JP6372103B2 - Switch device, data transfer control method, and data transfer control program - Google Patents

Description

  The present invention relates to a switch device, a data transfer control method, and a data transfer control program, and more particularly to a technique for determining a transfer destination of communication data by searching a transfer table.

  Japanese Patent Application Laid-Open No. 2004-228561 discloses an apparatus for realizing a pattern matching process having a high-speed search performance corresponding to a network band. In this multi-core CPU, a flow table is held in a shared memory shared by a plurality of processor cores. In this device, when a packet arrives, header information is extracted and the search is assigned to a specific processor core in the multi-core CPU. The processor core calculates a hash value from header information (search key) extracted from the packet, and searches for an entry from the flow table using the hash value.

  Patent Document 1 discloses an apparatus including M multi-core CPUs as another aspect. In this apparatus, the flow table is divided into M pieces and held in a shared memory in each multi-core CPU.

International Publication No. 2011/078108

  However, Patent Document 1 does not disclose any way of dividing the flow table. Therefore, the patent document 1 may have the following problem. Hereinafter, the problem will be described with reference to the drawings. In addition, the content demonstrated below is the content which this inventor newly examined, and does not demonstrate a prior art.

  FIG. 11 shows a case where a corresponding record is retrieved from a transfer table including one or more records by a plurality of SIMD cores 301 to 303 belonging to the same instruction group in the SIMD type accelerator device. Each of the SIMD cores 301 to 303 searches for a corresponding record from a transfer table independent for each communication port for a packet received via each of the communication ports A to C. In addition, the transfer table referred to by each of the SIMD cores 301 to 303 has a different number of records for each communication port. In FIG. 11, the size (number of records) of the transfer table referred to by the SIMD core is 1, the size of the transfer table (number of records) referred to by the SIMD core is 2, and the size of the transfer table referred to by the SIMD core ( The case where the number of records) is 3 is shown.

  In the SIMD type architecture, all the SIMD cores 301 to 303 belonging to the same instruction group need to execute the same instruction, and even if their own processing is completed, it is necessary to wait for the completion of the processing of other SIMD cores. should not. Therefore, in the case illustrated in FIG. 11, the SIMD core having a large number of search processing steps (the number of transfer table records to be referred to) is waited by the SIMD core having a small number of search processing steps.

  That is, as shown in FIG. 12, the SIMD core 301 that references the size 1 transfer table finishes the search process in 1 step, and the SIMD 302 core that references the size 2 transfer table finishes the search process in 2 steps. The SIMD core 303 that refers to the transfer table of size 3 finishes the search process in 3 steps. Therefore, the SIMD core 301 that refers to the size 1 transfer table waits 2 steps by the SIMD core 303, and the SIMD core 302 that refers to the size 2 transfer table waits 1 step by the SIMD core 303.

  As described above, when a transfer table is searched in parallel by being distributed among a plurality of SIMD cores, if the original transfer table is divided without consideration and assigned to each of the SIMD cores, some SIMD cores There is a problem of waiting. That is, there is a problem that the processing efficiency of the SIMD core (processor core) decreases.

  In order to solve the above-described problems, an object of the present invention is to provide a switching device and data transfer control capable of improving the processing efficiency of a processor core when a transfer table is searched by being distributed among a plurality of processor cores. A method and a data transfer control program are provided.

  The switch device according to the first aspect of the present invention is a switch device that determines a transfer destination of communication data received from the outside and transfers the communication data to the determined transfer destination, and is extracted from the communication data A transfer table storage unit that stores a search key and a transfer table including a plurality of records indicating the transfer destination of communication data from which the search key is extracted, and among the plurality of records included in the transfer table, A plurality of processor cores that determine the transfer destination of the communication data by searching the assigned record using a search key extracted from the communication data, the number of records included in the transfer table, and the processor core The transfer table is divided and the records are allocated to each of the plurality of processor cores on an average basis. A table subdivision unit, those having a.

  A data transfer control method according to a second aspect of the present invention is a data transfer control method for determining a transfer destination of received communication data and transferring the communication data to the determined transfer destination, which is extracted from the communication data Storing a transfer table including a plurality of records indicating a search key to be transmitted and a transfer destination of communication data from which the search key is extracted in a transfer table storage unit; and a plurality of records included in the transfer table. Allocating to a plurality of processor cores, and the plurality of processor cores search for a record allocated to itself among a plurality of records included in the transfer table with a search key extracted from the communication data. A search step for determining a transfer destination of the communication data, and in the assignment step, the transfer table The number of records included, based on the number of processor cores, in which average assigns the record to each of the plurality of processor cores by dividing the forwarding table.

  A data transfer control program according to a third aspect of the present invention is a data transfer control program for determining a transfer destination of received communication data and transferring the communication data to the determined transfer destination, wherein the communication data A storage process for storing a transfer table including a plurality of records indicating a search key extracted from the communication key and a transfer destination of communication data from which the search key is extracted, in a transfer table storage unit; and a plurality of processes included in the transfer table An allocation process for allocating records to a plurality of processor cores, and the plurality of processor cores search for a record allocated to itself among a plurality of records included in the transfer table with a search key extracted from the communication data By performing the search process for determining the transfer destination of the communication data, the computer executes the search process. , The number of records contained in the forwarding table, based on the number of the processor core, in which averagely assigning the record to each of the plurality of processor cores by dividing the forwarding table.

