JP6850618B2 - Relay device and relay method - Google Patents

Description

The present invention relates to a relay device that relays a packet group.

In recent years, so-called streaming technology, which delivers electronic contents such as video and audio in real time via a network, has become widespread. When such streaming is realized via the Internet, the same line is shared among a large number of users. Therefore, the cost per band can be kept low. On the other hand, communication quality (Quality of Service) is ensured by suppressing the amount of communication packets discarded. For this reason, a technique is known in which a relay device is equipped with a shaper and a large amount of data transmitted at one time is leveled at a constant rate and transferred (see, for example, Patent Document 1 below).

In Patent Document 1, flow conditions corresponding to users and services are set in advance as configuration definition information, and received communication packets are temporarily stored in a queue prepared for each flow, and predetermined shaping conditions (transmission band) are provided. Disclose a technology for transferring communication packets based on (value and priority).

Patent Document 2 discloses a technique for leveling communication by converting an input flow into a hash value by a constant hash function based on packet header information and the like, and automatically allocating the input flow to a queue corresponding to the hash value. Further, Patent Document 2 discloses a technique for leveling again by changing the hash function when disposal occurs due to collision of hash values or the like.

Japanese Unexamined Patent Publication No. 11-346246 Japanese Unexamined Patent Publication No. 2009-182546

However, in the technique of Patent Document 1, it is difficult to deal with a large number of users and services because it is necessary to set a flow for each user and service. For example, when setting a flow for every millions of users, it is difficult to secure the same number of queues as the number of users. Even if millions of queues are secured, the enormous capacity of the storage medium or the excessive load caused by the distribution processing of received packets causes a significant cost increase, which is not realistic. Moreover, it is inefficient to secure a queue of users with a low active rate.

Further, in the technique of Patent Document 2, collision of hash values may occur even after the hash function is changed. Further, as a result, there is a possibility that the order of the transmitted packets is frequently reversed due to frequent changes in the hash function.

An object of the present invention is to improve communication quality while suppressing an increase in storage capacity.

The relay device and the relay method which are one aspect of the invention disclosed in the present application are the relay device and the relay method for relaying the packet group, and the relay device belongs to the first flow and specifies the first flow. 1 A second including one or more first queues for storing a first packet containing specific information, and a second specific information converted into the same hash value as the hash value converted from the first specific information and specifying a second flow. A queue group having one or more second queues for storing two packets, a first storage unit for storing first correspondence information in which the hash value is associated with the first queue, and the second specific information. It has a second storage unit that stores the second correspondence information in which the second queue is associated with the second queue, and the relay device receives the packet group by the receiving unit and the receiving unit by the distribution unit. It is determined whether or not the specific information included in each packet of the packet group received by the packet and specifying the flow to which the packet belongs matches the second specific information, and if they match, the second correspondence is performed. With reference to the information, the packet is distributed to the second queue as the second packet, and if they do not match, the specific information is converted into the hash value, and the packet is referred to the first correspondence information to be the first packet. The packet group is distributed to the first queue as one packet, and the transmitting unit transmits the packet group distributed to the first queue and the second queue by the distribution unit.

According to a typical embodiment of the present invention, it is possible to improve the communication quality while suppressing the increase in the storage capacity. Issues, configurations and effects other than those described above will be clarified by the description of the following examples.

FIG. 1 is an explanatory diagram showing an example of a communication system. FIG. 2 is a block diagram showing a configuration example of a router. FIG. 3 is an explanatory diagram showing a detailed configuration example of the distribution unit. FIG. 4 is an explanatory diagram showing a detailed configuration example of the learning unit. FIG. 5 is a flowchart showing an example of the distribution processing procedure by the distribution unit. FIG. 6 is a flowchart showing an example of a learning processing procedure by the learning unit. FIG. 7 is a flowchart showing an example of the learning result processing procedure by the learning result processing unit. FIG. 8 is a flowchart showing a detailed processing procedure example of the preliminary area update processing 1 by the learning result processing unit. FIG. 9 is a flowchart showing a detailed processing procedure example of the preliminary area update processing 2 by the learning result processing unit.

<System configuration example>
FIG. 1 is an explanatory diagram showing an example of a communication system. The communication system 100 is a network system that provides users with contents such as moving images by streaming. As shown in the figure, the communication system 100 includes servers SV1 and SV2, routers RT1 to RT5 as relay devices, and personal computers PC1 to PC3 as terminals.

