JP6060050B2 - Communication apparatus and packet processing method - Google Patents

Description

  The present invention relates to a communication device, and more particularly, to a communication device that processes packets.

In recent years, the traffic in network systems has increased due to the digitization of TV content or the spread of smartphones. As a result, demand for lines exceeding 100 Gbps is increasing.

  A current packet processing FPGA (Field-Programmable Gate Array) processes a packet sequentially for each packet with a single core. In FPGA development so far, in order to cope with an increase in traffic, the operating frequency inside the FPGA is increased, and the number of parallel development bits is increased to speed up single-core packet processing. However, the operating frequency has reached its peak, and it is necessary to further increase the number of parallel development bits in order to realize packet processing exceeding 100 Gbps. Specifically, it is necessary to develop in parallel to a minimum size of 64 bytes (512 bits) of Ethernet (registered trademark), and information for a plurality of packets arrives at the FPGA within 1 clk. Since conventional single-core sequential processing for each packet cannot be performed, there is a large flow toward multi-core packet processing engines.

  When a packet management function such as a bandwidth control function or a statistic counting function is implemented in an FPGA, the FPGA stores packet information for each flow in one information storage medium (hereinafter referred to as a table), and the received packet The value of the corresponding table for each flow is updated every time a packet arrives.

  For example, when realizing the statistical count function, the FPGA manages statistical information of each flow in a table. Specifically, each time the packet arrives, the FPGA reads the number of received packets and the number of received bytes from the corresponding flow table, updates the number of received packets and the number of received bytes, and updates the results of updating the number of received packets and the number of received bytes. Write to the table.

  As a background art in this technical field, there is JP 2000-358066 A (Patent Document 1). The gazette states, “For each arriving packet, it checks whether a previous packet from the same source is pending on the processing element, and the check reveals that such a packet is pending. If this is the case, it forwards this arriving packet to the same processing element as the previous packet ”(see summary). That is, Patent Document 1 proposes a distributed processing technique in which a plurality of distributed processing units process packets of a plurality of flows.

JP 2000-358066 A

  However, according to the technique described in Patent Document 1, packets are distributed to a plurality of distributed processing units for each flow, and packet processing is executed by each distributed processing unit. For this reason, the technique described in Patent Document 1 requires complicated processing such as a Hash function for distributing the same flow to the same distributed processing unit.

  Further, when two or more packets of the same flow arrive at the same time, two packets are transmitted within one time interval shorter than the time interval required for processing one packet to one distributed processing unit by the packet distribution. Is transferred. When the packet to be processed next arrives before the packet processing that arrived at the distributed processing unit is not completed, the distributed processing unit cannot execute the packet processing that arrives at the same time and overflow occurs. There is. That is, there is a problem that the processing speed of a single distributed processing unit is limited in the processing speed of packets of the same flow.

  The present invention has been made in view of the above-described problems, and simplifies packet distribution processing to a distributed processing unit in a multi-core packet processing engine. It is another object of the present invention to provide a method capable of processing packets that arrive at the same time even when it is necessary to update the same flow information every time a packet arrives, such as a statistical count function.

The distribution unit distributes each packet to each distributed processing unit in the order in which a plurality of packets arrived at the input interface, and each distributed processing unit stores information on processing of each packet received by the distributed processing unit in a central processing unit. The central processing unit aggregates the operation results of a plurality of packets of the same flow received simultaneously from the distributed processing unit among the plurality of received packets, and updates the operation results of the corresponding flows held in the table.

  According to an embodiment of the present invention, in a multi-core packet processing engine, it is possible to suppress performance degradation due to load concentration on one distributed processing unit. Furthermore, even when it is necessary to update the same flow information every time a packet arrives, such as a statistical count function, packets that arrive at the same time can be processed.

It is explanatory drawing which shows the outline | summary of the packet process in the case of identifying the conventional flow and distributing a packet to a distributed processing part. It is explanatory drawing which shows the outline | summary of the packet process of the present Example 1. FIG. 1 is a block diagram illustrating a physical configuration and a logic circuit of a packet communication device according to a first embodiment. It is explanatory drawing which shows the format of the header in a device added to the packet of the present Example 1. FIG. It is explanatory drawing which shows the format of the packet used in the communication of the present Example 1. It is a block diagram which shows the logical structure of the input packet control part of the present Example 1. FIG. 3 is a block diagram illustrating a logical configuration of a distributed processing unit according to the first embodiment. FIG. 3 is a block diagram illustrating a logical configuration of a central processing unit according to the first embodiment. It is explanatory drawing which shows the statistical information table of the present Example 1. It is a flowchart which shows the distribution process which the distribution part of a present Example 1 performs. It is a flowchart which shows the header analysis process which the input header analysis part of the present Example 1 performs. It is a flowchart which shows the output determination process which the output determination process part of the present Example 1 performs. It is a flowchart which shows the calculating part selection arbitration process which the arbitration part of the present Example 1 performs. It is explanatory drawing which shows the calculation flow management list | wrist of the present Example 1. It is a flowchart which shows the calculation flow management process which the mediation part of the present Example 1 performs. It is a flowchart which shows the statistical information processing which the calculating part of the present Example 1 performs. It is a flowchart which shows the update mediation process which the mediation part of the present Example 1 performs. It is explanatory drawing which shows the format of the fragment header added to the packet of the present Example 1. FIG. It is explanatory drawing which shows the outline | summary of the fragment processing of the packet of the present Example 1. FIG. It is a block diagram which shows the logical structure of the input packet control part of the present Example 2. It is a flowchart which shows the bandwidth monitoring buffer write process which the bandwidth monitoring part of the present Example 2 performs. It is explanatory drawing which shows the relationship between the packet accumulation amount of the bandwidth monitoring buffer of the present Example 2, and the number of utilization distribution processing parts. It is a flowchart which shows the bandwidth monitoring buffer read-out process which the bandwidth monitoring part of the present Example 2 performs. It is a flowchart which shows the distribution process which the distribution part of the present Example 2 performs. It is explanatory drawing which shows the distributed processing part management list of the present Example 2. It is a block diagram which shows the logical structure of the central processing part of the present Example 3. It is explanatory drawing which shows the band setting table of the present Example 3. It is explanatory drawing which shows the calculation result holding table of the present Example 3. It is a flowchart which shows the calculating part selection arbitration process which the arbitration part of the present Example 3 performs. It is a flowchart which shows the process which calculates the token amount which should be accumulate | stored before receiving an input packet among the band calculation processes which the calculating part of the present Example 3 performs. It is a flowchart which shows the process which determines whether an input packet is within a guarantee among the band calculation processes which the calculating part of the present Example 3 performs. It is a flowchart which shows the update mediation process which the mediation part of the present Example 3 performs. In the selection of the result to be updated to the calculation result holding table in the update arbitration process executed by the arbitration unit of the third embodiment, all the bandwidth control results notified simultaneously from a plurality of calculation units that process the same ID were within the guarantee. It is explanatory drawing which shows the example of a case. In selecting a result to be updated to the calculation result holding table in the update arbitration process executed by the arbitration unit according to the third embodiment, all the bandwidth control results notified simultaneously from a plurality of calculation units that process the same ID are out of the guarantee. It is explanatory drawing which shows the example of a case. In the selection of the result to be updated to the calculation result holding table in the update arbitration process executed by the arbitration unit of the third embodiment, the bandwidth control results notified simultaneously from a plurality of calculation units that process the same ID are within and out of the guarantee. It is explanatory drawing which shows the example in the case of including. It is a flowchart which shows the output determination process which the output determination process part of the present Example 3 performs. It is explanatory drawing which shows the timing of the packet processing when the packet of the same flow arrives simultaneously in the packet processing in the case of identifying the conventional flow and distributing the packet to the distributed processing unit. In the packet processing of the first embodiment, it is an explanatory diagram showing the timing of packet processing when a plurality of packets of the same flow arrive simultaneously. In the packet processing of the present Example 3, it is explanatory drawing which shows the timing of the packet processing when the packet of the same flow arrives simultaneously.

  Embodiments of the present invention will be described below with reference to the drawings.

  An outline of packet processing in this embodiment is shown below.

  FIG. 1A is an explanatory diagram showing an outline of packet processing when a conventional flow is identified and packets are distributed to distributed processing units.

  For example, when packets of the same flow arrive, the distribution processing of the packet communication apparatus identifies the flow of the packet, and distributes the packets so that the same flow performs the same distributed processing. In this case, the performance of packet processing for the same flow depends on the performance of distributed processing (for example, # 1).

  FIG. 1B is an explanatory diagram illustrating an outline of packet processing according to the first embodiment.

  For example, when packets of the same flow arrive, the packet communication apparatus according to the present embodiment distributes packets to distributed processing (# 1 to #K) in round-robin order in the arrival order without the flow being identified. The packet communication apparatus according to the present embodiment aggregates information on packets distributed to each distributed process (# 1 to #K) in a central processing unit, and processes packets of the same flow collectively.

  An overview of the relationship between the packet arrival interval and the packet processing time (timing) in the present embodiment is shown below using FIGS. 33A and 33B.

  FIG. 33A is an explanatory diagram showing the timing of packet processing when a plurality of packets of the same flow arrive at the distribution process simultaneously in packet processing when identifying a flow of the prior art and distributing packets to the distributed processing unit. .

  For simplification, in FIG. 33A as an example, the distribution processing analyzes the headers of a plurality of packets simultaneously to identify the flow, and one distributed processing unit processes the packets with 1 clk for each packet (in this embodiment, (Statistical processing) An example in which two packets of the same flow arrive at the same time in the distribution process when it is possible.

  In FIG. 33A and FIGS. 33B and 33C described later, the input packet received first by the distribution processing unit is the first packet (S1), and the input packet received simultaneously with the first packet is the second packet. The packet (S2), the input packet received after 1 clk after the second packet is described as a third packet (S3), and the input packet received simultaneously with the third packet is described as a fourth packet (S4). In FIG. 33, the values of the flow IDs of the first packet, the second packet, the third packet, and the fourth packet are all “10”.

  In the prior art, as shown in FIG. 33A, when the distribution process receives the first packet and the second packet at the same time, the header is analyzed to identify the flow (T3301). Since both the first packet and the second packet have the flow ID value “10”, they are distributed to the same distributed processing unit (for example, # 1). Since the distributed processing unit of the prior art processes a packet for each packet at 1 clk, the distribution processing unit distributes the first packet and the second packet for each packet at the timing of T3302 and T3303 (# 1 ). The distributed processing unit starts statistical processing of the first packet at T3302 and starts statistical processing of the second packet at T3303. In the distribution process, when the third packet and the fourth packet are received, the header is analyzed to identify the flow (T3302). Since both the third packet and the fourth packet have the flow ID value “10”, they are distributed to the same distributed processing unit (for example, # 1) as the first packet and the second packet. In order to wait for the processing of the packet and the second packet, the distribution process distributes the third packet and the fourth packet to the distributed processing unit (# 1) at the timing of T3304 and T3305. (The statistical processing of the third packet cannot be executed at T3303 one clock after the processing timing T3302 of the header analysis of the third packet, so the statistical processing of the third packet cannot be executed.) , T3304 starts statistical processing of the third packet, and T3305 starts statistical processing of the fourth packet. Therefore, conventionally, when a plurality of packets of the same flow arrive at the same time, a packet to be processed next arrives before the processing of the previously arrived packet is finished, so that packet processing cannot be executed normally and an overflow occurs. There is a case.

  FIG. 33B is an explanatory diagram illustrating packet processing timing when a plurality of packets having the same flow arrive at the same time in the packet processing according to the first embodiment.

  In this embodiment, in the distribution process, when the first packet and the second packet are received, the first packet and the second packet are distributed to different distributed processing units (for example, # 1 and # 2), respectively. (T3301). Since no flow is identified, there is no processing delay here.