  According to each aspect of the present invention described above, a switching device, a data transfer control method, and data transfer control that can improve the processing efficiency of a processor core when a transfer table is searched by being distributed among a plurality of processor cores. A program can be provided.

It is a block diagram of the switch apparatus which concerns on embodiment. It is a functional block diagram of the switch apparatus concerning an embodiment. It is a functional block diagram of the communication data input device concerning an embodiment. It is a functional block diagram of a data processing device concerning an embodiment. It is a figure which shows an example of the transfer table memorize | stored in the transfer table division part which concerns on embodiment. It is a figure which shows the relationship between the functional block of the data processor which concerns on embodiment, and hardware. It is a flowchart which shows operation | movement of the switch apparatus which concerns on embodiment. It is a figure which shows an example of the relationship between the transfer table of the fragmentation transfer table memory | storage part which concerns on embodiment, and a SIMD core. It is a figure which shows an example of the processing condition of the switch apparatus which concerns on embodiment. It is a schematic block diagram of the switch apparatus which concerns on embodiment. It is a figure which shows an example of the relationship between the transfer table in an inventor's examination content, and a SIMD core. It is a figure which shows an example of the processing condition in the inventor's examination content.

<Embodiment of the Invention>
Hereinafter, preferred embodiments of the present invention will be described with reference to the drawings. Specific numerical values and the like shown in the following embodiments are merely examples for facilitating understanding of the invention, and are not limited thereto unless otherwise specified. In the following description and drawings, matters obvious to those skilled in the art are omitted or simplified as appropriate for the sake of clarity.

  With reference to FIG. 1, the structure of the switch apparatus 100 which concerns on embodiment of this invention is demonstrated. FIG. 1 is a configuration diagram of a switch device 100 according to an embodiment of the present invention.

  The switch device 100 includes a CPU (Central Processing Unit) 101, a main memory 102, a video memory 103, and a GPU (Graphics Processing Unit) 104.

  The CPU 101 controls the switch device 100 in an integrated manner. The CPU 101 implements a function as a software switch (virtual switch) on the switch device 100 by executing switching software (switching program) 110 stored in the main memory 102.

  The main memory 102 stores various information used by the CPU 101, calculation results of the CPU 101, and the like. The main memory 102 is also used for information exchange between the CPU 101 and the GPU 104. The CPU 101 stores information transferred to the GPU 104 in the main memory 102, thereby transferring the information from the main memory 102 to the video memory 103. This information transfer is performed by, for example, DMA transfer.

  The video memory 103 stores various information used by the GPU 104, calculation results of the GPU 104, and the like. The video memory 103 is also used for information exchange between the CPU 101 and the GPU 104. The GPU 104 stores information to be transferred to the CPU 101 in the video memory 103, thereby transferring the information from the video memory 103 to the main memory 102. This information transfer is performed by, for example, DMA transfer.

  The GPU 104 executes various processes in accordance with instructions from the CPU 101. The GPU 104 is a SIMD type accelerator device that operates in a SIMD type architecture. In this embodiment, an example in which a GPU is used as a SIMD type accelerator device will be described. However, the present invention is not limited to this as long as it is a SIMD type accelerator device capable of executing the same processing. The accelerator device is not limited to the SIMD type as long as similar processing can be performed.

  The communication ports 201 to 203 correspond to communication ports connected to the above-described software switch. The communication ports 201 to 203 transmit and receive communication data (packets). The communication ports 201 to 203 function as an input port that receives communication data and transmits it to the software switch, and an output port that transmits the data to the outside. In the present embodiment, the case where the number of communication ports is three is illustrated, but the number is not limited to this example. The communication port may include a reception-only communication port and a transmission-only communication port.

  With the above-described configuration, the switch device 100 operates as follows. When the CPU 101 receives communication data via the communication ports 201 to 203, the CPU 101 instructs the GPU 104 to determine the transfer destination of the communication data. In response to an instruction from the CPU 101, the GPU 104 searches for a transfer destination of the communication data from the transfer table based on information extracted from the communication data. Here, the GPU 104 has a plurality of SIMD cores as will be described later, and performs a search in parallel by the plurality of SIMD cores. The GPU 104 notifies the CPU 101 of the searched transfer destination. The CPU 101 transfers the communication data to the transfer destination notified from the GPU 104.

  Here, the switch device 100 is, for example, a server capable of constructing one or more virtual machines. And the switch apparatus 100 implement | achieves the software switch which connects between virtual machines with the above-mentioned apparatuses 101-104.

  Then, with reference to FIGS. 2-6, the functional block of the switch apparatus 100 which concerns on embodiment of this invention is demonstrated. FIG. 2 is a functional block diagram of the switch device 100 according to the embodiment of the present invention. FIG. 3 is a functional block diagram of the communication data input device 1 according to the embodiment of the present invention. FIG. 4 is a functional block diagram of the data processing apparatus 2 according to the embodiment of the present invention. FIG. 5 is a diagram showing an example of a transfer table stored in the transfer table dividing unit 29 according to the embodiment of the present invention. FIG. 6 is a diagram showing a relationship between functional blocks and hardware of the data processing apparatus 2 according to the embodiment of the present invention.