The servers SV1 and SV2 are servers that provide contents and store various contents. The servers SV1 and SV2 are directly connected to the router RT1. The number of servers connected to the router RT1 is shown as two for the sake of simplicity, but may be three or more.

Router RT1 is connected to network NT. The network NT is, for example, the Internet. However, the type of network NT is not particularly limited, and may be a WAN (Wide Area Network) such as a dedicated line, or a LAN (Local Area Network).

Further, the router RT2 is connected to the network NT. The servers SV1, SV2, routers RT1 and RT2 are installed, for example, by an ISP (Internet Service Provider) that provides a content distribution service. The router RT2 is connected to the personal computers PC1 to PC3 via the routers RT3 to RT5 installed in the homes A to C, respectively. Another relay device or line concentrator may be interposed between the routers RT2 and the routers RT3 to RT5.

The personal computers PC1 to PC3 are general-purpose personal computers. The terminals connected to the routers RT3 to RT5 may be any computer that can use the contents distributed by the servers SV1 and SV2, and may be, for example, a network-compatible television. Further, a plurality of terminals may be connected to the routers RT3 to RT5. Further, the router RT2 is assumed to be connected to the networks of homes A to C for the sake of simplicity, but in reality, it is connected to the networks of thousands or tens of thousands of homes, for example.

In the communication system 100, when the servers SV1 and SV2 receive the content usage request from the personal computers PC1 to PC3, the server SV1 and SV2 deliver the requested content to the personal computers PC1 to PC3 of the usage request source. When the routers RT1 and RT2 receive the packets delivered by the servers SV1 and SV2, the routers RT1 and RT2 temporarily store the packets in a buffer, level them, and transfer them to the personal computers PC1 to PC3. The personal computers PC1 to PC3 receive the transferred packet via the routers RT3 to RT5 and use it in the pre-installed application.

<Router configuration example>
FIG. 2 is a block diagram showing a configuration example of a router. Since the routers RT1 and RT2 shown in FIG. 1 have the same configuration in this embodiment, the configuration of the router RT1 will be described below unless otherwise specified. The router RT1 includes a receiving unit 201, a distribution unit 202, a learning unit 203, a buffer 204, a band control unit 205, a transmitting unit 206, and a learning result processing unit 207.

The receiving unit 201 is an interface for receiving the packet P from the outside. In this embodiment, the receiving unit 201 includes two physical ports (not shown). Servers SV1 and SV2 are connected to these two physical ports. The receiving unit 201 receives the packet P from the servers SV1 and SV2 and outputs the packet P to the distribution unit 202. The receiving unit 201 may include three or more physical ports. In this case, the receiving unit 201 may be connected to three or more servers.

The distribution unit 202 describes the packet P received by the reception unit 201 as queue Q1 to Qn secured in the buffer 204 (n is an arbitrary integer of 2 or more. When queues Q1 to Qn are not distinguished, it is expressed as queue Q. .) Sort to one of them. Specifically, for example, the distribution unit 202 converts the header information included in the header h of the packet P into a hash value by a predetermined hash function, and converts the packet P into a queue Q corresponding to the hash value. Sort.

The distribution unit 202 has a determination unit 221 and a hash value table 222. The determination unit 221 gives the header information included in the header h of the packet P to the hash function and converts it into a hash value. Here, the information given to the hash function is referred to as a "hash conversion key k". The header information used as the hash conversion key k includes, for example, one or more fields of a layer 2 header such as a MAC header, a layer 3 header such as an IPv4 header and an IPv6 header, and a layer 4 header such as a TCP header and a UPC header. May be used. The selection of the field may be fixed to the device or may be specified by the device operator. The hash conversion key k is flow-specific information that identifies the flow F to which the packet P belongs. In this example, the source IP address is used as an example of the header information in order to simplify the explanation.

The determination unit 221 holds the hash conversion key kb of the packet P belonging to the normal flow F (hereinafter, normal flow Fa) and the collision flow F (hereinafter, collision flow Fb), and the packet P from the receiving unit 201. It is determined whether or not it matches the hash conversion key k of.

Collision is a collision of hash values. That is, the hash value H obtained from the hash conversion key k included in the packet P belonging to the normal flow Fa (hereinafter, the hash conversion key ka is expressed) and the hash conversion key k included in the packet P belonging to the collision flow Fb (hereinafter referred to as the hash conversion key ka). , The hash conversion key kb is expressed. The hash value H obtained from kb ≠ ka) is the same value. That is, the normal flow Fa and the collision flow Fb are flows F through which packets P having different hash conversion keys k but having the same hash value flow.