  At the timing of T3301, the distributed processing unit (# 1) analyzes the header of the first packet to identify the flow, and the distributed processing unit (# 2) analyzes the header of the second packet to identify the flow. . Thereafter, each distributed processing unit transfers information such as the identified flow ID and packet length to the central processing unit. The central processing unit collectively executes statistical processing on the first packet and the second packet based on the information acquired from each distributed processing unit (# 1 and # 2) (T3302). For example, when both the first packet length and the second packet length are 100 bytes, 200 bytes are counted as the number of received bytes. Since the distributed processing unit in FIG. 33A processes a packet for each packet, statistical processing is performed for one packet in one clock. However, the central processing unit in FIG. 33B uses the first packet and the second packet. Since statistical processing relating to packets is collectively executed, statistical processing of the first packet and the second packet can be executed in one clock.

  In the distribution process, when the third packet and the fourth packet are received, the distribution process similarly separates the third packet and the fourth packet from different distributed processing units (for example, # 3 and # 4). ) (T3302).

  At the timing of T3302, the distributed processing unit (# 3) analyzes the header of the third packet to identify the flow, and the distributed processing unit (# 4) analyzes the header of the fourth packet to identify the flow. . Thereafter, each distributed processing unit transfers information such as the identified flow ID and packet length to the central processing unit. The central processing unit collectively executes statistical processing on the third packet and the fourth packet based on the information acquired from each distributed processing unit (# 3 and # 4) (T3303). In FIG. 33A, since the timing of T3303 is assigned to the statistical processing of the second packet, the statistical processing of the third packet is assigned to T3304. In FIG. 33B, the third processing is performed at the timing of T3303. Statistical processing of the packet and the fourth packet can be executed.

  Therefore, in this embodiment, when the packet information of the same flow arrives at the same time by aggregating the packet information in the central processing unit and processing the packets of the same flow that have arrived at the same time in the sorting process. However, a function that needs to update the same flow information every time a packet arrives, such as the statistical count function, can be realized without causing an overflow.

  Hereinafter, the first embodiment will be described with reference to FIGS. In addition, Example 1 described below is one of the examples of the present invention, and does not limit the present invention.

  In the first embodiment, the present invention is applied to the statistical count function. In the first embodiment, the number of received packets and the number of received bytes are counted as statistical information of the statistical count function.

  FIG. 2 is a block diagram illustrating a physical configuration of the packet communication apparatus 2 and the logic circuit 202 according to the first embodiment.

  The packet communication device 2 of the present embodiment is a device that transfers a packet to another packet communication device. The packet communication apparatus 2 includes a plurality of network interface boards (NIF) (200-1 to 200-M) and a switch unit 206.

  The NIF 200 is an interface for receiving a packet from the network and transmitting the packet to the network. The switch unit 206 is a device connected to each NIF 200, and is a device that distributes a packet received from the NIF 200 to the NIF 200 to which the packet is to be transmitted.

  Each NIF 200 includes an input / output line interface 201, a logic circuit 202, a SW interface 205, and an NIF management unit 208.

  The input / output line interface 201 is a communication port. The packet communication device 2 is connected to other packet communication devices via the input / output line interface 201 and the network. The input / output line interface 201 in the first embodiment is a line interface for Ethernet (registered trademark).

  The SW interface 205 is a device for connecting to the switch unit 206.

  The NIF management unit 208 is a processor such as a CPU, for example. The NIF management unit 208 controls processing in the logic circuit 202.

  The logic circuit 202 is a circuit that performs processing on the packet of this embodiment. The logic circuit 202 includes at least one memory and at least one arithmetic device (for example, a processor).

  The logic circuit 202 includes packet processing units such as an input packet control unit 203 and an output packet control unit 204, and includes a setting register 207.

  Packets received by the packet communication apparatus 2 are transmitted in the order of the input / output line interface 201, the input packet control unit 203, the SW interface 205, the switch unit 206, the SW interface 205, the output packet control unit 204, and the input / output line interface 201. Is done.

  Each processing unit included in the logic circuit 202 may be implemented by a physical device such as an integrated circuit, or may be implemented by a program executed by at least one processor. In addition, a plurality of processing units (for example, the input packet control unit 203 and the output packet control unit 204) may be implemented by a single device or program, or a single processing unit may be implemented by a plurality of devices or a plurality of programs. May be.

  The NIF management unit 208 controls the setting register 207. The setting register 207 has a storage area for temporarily storing data, and holds register values of the respective processing units included in the logic circuit 202.

  The setting register 207 is connected to all processing units included in the logic circuit 202 via the logic circuit 202. In the following, description of the processing contents of the setting register 207 is omitted, but all the processing units included in the logic circuit 202 execute processing using the setting register 207.

  The input / output line interface 201 adds a later-described apparatus header 3 to the received packet 4.

  FIG. 4 is an explanatory diagram illustrating the format of the packet 4 used in the communication according to the first embodiment. The format of the packet 4 shown in FIG. 4 is an example, and the packet used in this embodiment may include any information. For example, an IP address may be included instead of the MAC address. The packet 4 includes a destination MAC address 401, a source MAC address 402, a VLAN header 403, an ether type value 404, a payload 405, and a frame check sequence (FCS) 406. In the destination MAC address 401 or the source MAC address 402, the MAC address of the packet communication device 2 is set. In the VLAN header 403, a VLAN ID serving as a flow identifier is set.

  Note that an MPLS header or another protocol header may be set in the payload 405, and an MPLS label value or the like may be set as a flow identifier. A value for detecting a frame error is set in the frame check sequence (FCS) 406.

  FIG. 3 is an explanatory diagram showing the format of the in-device header 3 added to the packet 4 of the first embodiment. The in-device header 3 includes an output network interface board identifier (output NIF ID) 301, a flow ID 302, and a packet length 303. The output NIF ID 301 is internal routing information. The internal routing information in this embodiment is information indicating which NIF 200 port of the packet communication device 2 outputs the packet 4 received by the packet communication device 2. The switch unit 206 transfers the input packet transmitted to the switch unit 206 to the specific SW interface 205 of the specific NIF 200 according to the internal routing information.

  The input / output line interface 201 adds the in-device header 3 to the received packet 4 in order to process the received packet 4 in the packet communication device 2. The packet 4 to which the in-device header 3 is added and transmitted from the input / output line interface 201 to the switch unit 206 is hereinafter referred to as an input packet.

  The input / output line interface 201 stores values such as a null value in the output NIF ID 301 and the flow ID 302 when adding the in-device header 3 to the received packet 4. That is, the input / output line interface 201 does not determine the values stored in the output NIF ID 301 and the flow ID 302. Values are stored in the output NIF ID 301 and the flow ID 302 by the input packet control unit 203.

  The input / output line interface 201 acquires the packet length of the received packet 4 and stores the acquired packet length in the frame length 303 of the in-device header 3. Thereafter, the input / output line interface 201 transmits the packet 4 to the input packet control unit 203.

  The input packet control unit 203 performs packet processing described later on the packet 4 received from the input / output line interface 201. At the time of this packet processing, the output NIF ID 301 and the flow ID 302 of the in-device header 3 are added.

  After the packet processing, the input packet control unit 203 transmits the input packet to the SW interface 205. The SW interface 205 transfers the received input packet to the switch unit 206.

  After receiving the input packet from the SW interface 205 of each NIF 200, the switch unit 206 refers to the output NIF ID 301 of the input packet to identify the NIF 200 that is the transfer destination of the received input packet. Next, the switch unit 206 transfers the received input packet as an output packet to the SW interface 205 corresponding to the specified NIF 200.

  In the present embodiment, the packet 4 transmitted from the switch unit 206 to the input / output line interface 201 is referred to as an output packet.

  The SW interface 205 transfers the received output packet to the output packet control unit 204.

  In the above example, the input packet control unit 203 executes packet processing. However, the output packet control unit 204 may execute packet processing instead of the input packet control unit 203.

  When the output packet control unit 204 executes packet processing, the output packet control unit 204 transmits the output packet to the input / output line interface 201 after packet processing.

  When the output packet control unit 204 does not perform packet processing, the output packet control unit 208 transfers the output packet received from the SW interface 205 to the input / output line interface 201 as it is.

  The input / output line interface 201 removes the in-device header 3 from the received output packet. Thereafter, the input / output line interface 201 transfers the output packet to another device in the format of the packet 4 shown in FIG.

  FIG. 5 is a block diagram illustrating a logical configuration of the input packet control unit 203 according to the first embodiment.

  The input packet control unit 203 includes a distribution unit 502, a plurality of distributed processing units 503 (503-1 to 503-N) (N is an arbitrary natural number), a multiplexing unit 504, and a centralized processing unit 505.

  The distribution unit 502 is a processing unit that executes a distribution process S950 described later when the input packet control unit 203 receives an input packet from the input / output line interface 201. In the distribution process S950, the distribution unit 502 determines a distributed processing unit 503 that executes packet processing from the distributed processing units 503-1 to 503-N.

  The distributed processing unit 503 executes packet processing to be described later. The distributed processing unit 503 transfers packet information such as a flow ID and a packet length to the central processing unit 505 at the time of this packet processing. After the packet processing, the distributed processing unit 503 transmits the input packet to the multiplexing unit 504.

  The central processing unit 505 executes packet processing to be described later. Packet information transferred from a plurality of distributed processing units 503 (503-1 to 503-N) is aggregated and processed, and a statistical information table 704 described later is updated.

  The multiplexing unit 504 multiplexes the packets received from the respective distributed processing units 503 and transmits them to the SW interface 205 in the order of arrival.

  In this embodiment, the time required for packet processing by the input packet control unit 203 is determined in advance.

  Further, in the present embodiment, the number N of the distributed processing units 503 is the packet arrival interval received by the input packet control unit 203 (the number of packets when a plurality of packets arrive simultaneously within 1 clk) and the number of distributed processing units 503. It is determined based on the packet processing time. The shortest packet arrival interval is a value determined in advance according to the line to which the packet communication device 2 is connected.

  For example, when two packets arrive at the input packet control unit 203 within 1 clk at the same time and the packet processing time of the distributed processing unit 503 is 20 clk, the number of simultaneous arrival packets is multiplied by the packet processing time of the distributed processing unit 503 (remainder The result of the advance is “40”. For this reason, the number N of distributed processing units 503 is determined in advance to “40”.

  FIG. 6 is a block diagram illustrating a logical configuration of the distributed processing unit 503 according to the first embodiment.

  The distributed processing unit 503 includes an input header analysis unit 601, an input header processing unit 602, and an output determination processing unit 603.

  The input header analysis unit 601 is a processing unit that executes a header analysis process S1050 described later when an input packet is received from the distribution unit 502. In the header analysis processing S1050, the input header analysis unit 601 identifies the flow of the packet 4, transfers packet information required when the central processing unit 505 processes the packet, and transfers the packet to the header processing unit 602.

  When the input header processing unit 602 receives an input packet from the input header analysis unit 601, the input header processing unit 602 executes VLAN tag processing according to the flow identified by the input header analysis unit 601. The VLAN tag processing is, for example, transparency, conversion, addition, and deletion of the VLAN tag. That is, the VLAN tag processing is processing for converting the VLAN header of the input packet into the VLAN header of the packet output from the packet communication device 2. In the VLAN tag processing, the input header processing unit 602 may overwrite the destination MAC address 401 and the transmission source MAC address 402 with values set in the input / output line interface 201 or for each flow registered in advance. It may be overwritten by the value of. After processing the VLAN tag, the input header processing unit 602 transmits the input packet to the output determination processing unit 603. At this time, the input header processing unit 602 transmits the packet to the output determination processing unit 603 at the same timing as the calculation completion notification timing of the central processing unit 505 described later.

  When receiving an input packet from the input header processing unit 602, the output determination processing unit 603 is a processing unit that executes output determination processing S1150 described later. In the output determination processing S1150, the output determination processing unit 603 transfers the packet to the multiplexing unit 504 at the same time as the calculation of the central processing unit 505.

  FIG. 7 is a block diagram illustrating a logical configuration of the central processing unit 505 according to the first embodiment.

  The central processing unit 505 includes an arbitration unit 701, a calculation flow management list 702, calculation units 703 (703-1 to 703-N) (N is an arbitrary natural number), and a statistical information table 704.