  As illustrated in FIG. 2, the switch device 100 includes a plurality of communication data input devices 1, a data processing device 2, and a plurality of communication data output devices 3.

  The number of communication data input devices 1 is the same as that of communication ports 201 to 203 having a reception function. Similarly, the number of communication data output devices 3 is the same as the number of communication ports having a transmission function. When all of the communication ports 201 to 203 have a transmission / reception function, the number of communication data input devices 1 and the number of communication data output devices 3 are the same. Hereinafter, the three communication data input devices 1 are also referred to as a communication data input device 1a, a communication data input device 1b, and a communication data input device 1c, respectively. The three communication data output devices 3 are also referred to as communication data output devices 3a, 3b, and 3c, respectively.

  As illustrated in FIG. 3, the communication data input device 1 includes a communication data input control unit 11 and a data transfer unit 12. Hereinafter, each of the communication data input control units 11 in the communication data input devices 1a to 1c is also referred to as communication data input control units 11a, 11b, and 11c. In addition, each of the data transfer units 12 in the communication data input devices 1a to 1c is also referred to as a data transfer unit 12a, a data transfer unit 12b, and a data transfer unit 12c.

  The communication data input control unit 11 corresponds to a NIC (Nework Interface Card) emulated on a virtual machine operating on the switch device 100. The communication data input control unit 11 transmits communication data from the virtual machine to the data transfer unit 12. The communication data input control units 11a to 11c are not limited to virtual NICs as described above, and may include physical NICs.

  The data transfer unit 12 receives the communication data transmitted from the communication data input control unit 11 and transmits it to the data processing device 2. Each of the data transfer units 12a to 12c is associated with each of the communication data input control units 11a to 11c, and receives the communication data from the corresponding communication data input control unit 11 independently. Further, each of the data transfer units 12a to 12c is associated with each of input data storage units 21a to 21c included in the data processing apparatus 2 described later, and the communication data received by the associated input data storage unit is independent. And transfer. Each of the data transfer units 12a to 12c corresponds to the communication ports 201 to 203.

  As shown in FIG. 4, the data processing device 2 includes a plurality of input data storage units 21, a transfer table storage unit 22, a transfer table subdivision unit 23, a plurality of subdivision transfer table storage units 24, a transfer table search unit 25, A plurality of match priority determination units 26, a transfer table control information reception processing unit 27, a plurality of action execution units 28, a transfer table division unit 29, and a plurality of output data storage units 30 are provided.

  The number of input data storage units 21, match priority determination units 26, and action execution units 28 is the same as the number of data transfer units 12. The number of output data storage units 30 is the same as the number of communication data output devices 3a to 3c. In the present embodiment, a case where the number of subdivision transfer table storage units 24 is six will be described. Hereinafter, the three input data storage units 21 are also referred to as three input data storage units 21a, 21b, and 21c, respectively. The six subdivision transfer table storage units 24 are also referred to as subdivision transfer table storage units 24a, 24b, 24c, 24d, 24e, and 24f, respectively. The match priority determination unit 26 is also referred to as match priority determination units 26a, 26b, and 26c, respectively. The three action execution units 28 are also referred to as action execution units 28a, 28b, and 28c, respectively. The three output data storage units 30 are also referred to as output data storage units 30a, 30b, and 30c, respectively. As described above, the input data storage unit 21, the match priority determination unit 26, and the action execution unit 28 are prepared in a number corresponding to the communication data input control unit 11. The number of communication data storage units 30 corresponding to the number of communication data output devices 3 is prepared. The number of subdivision transfer table storage units 24 is the same as the number of SIMD cores to be used determined by the transfer table subdivision unit 23, as will be described later.

  The input data storage unit 21 receives and stores the communication data transmitted from the data transfer unit 12. Each of the input data storage units 21a to 21c is associated with each of the data transfer units 12a to 12c, and receives communication data independently from the corresponding data transfer unit 12.

  The transfer table control information reception processing unit 27 processes transfer table control information received from an external device. This transfer table control information is communication data transmitted from the outside to this software switch. When the software switch in the switch device 100 is realized as an OpenFlow switch, the external device corresponds to an OpenFlow controller. The transfer table control information includes information that instructs to add, change, or delete a record in the transfer table. When the transfer table control information is information that instructs to add or change a record, the transfer table control information includes information indicating the record. The transfer table control information reception processing unit 27 transfers the received transfer table control information to the transfer table storage unit 22.

The transfer table storage unit 22 stores a transfer table in advance. The transfer table storage unit 22 updates the transfer table based on the transfer table control information transferred from the transfer table control information reception processing unit 27. The record in the transfer table includes the following information (1) to (3).
(1) Priority (2) Match condition (3) Action

  The priority indicates the priority of a record (action) including itself. The priority is used for comparison for determining which record should be preferentially adopted when a plurality of records are hit when a record corresponding to the received communication data is searched. As long as the priority is a comparable value, any value may be adopted. For example, the priority may be indicated by a positive integer value, and a higher value may have a higher priority. In the present embodiment, it is assumed that priorities are set for each record so that the same priority does not exist in a plurality of records included in the transfer table.