In the case of a mismatch, the determination unit 221 converts the hash conversion key k of the packet P from the reception unit 201 into the hash value H as described above. On the other hand, if they match, the determination unit 221 does not hash-convert the hash conversion key k of the packet P from the reception unit 201, and queues Qb in the spare area prepared separately from the queue Qa that queues the normal flow Fa. The packet P is sent to.

The hash value table 222 is a table in which the hash value H obtained from the determination unit 221 and the queue Q of the buffer 204 are associated with each other. For example, when the hash value H is 1, the corresponding queue Q stores information such as Q1. The distribution unit 202 is realized by, for example, an integrated circuit. The hash value table 222 is stored in the memory in the integrated circuit.

The learning unit 203 monitors the flows F1 to Fn (referred to as flow F when the flows F1 to Fn are not distinguished) distributed to the plurality of queues Q corresponding to the distribution unit 202, and which flow F Learn which queue Q the packet is assigned to. Specifically, for example, the learning unit 203 monitors the flow rate of the packet P for each queue Q, and based on the hash conversion key k, the packet P of a plurality of flows F is stored in the same queue Q. To detect. When the learning unit 203 detects that packets P of a plurality of flows F are stored in the same queue Q, the learning unit 203 learns any one of the plurality of flows F F as a normal flow Fa, and the remaining flows. F is learned as a collision flow Fb.

The learning unit 203 sends a packet P to the queue Q according to the distribution by the distribution unit 202, and outputs the learning result to the learning result processing unit 207. The learning result includes, for example, the hash conversion key k of the packet P belonging to the collision flow Fb. The learning unit 203 is realized, for example, by causing a processor to execute a program stored in an integrated circuit or a memory.

The buffer 204 is a storage area for temporarily storing the packet P received by the receiving unit 201, and has a plurality of queues Q1 to Qn. In this embodiment, for example, the same storage capacity is assigned to each of the queues Q1 to Qn. As described above, each of the received packets P is stored in the plurality of queues Q1 to Qn according to the distribution result determined by the distribution unit 202. At the time of congestion, the packet P exceeding the capacity of the queues Q1 to Qn is discarded. For example, when the capacity of the packet P is already accumulated in the queue Q, the packet P newly input to the buffer 204 is discarded without being stored in the queue Q.

The band control unit 205 controls the band for each of the queues Q1 to Qn, and outputs the packet P accumulated in the queues Q1 to Qn to the transmission unit 206. Specifically, for example, the bandwidth control unit 140 performs scheduling based on a preset bandwidth control rule, sequentially reads packets P from queues Q1 to Qn, and outputs them to the transmission unit 206. The band control unit 205 is realized by, for example, an integrated circuit.

The transmission unit 206 is an interface for transmitting the packet P to the outside. In this embodiment, scheduling is performed by the equal allocation method. Specifically, for example, the packets P accumulated in the queues Q1 to Qn are read out at an equal frequency among the queues Q1 to Qn. However, the scheduling algorithm is not particularly limited, and may be, for example, a weighted equal allocation method. Further, a minimum band guarantee method, a maximum band regulation method, a surplus band weighting distribution method, or the like may be used.

The learning result processing unit 207 receives the learning result from the learning unit 203 and executes various processes. For example, the learning result processing unit 207 reflects the learning result in the distribution unit 202. Specifically, for example, the learning result processing unit 207 controls the distribution unit 202 to associate the queue Q of the spare area described above with the hash conversion key kb included in the learning result. As a result, when the distribution unit 202 receives the packet P including the hash conversion key, it sends it to the corresponding queue Q without converting it into a hash value.

Further, the learning result processing unit 207 controls the distribution unit 202 to delete the hash conversion key kb associated with the queue Q in the spare area, for example, by aging. The learning result processing unit 207 controls the distribution unit 202 to use the hash conversion key kb associated with the queue Q in the spare area, for example, to notify the learning unit 203 that the flow rate of the collision flow F is equal to or less than a predetermined amount. It may be deleted when it is obtained from.

Further, the learning result processing unit 207 transmits the learning result from the transmitting unit 206 to the terminal of the operator by using a message such as a trap or a system log, or displays the learning result on a display unit (not shown). Notify the operator of the learning result. When the learning result is registered or deleted, the learning result processing unit 207 may notify the operator of the information. The learning result processing unit 207 is specifically realized, for example, by causing the processor to execute a program that executes the processing of the learning result processing unit 207.

