continuously transmitted from the distribution servers 21 and 22 to the user A 311 and the user B 321. The queue 3112 in the edge node 31 for the line 3111 connected to the user A 311 and the line 3211 connected to the user B 321 are sent at the speed of the lines 211 and 221 connected to the distribution servers 21 and 22. The queue 3212 in the edge node 32 may overflow.

In FIGS. 2 and 2, a method for detecting the overflow of the queues 3112 and 3212 in the edge node in the core node 1 and a configuration for detecting the overflow, and a method for avoiding the overflow of the queue in the edge node after the detection and for avoiding the overflow. The configuration will be described.
FIG. 2 is a diagram showing a configuration of a virtual queue length calculation unit 12 in a core node according to an embodiment of the present invention.
Every time the packet relay unit 11 of FIG. 1 relays a packet, the packet relay unit 11 sends packet-related data 14 which is data related to the relayed packet to the virtual queue length calculation unit 12.

Of the packet-related data, the destination address 141 is sent to the user identification search unit 121, the user identification search unit 121 identifies the user from the destination address, and outputs a user identifier 123 unique to each user. The user identifier 123 is sent to the virtual queue length update unit 122. Of the packet-related data, the packet length 142 is sent to the virtual queue length update unit 122. The virtual queue length update unit 122 stores the virtual queue length for each user based on the user identifier 123 and the packet length 142 output from the user identification search unit. The virtual queue of the corresponding user in the virtual queue length table 1222. Update the length. Further, the virtual queue length calculation unit 12 stores the maximum virtual queue length indicating the maximum value of the virtual queue length so far. When the CPU unit 13 reads the maximum virtual queue length indicating the maximum value of the virtual queue length up to now via the access interface 15, the virtual queue length calculation unit 12 sets the maximum virtual queue length read / clear mode register 1222. To switch the operation of whether or not to clear the maximum virtual queue length.

FIG. 3 is a diagram showing an example of the virtual queue length table 1221 in the virtual queue length update unit 122.
The virtual queue length table 1221 uses the user identifier 12211 as an index, and as table elements, the virtual queue length update time 12212, the virtual queue length 12213, the maximum virtual queue length 12214 indicating the maximum value of the virtual queue length so far, and virtual. Indicates the output band 12215 of the queue, the limit virtual queue length 12216 which is the maximum value that the virtual queue length can take, and the number of packets whose updated virtual queue length exceeds the limit virtual queue length and cannot be loaded in the virtual queue at the time of packet relay. It is composed of discard statistics 12217 and relay statistics 12218 showing the number of packets loaded in the virtual queue at the time of packet relay.

The output band 12215 and the limit virtual queue length 12216 of the user are managed by the network administrator, and the values may change depending on the contract menu between the network administrator and the user. For example, a cheap menu may have an output band of 100 Mbps, and a high menu may have an output band of 1 Gbps. The limit virtual queue length 12216 is generally difficult for the user to understand, and is considered to be rarely disclosed to the user. For users who frequently drop packets, replace edge nodes 31 and 32 with devices with longer queues 3112 and 3212 to the user, or replace network interface cards with longer queues 3112 and 3212 with edge nodes 31 and 32. By putting it in 32, it is possible to avoid discarding the packet addressed to the user. In such a case, the limit virtual queue length may differ for each user.

FIG. 4 is a diagram showing a model of a virtual queue for updating the virtual queue length 12213 in the virtual queue length table 1221 shown in FIG.
The virtual queue simulates the actual queue length of an edge node by the virtual queue length representing the amount of packet data stored in the virtual queue. When relaying packets addressed to a user, the state in which packets are piled up is simulated by adding the packet length 142 of the relayed packet to the value of the virtual queue length of the virtual queue of the user, and the time per time obtained from the output band 12215. This model obtains an estimated value of the actual queue length at the edge node by simulating a state in which packets are pulled out by subtracting the value of the virtual queue length over time based on the amount of packet output. The maximum value that can be taken as the virtual queue length is set as the limit virtual queue length 12216. The virtual queue manages the length of the virtual queue but does not have the packet data itself, so no memory is required to hold the packet data.