  The match condition is information that is compared with a search condition (search key) extracted from the communication data when searching for a record corresponding to the received communication data. That is, in the search, a record including a matching condition that matches the search condition derived from the received communication data is extracted as a record corresponding to the communication data. The match condition is, for example, information indicating an input port from which communication data is received, a destination MAC address of communication data, a source MAC address of communication data, a destination IP address of communication data, a source IP address of communication data, and the like. is there. Note that the match condition is not limited to the information exemplified here, but may be only a part of the information exemplified here, and may include at least one of the information exemplified here and other information.

  The action is information indicating the content of processing performed on the received communication data. The action is, for example, at least one of the processing contents such as outputting the communication data to the communication data output device 3 (communication port), discarding the communication data, and updating the field of the communication data as the processing contents for the communication data. Is shown. The action of outputting communication data to the communication data output device 3 also indicates which of the communication data output devices 3a to 3c (communication ports 201 to 203) is the transfer destination (output destination) of the communication data.

  For example, the following contents are specifically shown in the record.

[Priority] 100
[Match condition] Input port (communication port): communication data input device 1a, destination address: 192.168.1.1, others: anything
[Action] Output to communication data output device 3c

  For example, when the received communication data is received via the communication data input device 1a (communication port 201) and “192.168.1.1” is indicated as the destination address (for example, in the header), the communication The record in the above example is hit as a record corresponding to the data.

  The OpenFlow protocol is widely known as a representative protocol for creating such a transfer table. That is, in the present embodiment, the OpenFlow protocol may be adopted and the flow table may be used as the transfer table, but the present invention is not limited to this.

  As illustrated in FIG. 5, the transfer table dividing unit 29 acquires the transfer table stored in the transfer table storage unit 22, and stores the transfer table divided for each input port (communication port) of the match condition. Thereby, the transfer table shows the transfer table 291 indicating the communication port 201 (communication data input device 1a) as the input port of the match condition, and the transfer table indicating the communication port 202 (communication data input device 1b) as the input port of the match condition. 292 and a transfer table 293 indicating the communication port 203 (communication data input device 1c) as an input port of the match condition.

  The transfer table subdivision unit 23 compares the number of records in the transfer table stored in the transfer table division unit 29 with the number of SIMD cores, and determines a record to be assigned to each SIMD core. The transfer table subdivision unit 23 further subdivides the transfer table divided by the transfer table division unit 29 in accordance with the determined assignment, and stores it in each of the subdivision transfer table storage units 24a to 24f. The transfer table subdivision unit 23 notifies the transfer table search unit 25 of SIMD core allocation information indicating the subdivision transfer table storage unit 24 and the input data storage unit 21 corresponding to each SIMD core.

  Here, the transfer table subdivision unit 23 subdivides each transfer table (each transfer table divided by the transfer table division unit 29) by averaging so that the number of records processed by the SIMD core is approximately the same. . As long as the number of records allocated to each SIMD core is determined to be closer to the same number, an arbitrary method may be adopted. For example, first, one SIMD core is assigned to each of the transfer tables 291 to 293, and SIMD cores are sequentially added to the transfer tables 291 to 293 in which the result of dividing the number of records by the assigned SIMD core is the largest. The processing to be performed may be determined by a method of repeating until there are no assignable SIMD cores.

  In the SIMD core assignment information, the input data storage unit 21a is assigned to the SIMD core to which the transfer table obtained by dividing the transfer table 291 is assigned, and the input to the SIMD core to which the transfer table obtained by dividing the transfer table 292 is assigned. The input data storage unit 21c is allocated to the SIMD core to which the data storage unit 21b is allocated and the transfer table obtained by dividing the transfer table 291 is allocated.

  As shown in FIG. 6, the transfer table search unit 25 has a plurality of SIMD cores 31. The plurality of SIMD cores 31 are included in the GPU 104 as described above. As described above, the number of the plurality of SIMD cores 31a to 31f used by the transfer table search unit 25 is the same as the number of the subdivision transfer table storage units 24a to 24f.

  Here, in this embodiment, as a result of processing the transfer tables 291 to 293 based on the number of records in the transfer table and the number of SIMD cores in the transfer table subdivision unit 23, the transfer table 291 corresponding to the communication port 201 is The transfer table 292 corresponding to the communication port 202 is stored in the subdivision transfer table storage units 24b and 24c and stored in the subdivision transfer table storage units 24b and 24c. The case where the corresponding transfer table 293 is divided into three and stored in the subdivision transfer table storage units 24d to 24f will be described.

  In this case, in the transfer table subdivision unit 23, the subdivision transfer table storage unit 24a corresponds to the SIMD core 31a and the input data storage unit 21a, and each of the subdivision transfer table storage units 24b and 24c corresponds to the SIMD cores 31b and c. Respectively, corresponding to the input data storage unit 21b, generating SIMD core allocation information indicating that each of the subdivision transfer table storage units 24d to 24f corresponds to each of the SIMD cores 31a to 24f and the input data storage unit 21c, The data is output to the transfer table search unit 25.

  As shown in FIG. 6, the segmented transfer table storage unit 24, the transfer table search unit 25, and the match priority determination unit 26 are realized by a SIMD type architecture (video memory 103, GPU 104, etc.). Others (input data storage unit 21, transfer table storage unit 22, transfer table segmentation unit 23, transfer table control information reception processing unit 27, action execution unit 28, transfer table division unit 29, output data storage unit 30, etc. ) Is realized by the CPU 101, the main memory 102, and the like.