  When receiving a processing request from the distributed processing unit 503 (503-1 to 503-N), the arbitrating unit 701 is a processing unit that executes arithmetic unit selection arbitration processing S1250 described later. The arbitration unit 701 identifies the flow based on the flow ID received from the distributed processing unit 503, and further refers to the calculation flow management list 702. And thereby, the calculating part 703 which performs the statistics count process regarding an input packet is determined from the calculating part 703-1-calculating part 703-N.

  The arbitration unit 701 holds the calculation flow management list 702 in a memory or the like provided in the logic circuit 210. The arbitration unit 701 then updates the computation flow management list 702 in the computation unit selection arbitration process S1250 and the computation flow management process S1450.

  Further, the arbitration unit 701 is a processing unit that executes an update arbitration process S1650, which will be described later, when a computation end notification is received from the computation unit 703. The arbitration unit 701 writes the statistical information updated by the calculation unit 703 back to the statistical information table 704, and transfers the calculation end notification to the distributed processing unit 503 that has transmitted the processing request.

  The calculation flow management list 702 indicates the flow ID of the input packet processed by the calculation unit 703 and the processing status of the calculation unit 703.

  The arithmetic unit 703 updates information of each flow included in the statistical information table 7044 in statistical information processing S1550 described later. In addition, the calculation unit 703 transmits the result updated in the statistical information processing S1550 to the arbitration unit 701. The arithmetic unit 703 includes N (N is an arbitrary natural number) in the central processing unit 505.

  FIG. 8 is an explanatory diagram illustrating the statistical information table 704 according to the first embodiment.

  The statistical information table 704 is a table having information about packets received by the packet communication device 2. The statistical information table 704 includes a flow ID 801, the number of received packets 802, and the number of received bytes 803.

  The flow ID 801 includes the flow ID of the input packet. The number of received packets 802 includes the number of input packets received by the distribution unit 502. The number of received bytes 803 indicates the total number of bytes of input packets received by the distribution unit 502.

  FIG. 13 is an explanatory diagram illustrating the calculation flow management list 702 according to the first embodiment.

  The calculation flow management list 702 includes a calculation unit 1301, a flow ID 1302, and a processing status counter 1303.

  The calculation unit 1301 includes an identifier that uniquely indicates the calculation units 703-1 to 703-N. The calculation unit 1301 of this embodiment includes a value that identifies the calculation unit 703 by a numerical value. The value stored in the calculation unit 1301 is set in advance by an administrator or the like. The flow ID 1302 includes a flow ID that uniquely indicates the flow assigned to the input packet processed by the arithmetic unit 703. The value stored in the flow ID 1302 is updated by the arbitration unit 701. The value stored in the flow ID 1302 corresponds to the value of the flow ID identified in the input header analysis unit 601.

  The processing status counter 1303 includes a value indicating the status of arithmetic processing (statistical information processing) in the arithmetic unit 703. The processing status counter 1303 of the present embodiment includes a value related to the execution time that has elapsed since the arithmetic processing started.

  A processing status counter 1303 shown below indicates the remaining time until the arithmetic processing is completed by the subtraction counter. Specifically, when the calculation process is started, the calculation time is stored in the processing status counter 1303. Then, the arbitration unit 701 subtracts the value of the processing status counter 1303 every unit time (1 clk in the present embodiment). An entry in which “0” is stored in the processing status counter 1303 indicates that the calculation unit 703 indicated by the calculation unit 1301 has finished the calculation process. An entry in which a positive number other than “0” is stored in the processing status counter 1303 indicates that the calculation unit 703 indicated by the calculation unit 1301 is performing calculation processing.

  In the following, the processing in the case where the processing status counter 1303 is a subtraction counter will be described, but the processing status counter 1303 may indicate the time that has elapsed since the arithmetic processing was started by the addition counter. In this case, the arbitration unit 701 may acquire the status of the calculation process by comparing a predetermined calculation time with the processing status counter 1303.

  The distribution process S950 executed by the distribution unit 502 will be described below. FIG. 9 is a flowchart illustrating the distribution process S950 executed by the distribution unit 502 according to the first embodiment.

  When receiving an input packet from the input / output line interface 201 (a plurality of input packets may be received simultaneously) (S900), the allocating unit 502 takes an internal counter value (this counter takes a value of 1 to N). , N returns to 1), and the packets are transferred to the distribution processing unit 503 (503-1 to 503-N) in the order in which the packets are received (S901). After S901, the distribution unit 502 counts up the value of the internal counter by the number of received packets (S902), and ends the distribution process S950 shown in FIG. 9 (S903).

  Hereinafter, the header analysis process S1050 executed by the input header analysis unit 601 will be described. FIG. 10 is a flowchart illustrating the header analysis processing S1050 executed by the input header analysis unit 601 according to the first embodiment.

  When the input header analysis unit 601 receives an input packet from the distribution unit 502 (S1000), the input header analysis unit 601 acquires the value of the packet length 303 from the in-device header 3 included in the input packet (S1001). After S1001, the flow of the packet 4 is identified based on the received input packet 403, thereby determining the output NIF ID (S1002). After S1002, the input header analysis unit 601 outputs the processing start trigger, the flow ID, and the acquired packet length to the central processing unit 505 (S1003). After S1003, the input header analysis unit 601 stores values in the output NIF ID 301 and the flow ID 302 of the in-device header 3 of the input packet, and forwards the packet to the input header processing unit 602 (S1004). The header analysis process S1050 shown is terminated (S1005).

  The output determination process S1150 executed by the output determination processing unit 603 according to the first embodiment will be described below. FIG. 11 is a flowchart illustrating the output determination processing S1150 executed by the output determination processing unit 603 according to the first embodiment.

  When the output determination processing unit 603 receives an input packet from the input header processing unit 602 (S1100), the output determination processing unit 603 determines whether there is a calculation end notification from the central processing unit 505 (S1101). If it is determined in S1101 that there is an operation end notification, the packet is transferred to the multiplexing unit 504 (S1102), and the output determination process S1150 shown in FIG. 11 is ended (S1104). If it is determined in S1101 that there is no calculation end notification, the input packet is discarded (S1103), and the output determination process S1150 shown in FIG. 11 is ended (S1104).

  Hereinafter, the calculation unit selection arbitration process S1250 executed by the arbitration unit 701 according to the first embodiment will be described. FIG. 12 is a flowchart illustrating the calculation unit selection arbitration process S1250 executed by the arbitration unit 701 according to the first embodiment.

  The arbitration unit 701 receives the processing start trigger from the distributed processing unit 503 (503-1 to 503-N) (may receive a plurality of processing start triggers at the same time) (S1200) and receives the processing start trigger at the same time. The flow ID and the packet length to be acquired are acquired, and the arrival order of the processing start trigger is held (S1201). After S1201, the arbitration unit 701 determines whether or not the values of the processing status counters 1303 of all entries in the computation flow management list 702 are different from “0” (S1202).

  If it is determined in S1202 that the processing status counters of all entries in the calculation flow management list 702 are different from “0”, all the calculation units 703 (703-1 to 703-N) are executing calculation processing. is there. Since the arithmetic processing in the case of a new flow cannot be accepted, the arbitration unit 701 determines to discard the input packet (S1203), and ends the arithmetic unit selection arbitration processing S1250 shown in FIG. 12 (S1210).

  If there is at least one entry whose processing status counter 1303 is “0” among the entries in the computation flow management list 702 in S1202, at least one of all the computation units 703 (703-1 to 703-N). One arithmetic unit 703 does not execute arithmetic processing. Therefore, the arbitration unit 701 searches the statistical information table 704 using the flow ID value acquired in S1201 as a search key (flow ID 801), and acquires information on the number of received packets 802 and the number of received bytes 803 (S1204). ).

  In S1202, when a plurality of processing start triggers are received simultaneously in S1200, the arbitration unit 701 checks in the calculation unit 703 whether the number of calculation units corresponding to the number of flows that have received the process start trigger can be processed. . The arbitration unit 701 performs processing in S1204 when there are a plurality of entries in the computation flow management list 702 whose processing status counter 1303 is “0” for SFlow and the arbitration unit 701 has received a processing start trigger at the same time. move on. The arbitration unit 701 proceeds to the processing of S1203 when there is no entry whose processing status counter 1303 is “0” for a plurality of flows that simultaneously received the processing start trigger in S1200. If there is no entry for which the process status counter 1303 is “0” for a plurality of flows that simultaneously received the process start trigger in S1200, the process for the flow for the entry for which the process status counter 1303 is “0” is performed in S1204. The processing may proceed to S1203 for the remaining flow processing.

  After S1204, it is determined whether or not the value of the flow ID acquired in S1201 and the processing status counter 1303 match the value of the flow ID 1302 of an entry different from “0” (S1205).

  In S1205, when the value corresponding to the flow ID acquired in S1201 matches the value of the flow ID 1302 of the entry whose processing status counter 1303 is different from “0”, the arbitration unit 701 determines the flow ID acquired. It may be determined that the value corresponding to the value matches the value of the flow ID 1302 of the entry whose processing status counter 1303 is different from “0”.

  In S1205, when it is determined that the value of the flow ID acquired in S1201 does not match the value of the flow ID 1302 of the entry whose processing status counter 1303 is different from “0”, the arbitration unit 701 has not performed the arithmetic processing. In order to cause the calculation unit to newly execute a calculation process related to the input packet, the process of S1208 is executed.

  In S1205, when it is determined that the value of the flow ID acquired in S1201 matches the value of the flow ID 1302 of the entry whose processing status counter 1303 is different from “0”, the calculation unit 703 corresponding to the flow ID of the input packet. Can continue the arithmetic processing related to the input packet. For this reason, the arbitration unit 701 determines the arithmetic unit 703 indicated by the arithmetic unit 1301 of the entry whose acquired flow ID value matches the flow ID 1302 to be the arithmetic unit 703 that continuously executes the arithmetic processing related to the input packet. .

  Then, the arbitrating unit 701 updates the value of the processing status counter 1303 of the entry in which the acquired flow ID value matches the flow ID 1302 to a value indicating the calculation time (S1206). For example, when the calculation time is 10 clk, the arbitration unit 701 updates the value of the processing status counter 1303 of the entry in which the acquired flow ID value matches the flow ID 1302 to “10”.

  By updating the processing status counter 1303 in S1206, the arbitration unit 701 indicates that each calculation unit 703 is executing until the calculation process is completed.

After S1206, the arbitration unit 701 calculates the number of acquired packets and the acquired packet length by using the number of flows with the same flow ID and the sum of the packet lengths of the same flow ID among the flows acquired simultaneously with the processing start trigger in S1201. Is held for each flow (S1211). The number of acquired packets is the number that the flow IDs acquired in S1201 are the same, that is, the number of simultaneously input packets of the same flow ID. The acquired packet length is the sum of the packet lengths of the same flow ID acquired in S1201, that is, the total packet length of the simultaneously input packets of the same flow ID.
After S1211, the arbitration unit 701 transfers the processing start trigger and the new / continuation parameter to the calculation unit 703 indicated by the calculation unit 1301 of the entry of the calculation flow management list 702 in which the acquired flow ID value matches the flow ID 1302 value. An operation notification including the number of acquired packets, the acquired packet length, the number of received packets, and the number of received bytes is output (S1207).

  In S1207, the arbitration unit 701 determines the new / continuation parameter as a value indicating continuation, and includes the new / continuation parameter (“1” in the present embodiment) indicating continuation in the calculation notification. The new / continuation parameter in the present embodiment is a value for notifying each arithmetic unit 703 what kind of arithmetic processing is to be executed. After S1207, the calculation unit selection arbitration process S1250 shown in FIG. 12 is terminated (S1210).

  In S1205, when it is determined that the value of the flow ID acquired in S1201 does not match the value of the flow ID 1302 of the entry whose processing status counter 1303 is different from “0”, the arbitration unit 701 sets the processing status counter 1303 to “0”. Among the entries in the computation flow management list 1302 that is “,” the entry having the smallest number in the computation unit 1301 is extracted. The arbitration unit 701 determines the calculation unit 703 indicated by the calculation unit 1301 of the extracted entry to be a calculation unit 703 that newly executes a calculation process related to the input packet.

  Then, the arbitrating unit 701 updates the flow ID 1302 of the extracted entry with the value of the flow ID acquired in S1201. Further, the arbitrating unit 701 updates the processing status counter 1303 of the extracted entry to a value indicating the calculation time (S1208).