<Detailed configuration example of distribution unit 202>
FIG. 3 is an explanatory diagram showing a detailed configuration example of the distribution unit 202. The distribution unit 202 has a determination unit 221 and a hash value table 222. The determination unit 221 has a conversion unit 301 and a learning result table 302. The conversion unit 301 converts the hash conversion key ka of the packet P into the hash value H. The learning result table 302 is a table in which the spare area hash conversion key 321 and the spare area queue 322 are associated with each other. The spare area hash conversion key 321 is a storage area for storing the hash conversion key kb of the packet P belonging to the collision flow F. The spare area queue 322 is a storage area for storing the identification information of the queue Qb to which the collision flow F is sent.

The entry in the learning result table 302 is deleted by the learning result processing unit 207 after a certain period of time has elapsed due to aging. Further, the entry in the learning result table 302 may be deleted by the learning result processing unit 207 when the flow rate of the collision flow F corresponding to the spare area hash conversion key 321 becomes a predetermined amount or less. The learning result table 302 is stored in the memory in the integrated circuit.

The hash value table 222 has a hash value 303, a queue 304, and an in-use flag 305. The hash value 303 is a storage area for storing hash values H1 to Hi (i is an integer smaller than n. n is the total number of queues Q. When hash values H1 to Hi are not distinguished, it is expressed as hash value H). is there.

The queue 304 is a storage area for storing the identification information of the queues Qa1 to Qai (referred to as queue Qa when the queues Qa1 to Qai are not distinguished). Specifically, for example, the queues Qa1 to Qai correspond to the hash value H, respectively. The queues Qa1 to Qi are, for example, queues Q1 to Qi. The queue Qa is referred to as a normal area.

The queues Qb1 to Qbj (j is an integer to be n−i. When the queues Qb1 to Qbj are not distinguished, they are referred to as queues Qb) are not associated with the hash value 303. The queues Qb1 to Qbj are, for example, queues Qi + 1 to Qn. The queue Qb is referred to as a spare area. The packet P is sent to the queue Qa in the normal area, but when the collision is found, the packet P belonging to the collision flow Fb is sent to the queue Qb in the spare area. When the queues Qa and Qb are not distinguished, they are referred to as queue Q.

The in-use flag 305 is a storage area that stores the in-use flag for the entry of the spare area. When searching for the queue Q to send the packet P belonging to the collision flow Fb, the determination unit 221 refers to the in-use flag 305 of the spare area. The values fb1 to Fbj of the in-use flag 305 (when not specified as the values fb1 to Fbj of the in-use flag 305, they are expressed as the value fb of the in-use flag 305 or the in-use flag fb) are two values of ON or OFF. Is. "ON" means in use and "OFF" means not in use. When sending the packet P belonging to the collision flow F to the queue Q, the determination unit 221 identifies the queue 304 (Qb) in which the value fb of the in-use flag 305 is OFF, and the packet P is sent to the queue 304 (Qb). Is sent.

<Detailed configuration example of learning unit 203>
FIG. 4 is an explanatory diagram showing a detailed configuration example of the learning unit 203. The learning unit 203 has learning mechanisms LRNt1 to LRNtm (m is an integer satisfying 0 ≦ m ≦ n. When learning mechanisms LRNt1 to LRNtm are not distinguished, it is referred to as learning mechanism LRNt). For example, one learning mechanism LRNt corresponds to a queue Q to which one or more flows F are distributed. In FIG. 4, since the flows F1 and F2 are distributed to the same queue Q, the learning mechanism LRNt1 monitors and learns the flows F1 and F2. Similarly, the learning mechanism LRNt2 monitors and learns the flows F3 and F4. The learning mechanism LRNtm-1 monitors and learns the flows Fn-3 and Fn-2. The learning mechanism LRNtm monitors and learns the flows Fn-1 and Fn.

Further, in FIG. 4, for example, in order to reduce the load on the learning unit 203, the learning unit 203 describes the processing by the learning mechanism LRNt1 to LRNtm as time t1 to tm (when time t1 to tm is not distinguished, time t. ), Execute in time division. When the processing at the time tm is completed, the process is repeated again from the learning mechanism LRNt1 corresponding to the time t1. It should be noted that the operator of the router RT1 may be able to fixedly specify how many flows F one learning mechanism LRNt is in charge of, and how many hours t1 to tm are set, and the learning unit. It may be automatically determined in conjunction with the CPU usage rate in 203. For example, the learning unit 203 reduces the number of flows in charge of one learning mechanism LRNt or shortens the time t when the CPU usage rate becomes high. On the contrary, when the CPU usage rate becomes low, the learning unit 203 increases the number of flows in charge of one learning mechanism LRNt or lengthens the time t.