  The transfer table search unit 25 matches the communication data stored in the input data storage unit 21 with the record stored in the subdivision transfer table storage unit 24 according to the SIMD core allocation information output from the transfer table subdivision unit 23. Search by comparing the conditions.

  That is, the SIMD core 31a searches the transfer table stored in the subdivision transfer table storage unit 24a corresponding to itself, and the SIMD core 31b transfers the transfer table stored in the subdivision transfer table storage unit 24b corresponding to itself. The SIMD core 31c searches the transfer table stored in the segmented transfer table storage unit 24c corresponding to itself, and the SIMD core 31d is stored in the segmented transfer table storage unit 24d corresponding to itself. The transfer table is searched, the SIMD core 31e searches for the transfer table stored in the subdivision transfer table storage unit 24e corresponding to itself, and the SIMD core 31f stores it in the subdivision transfer table storage unit 24f corresponding to itself. Search the forwarded transfer table. Further, the SIMD core 31a acquires the communication data stored in the input data storage unit 21a corresponding to itself, and searches for the communication data. The SIMD cores 31b and 31c are input data storage units corresponding to themselves. The communication data stored in 21b is acquired and the communication data is searched, and the SIMD cores 31d to 31f acquire the communication data stored in the input data storage unit 21c corresponding to the communication data, and the communication data Search for.

  Therefore, in this example, there are an instruction group including the SIMD core 31a, an instruction group including the SIMD cores 31b and 31c, and an instruction group including the SIMD cores 31d to 31f. That is, the transfer table search unit 25 divides the SIMD cores 31a to 31f into different instruction groups based on the SIMD core assignment information.

  Further, the transfer table search unit 25 stores a dummy record used when there is no record hit in the match condition. When the transfer table search unit 25 detects a record that hits the match condition, the transfer table search unit 25 outputs the hit record and communication data to the match priority determination unit 26. The transfer table search unit 25 outputs a dummy record to the match priority determination unit 26 when a record hitting the match condition cannot be detected. For example, this record may indicate that the priority is the lowest and the communication data is discarded as an action.

  Here, each of the input data storage units 21a to 21c corresponds to each of the match priority determination units 26a to 26c. Therefore, the SIMD core 31a corresponding to the input data storage unit 21a outputs the search result (record and communication data) to the match priority determination unit 26a, and the SIMD cores 31b and 31c corresponding to the input data storage unit 21b The result (record and communication data) is output to the match priority determination unit 26b, and the SIMD cores 31d to 31f corresponding to the input data storage unit 21c output the search result (record and communication data) to the match priority determination unit 26c. To do.

  The match priority determination unit 26 receives the record and communication data output from the transfer table search unit 25. When the match priority determination unit 26 receives records and communication data output from a plurality of SIMD cores, the match priority determination unit 26 compares the priorities of the records and selects the action of the record having the highest priority. Further, the match priority determination unit 26 outputs action information indicating the selected action and communication data to the action execution unit 28. Each of the match priority determination units 26a to 26c corresponds to each of the action execution units 28a to 28c, and outputs action information and communication data to the corresponding action execution unit 28.

  Here, as the match priority determination unit 26, an SIMD core belonging to an instruction group that generates a search result to be processed by itself may be assigned, or an SIMD core belonging to an instruction group different from that may be assigned. Also good. The match priority determination unit 26c will be described as an example. For example, the match priority determination unit 26c may allocate at least one SIMD core among the SIMD cores 31d to 31f that belong to the instruction group that generates the search result to be processed by the match priority determination unit 26c. . In addition, the match priority determination unit 26c may assign a SIMD core (not shown) that belongs to an instruction group different from the instruction group that generates a search result to be processed by the match priority determination unit 26c.

  The action execution unit 28 processes the action indicated by the action information output from the match priority determination unit 26 for the communication data output from the match priority determination unit 26. For example, when communication data is output to the communication data output device 3 (communication port) as an action, communication data is output to the output data storage unit 30 corresponding to the action among the output data storage units 30a to 30c. Remember.

  The output data storage unit 30 outputs the communication data stored by the action execution unit 28 to the communication data output device 3 corresponding to itself. The output data storage unit 30a corresponds to the communication data output device 3a, the output data storage unit 30b corresponds to the communication data output device 3b, and the output data storage unit 30c corresponds to the communication data output device 3c.

  That is, the CPU 101 that functions as the output data storage unit 30 outputs the communication data stored in the memory 102 that also functions as the output data storage unit 30 to the communication data output device 3 corresponding to the output data storage unit 30.

  Therefore, the action execution unit 28 outputs the communication data to the output data storage unit 30a when the communication data output device 3a is indicated as the transfer destination in the action of outputting the communication data to the communication data output device 3 (communication port). When the communication data output device 3b is shown, the communication data is stored in the output data storage unit 30b. When the communication data output device 3c is shown, the communication data is stored in the output data storage unit 30c. To do.