In S1208 described above, the arbitration unit 701 extracts the entry with the smallest number (smallest number) in the computation unit 1301 from the computation flow management list 1302 whose processing status counter 1303 is “0”. However, the arbitration unit 701 may extract an entry according to any rule as long as it is one of the entries in the computation flow management list 702 whose processing status counter 1303 is “0”. For example, the arbitrating unit 701 may extract the entry having the largest value, or the order of entries to be extracted in advance may be determined by the administrator.
After S1208, the arbitrating unit 701 holds the number of the same flow ID acquired in S1201 and the sum of the packet lengths of the same flow ID acquired in S1201 as the acquired packet number and the acquired packet length, respectively (S1212). ).
After S <b> 1212, the arbitration unit 701 performs an operation including the processing start trigger, the new / continuation parameter, the number of acquired packets, the acquired packet length, the number of received packets, and the number of received bytes, to the calculation unit 703 determined in S <b> 1207 and S <b> 1209. A notification is output (S1209). In S1209, the arbitration unit 701 determines the new / continuation parameter as a value indicating new, and includes the new / continuation parameter indicating “new” (“0” in this embodiment) in the calculation notification.

  In S1207 and S1209, the processing start trigger included in the output calculation notification is a value indicating that the calculation unit 703 starts statistical information processing S1550 described later. The number of acquired packets included in the calculation notification is the number of acquired packets held in S1211 or S1212. The acquisition packet length included in the calculation notification is the acquisition packet length held in S1211 or S1212. The number of received packets and the number of received bytes are the number of received packets and the number of received bytes acquired in S1204.

  In S1207 and S1209, the number of input packets and the total packet length of the flows with the same flow ID that received the processing start trigger at the same time are output to one arithmetic unit 703 as the number of acquired packets and the acquired packet length. . Therefore, even if the central processing unit 505 receives processing start triggers for input packets with the same flow ID at the same time, statistical information processing can be executed without causing overflow.

  After S1209, the calculation unit selection arbitration process S1250 shown in FIG. 12 is terminated (S1210).

  The calculation flow management process S1450 executed by the arbitration unit 701 will be described below.

  FIG. 14 is a flowchart illustrating the calculation flow management process S1450 executed by the arbitration unit 701 according to the first embodiment.

  The arbitration unit 701 determines whether or not the processing status counters 1303 of all entries in the calculation flow management list 702 are 0 every clock (S1400).

  If it is determined in S1401 that the processing status counter 1303 of all entries in the computation flow management list 702 is 0, the arbitration unit 701 ends the computation flow management processing S1450 (S1403).

  If it is determined in S1401 that the processing status counter 1303 has an entry other than 0 in the calculation flow management list 702, the arbitration unit 701 subtracts 1 from a value other than 0 included in the processing status counter 1303 column. (S1402). After S1402, the arbitrating unit 701 ends the calculation flow management process S1450 (S1403).

  The statistical information processing S1550 executed by the calculation unit 703 (703-1 to 703-N) will be described below.

  FIG. 15 is a flowchart illustrating the statistical information processing S1550 executed by the calculation unit 703 according to the first embodiment.

  When the calculation unit 703 receives a calculation notification including a process start trigger from the arbitration unit 701 (S1500), the new / continuation parameter, the acquired packet number, the acquired packet length, the received packet number, and the received calculation notification, The number of received bytes is acquired (S1501).

  After S1501, the calculation unit 703 determines whether the new / continuation parameter is “0”, that is, whether the new / continuation parameter indicates new (S1502).

  When the new / continuation parameter is “0”, that is, when it is determined in S1502 that the new / continuation parameter indicates new, the arithmetic unit 703 proceeds to the processing of S1504.

  If the new / continuation parameter is not “0”, that is, if it is determined in S1502 that the new / continuation parameter indicates continuation, the calculation unit 703 causes the calculation unit 703 to internally retain the previous calculation result for continuation processing ( The previous received packet count and the previous received byte count) are acquired (S1503). The previous calculation result held internally for the continuation processing is the calculation result held by the arithmetic unit 703 in the logic circuit 202 in order to continuously execute the next statistical information processing before all statistical information processing ends. It is. In S1503, the calculation unit 703 updates the number of received packets and the number of received bytes included in the calculation notification received in S1501 by acquiring the previous calculation result held internally, and the process proceeds to S1504. The arbitration unit assigns processing to the processing unit that is processing the same flow, and uses the previous calculation result held inside the calculation unit, so that even if the input packets of the same flow are received continuously, overflow will occur. Statistical information processing can be executed normally without occurrence.

  After S1502 (when the new / continuation parameter is “0”) or S1503, the calculation unit 703 determines the number of received packets (the number of received packets acquired simultaneously with the calculation notification in S1501 or the number of previous received packets acquired in S1503). ) Is added to the number of acquired packets acquired at the same time as the calculation notification in S1501. In addition, the calculation unit 703 adds the acquired packet length acquired simultaneously with the calculation notification in S1501 to the number of received bytes (the number of received bytes acquired in S1501 or the previous number of received bytes acquired in S1503).

  Then, the calculation unit 703 acquires the addition result as the updated number of received packets and the number of received bytes, and arbitrates the calculation end notification including the acquired updated number of received packets and the number of received bytes. The data is transferred to 701 (S1504).

  After S1504, the calculation unit 703 internally stores the updated number of received packets and the updated number of received bytes as the previous received packet number and the previous received byte number for subsequent processing of the next packet (S1505).

  After S1505, the calculation unit 703 ends the statistical information processing S1550 (S1506).

  Below, update mediation processing S1650 which mediation part 701 performs is explained.

  FIG. 16 is a flowchart illustrating the update arbitration process S1650 executed by the arbitration unit 701 according to the first embodiment.

  When the arbitration unit 701 receives a calculation end notification from the calculation unit 703 (703-1 to 703-N) (S1600), the arbitration unit 701 acquires the number of received packets and the number of received bytes included in the received calculation end notification, The data is written back to the statistical information table 704 (S1601). Here, in the prior art, writing back to the statistical information table is performed for the number of received flows, but in this embodiment, the number of acquired packets of the flow with the same flow ID that received the processing start trigger at the same time, and One arithmetic unit 703 performs statistical information processing using the total value of the acquired packet lengths. Therefore, the number of write-backs can be reduced as compared with the prior art.

  After S1601, the arbitration unit 701 specifies the distributed processing unit 503 (503-1 to 503-N) that is the transmission source of the process start trigger from the arrival order of the process start trigger held in the calculation unit selection arbitration process S1250, An operation end notification is transferred to the identified distributed processing unit 503 (S1602).

  After S1602, the arbitration unit 701 ends the update arbitration process S1650 (S1603).

  As described above, according to the first embodiment, the packet communication device 2 aggregates the information of the packets distributed to the distributed processing units 503 (503-1 to 503-N) by the round robin in the arbitration unit 701. Notify the arithmetic unit 703 (703-1 to 703-N). As a result, the calculation unit 703 according to the first embodiment can simultaneously execute the statistical count processing regarding packets of the same flow.

  Therefore, according to the first embodiment, performance degradation due to load concentration on one distributed processing unit can be suppressed in the multi-core packet processing engine. Furthermore, by increasing the number of distributed processing units, it is possible to speed up the packet processing engine in a scalable manner.

  In the first embodiment, the distribution unit 502 distributes the input packet to the distributed processing units 503 (503-1 to 503-N) in units of input packets. The input packet may be distributed to the distributed processing unit 503 after being fragmented.

  FIG. 17 is an explanatory diagram showing the format of the fragment header 5 added to the packet of the first embodiment.

  The format of the fragment header 5 shown in FIG. 17 is an example, and the fragment header used in this embodiment may include any information.

  The fragment header 5 includes a total fragment number 1701, a sequence number 1702, and a fragment length 1703.

  The total number of fragments 1701 is set to the total number of fragmented input packets (packet 4 with in-device header 3 attached). In the sequence number 1702, the order of the fragmented input packet is set. In the fragment length 1703, the length of the fragmented input packet is set.

  FIG. 18 is an explanatory diagram illustrating an outline of packet fragment processing according to the first embodiment.

  The distribution unit 702 fragments the received input packet, adds the fragment header 5, and transfers the fragmented packet to the distributed processing unit 503 (503-1 to 503 -N).

  When the multiplexing unit 504 receives the fragmented input packet from the distributed processing unit 503 (503-1 to 503-N), the multiplexing unit 504 sets the total number of fragments 1701, the sequence number 1702, and the fragment length 1703 of the in-device header 5. The input packet may be reconstructed by referring to it.

  The distribution unit 502 distributes only the header portion (for example, from the destination MAC address 401 to the ethertype value 404) of the input packet to the distribution processing unit 503, and transfers the remaining payload portion to the multiplexing unit 504 for distributed processing. You may wait until the process of the part 503 is completed.

  When the multiplexing unit 504 receives the header portion of the input packet from the distributed processing unit 503 (503-1 to 503-N), the multiplexing unit 504 may receive the remaining payload portion that has been queued and reconstruct the input packet. .

  Hereinafter, the second embodiment will be described with reference to FIGS. The second embodiment described below is one of the embodiments of the present invention and does not limit the present invention.

  The packet communication apparatus 2 according to the first embodiment implements the statistical count process by implementing the centralized / distributed processing method of the present invention. The packet communication apparatus according to the second embodiment further changes the number of distributed processing units to be used according to the bandwidth of the input packet. In the second embodiment, in order to monitor the bandwidth of the input packet, the input packet control unit of the packet communication device includes a bandwidth monitoring unit.

  FIG. 19 is a block diagram illustrating a logical configuration of the input packet control unit 1900 according to the second embodiment.

  The input packet control unit 1900 according to the second embodiment includes a bandwidth monitoring buffer 19, a bandwidth monitoring unit 1901, a distribution unit 1902, a distributed processing unit management list 119, and a plurality of distributed processing units 503 (503-1 to 503-N) (N Is an arbitrary natural number), a multiplexing unit 1904, and a central processing unit 505.

  The distributed processing unit 503 and the central processing unit 505 in the second embodiment are the same as the distributed processing unit 503 and the central processing unit 505 in the first embodiment.

  The bandwidth monitoring buffer 19 is used for storing input packets.

  When the input packet control unit 1900 receives an input packet from the input / output line interface 201, the bandwidth monitoring unit 1901 executes a bandwidth monitoring buffer write process S2050, which will be described later, and stores the input packet in the bandwidth monitoring buffer 19. is there.

  When the bandwidth monitoring unit 1901 receives a packet read request from the allocating unit 1902, the bandwidth monitoring unit 1901 executes a bandwidth monitoring buffer reading process S 2250 described later, reads the input packet from the bandwidth monitoring buffer 19, and distributes the input packet to the allocating unit 1902 It is a processing part to transfer to.

  Further, the bandwidth monitoring unit 1901 determines the number of distributed processing units to be used (X described later) (X is an arbitrary natural number from 1 to N) according to the amount of packets accumulated in the bandwidth monitoring buffer 19.

  The distribution unit 1902 is a processing unit that executes a distribution process S2350 described later when a packet transmission request is received from the bandwidth monitoring unit 1901. In the distribution process S2350, the distribution unit 1902 determines a distributed processing unit 503 that executes packet processing from the distributed processing units 503-1 to 503-X. X is determined by the bandwidth monitoring unit 1901.

  The distributed processing unit management list 119 indicates the ID of the distributed processing unit 503 (503-1 to 503-N) to which the distribution unit 1902 distributes the input packet, and the processing status of the distributed processing unit 503.

  The multiplexing unit 1904 multiplexes the packets received from the respective distributed processing units 503 and transmits them to the SW interface 205 in the order of arrival. In addition, the multiplexing unit 1904 manages the arrival order according to the number X of the distributed processing units to be used determined by the bandwidth monitoring unit 1901 and whether or not the packets to be multiplexed are received from the used distributed processing units. To do.

  FIG. 24 is an explanatory diagram illustrating the distributed processing unit management list 119 according to the second embodiment.