FIG. 5 is a diagram showing a time change of the virtual queue length 12213 of the virtual queue shown in FIG.
FIG. 5 shows the time change of the virtual queue length when the packet was relayed at the previous time 122121 and then relayed again at the current time 122122. At the previous time 122121, the virtual queue length 12213 increases by the packet length 142 minutes. At the current time 122122, the packet length is only 142 minutes from the state where the virtual queue length is reduced by the band * (current time-previous time), which is the amount of packets output from the virtual queue during the time from the previous time 122121 to the current time 122122. The virtual queue length 12213 increases by the amount of the virtual queue.

FIG. 6 is a flowchart showing a method of updating the virtual queue length table in the virtual queue length update unit 122 at the time of packet relay.
When the packet relay unit 11 of FIG. 1 relays the packet and the virtual queue length calculation unit 12 receives the packet-related data 14, the virtual queue length calculation unit 12 executes START S500 or less in the flowchart of FIG. In S501, the virtual queue length immediately before the packet arrives (Equation 1) Virtual queue length immediately before the packet arrival =
Virtual queue length- (current time-update time) * Calculated based on bandwidth. Next, in S502 (Equation 2), the virtual queue length <0 immediately before the packet arrives.
Judge whether or not. If YES, it means that the virtual queue has already been swept when the packet arrives. Therefore, in S503 (Equation 3), the virtual queue length immediately before the packet arrives = 0.
To execute. After that, in S504 (Equation 4) Virtual queue length = Virtual queue length immediately before the packet arrives + Packet length, the packet length is added to the virtual queue length, and in S505 (Equation 5) Virtual queue length> Limit virtual queue length Is determined. If YES, the limit virtual queue length is exceeded when the packet is loaded on the virtual queue. In this case, the packet cannot be loaded in the virtual queue, and the queue length of the packet is not added to the virtual queue length as in (Equation 6) virtual queue length = virtual queue length immediately before the packet arrives in S506, and in S507. The virtual queue discard statistic, which indicates the number of packets discarded without being loaded on the virtual queue, is incremented by one.

When the determination result of S505 is NO, the packet can be loaded in the virtual queue because the limit virtual queue length is not exceeded when the packet is loaded in the virtual queue, and the virtual queue length is shown in S504 and Equation 4. As described above, the packet length of the packet is added to the virtual queue length. If the packet can be loaded on the virtual queue, in S508 (Equation 7) Maximum virtual queue length = MAX (maximum virtual queue length, virtual queue length)
The maximum virtual queue length indicating the maximum value of the virtual queue length so far is updated as in the above, and the virtual queue relay statistics indicating the number of packets loaded in the virtual queue length in S509 is incremented by one.

Finally, the update time is updated as (Equation 8) update time = current time at 510, and this flowchart ends at 511.

FIG. 7 is a flowchart showing a method of updating the virtual queue length table 1221 in the virtual queue length update unit 122 when the maximum virtual queue length 12214 is read from the CPU unit 13 of FIG.
When reading the maximum virtual queue length 12214 from the CPU unit 13 of FIG. 1, the virtual queue length calculation unit 12 executes START S520 or less in the flowchart of FIG. This flowchart is executed when the set value of the maximum virtual queue length read / clear mode register 1222 in FIG. 2 is a set value indicating clearing. Here, for example, it is assumed that the setting is cleared when the set value of the maximum virtual queue length read / clear mode register 1222 is 1. When the set value of the maximum virtual queue length read clear mode register 1222 is set to 1, the maximum virtual queue length 12214 needs to be updated even when the maximum virtual queue length 12214 is read. The process shown in the flowchart of FIG. 7 is executed.