The learning mechanism LRNt monitors whether or not the flow F to be learned collides. Specifically, for example, the learning mechanism LRNt determines that a collision has occurred when a plurality of packets P having different hash conversion keys k pass through the learning mechanism LRNt. For example, when a plurality of packets P having different hash conversion keys k pass through the learning mechanism LRNt, the learning mechanism LRNt determines the flow F corresponding to the hash conversion key k of the first-come-first-served packet P as the normal flow Fa, and the remaining flow. Determine F as collision flow Fb. After this learning, the normal flow Fa flows to the queue Qa that is the distribution destination, and the collision flow Fb flows to the queue Qb in the preliminary region.

By determining the normal flow Fa in the first-come-first-served packet P in this way, it is possible to speed up the learning process and reduce the processing load in the learning mechanism LRNt. The learning mechanism LRNt may determine the flow F, which is the maximum number of packets for each hash conversion key k within the time t, as the normal flow Fa, instead of the first-come-first-served packet P. As a result, it is possible to improve the accuracy of the learning process in the learning mechanism LRNt. When there are a plurality of flows F having the maximum number of packets, the normal flow Fa may be determined by the first-come-first-served packet P. As a result, it is possible to improve the efficiency of the learning process while improving the accuracy of the learning process as much as possible.

<Example of distribution processing procedure by distribution unit 202>
FIG. 5 is a flowchart showing an example of the distribution processing procedure by the distribution unit 202. When the distribution unit 202 acquires the packet P from the reception unit 201 (step S501), the distribution unit 202 receives at least a part of the header information fields (for example, transmission) included in the header h of the acquired packet P. The original IP address) is acquired as the hash conversion key k (step S502).

The distribution unit 202 determines whether or not the acquired packet P is the packet P of the collision flow F (step S503). Specifically, for example, the distribution unit 202 refers to the spare area hash conversion key 321 of the learning result table 302, and determines whether or not it matches the hash conversion key k of the acquired packet P. If the hash conversion key k of the acquired packet P is matched, the acquired packet P becomes the packet P of the collision flow F (step S503: Yes). Therefore, the process proceeds to step S507. On the other hand, if the hash conversion key k of the acquired packet P does not match, the acquired packet P becomes the packet P of the normal flow Fa (step S503: No). Therefore, the process proceeds to step S504.

In step S504, the distribution unit 202 gives the hash conversion key k acquired in step S502 to the hash function to calculate the hash value H (step S504). Then, the distribution unit 202 refers to the entry group of the normal area of the hash value table 222, and identifies the queue Qa corresponding to the hash value 303 (calculated hash value H) from the queue 304. The distribution unit 202 outputs the packet P acquired in step S501 to the specified queue Qa (step S506). As a result, the distribution process of the distribution unit 202 is completed.

Further, in step S507, the distribution unit 202 specifies the queue Qb corresponding to the hash conversion key kb acquired in step S503 from the learning result table 302 (step S507). The distribution unit 202 outputs the packet P acquired in step S501 to the specified queue Qb (step S506). As a result, the distribution process of the distribution unit 202 is completed.

<Example of learning process procedure by the learning department>
FIG. 6 is a flowchart showing an example of a learning processing procedure by the learning unit 203. The learning unit 203 starts the time division processing by the learning mechanism LRNt (step S601). Steps S602 to S604 are learning processes by each learning mechanism LRNt. The learning mechanism LRNt learns the passing flow F within the time t (step S602). The learning mechanism LRNt determines whether or not a plurality of flows F have passed (step S603). When a plurality of flows F have not passed (step S603: No), the learning process of the learning mechanism LRNt ends, and the process proceeds to step S605.

On the other hand, when a plurality of flows F have passed (step S603: Yes), the learning mechanism LRNt determines the hash conversion key ka of the packet P of the normal flow Fa depending on the maximum number of packets for each first-come-first-served packet P or the hash conversion key k. And the hash conversion key kb of the collision flow Fb (step S604). As a result, the learning process of the learning mechanism LRNt is completed, and the process proceeds to step S605.

When the learning process by each learning mechanism LRNt is completed, the learning unit 203 is not currently used in the hash conversion key kb of the collision flow Fb specified by each learning mechanism LRNt and the spare area of the hash value table 222 ( The in-use flag 305 is OFF) and the queue Qb (identification information) is output to the learning result processing unit 207 as a learning result (step S605). As a result, the learning process by the learning unit 203 ends.