  The distributed processing unit management list 119 includes a distributed processing unit 2401 and a processing status counter 2402.

  The distributed processing unit 2401 includes an identifier that uniquely indicates the distributed processing units 503-1 to 503-N. The distributed processing unit 2401 of this embodiment includes a value that identifies the distributed processing unit 503 by a numerical value. The value stored in the distributed processing unit 2401 is set in advance by an administrator or the like.

  The processing status counter 2402 includes a value indicating the status of packet processing in the distributed processing unit 503. The processing status counter 2402 of the present embodiment includes a value related to the execution time that has elapsed since the packet processing was started.

  The processing status counter 2402 shown below indicates the remaining time until the packet processing is completed by the subtraction counter, similarly to the calculation flow management processing S1450 of the first embodiment. Specifically, when packet processing is started, the packet processing time is stored in the processing status counter 2402. Then, the distribution unit 1902 subtracts the value of the processing status counter 2402 every unit time (1 clk in the present embodiment).

  An entry in which “0” is stored in the processing status counter 2402 indicates that the distributed processing unit 503 indicated by the distributed processing unit 2401 has finished packet processing. An entry in which a positive number other than “0” is stored in the processing status counter 2402 indicates that the distributed processing unit 503 indicated by the distributed processing unit 2401 is processing a packet.

  In the following, the processing when the processing status counter 2402 is a subtraction counter will be described, but the processing status counter 2402 may indicate the time elapsed since the packet processing was started by the addition counter. In this case, the allocating unit 1902 may acquire the packet processing status by comparing a predetermined packet processing time with the processing status counter 2402.

  The bandwidth monitoring buffer write process S2050 executed by the bandwidth monitoring unit 1901 will be described below.

  FIG. 20 is a flowchart illustrating the bandwidth monitoring buffer write processing S2050 executed by the bandwidth monitoring unit 1901 according to the second embodiment.

  When receiving an input packet from the input / output line interface 201 (a plurality of input packets may be received simultaneously) (S2000), the bandwidth monitoring unit 1901 acquires the packet length 303 from the in-device header 3 (S2001).

  After S2001, the bandwidth monitoring unit 1901 determines whether the sum of the packet length acquired in S2001 and the accumulated amount of the bandwidth monitoring buffer 19 (packet accumulated amount 21 described later) is equal to or smaller than the buffer size of the bandwidth monitoring buffer 19. Determination is made (S2002).

  In S2002, when the sum of the packet length acquired in S2001 and the accumulated amount of the bandwidth monitoring buffer 19 is equal to or smaller than the buffer size of the bandwidth monitoring buffer 19, the bandwidth monitoring unit 1901 adds the number of received packets to the accumulated packet number. Then, the packet length is updated by adding the acquired packet length to the packet accumulation amount 21, and the received packet is stored in the bandwidth monitoring buffer 19 (S2003).

  After S2005, the bandwidth monitoring unit 1901 ends the bandwidth monitoring buffer write processing S2050 (S2005).

  In S2002, if the sum of the packet length acquired in S2001 and the accumulated amount of the bandwidth monitoring buffer 19 is not less than or equal to the buffer size of the bandwidth monitoring buffer 19, the bandwidth monitoring unit 1901 discards the packet (S2004) and performs bandwidth monitoring. The buffer writing process S2050 is terminated (S2005).

  FIG. 21 is an explanatory diagram illustrating the relationship between the packet accumulation amount 21 of the bandwidth monitoring buffer 19 and the number of usage distribution processing units according to the second embodiment.

  The bandwidth monitoring unit 1901 notifies the distribution unit 1902 of a packet transmission request when the number of packets accumulated in the bandwidth monitoring buffer 19 is 1 or more.

  Further, the bandwidth monitoring unit 1901 determines X (X is an arbitrary natural number from 1 to N) indicating the number of distributed processing units 503 to be used in accordance with the packet accumulation amount 21 of the bandwidth monitoring buffer 19, and a distribution unit 1902 and the multiplexing unit 1904 are notified of the number X of usage sharing processing units. The bandwidth monitoring unit 1901 decreases the value of X when the packet accumulation amount 21 is small (for example, N / 4), and increases the value of X when the packet accumulation amount 21 is large and approaches the buffer size ( For example, N).

  Hereinafter, the bandwidth monitoring buffer read process S2250 executed by the bandwidth monitoring unit 1901 will be described.

  FIG. 22 is a flowchart illustrating the bandwidth monitoring buffer read processing S2250 executed by the bandwidth monitoring unit 1901 of the second embodiment.

  When the bandwidth monitoring unit 1901 receives a packet read request from the distribution unit 1902 (S2200), the bandwidth monitoring unit 1901 subtracts the read packet number from the accumulated packet number, and subtracts the packet length of the read packet from the packet accumulation amount 21. The packet is updated, and the packet is read from the bandwidth monitoring buffer 19 and sent to the distribution unit 1902 (S2201).

After S2201, the bandwidth monitoring unit 1901 ends the bandwidth monitoring buffer read processing S2250 (S2202).
The distribution process S2350 executed by the distribution unit 1902 will be described below.

  FIG. 23 is a flowchart illustrating the distribution process S2350 executed by the distribution unit 1902 of the second embodiment.

  If the distribution unit 1902 receives a packet transmission request from the bandwidth monitoring unit 1901 (S2300), the values of the processing status counters 2402 of all entries # 1 to #X in the distributed processing unit management list 119 are “0”. It is determined whether or not they are different (S2301).

  If it is determined in S2301 that the processing status counters 2402 of all entries # 1 to #X in the distributed processing unit management list 119 are different from “0”, all the distributed processing units 503 (503) of # 1 to #X -1 to 503-X) are executing packet processing. Since the new packet processing cannot be accepted, the distribution unit 1902 returns to the processing of S2301 and waits for the processing of any of the distributed processing units 503 of # 1 to #X to end.

  If there is at least one entry whose processing status counter 2402 is “0” among the entries of # 1 to #X in the distributed processing unit management list 119 in S2301, all the distributed processing units of # 1 to #X. At least one distributed processing unit 503 among 503 (503-1 to 503-X) is in a state where packet processing can be accepted. Therefore, the allocating unit 1902 notifies the bandwidth monitoring unit 1901 of a packet read request and receives a packet from the bandwidth monitoring unit 1901 (S2302).

  After S2302, from the number indicated by the value of the internal counter (this counter takes a value of 1 to X and returns to 1 next to X), the distributed processing units 503 (503-1 to 503-503) are received in the order in which the packets are received. The packet is transferred to (X) (S2303).

  After S2303, the distribution unit 1902 counts up the value of the internal counter by the number of received packets (S2304).

  X is a value determined by the number of usage distribution processing units notified from the bandwidth monitoring unit 1901 and varies according to the packet accumulation amount 21 of the bandwidth monitoring buffer 19. In the distribution process S2350, the timing for reflecting the change in the value of X may be the time when the value of an internal counter (to be described later) returns from X to 1 (wraps around). Similarly to the distribution unit 1902, the multiplexing unit 1904 also includes an internal counter (this counter takes a value of 1 to X and returns to 1 after X), and the distributed processing unit 503 indicated by the number of the internal counter. The timing when the packets are multiplexed in order from (503-1 to 503-X) and the change in the value of X is reflected may be the time when the value of the internal counter described later returns from X to 1 (wraps around). In addition, when the multiplexing unit 1904 receives a packet from a distributed processing unit different from the distributed processing unit indicated by the number of the internal counter, the multiplexing unit 1904 may discard the packet.

  After S2304, it is determined whether or not the packet transmission request from the bandwidth monitoring unit 1901 has ended (S2305).

  If the packet transmission request from the bandwidth monitoring unit 1901 has not ended in S2305, the allocating unit 1902 returns to the processing of S2301.

  If the packet transmission request from the bandwidth monitoring unit 1901 has ended in S2305, the distribution unit 1902 ends the distribution process S2350 shown in FIG. 23 (S2306).

  As described above, according to the second embodiment, the packet communication device 2 includes the distribution processing unit 503 (503-1 to 503-X) (X) in the bandwidth monitoring unit 1901 in which the distribution unit 1902 distributes packets in round robin. The number of distributed processing units 503 is notified to the distribution unit 1902 and the multiplexing unit 1904. As a result, the input packet control unit 1900 according to the second embodiment changes the number of distributed processing units to be used according to the bandwidth of the input packet, and the calculation unit 703 simultaneously executes the statistical count processing regarding the packets of the same flow. It is possible.

  Therefore, according to the second embodiment, performance degradation due to load concentration on one distributed processing unit can be suppressed in a multi-core packet processing engine. Furthermore, by increasing the number of distributed processing units, it is possible to speed up the packet processing engine in a scalable manner. Furthermore, when the input bandwidth is small, it is possible to reduce power consumption by reducing the number of distributed processing units to be used.

  The packet communication apparatus 2 according to the first embodiment implements the statistical count process by implementing the centralized / distributed processing method of the present invention. The packet communication apparatus according to the third embodiment performs bandwidth control based on each input packet. In the third embodiment, the packet communication apparatus uses a token bucket algorithm in order to realize the bandwidth control function.

  The operation outline of the token bucket algorithm will be described below.

  When the token bucket algorithm is used, the packet communication apparatus adds tokens to the token bucket (a predetermined area held in the memory) at a speed corresponding to the set bandwidth. When the token for the packet length of the packet scheduled to be transmitted by the packet communication apparatus is accumulated in the token bucket, the packet communication apparatus determines that the token is within the guaranteed bandwidth, and subtracts the token for the packet length from the token bucket. . Further, the packet communication device determines that the token for the packet length of the packet to be transmitted by the packet communication device is not stored in the token bucket, and determines that the packet is out of the guaranteed bandwidth, holds the token, and stores the token for the packet length. Do not subtract tokens from token buckets.

  An outline of the relationship between the packet arrival interval and the packet processing time (timing) in this embodiment is shown below.

  FIG. 33C is an explanatory diagram illustrating packet processing timing when a plurality of packets of the same flow arrive at the sorting process simultaneously in the packet processing of the third embodiment.

  In the third embodiment, the processing of the central processing unit is different from that in FIG. 33B of the first embodiment.

  Based on the information acquired from each of the distributed processing units (# 1 and # 2), the central processing unit simultaneously performs the bandwidth calculation for the first packet and the bandwidth calculation for the second packet (T3302). Here, in this embodiment, bandwidth calculation is performed using the token bucket algorithm for each packet, so the bandwidth calculation for the second packet subtracts the token for the packet length of the first packet and corrects the token. And calculate. Details regarding this correction will be described later.

  Further, the central processing unit simultaneously performs the bandwidth calculation for the third packet and the bandwidth calculation for the fourth packet based on the information acquired from each of the distributed processing units (# 3 and # 4) (T3303). ).

  Therefore, in this embodiment, the centralized processing unit collects packet information and processes the packets of the same flow collectively, so that even when a plurality of packets of the same flow arrive at the distribution process at the same time, the bandwidth control function Thus, a function that needs to update the same flow information every time a packet arrives can be realized.

  Hereinafter, the third embodiment will be described with reference to FIGS. The third embodiment described below is one of the embodiments of the present invention and does not limit the present invention.

  FIG. 25 is a block diagram illustrating a logical configuration of the central processing unit 2505 according to the third embodiment.

  The central processing unit 2505 of the third embodiment includes an arbitration unit 2501, a calculation flow management list 702, a plurality of calculation units 2503 (2503-1 to 2503-N) (N is an arbitrary natural number), a bandwidth setting table 2504, and a calculation A result holding table 2514 is provided.

  The calculation flow management list 702 in the third embodiment is the same as that in the first embodiment except for the execution time held in the processing status counter 1303 (the difference between the statistical information processing in the first embodiment and the execution time of the bandwidth control processing in the third embodiment). The same as the calculation flow management list 702 in FIG.

  When receiving a processing request from the distributed processing unit 5503 (5503-1 to 5503-N), the arbitrating unit 2501 is a processing unit that executes arithmetic unit selection arbitration processing S2850 described later. The arbitration unit 2501 identifies the flow by the flow ID received from the distributed processing unit 5503, and further refers to the calculation flow management list 702. And thereby, the calculating part 2503 which performs the bandwidth control process regarding an input packet is determined from the calculating parts 2503-1 to 2503-N.