The maximum virtual queue length 12214 in the virtual queue length table 1221 is returned as the read value of the maximum virtual queue length in S521.
In S522, the virtual queue length is calculated by (Equation 9) Virtual queue length = Virtual queue length- (Current time-Last update time) * Band, and in S523 (Equation 10) Virtual queue length <0
Judge whether or not. If YES, it means that the virtual queue has already been swept at the time of reading the maximum virtual queue length. Therefore, in S524 (Equation 11), the virtual queue length = 0.
To execute. After that, in S525, it is determined whether or not the maximum virtual queue length is in the read clear mode, and if YES, since it is in the read clear mode, (Equation 12) maximum virtual queue length = virtual queue length is executed in S526. This is the value when it is considered that the value of the current virtual queue length is entered in the maximum virtual queue length immediately after the maximum virtual queue length is cleared. If NO, the read clear mode is not set, so the value of the maximum virtual queue length is saved.
Finally, the update time is updated in S527 as (Equation 13) update time = current time, and this flowchart ends in S528.

Here, reading the maximum virtual queue length 12214 instead of the virtual queue length 12213 from the CPU unit 13 is the maximum virtual queue length 12214, which is the maximum value of the virtual queue length so far, rather than the virtual queue length 12213 itself. Is more useful for determining the overflow of virtual queues. The virtual queue length 12213 is expected to change drastically with time, and if it happens to be a small value when read from the CPU unit 13, there is a risk of underestimating the possibility of overflow, but the maximum virtual queue Since the length 12214 is the maximum value of the virtual queue length 12213 so far, the possibility of overflow can be evaluated more accurately.

Further, when the value of the maximum virtual queue length read clear mode register 1222 is set to the mode of clearing at the time of reading, the maximum virtual queue length 12214 is cleared every time the maximum virtual queue length 12214 is read from the CPU unit 13. The value read from the CPU unit 13 is the maximum value between the time of the previous reading and the time of the current reading. By repeating this periodically, there is no timing at which the time change of the maximum value of the virtual queue length is not captured as the maximum value from the CPU unit 13, and the same maximum value is not captured multiple times in a plurality of overlapping periods. , Can be grasped.

FIG. 8 is a diagram illustrating an operation of avoiding discarding the actual queue by loading only the overflowed user packets in the shaper queue in the core node when the virtual queue in the core node 1 overflows.
The shaper queue is a queue having an output band control unit 112 on the output side, such as 111. In the packet relay unit 11, there are a plurality of queues having an output bandwidth control unit 112 such as 111, and there is also a queue 113 having no output bandwidth control unit. The output of each queue is arbitrated by 114, and the line 16 is used. Output the packet to.
When the virtual queue of a certain user, for example, user A, overflows in the virtual queue length calculation unit 12, the packet destined for user A is output bandwidth controlled by loading the packet destined for user A on the shaper queue 111. The bandwidth is limited by the unit 112, and it is possible to avoid discarding in the real queue 3112 destined for the user A in the edge node 31.

FIG. 9 is a flowchart showing a method of replacing users registered in the shaper in order to realize the method of avoiding packet discard shown in FIG.
The process shown in FIG. 9 is executed by the CPU unit 13 in the core node 1.
In the process of FIG. 9, START S540 is started after the device is started, and the processes from S541 to S551 are repeated.
In the loop between S541 and S551, iterative processing is executed for all user identifiers between S542 and S550.
S543 means an interval from executing a process for a certain user to performing a process for the next user. For example, assuming that the waiting time is 1 ms and the total number of users is 100,000, the time to exit the loop from S542 to S550. That is, the time required to perform the processing of all users is 100 seconds. In this case, the traffic destined for the user is registered in the shaper queue within 100 seconds after the packet of the user starts to be discarded.
In S544, the value of the maximum virtual queue length read / clear mode register is set to the mode for clearing the maximum queue length, and the maximum virtual queue length and the discard statistics value of the user identifier are read from the CPU unit 13.

In S545, it is determined whether or not the user has been registered in the shaper queue. Here, "the user has been registered in the shaper queue" means that one queue with a shaper is secured, the bandwidth of the user is set in the queue, and the destination address range addressed to the user is described in FIG. It means that it is registered in the shaper queue search table 117 of.
If NO (the user is not already registered in the shaper queue)
(Number 14) Increase in virtual queue discard statistics =
Virtual queue discard statistics-The increase in virtual queue discard statistics for the user is calculated from the previous virtual queue discard statistics, and in S547,
(Equation 15) Increase in virtual queue discard statistics> 0
Therefore, it is determined whether or not the virtual queue discard statistics of the user are increasing.