<Example of learning result processing procedure by learning result processing unit 207>
FIG. 7 is a flowchart showing an example of the learning result processing procedure by the learning result processing unit 207. The learning result processing unit 207 acquires the hash conversion key kb which is the learning result from the learning unit 203 (step S701). The learning result processing unit 207 takes out the hash conversion key kb and the queue Qb (identification information) one by one from the learning result, associates them with each other, and generates a set (step S702). The learning result processing unit 207 determines whether or not there is an unselected set (step S703). When there is an unselected set (step S703: Yes), the learning result processing unit 207 selects one unselected set (step S704).

The learning result processing unit 207 registers the selected set in the learning result table 302 (step S705). The learning result processing unit 207 turns on the value fb of the in-use flag 305 in the hash value table 222 for the selected set of queues Qb (step S706). Then, the process returns to step S703. If there is no unselected pair in step S703 (step S703: No), the learning result processing ends.

<Preliminary area update process 1 by the learning result processing unit 207>
Next, the preliminary area update process 1 by the learning result processing unit 207 will be described. In the preliminary area update process 1, the learning result processing unit 207 controls the distribution unit 202 to delete the entry in the learning result table 302 by aging. As a result, the queue Qb of the spare region can be allocated from the old collision flow Fb to the latest collision flow Fb. A detailed processing procedure example will be described below.

FIG. 8 is a flowchart showing a detailed processing procedure example of the preliminary area update processing 1 by the learning result processing unit 207. First, the learning result processing unit 207 waits for the completion of registration of the hash conversion key kb in step S706 (step S801: No). When completed (step S801: Yes), the learning result processing unit 207 starts timing the registered hash conversion key kb (step S802). Then, the learning result processing unit 207 waits for the elapse of a predetermined time (step S803: No). When the predetermined time has elapsed (step S803: Yes), the learning result processing unit 207 deletes the entry of the hash conversion key kb registered in step S706 from the learning result table 302 (step S804).

Then, the learning result processing unit 207 turns off the value fb of the in-use flag 305 of the spare area corresponding to the hash conversion key kb deleted in step S804 (step S805), and the spare area update process 1 ends. As a result, the queue Qb of the spare region can be allocated from the old collision flow Fb to the latest collision flow Fb. Further, since it is not necessary to monitor the flow rate of the collision flow F, the processing load in the learning processing unit can be reduced.

<Preliminary area update process 2 by the learning result processing unit 207>
Next, the preliminary area update process 2 by the learning result processing unit 207 will be described. In the preliminary area update process 2, the learning result processing unit 207 monitors the flow rate of the collision flow F from the learning unit 203, controls the distribution unit 202, and deletes the entry of the collision flow F that is equal to or less than a predetermined amount. .. As a result, the larger the flow rate, the more the cue Qb in the spare region can be allocated to the collision flow F. A detailed processing procedure example will be described below.

FIG. 9 is a flowchart showing a detailed processing procedure example of the preliminary area update processing 2 by the learning result processing unit 207. First, the learning result processing unit 207 acquires the flow rate of the collision flow F from the learning mechanism LRNt that sends the collision flow F to the queue Qb in the preliminary region (step S901). The learning result processing unit 207 determines whether or not there is a collision flow Fb whose acquired flow rate is equal to or less than a predetermined amount (step S902). The predetermined amount is set in advance by the operator.

When there is no collision flow Fb whose acquired flow rate is equal to or less than a predetermined amount (step S902: No), the preliminary area update process 2 ends. On the other hand, when there is a collision flow Fb whose acquired flow rate is equal to or less than a predetermined amount (step S902: Yes), the learning result processing unit 207 deletes an entry corresponding to the collision flow Fb of a predetermined amount or less from the learning result table 302 (step S902: Yes). Step S903). The learning result processing unit 207 identifies the spare area queue 322 (Qb) corresponding to the deleted hash conversion key kb, and turns off the value fb of the in-use flag 305 of the spare area corresponding to the specified queue Qb (step S904). ), The spare area update process 2 ends. As a result, the larger the flow rate, the more the cue Qb in the spare region can be allocated to the collision flow Fb. Therefore, highly accurate distribution can be realized.

In FIG. 9, the entry corresponding to the collision flow Fb of the predetermined amount or less is deleted from the learning result table 302 (step S903), but when the collision flow Fb is continuously equal to or less than the predetermined amount for a certain period of time, the entry is deleted. The entry corresponding to the collision flow Fb may be deleted from the learning result table 302. As a result, more accurate distribution can be realized.