  In addition, the arbitration unit 2501 holds the calculation flow management list 702 in a memory or the like provided in the logic circuit 210. Then, the arbitration unit 2501 updates the computation flow management list 702 in the computation unit selection mediation process S2850 and the computation flow management process S1450.

  Further, the arbitration unit 2501 is a processing unit that executes update arbitration processing S3050, which will be described later, when a computation end notification is received from the computation unit 2503. The arbitration unit 2501 selects information to be written back to the calculation result holding table 2514 from the band control information updated by the calculation unit 2503, and transfers the calculation end notification to the distributed processing unit 5503 that has transmitted the processing request.

  The calculation unit 2503 updates information of each flow included in the calculation result holding table 2514 in a band calculation process S2950 described later. In addition, the calculation unit 2503 transmits the result updated in the band calculation processing S2950 to the arbitration unit 2501. The calculation unit 2503 includes N (N is an arbitrary natural number) in the central processing unit 2505.

  FIG. 26 is an explanatory diagram of a bandwidth setting table 2504 according to the third embodiment.

  The value stored in the bandwidth setting table 2504 is a setting value used by the calculation unit 2503 to execute the token algorithm, and is a value determined in advance by an administrator or the like. The bandwidth setting table 2504 includes a flow ID 2601, a period addition token value 2602, and a token bucket depth 2603.

  The flow ID 2601 includes an identifier that uniquely indicates the flow assigned to the input packet. The period addition token value 2602 indicates the amount of tokens added to the token bucket per unit time (corresponding to the guaranteed bandwidth accumulated per unit time). The token bucket depth 2603 indicates the maximum amount of tokens that the token bucket can hold.

  FIG. 27 is an explanatory diagram of an operation result holding table 2514 according to the third embodiment.

  The calculation result holding table 2514 is a table that holds calculation results for determining whether or not a packet can be transmitted in the guaranteed bandwidth. The calculation result holding table 2514 includes a flow ID 2701, a previous packet arrival time 2702, and a holding token value 2703. The flow ID 2701 includes a flow ID that uniquely indicates the flow assigned to the input packet. The previous packet arrival time 2702 indicates the time when the bandwidth calculation process executed last time in the calculation unit 2503 is executed. The retained token value 2703 indicates the amount of tokens accumulated in the token bucket, which is calculated by the bandwidth calculation process executed last time. That is, the previous packet arrival time 2702 indicates the time when the value stored in the retained token value 2703 is calculated.

  Hereinafter, the calculation unit selection arbitration process S2850 executed by the arbitration unit 2501 will be described.

  FIG. 28 is a flowchart illustrating the calculation unit selection arbitration process S2850 executed by the arbitration unit 2501 according to the third embodiment.

  The arbitration unit 2501 receives the processing start trigger from the distributed processing unit 5503 (5503-1 to 5503-N) (may receive a plurality of processing start triggers at the same time) (S2800), and receives the processing start trigger at the same time. The flow ID and packet length to be acquired are acquired, and the arrival order of the processing start trigger is held (S2801).

  After S2801, the arbitration unit 2501 determines whether or not the values of the processing status counters 1303 of all entries in the computation flow management list 702 are different from “0” (S2802).

  If it is determined in S2802 that the processing status counters of all entries in the calculation flow management list 702 are different from “0”, all the calculation units 2503 (2503-1 to 2503-N) are executing calculation processing. is there. Since the arithmetic processing in the case of a new flow cannot be accepted, the arbitration unit 2501 determines to discard the input packet (S2803), and ends the arithmetic unit selection arbitration processing S2850 shown in FIG. 28 (S2811).

  If there is at least one entry whose processing status counter 1303 is “0” among the entries in the calculation flow management list 702 in S2802, at least one of all the calculation units 2503 (2503-1 to 2503-N). One arithmetic unit 2503 does not execute arithmetic processing. Therefore, the arbitration unit 2501 searches the bandwidth setting table 2504 using the flow ID value acquired in S2801 as a search key (flow ID 2601), and acquires information on the period addition token value 2602 and the token bucket depth 2603. (S2804).

  After S2804, the arbitration unit 2501 searches the calculation result holding table 2514 using the flow ID value acquired in S2801 as a search key (flow ID 2701), and acquires information on the previous packet arrival time 2702 and the holding token value 2703. (S2805).

  In S2802, if the arbitration unit 2501 receives a plurality of processing start triggers in S2800 at the same time, it is checked whether the calculation for the plurality of flows is possible. If there are as many entries in the computation flow management list 702 that the processing status counter 1303 is not “0”, as many as the number of processing start triggers received at S2800, the process proceeds to S2804. If there is no entry whose processing status counter 1303 is “0” for a plurality of flows that have received the processing start trigger at S2800, the process proceeds to S2803. If there are not enough entries for which the processing status counter 1303 is “0” for the number of processing start triggers simultaneously received in S2801, the entry for which the processing status counter 1303 is “0” proceeds to the processing of S2804, and the remaining The flow may proceed to S2803.

  After S2805, it is determined whether or not the value of the flow ID acquired in S2801 matches the value of the flow ID 1302 of an entry whose processing status counter 1303 is different from “0” (S2806).

  In S2806, if the value corresponding to the value of the flow ID acquired in S2801 matches the value of the flow ID 1302 of the entry whose processing status counter 1303 is different from “0”, the arbitration unit 2501 It may be determined that the value corresponding to the value matches the value of the flow ID 1302 of the entry whose processing status counter 1303 is different from “0”.

  In S2806, when it is determined that the value of the flow ID acquired in S2801 does not match the value of the flow ID 1302 of the entry whose processing status counter 1303 is different from “0”, the arbitration unit 2501 does not execute the arithmetic processing. In order to cause the calculation unit to newly execute a calculation process related to the input packet, the process of S2809 is executed.

  In S2806, when it is determined that the value of the flow ID acquired in S2801 does not match the value of the flow ID 1302 of the entry whose processing status counter 1303 is different from “0”, the arbitration unit 2501 sets the processing status counter 1303 to “0”. Among the entries in the computation flow management list 702 that are "", entries are extracted in order from the smallest number of the computation unit 1301. The arbitration unit 2501 determines the arithmetic unit 2503 indicated by the arithmetic unit 1301 of the extracted entry as the arithmetic unit 2503 that newly executes the arithmetic processing related to the input packet.

  In the present embodiment, in order to acquire the bandwidth control result for each packet, when the arbitration unit 2501 acquires a plurality of packets having the same flow ID at the same time, each packet is processed by a separate calculation unit 2503. For this reason, the arbitration unit 2501 manages each packet of the same flow ID acquired at the same time with separate entries in the calculation flow management list 702.

  For example, when the arbitration unit 2501 acquires processing triggers for the first packet and the second packet with the same flow ID at the same time and determines to newly execute the arithmetic processing, the arbitration unit 2501 extracts two entries from the arithmetic flow management list 702 Then, the calculation unit 2503 indicated by the calculation unit 1301 of the extracted entry is determined to be a calculation unit 2503 that newly executes a calculation process related to the first packet or the second packet. Then, the arbitration unit 2501 updates the flow ID 1302 of the extracted entry with the value of the flow ID acquired in S2801. Further, the arbitrating unit 2501 updates the processing status counter 1303 of the extracted entry to a value indicating the calculation time (S2809). For example, when the calculation time is 10 clk, the arbitration unit 2501 updates the value of the processing status counter 1303 of the entry in which the acquired flow ID value matches the flow ID 1302 to “10”.

  By updating the processing status counter 1303 in S2809, the arbitration unit 2501 indicates that each calculation unit 2503 is executing until the calculation process is completed.

In S2809 described above, the arbitration unit 2501 extracts entries in order from the smallest number (the smallest number) of the computation unit 1301 among the entries in the computation flow management list 1302 whose processing status counter 1303 is “0”. However, the arbitration unit 2501 may extract an entry according to any rule as long as it is one of the entries in the computation flow management list 702 whose processing status counter 1303 is “0”. For example, the arbitration unit 2501 may extract the entries in order from the largest value, or the order of entries to be extracted in advance may be determined by the administrator.
After S2809, the arbitrating unit 2501 determines and holds a token correction amount to be used when a plurality of processing start triggers having the same flow ID are acquired simultaneously (S2813). The token correction amount is used when the arbitration unit 2501 acquires a plurality of packets having the same flow ID at the same time. The token correction amount is calculated using a bandwidth calculation process S2950, which will be described later, so that a token calculation is performed each time an input packet arrives. Therefore, this is a value for correction.

  For example, the arbitration unit 2501 sets the token correction amount to 0 when it does not simultaneously acquire processing start triggers regarding a plurality of packets having the same flow ID. When processing start triggers regarding a plurality of packets having the same flow ID (referred to as a first packet and a second packet) are acquired at the same time, the arbitration unit 2501 is determined to perform processing related to the first packet. The token correction amount for 2503 is set to 0, and the token correction amount for the arithmetic unit 2503, which is determined to be processed for the second packet, is set as the packet length of the first packet. Furthermore, when processing triggers related to a plurality of packets having the same flow ID (first packet, second packet, and third packet) are acquired at the same time, the arbitration unit 2501 performs processing related to the third packet. Then, the token correction amount for the arithmetic unit 2503 determined to be the value obtained by adding the packet length of the second packet to the packet length of the first packet. Hereinafter, similarly, when processing triggers related to a plurality of packets having the same flow ID (first packet, second packet, third packet, and fourth packet) are acquired at the same time, the arbitration unit 2501 Is a value obtained by adding the packet length of the second packet and the packet length of the third packet to the packet length of the first packet, and the token correction amount for the arithmetic unit 2503 determined to process the fourth packet. To do.

  After S2813, the arbitration unit 2501 outputs a calculation notification including the processing start trigger, the new / continuation parameter, the acquired packet length, the table acquisition result, and the token correction amount to the determined calculation unit 2503 (S2810).

  In S2810, the arbitration unit 2501 determines the new / continuation parameter as a value indicating new, and includes the new / continuation parameter indicating “new” (“0” in this embodiment) in the calculation notification. The new / continuation parameter in the present embodiment is a value for notifying each arithmetic unit 2503 what kind of arithmetic processing is to be executed.

  In S2810, the process start trigger included in the output calculation notification is a value indicating that the calculation unit 2503 starts a band calculation process S2950 described later. The acquired packet length included in the calculation notification is the packet length acquired in S2801. In this embodiment, even when a plurality of packets having the same flow ID are acquired at the same time, the packet length is not added as in the first embodiment because the processing is executed by different calculation units.

  In S2810, the table acquisition results included in the calculation notification are the periodic addition token value, the token bucket depth, the previous packet arrival time, and the retained token value acquired in S2804 and S2805. In S2810, the token correction amount included in the calculation notification is the token correction amount held in S2813.

  The arbitration unit 2501 outputs the token correction amount to the calculation unit 2503, so that even when input packets with the same flow ID are received simultaneously, a plurality of calculation units 2503 can perform token calculation for each packet. The bandwidth calculation process can be executed normally.

  After S2810, the calculation unit selection arbitration process S2850 shown in FIG. 28 ends (S2811).

  If it is determined in S2806 that the value of the flow ID acquired in S2801 matches the value of the flow ID 1302 of the entry whose processing status counter 1303 is different from “0”, the calculation unit 2503 corresponding to the flow ID of the input packet. Can continue the arithmetic processing related to the input packet. For this reason, the arbitration unit 2501 determines the calculation unit 2503 indicated by the calculation unit 1301 of the entry whose flow ID 1302 matches the acquired flow ID value as the calculation unit 2503 that continuously executes the calculation process related to the input packet. .

  Then, the arbitration unit 2501 updates the value of the processing status counter 1303 of the entry in which the acquired flow ID value matches the flow ID 1302 to a value indicating the calculation time (S2807).