  Similar to the communication data input device 1, the communication data output device 3 corresponds to a NIC and a communication port emulated on a virtual machine. The communication data output device 3a corresponds to the NIC and communication port 201 of the communication data input device 1a, the communication data output device 3b corresponds to the NIC and communication port 202 of the communication data input device 1b, and the communication data output device 3c This corresponds to the NIC and communication port 203 of the communication data input device 1c. As described above, in this embodiment, since the communication ports 201 to 203 have the transmission / reception function, the number of the communication data output devices 3 is the same as the number of the communication data input devices 1, but the communication ports only have the transmission function. May also exist. The number is not necessarily the same. Similarly, for the communication data input device 1, in the case of the communication data input device 1, a communication port having only a reception function may exist.

  That is, the transfer destination of the communication data received from each input port (communication ports 201 to 203) is searched by the data processing device 2, and the communication data output device 3 (the communication port 201 to 203 of the communication ports 201 to 203) serving as the transfer destination according to the search result. By transferring to any one of them, the communication data can be transferred to a desired transfer destination (for example, a virtual machine).

  As described above, in this embodiment, the communication data received by each of the communication ports 201 to 203 (data transfer units 12a to 12c) is input data storage corresponding to each of the communication ports 201 to 203. Units 21a to 21c, SIMD cores 31a to 31f (the SIMD core 31 is the SIMD core 31a, the SIMD cores 31b and 31c, and the SIMD cores 31d to 31f), the match priority determination units 26a to 26c, and the action execution It is possible to perform processing independently via each of the units 28a to 28c and the output data storage units 30a to 30c. Therefore, it is possible to process the communication data received from the plurality of communication ports 201 to 203 without serializing it into one.

  In addition, since the communication port maintains independence in this way, even if a certain communication port is heavily loaded, there is no possibility that even communication data received by another communication port will stay. .

  Subsequently, the operation of the switch device 100 according to the embodiment of the present invention will be described with reference to FIG. FIG. 7 is a flowchart showing the operation of the switch device according to the embodiment of the present invention.

  The switch device 100 performs the following operations in advance before processing communication data.

(1) Each of the data transfer units 12 stores the position of the input data storage unit 21 corresponding to itself. That is, in the above example, the data transfer unit 12a stores the input data storage unit 21a as a communication data transfer destination, stores the input data storage unit 21b as a communication data transfer destination, and communicates the input data storage unit 21b with the communication data. Store as data transfer destination.

(2) The transfer table control information reception processing unit 27 receives transfer table control information from the outside, creates a transfer table record including priority, match condition, and action, and stores it in the transfer table storage unit 22.

(3) The transfer table dividing unit 29 receives a record from the transfer table storage unit 22 and divides the transfer table for each table group having the same input port in the match condition. That is, the transfer table is divided as shown in FIG.

(4) The transfer table subdivision unit 23 reads the transfer table for each input port created by the transfer table division unit 29, compares the number of records with the number of SIMD cores, and records the number of records processed by each SIMD core. Further subdivide the forwarding table so that. At the same time, the transfer table fragmentation unit 23 determines the assignment of SIMD cores that are responsible for the respective processes of the fragmented transfer table. The transfer table subdivision unit 23 stores the subdivided transfer table in the subdivision transfer table storage unit 24 corresponding to the SIMD core in charge of processing of the transfer table. At the same time, the transfer table subdivision unit 23 notifies the transfer table search unit 25 of SIMD core allocation information.

  In the example shown in FIG. 8, the total number of records in the transfer tables A (291), B (292), and C (293) is 6, and the number of SIMD cores is 6. Therefore, if the number of records handled by each SIMD core is averaged, it can be averaged by assigning one record to one core. In this case, the transfer table subdivision unit 23 stores the information of each record in each of the subdivision transfer table storage units 24a to 24f, and notifies the transfer table search unit 25 of the SIMD core allocation information corresponding thereto.

  According to this, as illustrated in FIG. 9, in the search in the transfer table search unit 25, the search processing in each of the SIMD cores 31a to 31f can be completed in one step. That is, since the processing time in each SIMD core 31a-31f is averaged, the waiting time of each SIMD core 31a-31f can be eliminated, and the processing efficiency of each SIMD core 31a-31f can be improved.

  That is, the SIMD core allocation information in this case indicates that the SIMD core 31a is in charge of the transfer record of the subdivision transfer table storage unit 24a, and the SIMD core 31b is in charge of the transfer record of the subdivision transfer table storage unit 24b. , Indicates that the SIMD core 31c is responsible for the transfer record of the subdivision transfer table storage unit 24c, indicates that the SIMD core 31e is in charge of the transfer record of the subdivision transfer table storage unit 24e, and stores the subdivision transfer table storage The SIMD core 31f is in charge of the transfer record of the unit 24f, and the SIMD core 31g is in charge of the transfer record of the subdivision transfer table storage unit 24g.

  Hereinafter, processing of communication data will be described.

(1) The communication data input control unit 11 transmits communication data to the data transfer unit 12. As described above, the communication data input control unit 11 corresponds to a NIC on a virtual machine, but other operations than the operation of transmitting communication data to the data transfer unit 12 are general virtual machines. The operation may be the same as that of the above NIC. Thereby, the data transfer part 12 receives communication data (S1).

(2) The data transfer unit 12 stores the received communication data in the input data storage unit 21 corresponding to itself (S2). As described above, each of the data transfer units 12a to 12c processes communication data independently of each other.