As described above, the router RT1 of the present embodiment is a relay device that relays the packet group, and has a queue group (queue Qa) having one or more first queues (queue Qa) and one or more second queues (queue Qb). It has a buffer 204), a first correspondence information (hash value table 222), and a second correspondence information (learning result table 302).

The first queue stores the first packet. The first packet belongs to the first flow (normal flow Fa) and includes the first specific information (hash conversion key ka) that specifies the first flow. The second queue stores the second packet. The second packet is converted into the same hash value as the hash value converted from the first specific information, and includes the second specific information (hash conversion key kb) that specifies the second flow (collision flow Fb).

The first correspondence information is stored in association with the hash value and the first queue (Qa) that specifies the first queue. The second correspondence information is stored in association with the second specific information and the second queue (Qb) that specifies the second queue.

The router RT1 has a receiving unit 201 for receiving a packet group, a distribution unit 202, and a transmitting unit 206. The distribution unit 202 includes whether or not the specific information (hash conversion key k) included in each packet of the packet group received by the reception unit 201 and specifying the flow to which the packet belongs matches the second specific information. Is determined by the determination unit 221. If they match, the distribution unit 202 refers to the second correspondence information and distributes the packet as a second packet to the second queue. If they do not match, the distribution unit 202 converts the specific information into a hash value and converts the first correspondence information into a hash value. With reference to the packet, the packet is distributed to the first queue as the first packet.

The transmission unit 206 transmits the packet group distributed to the first queue and the second queue by the distribution unit 202.

As described above, in the router RT1 described above, when a collision occurs in a plurality of flows F, the collision flow Fb is saved in the queue Q of the spare area, so that it is not necessary to set the flow for each user or service. It is not necessary to secure the queue Q of the user with a low active rate. As a result, it is possible to suppress an enormous increase in the capacity of the storage medium. Further, when a collision occurs in a plurality of flows F, the collision flow Fb is saved in the queue Q of the spare area, so that the collision of the hash value can be avoided, the order reversal of the packet P is suppressed, and as a result, the collision causes. It is possible to suppress the influence of congestion on the entire traffic and ensure communication quality.

Further, the router RT1 may have a learning unit 203 and a learning result processing unit 207. The learning unit 203 detects that, as a result of distribution by the distribution unit 202, a plurality of flows F to which packets having different specific information belong are output to the same first queue, and among the plurality of flows F, a specific flow F is specified. The flow is determined as the first flow, and the flow other than the specific flow is determined as the second flow. The learning result processing unit 207 acquires the learning result including the specific information included in the packets belonging to the other flow determined by the learning unit 203 in the second flow and the identification information of the second queue of the second correspondence information. , Register in the second correspondence information.

As a result, the router RT1 can autonomously update the second correspondence information, and can continue to secure the communication quality.

Further, the learning result processing unit 207 adds information (in-use flag 305) indicating that the second queue included in the learning result is in use to the second correspondence information, and acquires the learning result after the addition. If so, select the second queue to which no information indicating that it is in use is given.

As a result, it is possible to prevent the selection of the second cue in use, and it is possible to suppress collision in the second cue as well.

Further, the learning unit 203 may determine a specific flow including the first-come-first-served packet P among the plurality of flows F as the first flow. As a result, the learning process in the learning unit 203 can be speeded up and the processing load can be reduced.

Further, the learning unit 203 may determine a specific flow in which the number of flow rate packets is maximum within a certain period among the plurality of flows F as the first flow. As a result, it is possible to improve the accuracy of the learning process in the learning unit 203.

Further, the learning result processing unit 207 may delete the learning result for which a predetermined period has passed since the registration of the second correspondence information. As a result, the second cue can be assigned from the old collision flow Fb to the latest collision flow Fb.

Further, the learning result processing unit 207 may delete the learning result corresponding to the second flow in which the number of flow rate packets after being registered is equal to or less than the threshold value in the second correspondence information. As a result, the larger the flow rate, the more the cue Qb in the spare region can be allocated to the collision flow Fb. Therefore, efficient queuing can be realized.

Further, the learning result processing unit 207 can prompt the operator to change the hash conversion key kb and review the network configuration by notifying the operator of the learning result.

The present invention is not limited to the above-described embodiment, and includes various modifications and equivalent configurations within the scope of the attached claims. For example, the above-described examples have been described in detail in order to explain the present invention in an easy-to-understand manner, and the present invention is not necessarily limited to those having all the described configurations. Further, a part of the configuration of one embodiment may be replaced with the configuration of another embodiment. Further, the configuration of another embodiment may be added to the configuration of one embodiment. In addition, other configurations may be added, deleted, or replaced with respect to a part of the configurations of each embodiment.