After S2807, the arbitration unit 2501 determines and holds a token correction amount to be used when a plurality of process start triggers having the same flow ID are simultaneously acquired, similarly to the process of S2813 (S2812).
After S2812, the arbitration unit 2501 transfers the processing start trigger and the new / continuation parameter to the calculation unit 2503 indicated by the calculation unit 1301 of the entry of the calculation flow management list 702 in which the acquired flow ID value matches the flow ID 1302 value. The calculation notification including the acquisition packet length, the table acquisition result, and the token correction amount is output (S2808).

  In S2808, when a plurality of processing triggers regarding packets having the same flow ID arrive at the same time, the arbitration unit 2501 outputs the token correction amount held in S2812, so that the plurality of arithmetic units 2503 can execute tokens for each packet in parallel. The calculation can be performed, and the band calculation process can be executed without causing an overflow.

  After S2808, the calculation unit selection arbitration process S2850 shown in FIG. 28 is terminated (S2811).

  Hereinafter, the band calculation processing S2950 executed by the calculation unit 2503 (2503-1 to 2503-N) will be described with reference to FIGS. 29A and 29B.

FIG. 29A is a flowchart illustrating a process of calculating the token amount to be accumulated until the input packet is received, in the band calculation process S2950 executed by the calculation unit 2503 of the third embodiment.
When the calculation unit 2503 receives the distribution notification including the processing start trigger from the arbitration unit 2501 (S2900), the new / continuation parameter, the acquired packet length, the table acquisition result (period addition token amount, token bucket) included in the calculation notification The depth, the previous packet arrival time, and the retained token value are included), and the token correction amount is acquired (S2901).

  After S2901, the calculation unit 2503 determines whether the new / continuation parameter is “0”, that is, whether the new / continuation parameter indicates new (S2902).

  If the new / continuation parameter is “0”, that is, if it is determined in S2902 that the new / continuation parameter indicates new, the arithmetic unit 2503 proceeds to the processing of S2905.

  If it is determined in S2902 that the new / continuation parameter is not “0”, that is, if it is determined that the new / continuation parameter indicates continuation, the calculation unit 2503 updates the previous table internally held for continuation processing. Time is acquired as the previous packet arrival time (S2903).

  After S2903, the calculation unit 2503 acquires the hold token value for continuation processing received from the arbitration unit 2501 as the hold token value (S2904), and proceeds to the process of S2905.

  In S2905, the calculation unit 2503 calculates the time when the bandwidth setting table 2504 was searched in the calculation unit selection arbitration process S2850 of the arbitration unit 2501 from the time difference from the process start trigger received in S2900, and the next table update time is Internally held for continued processing of received input packets (S2905).

  After S2905, the calculation unit 2503 calculates a difference between the time when the bandwidth setting table 2504 calculated in S2905 is searched and the previous packet arrival time acquired in S2901 as a packet arrival interval. Then, the arithmetic unit 2503 calculates the added token amount by multiplying the calculated packet arrival interval by the periodic addition token value acquired in S2901 (S2906).

  After S2906, the computing unit 2503 adds the added token amount calculated in S2906 to the value of the retained token value acquired in S2901 (S2907), and proceeds to the process of S2908 shown in FIG. 29B.

  The addition result in S2907 is the token amount stored in the token bucket until the input packet is received.

  In addition to the processing of S2902 to S2907, the calculation unit 2503 of the third embodiment may perform different processing according to the value of the new / continuation parameter. For example, when the new / continuation parameter indicates new, the calculation unit 2503 may add a value given in advance by the user to the addition result that is the result of S2907.

  FIG. 29B is a flowchart illustrating a process of determining whether or not an input packet is within the guarantee in the band calculation process S2950 executed by the calculation unit 2503 of the third embodiment.

  After S2907, the arithmetic unit 2503 determines whether or not the addition result in S2907 exceeds the value of the depth of the token bucket acquired in S2901 (S2908).

  In S2908, when it is determined that the addition result in S2907 is equal to or less than the token bucket depth value, the calculation unit 2503 obtains a value obtained by subtracting the token correction amount acquired in S2901 from the addition result in S2907, as the current token amount. (S2909). After S2909, the arithmetic unit 2503 proceeds to the process of S2911.

  In S2908, when it is determined that the addition result in S2907 exceeds the value of the token bucket depth, the calculation unit 2503 calculates a value obtained by subtracting the token correction amount acquired in S2901 from the token bucket depth. The current token amount is held (S2910). After S2910, the arithmetic unit 2503 proceeds to the process of S2911.

  When receiving the input packets with the same flow ID at the same time according to the token correction amounts in S2909 and S2910, the value of the token amount to be updated can be calculated according to the arrival order of the input packets.

  After S2909 or S2910, the calculation unit 2503 determines whether or not the current token amount held in S2909 or S2910 is equal to or larger than the acquired packet length value acquired in S2901 (S2911).

  If it is determined in S2911 that the current token amount is equal to or greater than the value of the acquired packet length, the calculation unit 2503 calculates a value obtained by subtracting the acquired packet length from the current token amount as an update holding token value (S2912). . The update holding token value indicates the amount of token remaining in the token bucket after transmitting the input packet.

  After S2912, the calculation unit 2503 generates a band control result indicating that the band is within the guarantee, and transmits the generated band control result to the arbitration unit 2501 (S2913).

  After S2913, the arithmetic unit 2503 proceeds to the process of S2916.

  If it is determined in S2911 that the current token amount is less than the value of the acquired packet length, the calculation unit 2503 holds the same value as the current token amount as the update holding token value (S2914).

  After S2914, the arithmetic unit 2503 generates a band control result indicating that the band is not guaranteed, and transmits the generated band control result to the arbitrating unit 2501 (S2915).

  After S2915, the arithmetic unit 2503 proceeds to the process of S2916.

  In S2913 or S2914, the calculation unit 2503 generates a band control result indicating that the band is within or not guaranteed, and in S2916 described later, the generated band control result and a notification indicating the end of the calculation. , It may be transmitted to the arbitration unit 2501.

  By transmitting or generating the bandwidth control result in S2913 and S2914, the calculation unit 2503 can cause the input packet to perform processing according to the bandwidth control result in output determination processing S3150 described later.

  After S2913 or S2914, the calculation unit 2503 generates a notification indicating the end of the calculation, and the generated notification using the time when the bandwidth setting table 2504 is searched as the previous packet arrival time and the updated hold token value as the hold token value. At the same time, the data is transmitted to the arbitration unit 2501 (S2916).

  After S2916, the processing shown in FIGS. 29A and 29B is terminated (S2917).

  The update arbitration process S3050 executed by the arbitration unit 2501 will be described below.

  FIG. 30 is a flowchart illustrating the update arbitration processing S3050 executed by the arbitration unit 2501 according to the third embodiment.

  When the arbitration unit 2501 receives the calculation end notification from the calculation unit 2503 (2503-1 to 2503-N) (S3000), the bandwidth control result, the previous packet arrival time, and the holding token included in the received calculation end notification A value is acquired (S3001).

  After S3001, the arbitration unit 2501 determines whether or not calculation end notifications are received from a plurality of calculation units 2503 that are processing the same flow (S3002).

  In S3002, when it is determined that the calculation end notification has not been received from the plurality of calculation units 2503 that are processing the same flow, the arbitration unit 2501 proceeds to the process of S3004.

  If it is determined in S3002 that calculation completion notifications are received from a plurality of calculation units 2503 that are processing the same flow, the arbitration unit 2501 selects a result to be updated in the calculation result holding table 2514 (S3003). ). A method of selecting a result to be updated in the calculation result holding table 2514 in S3003 will be described with reference to FIGS. 31A to 31C described later.

  After S3003, the arbitrating unit 2501 proceeds to the process of S3004.

  In step S3004, the arbitration unit 2501 writes back the previous packet arrival time and the stored token value in the calculation result holding table 2514.

  After S3004, the arbitration unit 2501 uses the retained token value written back to the computation result retaining table 2514 as the previous retained token value for the continuation processing of the next received input packet, and receives the computation end notification. (S3005).

  After S3005, the arbitration unit 2501 identifies the distributed processing unit 5503 (5503-1 to 5503-N) that is the transmission source of the process start trigger from the arrival order of the process start trigger held in the calculation unit selection arbitration process S2850, The calculation completion notification and the bandwidth control result are transferred to the identified distributed processing unit 5503 (S3006).

  After S3006, the arbitrating unit 2501 ends the update arbitration process S3050 (S3007).

  Hereinafter, a selection example of the previous packet arrival time and the retained token value will be described with reference to FIGS. 31A to 31C.

  FIG. 31A shows bandwidth control that is simultaneously notified from a plurality of arithmetic units that process the same flow ID in selecting a result to be updated in the arithmetic result holding table 2514 in the update arbitration processing S3050 executed by the arbitration unit 2501 of the third embodiment. It is explanatory drawing which shows the example when all the results are within guarantee.

  In FIG. 31A, four input packets arrive at the same time, and the calculation unit (1) to the calculation unit (4) process the first packet to the fourth packet of the four input packets in order, and all of the bandwidth control results are An example that was within the warranty is shown. When the bandwidth control result is within the guarantee, the calculation units ((1) to (4)) use the bandwidth calculation processing S2950 to set the update holding token value to a value obtained by subtracting the token for the packet length from the current token amount. . For this reason, the arbitration unit 2501 has the smallest value of the retained token value output from the calculation units ((1) to (4)), the previous packet arrival time of the calculation unit (4) that processed the fourth packet, and Select a retention token value. Note that since the previous packet arrival time is the same in any of the arithmetic units ((1) to (4)), the arbitration unit 2501 determines the previous packet arrival time output from the arithmetic unit (1) by the arithmetic unit (3). You may choose.

FIG. 31B shows bandwidth control that is simultaneously notified from a plurality of arithmetic units that process the same flow ID in selection of results to be updated in the arithmetic result holding table 2514 in the update arbitration processing S3050 executed by the arbitration unit 2501 of the third embodiment. It is explanatory drawing which shows the example when all the results are out of guarantee.
In FIG. 31B, four input packets arrive simultaneously, and the calculation unit (1) to calculation unit (4) process the first packet to the fourth packet of the four input packets in sequence, and all the band control results are An example that was out of warranty is shown. When the bandwidth control result is not guaranteed, the calculation units ((1) to (4)) set the update holding token value to the same value as the current token amount in the bandwidth calculation processing S2950. Since the values of the holding token values output from the calculation units ((1) to (4)) are all the same, the arbitration unit 2501 includes the previous packet arrival time of the calculation unit (1) that processed the first packet, and Select a retention token value. The arbitration unit 2501 may select the previous packet arrival time output from the calculation unit (2) and the hold token value output from the calculation unit (4).

  FIG. 31C shows bandwidth control that is simultaneously notified from a plurality of arithmetic units that process the same flow ID in selection of results to be updated in the arithmetic result holding table 2514 in the update arbitration processing S3050 executed by the arbitration unit 2501 of the third embodiment. It is explanatory drawing which shows the example when a result includes the guarantee inside and a guarantee outside.

  In FIG. 31C, four input packets arrive at the same time, and the calculation unit (1) to the calculation unit (4) sequentially process the first packet to the fourth packet of the four input packets, and the calculation unit (1) and An example is shown in which the bandwidth control result of the computation unit (2) is within the guarantee, and the bandwidth control results of the computation unit (3) and the computation unit (4) are not guaranteed. In this case, the calculation unit (1) and the calculation unit (2) subtract tokens for the packet length. The calculation unit (3) and the calculation unit (4) do not subtract tokens for the packet length, and the update holding token values of the calculation unit (3) and calculation unit (4) are the same as the update holding token value of the calculation (2). become. For this reason, the arbitration unit 2501 selects the previous packet arrival time and the retained token value of the calculation unit (2) that has processed the second packet. The arbitration unit 2501 may select the previous packet arrival time output from the calculation unit (3) and the hold token value output from the calculation unit (4). Further, since the previous packet arrival time is the same, the arbitration unit 2501 may select the previous packet arrival time output by the calculation unit (1).

  Accordingly, the arbitration unit 2501 selects a value to be updated from the calculated value of each holding token calculated according to the arrival order of the input packets when receiving the processing start trigger of the input packets having the same flow ID at the same time. And the value is updated in the calculation result holding table 2514. Therefore, in the conventional technique, it is necessary to update the calculation result holding table 2514 for the number of times the calculation processing notification is received. However, in this application, the calculation processing holding table 2514 is updated by consolidating calculation processing notifications of the same flow. I can do it.