Or,
(Equation 16) Increase in virtual queue relay statistics =
Virtual queue relay statistics-Calculate the increase in virtual queue relay statistics for the user based on the previous virtual queue relay statistics.
(Number 17) Virtual queue discard rate =
Virtual queue discard statistics increase /
(Increase in virtual queue relay statistics + increase in virtual queue discard statistics)
Calculates the virtual queue discard rate of the user, and sets S547 to
(Equation 18) It is also possible to determine whether or not to register the user in the shaper queue under the condition that the virtual queue discard rate> the discard rate threshold value. Here, by reducing the discard rate threshold value to, for example, about 10 minus 6th power, users with a very small amount of waste are prevented from being registered in the shaper queue, and the number of users registered in the shaper queue reaches the upper limit. It can also be prevented.

Alternatively, instead of determining S547 based on whether the user's virtual queue discard statistics are increasing,
(Equation 19) Judgment is made under the condition of maximum virtual queue length> limit actual queue length * judgment threshold.
(Equation 20) Judgment threshold <1
By doing so, it is also possible to detect a user whose virtual queue is growing and is likely to overflow before the overflow occurs. By determining under this condition, the packet of the user can be registered in the shaper queue before the actual queue overflows, and the actual queue overflow can be prevented.

Further, the virtual queue length simulates the real queue length, and the overflow of the virtual queue and the overflow of the real queue do not exactly match. Therefore, the margin of the accuracy of the simulation exceeds, for example, 50%. By adopting a method of determining that there is a high risk of the actual queue overflowing and loading it in the shaper queue, it is possible to more reliably avoid discarding the actual queue.

If YES in S547, it means that packets destined for the user are overflowing from the virtual queue, so the user is registered in the shaper queue at 548.
If YES in S545, the user is already registered in the virtual queue, and in S549, only the rank of the user in the virtual queue maximum occupancy order user identifier table is updated. Here, the maximum virtual queue occupancy rate is a value indicating how much the maximum virtual queue length has been increased with respect to the virtual queue length limit virtual queue length of the user.
(Number 21) Maximum occupancy rate of virtual queue
= Calculated by maximum virtual queue length / limit virtual queue length. When this value reaches 100%, it can be determined that the virtual queue has been discarded.

FIG. 10 is a flowchart showing the process of the subroutine “Registering the user in the shaper queue” S548 in FIG.
Enter this subroutine with S54800.
In S54801, it is determined whether or not there is a vacancy in the shaper queue.
In the case of NO, since there is no space in the shaper queue, the user is registered after performing the process of expelling another user already registered in the shaper queue shown in S54802.
S54802 is a process of expelling other registered users to create a vacancy, and selects the user with the smallest maximum occupancy rate from the user identifier table in order of the maximum occupancy rate of the virtual queue shown in FIG. 12, which will be described later, to maximize the virtual queue. Delete from the user identifier table in order of occupancy and the shaper queue.
In S54803, since there is a vacancy in the shaper queue or after the vacancy is created, the user is registered in the user identifier table in order of the maximum occupancy rate of the virtual queue in FIG. 12 and the shaper queue, and the subroutine is exited in S54806.

FIG. 11 is a diagram showing a shaper queue search table 117 and a shaper queue in the packet relay unit 11 for realizing the method of avoiding packet discard shown in FIG.
Reference numeral 117 denotes a search table for outputting the determination result of whether or not to load in the shaper queue at the time of packet relay and the queue number when loading in the shaper in the packet relay unit 11.
When the packet relay unit 11 relays a packet, the shaper search table 117 is searched using the destination address as a search key. When the destination address of the packet is within the destination address range of any of the entries registered in the shaper queue search table 117, the packet is loaded in the shaper queue corresponding to the shaper queue number of the entry. Here, the destination address range is used as the index of the shaper queue search table 117 in the example of FIG. 11 because it is assumed that the subnet address is stored as an example.

In the example of FIG. 11, as a result of searching the shaper search table 117 using the destination address as a search key, for example, when a packet addressed to the user A enters the destination address range A, the packet is loaded in the queue 111 of the shaper queue number 100. When a packet addressed to the user C enters the destination address range C, the packet is loaded in the queue 115 of the shaper queue number 200.