Further, each of the above-described configurations, functions, processing units, processing means, etc. may be realized by hardware by designing a part or all of them by, for example, an integrated circuit, and the processor realizes each function. It may be realized by software by interpreting and executing the program to be executed.

Information such as programs, tables, and files that realize each function is recorded in a memory, hard disk, storage device such as SSD (Solid State Drive), or IC (Integrated Circuit) card, SD card, DVD (Digital Versaille Disc). It can be stored in a medium.

In addition, the control lines and information lines indicate those that are considered necessary for explanation, and do not necessarily indicate all the control lines and information lines necessary for implementation. In practice, it can be considered that almost all configurations are interconnected.

100 Communication system 140 Bandwidth control unit 201 Reception unit 202 Distribution unit 203 Learning unit 204 Buffer 205 Bandwidth control unit 206 Transmission unit 207 Learning result processing unit 221 Judgment unit 222 Hash value table 301 Conversion unit 302 Learning result table

Claims (8)

A relay device that relays packets
It is converted into the same hash value as the hash value converted from the first specific information and one or more first queues that belong to the first flow and store the first packet containing the first specific information that specifies the first flow. A queue group having one or more second queues for storing a second packet including a second specific information that specifies a second flow, and a queue group.
A first storage unit that stores first correspondence information in which the hash value and the first queue are associated with each other.
A second storage unit that stores the second correspondence information in which the second specific information and the second queue are associated with each other.
A receiver that receives the packet group and
It is determined whether or not the specific information included in each packet of the packet group received by the receiving unit and specifying the flow to which the packet belongs matches the second specific information, and if they match, the said The packet is distributed to the second queue as the second packet with reference to the second correspondence information, and if they do not match, the specific information is converted into the hash value, and the packet is referred to with reference to the first correspondence information. As the first packet, and a distribution unit that distributes the packet to the first queue.
A transmission unit that transmits the packet group distributed to the first queue and the second queue by the distribution unit, and a transmission unit.
A relay device characterized by having. The relay device according to claim 1.
As a result of distribution by the distribution unit, it is detected that a plurality of flows to which packets having different specific information belong are output to the same first queue, and among the plurality of flows, a specific flow is selected as the first queue. A learning unit that determines one flow and determines other flows other than the specific flow as the second flow.
The learning result including the specific information included in the packet belonging to the other flow determined by the learning unit in the second flow and the second queue designating the second queue of the second correspondence information. The learning result processing unit that acquires and registers in the second correspondence information,
A relay device characterized by having. The relay device according to claim 2.
When the learning result processing unit adds information indicating that the second queue included in the learning result is in use to the second correspondence information and acquires the learning result after the information is given. A relay device comprising selecting another second queue to which information indicating that it is in use is not given. The relay device according to claim 2.
The learning unit is a relay device characterized in that a specific flow to which the first-come-first-served packet belongs among the plurality of flows is determined as the first flow. The relay device according to claim 2.
The learning unit is a relay device characterized in that a specific flow having the maximum number of flow rate packets within a certain period of time among the plurality of flows is determined as the first flow. The relay device according to claim 2.
The learning result processing unit is a relay device characterized by deleting the learning result for which a predetermined period has passed since the registration of the second correspondence information. The relay device according to claim 2.
The learning result processing unit is a relay device that deletes the learning result corresponding to the second flow in which the number of flow rate packets after being registered is equal to or less than the threshold value in the second correspondence information. It is a relay method by a relay device that relays a group of packets.
The relay device is
It is converted into the same hash value as the hash value converted from the first specific information and one or more first queues that belong to the first flow and store the first packet containing the first specific information that specifies the first flow. A queue group having one or more second queues for storing a second packet including a second specific information that specifies a second flow, and a queue group.
A first storage unit that stores first correspondence information in which the hash value and the first queue are associated with each other.
It has a second storage unit that stores the second correspondence information in which the second specific information and the second queue are associated with each other.
In the relay method,
The relay device is
Receive the packet group and
It is determined whether or not the specific information included in each packet of the packet group and specifying the flow to which the packet belongs matches the second specific information, and if they match, the second correspondence information is referred to. Then, the packet is distributed to the second queue as the second packet, and if they do not match, the specific information is converted into the hash value, and the packet is referred to as the first packet with reference to the first correspondence information. Sort to the first queue,
The packet group distributed to the first queue and the second queue is transmitted.
A relay method characterized by that.

Images