  The output determination processing S3250 executed by the output determination processing unit 603 of the distributed processing unit 5503 (5503-1 to 5503-N) will be described below.

  The output determination processing unit 603 according to the third embodiment rewrites the payload based on the bandwidth control result of each input packet received from the central processing unit 2505.

  FIG. 32 is a flowchart illustrating the output determination processing S3250 executed by the output determination processing unit 603 according to the third embodiment.

  When the output determination processing unit 603 receives an input packet from the input header processing unit 602 (S3200), the output determination processing unit 603 determines whether there is a calculation end notification from the central processing unit 2505 (S3201).

  If it is determined in S3201 that there is a calculation end notification, the packet payload (for example, the CoS value of the VLAN header 403) is overwritten according to the bandwidth control result (within guarantee or not guaranteed) received simultaneously with the calculation end notification. (S3202).

  After S3202, the packet is transferred to the multiplexing unit 504 (S3203), and the output determination process S3250 shown in FIG. 32 is terminated (S3205).

  If it is determined in S3201 that there is no calculation end notification, the input packet is discarded (S3204), and the output determination processing S3250 shown in FIG. 32 is ended (S3205).

  As described above, according to the third embodiment, the packet communication device 2 aggregates information on packets distributed to each distributed processing unit 5503 (5503-1 to 5503-N) in a round robin manner in the arbitrating unit 2501. Notify the arithmetic unit 2503 (2503-1 to 2503-N). As a result, the arithmetic unit 2503 according to the third embodiment can simultaneously execute the bandwidth control processing for packets of the same flow.

  Therefore, according to the third embodiment, similarly to the first and second embodiments, in the multi-core packet processing engine, it is possible to suppress performance degradation due to load concentration on one distributed processing unit. Furthermore, by increasing the number of distributed processing units, it is possible to speed up the packet processing engine in a scalable manner.

2 Packet communication device 3 In-device header 4 Communication packet 5 Fragment header 19 Band monitoring buffer 21 Packet accumulation amount 119 Distributed processing unit management list 200 Network interface board 201 Input / output line interface 202 Logic circuit 203 Input packet control unit 204 Output packet control unit 205 SW interface 206 Switch unit 207 Setting register 208 NIF management unit 301 Output network interface board identifier 302 Flow ID
303 Packet length 401 Destination MAC address 402 Source MAC address 403 VLAN header 404 Ethertype value 405 Payload 406 Frame check sequence 502 Distribution unit 503 Distribution processing unit 504 Multiplexing unit 505 Central processing unit 601 Input header analysis unit 602 Input header analysis unit 603 Output determination processing unit 701 Arbitration unit 702 Operation flow management list 703 Operation unit 704 Statistical information table 1701 Total fragment number 1702 Sequence number 1703 Fragment length 1900 Input packet control unit 1901 Bandwidth monitoring unit 1902 Distribution unit 1904 Multiplexing unit 2501 Arbitration unit 2503 Calculation unit 2504 Band setting table 2505 Central processing unit 2514 Calculation result holding table 5503 Distributed processing unit

Claims (11)

A communication device connected to a network,
An input interface for receiving a plurality of packets transmitted from the network ;
A plurality of distributed processing units that execute a first process of the received plurality of packets;
A distribution unit that distributes the received plurality of packets to any of the plurality of distributed processing units;
A centralized processing unit that aggregates information on the received plurality of packets processed by each of the plurality of distributed processing units and executes a second process ;
A multiplexing unit that aggregates and outputs the packets on which the first processing is performed in each of the plurality of distributed processing units based on the second processing result from the central processing unit,
The distribution unit is
According to the sequence number provided in the distributed processing unit, each of the plurality of packets is distributed to each distributed processing unit in the order of arrival at the input interface,
Each of the distributed processing units
Input header analysis that identifies a packet length and a flow ID from the header of the received packet, and outputs the information including the packet length, the flow ID, and a start trigger of the second processing to the central processing unit Received from the input header processing unit based on the result of the second processing of the central processing unit, and an input header processing unit that converts the header of the packet received from the input header analysis unit for packet transfer The output determination processing unit that controls the transfer of the received packet to the multiplexing unit performs the first process,
The central processing unit
A plurality of arithmetic units that perform statistical information processing of the received packet, a statistical information table that holds the result of the processing for each flow ID, and an arithmetic flow management list that manages the states of the plurality of arithmetic units,
When receiving the information from any of the plurality of distributed processing units,
Select a calculation unit capable of the statistical information processing from the calculation flow management list,
Each of the selected computing units is configured to perform the statistical information processing based on the information received for each flow ID,
The result of the statistical information processing performed by the arithmetic unit is accumulated in the statistical information table, and the second process is performed by an arbitration unit that notifies the distributed processing unit of the end of the calculation. A communication device. The communication device according to claim 1,
The bandwidth monitoring buffer for storing the packet and the input bandwidth of the packet are monitored, and the number of distributed processing units to which the allocating unit distributes processing related to the packet is determined according to the accumulated amount of the packet stored in the bandwidth monitoring buffer. A bandwidth monitoring unit to be changed,
The communication unit, wherein the distribution unit distributes the packet to each distributed processing unit based on the number of distributed processing units notified from the bandwidth monitoring unit. The communication device according to claim 1,
The central processing unit
A plurality of arithmetic units for executing processing relating to the packet;
Determine whether there is a packet of the same flow among a plurality of packets received simultaneously from the distributed processing unit, if there are a plurality of packets of the same flow, determine a correction amount for correcting the calculation result, An arbitration unit that assigns computation processing of each packet to each computation unit;
Each of the calculation units corrects the calculation results regarding the plurality of packets of the same flow based on the correction amount,
The arbitration unit selects any one of the corrected calculation results from each of the calculation units, and updates the calculation result of the corresponding flow held in the table based on the selected calculation result. . The communication device according to claim 3 ,
A setting table including the flow identifier, a token amount per unit time added to a token bucket corresponding to the flow, and a maximum amount of tokens accumulated in the token bucket;
The table holds a calculation result including a flow identifier, a previous packet arrival time, and a holding token value,
The arbitration unit determines the token correction amount of the second packet received simultaneously with the first packet for the first packet and the second packet of the same flow among the plurality of packets received simultaneously from the distributed processing unit. A communication apparatus characterized by having a packet length of the first packet. The communication device according to claim 4 ,
Among the plurality of arithmetic units, the arithmetic unit that executes processing related to the first packet is:
As a result of executing the processing related to the first packet, when it is determined that the token amount accumulated in the token bucket until the first packet is received is less than the packet length of the first packet, Output bandwidth control information indicating that transmission of the first packet is not guaranteed, and the token amount accumulated in the token bucket until the first packet is received;
As a result of executing the processing related to the first packet, when it is determined that the token amount accumulated in the token bucket until the first packet is received is equal to or greater than the packet length of the first packet, the first packet Subtract the token for the first packet length from the bandwidth control information indicating that transmission of one packet is within the guarantee and the token amount accumulated in the token bucket until the first packet is received. Output the token amount
Among the plurality of arithmetic units, the arithmetic unit that executes processing related to the second packet is:
As a result of executing the processing related to the second packet, which is obtained by correcting the processing related to the first packet, the first packet length is calculated from the token amount accumulated in the token bucket until the second packet is received. When it is determined that the token amount obtained by subtracting is less than the packet length of the second packet, the bandwidth control information indicating that the transmission of the first packet is not guaranteed, and the second packet Output the token amount obtained by subtracting the token for the first packet length from the token amount accumulated in the token bucket until reception;
As a result of executing the processing related to the second packet, which is obtained by correcting the processing related to the first packet, the first packet length is calculated from the token amount accumulated in the token bucket until the second packet is received. When it is determined that the token amount obtained by subtracting is greater than or equal to the packet length of the second packet, the bandwidth control information indicating that the transmission of the first packet is within the guarantee and the second packet are received. Output a token amount obtained by subtracting tokens for the first packet length and the second packet length from the token amount accumulated in the token bucket until
The mediation unit
When the bandwidth control information of the first packet is out of guarantee, the token amount output by the arithmetic unit that executed the processing related to the first packet is updated to the stored token value of the table,
When the bandwidth control information of the first packet is within the guarantee and the bandwidth control information of the second packet is not the guarantee, the token amount output by the arithmetic unit that has performed the processing related to the first packet is Update to the retention token value in the table,
If the bandwidth control information of the second packet is within the guarantee, the token amount output by the arithmetic unit that executed the processing related to the second packet is updated to the stored token value of the table,
Furthermore, the communication device includes:
A communication apparatus comprising: an output determination processing unit that rewrites headers of the first packet and the second packet based on the first and second band control information output from the arithmetic unit. The communication device according to claim 1,
The communication device according to claim 1, wherein the distribution unit fragments the packet and distributes the packet to the distributed processing units. The communication device according to claim 1,
The communication device, wherein the distribution unit distributes a header portion of the packet to each of the distributed processing units. A communication method by a communication device connected to a network, wherein the communication device is:
An input interface for receiving a plurality of packets transmitted from the network ;
A plurality of distributed processing units that execute a first process of the received plurality of packets;
A distribution unit that distributes the received plurality of packets to any of the plurality of distributed processing units;
A centralized processing unit that aggregates information on the received plurality of packets processed by each of the plurality of distributed processing units and executes a second process ;
A multiplexing unit that aggregates and outputs the packets on which the first processing is performed in each of the plurality of distributed processing units based on the second processing result from the central processing unit,
The distribution unit is
According to the sequence number provided in the distributed processing unit, each of the plurality of packets is distributed to each distributed processing unit in the order of arrival at the input interface,
Each of the distributed processing units
Input header analysis that identifies a packet length and a flow ID from the header of the received packet, and outputs the information including the packet length, the flow ID, and a start trigger of the second processing to the central processing unit Received from the input header processing unit based on the result of the second processing of the central processing unit, and an input header processing unit that converts the header of the packet received from the input header analysis unit for packet transfer The output determination processing unit that controls the transfer of the received packet to the multiplexing unit performs the first process,
The central processing unit
A plurality of arithmetic units that perform statistical information processing of the received packet, a statistical information table that holds the result of the processing for each flow ID, and an arithmetic flow management list that manages the states of the plurality of arithmetic units,
When receiving the information from any of the plurality of distributed processing units,
Select a calculation unit capable of the statistical information processing from the calculation flow management list,
Each of the selected computing units is configured to perform the statistical information processing based on the information received for each flow ID,
The result of the statistical information processing performed by the arithmetic unit is accumulated in the statistical information table, and the second process is performed by an arbitration unit that notifies the distributed processing unit of the end of the calculation. A characteristic communication method. The communication method according to claim 8 , comprising:
The mediation unit
Notifying the one arithmetic unit of information including the number of packets related to a plurality of packets of the same flow and the total value of the packet length,
The one arithmetic unit performs a statistical process based on information including the notified number of packets and a total value of packet lengths, and notifies the arbitration unit of a calculation result. The communication method according to claim 8 , comprising:
The bandwidth monitoring buffer for storing the packet and the input bandwidth of the packet are monitored, and the number of distributed processing units to which the allocating unit distributes processing related to the packet is determined according to the accumulated amount of the packet stored in the bandwidth monitoring buffer. A bandwidth monitoring unit to be changed,
The allocating unit distributes the packet to each distributed processing unit based on the number of distributed processing units notified from the bandwidth monitoring unit. The communication method according to claim 8 , comprising:
The central processing unit
A plurality of arithmetic units for executing processing relating to the packet;
Determine whether there is a packet of the same flow among a plurality of packets received simultaneously from the distributed processing unit, if there are a plurality of packets of the same flow, determine a correction amount for correcting the calculation result, An arbitration unit that assigns computation processing of each packet to each computation unit;
Each of the calculation units corrects the calculation results regarding the plurality of packets of the same flow based on the correction amount,
The arbitration unit selects any one of the corrected calculation results from each of the calculation units, and updates a calculation result of a corresponding flow held in the table based on the selected calculation result. .

Images