If the destination address of the packet is not in the destination address range of any entry registered in the shaper queue search table 117, the packet is loaded in the queue that is not shaped.
For example, if the packet addressed to user B is not included in the destination address range of any entry in the shaper queue search table 117, the packet is loaded in the queue 113 that is not shaped.

FIG. 12 is an example of the virtual queue maximum occupancy rate order user identifier table used for the processing shown in FIGS. 9 and 10. In this example, the maximum occupancy rate of the virtual queue reaches 100% for 10 users, and the virtual queue is overflowing, but the 11th and subsequent users indicate that the maximum occupancy rate of the virtual queue does not reach 100%. Since the traffic destined for each user changes over time, even a user registered in the shaper queue due to an overflow of the virtual queue in the past may decrease in traffic over time. Therefore, by always sorting the users in the order of the maximum occupancy rate of the virtual queue, the maximum occupancy rate is the smallest when the shaper queue is full, and the user who is least likely to be discarded even if it is expelled from the shaper queue. By expelling them in order from the beginning, it is possible to more reliably avoid discarding in the actual queue of all users.

13 and 14 show examples of tables other than FIG. 12, which are required for the CPU unit 13 to manage the queues of users and shapers in the core node 1. The example shown here is an example, and the table configuration does not necessarily have to be the same as this.
FIG. 13 is a user information table for the CPU unit 13 to manage users in the core node 1. From the user information table, the CPU unit 13 sets the output band 12215 and the limit virtual queue length 12216 in the entry of each user identifier in the virtual queue length table 1221 in FIG. 3, and enters each index in the shaper queue search table 117 in FIG. Set the destination address range, user identifier, and shaper queue number in.
FIG. 14 is a shaper queue management information table for the CPU unit 13 to manage the shaper queues 111, 113, and 115 in the core node 1. From the shaper queue management information table, the CPU unit 13 obtains an index for registering the shaper queue number as an index in the shaper queue search table 117 of FIG. 11 and an index of the user identifier table in order of the virtual queue maximum occupancy rate of FIG.

Alternatively, instead of automatically performing the processes shown in FIGS. 9 and 10 in the core node 1, the state of the virtual queue for each user is notified to the network administrator, and the network administrator sends the shaper queue in the packet relay unit 11. It is also possible to register the user in the configuration definition. This method is suitable when the management policy of the network administrator wants to manage the users registered in the shaper queue rather than automatically replacing the users registered in the shaper queue.
As a method of communicating the virtual queue status for each user to the network administrator, for example, the following tabular format can be used. As the display order, for example, the virtual queue is sorted in descending order of the maximum virtual queue occupancy rate, and the virtual queue is discarded for the user whose maximum virtual queue occupancy rate reaches 100%. Therefore, the virtual queue is discarded in descending order of the virtual queue discard rate. When the images are further sorted and displayed, it is easy for the network administrator to understand the user who has been discarded or the user who is likely to be discarded. Further, in the above sort order, the number of users to be displayed may be specified so that only the higher-ranking users are displayed.
Table 1 Virtual queue status display format −−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
Virtual Queue Virtual Queue User Maximum Occupancy Disposal Rate Identifier −−−−−−−−−−−−−−−−−−−−−−−−−−−−−
100% 10% User A
100% 1% User C
100% 0.1%:
99% 0%:
80% 0%:
50% 0%:
20% 0%:
5% 0%:
1% 0%:
0.1% 0%:
−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−–
Table 1 may also display the amount of increase in the virtual queue discard statistics.
Further, not only the user identifier but also the edge node to which the user belongs, the number of the line connected to the user in the edge node, the line speed, and other information of the user's contract form may be displayed together.