(3) The transfer table search unit 25 reads the communication data from the input data storage unit 21 based on the SIMD core assignment information notified from the transfer table subdivision unit 23, the search key extracted from the communication data, and the subdivision The matching condition of each record in the transfer table stored in the generalized transfer table storage unit 24 is compared, and the record with the highest priority is searched for among the matched records. The transfer table search unit 25 transmits the searched record and communication data to the match priority determination unit 26 (S3). If no matching record is detected, the forwarding table search unit 25 performs processing when the priority is the lowest and the action does not match the matching conditions of all the records in the forwarding table (for example, discarding communication data ), And the dummy record and the communication data are transmitted to the match priority determination unit 26.

(4) The match priority determination unit 26 determines the priority of each record transmitted from the transfer table search unit 25, and selects only one record with the highest priority. When only one record is received as a search result like the match priority determination unit 26a, the record is selected. The match priority determination unit 26 transmits communication data and action information indicating the action of the selected record to the action execution unit 28 (S4).

(5) The action execution unit 28 transmits communication data to the communication data output device 3 corresponding to the action indicated by the action information transmitted from the match priority determination unit 26 (S5).

  Hereinafter, effects of the present embodiment will be described.

  The first effect is that the search process is performed separately for each input port (communication port), and the independence of the communication data transfer process for each input port is maintained. Even if the load is high, the transfer processing of other input ports is not affected. For example, when serially processing communication data received from a plurality of input ports into one, when a certain input port receives a large amount of communication data and has a high load, it is received from another input port. There is also a problem that processing delay occurs in the communication data. On the other hand, in this embodiment, since the independence of communication data transfer processing is maintained for each input port, occurrence of such processing delay can be suppressed.

  The second effect is that the SIMD accelerator device can be operated efficiently because records in the transfer table are allocated to each SIMD core on average. That is, since the search processing time in each SIMD core can be averaged, it is possible to reduce the waiting time for completing the processing of other SIMD cores by the SIMD core.

  The third effect is that the cache hit rate increases because the assignment of the SIMD core to the forwarding table is statically determined. In other words, the transfer table is divided for each input port in the record matching condition, and the transfer table corresponding to the input port is searched for the communication data received from the input port. This is because a search for a record including a port in a match condition can be omitted.

  A fourth effect is that communication data that arrives at a certain communication port is guaranteed to be processed in the order of arrival. This is because, in some cases, a plurality of SIMD cores are assigned to one communication port and search processing is performed in parallel, but the communication data processed by the assigned plurality of SIMD cores is always one. . Therefore, subsequent communication data is not overtaken.

  The fifth effect is that the CPU idle time increases because the search processing of communication data is entrusted from the CPU to the SIMD accelerator device. That is, when other processes are executed in parallel by the CPU, the processing time can be increased. As described above, this is effective when a server is used as the switch device 100 and another process such as operating a virtual machine is executed.

<Other embodiments of the invention>
When a search is performed in each of a plurality of SIMD cores, it is preferable that for the same communication data, a record is detected only in one SIMD core, and no record is detected in other SIMD cores (zero search results). However, a record may be detected by each of a plurality of SIMD cores.

  In the above description, in order to cope with this case, the case has been described in which the record indicates the priority, and the match priority determination unit 26 selects the record indicating the highest priority. However, the method of narrowing down the search result when a plurality of records is detected is not limited to this.

  For example, instead of the match priority determination unit 26, a search result adjustment unit as described below may be provided. That is, the search result adjustment unit performs control so that one record is selected from records detected by a plurality of SIMD cores based on a predetermined rule, and other records are not adopted. The predetermined rule is, for example, a rule shown in the following (1) and (2).

(1) Numbers are assigned to records in advance, and the record with the smallest number is selected.
(2) A number is assigned to the SIMD core in advance, and the record detected by the SIMD core having the smallest number is selected from the plurality of SIMD cores from which the record is detected.

<Schematic configuration of the embodiment of the present invention>
The switch device 100 according to the embodiment of the present invention described above can also be understood as a schematic configuration described below. FIG. 10 is a configuration diagram of the switch device 9 which is a schematic configuration of the switch device 100 according to the embodiment of the present invention.

  The switch device 9 includes a transfer table storage unit 91, a transfer table subdivision unit 92, and a plurality of processor cores 93. The switch device 9 determines a transfer destination of communication data received from the outside, and transfers the communication data to the determined transfer destination.

  The transfer table storage unit 91 stores a transfer table including a plurality of records indicating a search key extracted from communication data and a transfer destination of communication data from which the search key is extracted.

  The transfer table segmentation unit 92 determines a transfer destination of the communication data by searching a record assigned to itself among a plurality of records included in the transfer table with a search key extracted from the communication data. .

  The plurality of processor cores 93 divide the transfer table based on the number of records included in the transfer table and the number of processor cores 93 and assign the records to each of the plurality of processor cores 93 on an average.

  According to the switching device 9 described above, since the search processing time in the processor core can be averaged, the waiting time for the search completion of another processor core by the processor core is reduced, and the processing efficiency of the processor core is improved. can do.

  Here, the switch device 100 according to the embodiment of the present invention supplies a program (including the switching software 110) for realizing the functions of the above-described embodiments to the computer (switch device 100), and the switch device 100 ( It can be configured by being executed by the CPU 101 and the GPU 104).