In FIG. 15, the virtual queue monitoring server 23 monitors the virtual queue length in the core node 1 to estimate the packet discard status of the real user queues at the edge nodes 31 and 32, and manually edge nodes 31 and 32 as needed. It is a figure which shows that it is possible to avoid discarding the actual cues 3112 and 3212 by exchanging the cues with a device having a long cue length.
The information of the virtual queue length table shown in FIG. 3 calculated by the virtual queue length calculation unit 12 is read from the CPU unit 13, and includes the maximum virtual queue length, virtual queue discard statistics, and virtual queue relay statistics from the virtual queue monitoring server 23. By reading out the statistical information, the user who is being discarded in the virtual queue is grasped, and the user who is being discarded, for example, the edge node 31 accommodating the user A is replaced with a device having a long queue, or the edge node. Discarding can be avoided by replacing the 31 network interface cards with network interface cards with long queues.

Furthermore, by making the limit virtual queue length for each user in the core node longer than the maximum value that can be taken as the actual queue length in the edge node (hereinafter referred to as the limit actual queue length), it is possible to grasp the degree of overflow. it can. For example, if the limit virtual queue length is set to about 4 times the limit actual queue length and the maximum value of the virtual queue length reaches twice the limit actual queue length, discarding is avoided by increasing the actual queue length. In that case, it turns out that it is necessary to double the number.

The degree of overflow can also be grasped by simultaneously calculating the virtual queue length for a plurality of transmission bands for each user in the core node. For example, when the actual transmission band of a user's line connected to the edge node is 100 Mbps, the edge node calculates the virtual queue length for the user at the same time at 100 Mbps, 200 Mbps, and 400 Mbps, and the virtual queue length is limited at 100 Mbps. When the virtual queue length is reached and it is not reached at 200 Mbps or more, it is found that it is necessary to increase the speed to 200 Mbps when avoiding discarding by accelerating the line to the user of the edge node. However, since it is not necessary to provide a real queue for each user, the memory required to provide a real queue is not required, and the virtual queue of many users, for example, about 100,000 users. You can simulate the length.
By simulating the queue length of the queue for the user line in the edge node to which the user line is connected at the core node in the network, the core node concentrates on which edge node needs to be enhanced such as extending the queue. Can be grasped.
This eliminates the need to memorize the maximum queue length and the like at the edge node and provide an interface that can aggregate from the aggregation device, and it is possible to use an inexpensive device for the edge node that requires a large amount as compared with the core node.
Furthermore, when the virtual queue length for each user simulated by the core node reaches the limit virtual queue length, packets destined for the user are stacked in one of a plurality of queues that are shaped within the core node. , Even if the number of queues to be shaped is smaller than the total number of accommodated users, for example, 100,000 users, for example, 10,000, 10,000 users may overflow the actual queue at the edge node. If it is within the range, it is possible to avoid the overflow of packets at the edge node for all users.

Next, the second embodiment will be described.
FIG. 16 is a diagram illustrating an operation of avoiding discarding the actual queue by loading only the overflowed user packets in the shaper queue in the core node when the virtual queue in the core node 1 overflows.
In the second embodiment, a configuration is shown in which a plurality of shaper queues are provided in the core node, and packets destined for users who are not registered in the shaper queue for each user are distributed to a plurality of shaper queues by hash and stacked.
In the core node 1 shown in FIG. 8 used in the description of the first embodiment, all packets destined for users not registered in the shaper queue for each user are loaded in the queue 113 having no output bandwidth control unit.

On the other hand, the core node 1 of FIG. 16 for explaining the second embodiment is characterized in that it includes a plurality of shaper cues. In the second embodiment, the packets destined for the user registered in the shaper queue for each user are loaded in one of the plurality of shaper queues 1110 for each user, and the packets destined for the other users are stored in the distribution unit 1131 by the user. The packet is distributed to a plurality of shaper queues 11320 according to the hash value obtained by the hash function using the identification information as a hash key, and is loaded on one of the plurality of shaper queues 11320.

In FIG. 16, the passage route of the packet destined for user A 311 is an example of the passage route of the packet destined for the user registered in the shaper queue for each user. The packet destined for user A is loaded in the shaper queue 111 for each user, and the output bandwidth control unit 112 performs output bandwidth control and outputs the packet. The passage route of the packet to user B 321 is an example of the passage route of the packet to other users. Packets destined for user B are loaded in queue 1132, which is one of a plurality of shaper queues 11320 distributed by hash by distribution unit 1131, and output band control is performed by output band control unit 1133 and output.

Since only packets destined for user A are loaded in the queue 111, the bandwidth controlled by the output bandwidth control unit 112 of the queue 111 is determined according to the bandwidth of the line 3111 connected to the user A. On the other hand, since the queue 1132 is loaded with packets not only for user B but also for a plurality of users, the bandwidth controlled by the output bandwidth control unit 11320 of queue 1132 should correspond to the bandwidth of the line 3211 connected to user B. can not decide. The band controlled by the output band control unit 11320 of the queue 1132 is generally set to a value larger than the band of the line 3211 connected to the user B.

According to the second embodiment, it takes time to determine the necessity of bandwidth control for each user. Therefore, when the traffic to a certain user suddenly increases, it is registered in the shaper queue for each user after the traffic increases. In the meantime, it is possible to prevent the overflow that may occur in the actual queue of the edge node. Alternatively, the overflow can be reduced.
In addition, when there is only one shaper queue as in the first embodiment, when the amount of traffic destined for a certain user changes, it has an effect such as an increase in fluctuations with respect to the traffic destined for another user. In the second embodiment, since the queue is divided by the hash, when the amount of traffic destined for one user changes, the traffic destined for the user loaded in the other queue can be prevented from being affected.

The present invention is not limited to the above-described embodiment, and various modifications are possible. Further, the present invention is not limited to the above-described embodiment, and it goes without saying that the present invention can be carried out in various embodiments. Further, each of the above configurations, functions, processes and the like may be realized by hardware or software in part or all of them.

1 ... Core node, 11 ... Packet relay unit, 12 ... Virtual queue length calculation unit, 13 ... CPU unit, 14 ... Packet-related data, 15 ... Access interface 21 ... Distribution server, 22 ... Distribution server, 31 ... Edge node, 32 ... Edge node, 311 ... User A, 321 ... User B, 3112 ... Real queue for user A, 3212 ... Real queue for user B, 121 ... User identification search unit, 122 ... Virtual queue length update unit, 1221 ... Virtual queue Length table, 1222 ... Maximum virtual queue length Read clear mode register, 142 ... Packet length, 12213 ... Virtual queue length, 12215 ... Output band, 12216 ... Limit virtual queue length, 111 ... Shaper queue, 112 ... Output band control unit, 113 ... Queue without output band control unit, 114 ... Queue output arbitration unit, 115 ... Shaper queue, 116 ... Output band control unit, 117 ... Shaper queue search table, 23 ... Virtual queue monitoring server, 1131 ... Distribution unit 11,10 ... Multiple shaper queues for each user, 11320 ... Multiple shaper queues for determining the distribution destination by a hash function.

Claims (14)