  The program executed by the above-described computer (switch device 100) is stored using various types of non-transitory computer readable media and supplied to the computer (switch device 100). be able to. Non-transitory computer readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (for example, flexible disks, magnetic tapes, hard disk drives), magneto-optical recording media (for example, magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-R / W, semiconductor memory (for example, mask ROM, PROM (Programmable ROM), EPROM (Erasable PROM), flash ROM, RAM (Random Access Memory)) are included. Further, the program may be supplied to the computer by various types of transitory computer readable media. Examples of transitory computer readable media include electrical signals, optical signals, and electromagnetic waves. The temporary computer-readable medium can supply the program to the computer via a wired communication path such as an electric wire and an optical fiber, or a wireless communication path.

  Note that the present invention is not limited to the above-described embodiment, and can be changed as appropriate without departing from the spirit of the present invention.

1 communication data input device 2 data processing device 3 communication data output device 11 communication data input control unit 12 data transfer unit 21 input data storage unit 22 transfer table storage unit 23 transfer table subdivision unit 24 subdivision transfer table storage unit 25 transfer table Search b
26 Match priority determination unit 27 Transfer table control information reception process 28 Action execution unit 29 Transfer table division unit 30 Output data storage unit 31 SIMD core 100 Switch device 101 CPU
102 main memory 103 video memory 104 GPU
110 Switching software 201, 202, 203 Communication port

Claims (9)

A switch device that determines a transfer destination of communication data received from outside, and transfers the communication data to the determined transfer destination,
A transfer table storage unit for storing a transfer table including a plurality of records indicating a search key extracted from the communication data and a transfer destination of the communication data from which the search key is extracted;
Among a plurality of records included in the transfer table, a plurality of processor cores that determine a transfer destination of the communication data by searching a record assigned to itself with a search key extracted from the communication data;
Based on the number of records included in the transfer table and the number of processor cores, a transfer table subdividing unit that divides the transfer table and assigns the records to each of the plurality of processor cores on average.
A switch device comprising: The switch device receives the communication data via each of a plurality of input ports,
The record further indicates the input port and, as the transfer destination, indicates a transfer destination of communication data received via the input port,
The switch device further includes a transfer record dividing unit that divides the transfer table into record groups having the same input port indicated by the record,
The transfer table subdivision unit further divides each of the record groups to assign the records to each of the plurality of processor cores so that the records assigned to the same processor core indicate the same input port. ,
Each of the plurality of processor cores determines a transfer destination of communication data received via an input port indicated by a record assigned to the processor core.
The switch device according to claim 1. The plurality of records further indicate a priority,
The switch device may further include a transfer destination in a record indicating the highest priority among the two or more transfer destinations when the transfer destination of the same communication data is determined by two or more processor cores. With a priority determination unit that determines the transfer destination of communication data,
The switch device according to claim 1 or 2. When the processor core detects a plurality of transfer destinations as the transfer destination of the communication data from the record assigned to itself, the processor core determines the transfer destination in the record indicating the highest priority among the plurality of transfer destinations. Determine the data transfer destination,
The switch device according to claim 3. The switch device further selects one transfer destination according to a predetermined rule from the two or more transfer destinations when the transfer destination of the same communication data is determined by two or more processor cores. It has a search result adjustment unit that determines the data transfer destination,
The switch device according to claim 1 or 2. The switch device further includes:
A SIMD type accelerator device having the plurality of processor cores;
When the communication data is received, the SIMD accelerator device is instructed to determine the transfer destination of the communication data, and the communication data is transferred to the transfer destination determined by the SIMD accelerator device according to the instruction CPU to
The switch device according to claim 1, comprising: The switch device further includes:
Transfer table control information reception processing for receiving transfer table control information for instructing update of a record in the transfer table from the outside and updating the transfer table stored in the transfer table storage unit based on the received transfer table control information Comprising a part,
The switch device according to claim 1. A data transfer control method for determining a transfer destination of received communication data and transferring the communication data to the determined transfer destination,
Storing a transfer table including a plurality of records indicating a search key extracted from the communication data and a transfer destination of the communication data from which the search key is extracted, in a transfer table storage unit;
Assigning a plurality of records included in the transfer table to a plurality of processor cores;
The plurality of processor cores determine a transfer destination of the communication data by searching for a record assigned to itself among a plurality of records included in the transfer table with a search key extracted from the communication data. And a search step to
In the allocating step, based on the number of records included in the transfer table and the number of processor cores, the transfer table is divided and the records are allocated to each of the plurality of processor cores on average.
Data transfer control method. A data transfer control program for determining a transfer destination of received communication data and transferring the communication data to the determined transfer destination,
A storage process for storing a transfer table including a plurality of records indicating a search key extracted from the communication data and a transfer destination of the communication data from which the search key is extracted;
An allocation process for allocating a plurality of records included in the transfer table to a plurality of processor cores;
The plurality of processor cores determine a transfer destination of the communication data by searching for a record assigned to itself among a plurality of records included in the transfer table with a search key extracted from the communication data. Search processing to be executed by a computer,
In the allocation process, based on the number of records included in the transfer table and the number of processor cores, the transfer table is divided and the records are allocated to each of the plurality of processor cores on average.
Data transfer control program.

Images