A communication device that transmits a packet addressed to a user delivered from a server to a network and also transmits a packet received via the network to a server, and the packet addressed to the user is from an edge device located at the edge of the network. It is sent to the user and
Communication equipment
A packet that has a number of first shaper queues smaller than the total number of users of the user, relays packets sent and received between the server and the network, controls the bandwidth of the packet, and extracts packet-related information contained in the packet. With the relay part
Based on the packet-related information and the band band information of the line between the edge device and the user, the virtual queue length, which is an estimated value of the queue length of the transmission queue addressed to the user in the edge device, is calculated and held for each user. Virtual queue length calculation unit and
The packet relay unit is determined by determining the necessity of bandwidth control for each user based on the virtual queue length and predetermined conditions, and assigning the first shaper queue to each user who is determined to require bandwidth control. A communication device having a control unit that controls the packet relay unit so as to control the bandwidth of a packet addressed to the user. The communication device according to claim 1, wherein the packet relay unit extracts a packet length and a destination address as packet-related information, and in the virtual queue length calculation unit, each user is based on the destination address. A unique identifier is set, the user is identified using the user identifier, and the virtual queue length is managed using the packet length.
The control unit sets the number of the first shaper queue and the user identifier assigned to the user determined to require bandwidth control in the packet relay unit.
A communication device in which the packet relay unit controls bandwidth for each user based on the settings. In the communication device according to claim 1, the determination as to whether or not bandwidth control is necessary for each user is made based on the maximum virtual queue length, which is the maximum value of the virtual queue length since it was cleared last time. A communication device characterized by. In the communication device according to claim 1, it is assumed that the determination of the necessity of bandwidth control for each user exceeds the limit virtual queue length, which is the maximum value that the virtual queue length can take, and is discarded. A communication device characterized in that a judgment is made based on a value obtained from the number of packets. In the communication device according to claim 1, the determination of the necessity of bandwidth control for each user is based on a value obtained by multiplying the limit virtual queue length, which is the maximum value that the virtual queue length can take, by a preset threshold value. A communication device characterized by making a judgment. The communication device according to claim 2.
The packet relay unit has a shaper queue search table in which the shaper queue, the user identifier, and the destination address range of the user identifier are associated with each other.
The virtual queue length calculation unit further has information on the limit virtual queue length, which is the maximum value that the virtual queue length can take, for each user identified by the user identifier.
The control unit adds or deletes a user assigned to the first shaper queue based on the virtual queue maximum occupancy rate, which is the ratio of the maximum virtual queue length to the shaper queue search table and the limit virtual queue length. A communication device characterized by performing on a search table. The communication device according to claim 1.
The packet relay unit further has a second plurality of shaper queues different from the first shaper queue.
The packet relay unit
When the control unit receives a packet addressed to a user to whom the first shaper queue is not assigned, the communication is characterized in that the packet is stored in any one of the second plurality of shaper queues. apparatus. It is a communication method in a communication device that transmits a packet addressed to a user delivered from a server to a network and also transmits a packet received via the network to a server.
The communication device has a packet relay unit having a number of first shaper queues smaller than the total number of users of the user.
The packet addressed to the user is transmitted to the user from an edge device located at the edge of the network.
The packet-related information contained in the packet is extracted, and the packet-related information is extracted.
Based on the packet-related information and the bandwidth information of the line between the edge device and the user, the virtual queue length, which is an estimated value of the queue length of the transmission queue addressed to the user in the edge device, is calculated and held for each user. ,
By determining the necessity of bandwidth control for each user based on the virtual queue length and predetermined conditions, and assigning any of the first shaper queues to each user determined to require bandwidth control, A communication method characterized in that bandwidth control of a packet addressed to the user is performed. The communication method according to claim 8, wherein the packet length and the destination address are extracted as packet-related information, a unique identifier is set for each user based on the destination address, and the user is used by using the user identifier. The user is identified and the virtual queue length is managed using the packet length, and the user is assigned based on the first shaper queue number and the user identifier assigned to the user who is determined to require bandwidth control. A communication method characterized in that band control is performed for each band. In the communication method according to claim 8, the determination of the necessity of bandwidth control for each user is made based on the maximum virtual queue length, which is the maximum value of the virtual queue lengths since the last time it was cleared. A communication method characterized by. In the communication method according to claim 8, it is assumed that the determination of the necessity of bandwidth control for each user exceeds the limit virtual queue length, which is the maximum value that the virtual queue length can take, and is discarded. A communication method characterized in that a judgment is made based on a value obtained from the number of packets. In the communication method according to claim 8, the determination of the necessity of bandwidth control for each user is based on a value obtained by multiplying the limit virtual queue length, which is the maximum value that the virtual queue length can take, by a preset threshold value. A communication method characterized by making a judgment. The communication method according to claim 9.
A shaper queue search table that associates the shaper queue with the user identifier and the destination address range of the user identifier.
For each user identified by the user identifier, information on the limit virtual queue length, which is the maximum value that the virtual queue length can take, is further provided.
Additions and deletions of users assigned to the first shaper queue are added to the shaper queue search table based on the virtual queue maximum occupancy rate, which is the ratio of the maximum virtual queue length to the shaper queue search table and the limit virtual queue length. A communication method characterized by performing. The communication method according to claim 8.
The packet relay unit of the communication device has a second plurality of shaper queues different from the first shaper queue.
A communication method characterized in that when a packet addressed to a user to whom the first shaper queue is not assigned is received, the packet is stored in any one of the second plurality of shaper queues.

Images