JP2020106869A - Information processing device, method, and program - Google Patents

Abstract

To provide an information processing device capable of improving or suppressing a decrease in bus utilization efficiency when performing data communication between modules via memory and bus.SOLUTION: If a start address and/or size of data to be communicated and/or both are less than a communication unit of a bus used for communication with memory, an information processing device has a communication control unit that performs packing so as to be the communication unit. Fractional data, which is the data to be communicated, which is a fractional part of the communication unit, is collectively stored in a continuous area of the memory for a plurality of data.SELECTED DRAWING: Figure 4A

Description

The present invention relates to an information processing device, method, and program, and more particularly, to an information processing device, method, and program for processing information by performing data communication between modules via a memory and a bus.

An information processing apparatus including a CPU (Central Processing Unit) may also include an accelerator HW (Hardware) for the purpose of, for example, improving processing performance. FIG. 1 shows an example of HW configuration of an information processing device including a CPU and an accelerator HW.

The memory stores a software program executed by the CPU and data to be processed by the software program and the accelerator HW. When the accelerator HW executes a process instead of the software program, that is, when the process is offloaded from the CPU to the accelerator HW, the software program notifies the accelerator HW of the data to be processed, and the accelerator HW targets the process. Perform processing by referring to the data. If there is processing result data, the accelerator HW writes the processing result data in the memory and notifies the software program. Similarly, when the accelerator HW itself generates data, the accelerator HW also writes the data in the memory and notifies the software program.

As one of the data communication methods between the software program and the accelerator HW as described above, there is a ring queue that mainly uses a pointer for an entry. This ring queue is a FIFO (First In First Out) realized on a memory, and as its entry, a pointer mainly to data to be processed is stored.

An example of a ring queue for communication packet data is a DPDK (Data Plane Development Kit) ring queue. DPDK is a high-speed packet processing library for communication software, which provides a data structure and a group of functions for high-speed packet processing. DPDK stores packets in memory in mbuf format. A data configuration example when packet data is communicated between modules (for example, between a software program and an accelerator HW) using the DPDK ring queue and the mbuf format will be described with reference to FIG.

In the mbuf format, a packet is stored in a memory by using a packet buffer that stores the packet data itself and an mbuf structure that stores metadata about the packet data. The mbuf structure stores the start address and buffer length of the packet buffer, information about the position where the packet data is actually stored in the packet buffer (offset, packet data length), and the like. The packet data is not always stored from the beginning of the packet buffer, but may be stored midway.

The DPDK ring queue is composed of state management data and an array of 8-byte data, for example, and functions as a FIFO. When packet data in the mbuf format is communicated using the DPDK ring queue, the pointer of the mbuf structure is enqueued and dequeued. The producer index that is the state management data of the DPDK ring queue indicates the array index that should store the next Enqueue element, and the consumer index that is also the state management data that stores the next Dequeue element. Indicates the array index.

In the example of FIG. 2, three mbuf structure pointers are stored in the locations of array indexes 1 to 3. Here, when the module that transmits the packet data enqueues the mbuf structure pointer of the packet data in the DPDK ring queue, the pointer is stored at the position of the array index 4 and the producer index is updated to 5. When the module that receives the packet data dequeues from the DPDK ring queue, the mbuf structure pointer is read from the position of array index 1 and passed to the module. At the same time, the consumer index is updated to 2. Each module performs processing on packet data by referring to and updating the mbuf structure on the memory and the packet buffer based on the mbuf structure pointer.

Patent Document 1 discloses a system including a memory for storing cyclic transfer data and aperiodic transfer data together with first and second pointers, and a data transfer circuit connected to the memory and a communication bus. ing. In Patent Document 1, when the data for aperiodic transfer is transferred to the communication bus, if the transfer of the data length designated by the second pointer cannot be completed, the data length is updated by the control unit and transferred to the communication bus. doing.

Patent Document 2 discloses an information processing device using the DPDK and mbuf formats.

JP, 2011-186918, A JP, 2017-046286, A

In the communication by the bus shown in FIG. 1, restrictions may be imposed on the size and the start address. For example, the size may have to be a multiple of some minimum unit size (eg 64 bytes, cache line size). Also, for example, the start address (the start address of read or write) at the time of communication is not a multiple of a certain minimum unit size (for example, 64 bytes, cache line size, it may be different from the minimum unit size related to the size). It may not happen.

On the other hand, the data exchanged between the software program and the accelerator HW does not always satisfy the above restrictions. If the constraint is not met, the bus needs to adjust the size of the data to communicate on the bus and the starting address to meet the constraint. As a result, the size of the data actually communicated on the bus is always larger than the size of the data originally intended to be communicated.

This will be described with reference to FIG. 3 by taking the case where the minimum unit size of the size and the start address are both 64 bytes and the case where packet data is communicated as an example.

FIG. 3A shows a case where the packet data size and the start address are both multiples of the minimum unit size (128 bytes, 64*n). In this case, since the bus constraint is satisfied, it is possible to communicate with the same size and start address.

FIG. 3B shows the case where the size of the packet data is not a multiple of the minimum unit size (136 bytes). In this case, the constraint concerning the size of the bus is not satisfied, and the data actually communicated on the bus is 192 bytes shown by the broken line rectangle.

FIG. 3C shows the case where the start address of the packet data is not a multiple of the minimum unit size (64*n+8). In this case, the constraint on the start address of the bus is not satisfied, and the data actually communicated on the bus is 192 bytes shown by the broken line rectangle.

In Ethernet, which is a widely used communication method, the size of a frame that is packet data can be set in units of 1 byte from 64 bytes to 1518 bytes (or 1522 bytes). Therefore, packet data that does not satisfy the above constraint may frequently occur.

In addition, in the process of packet processing, data is often added or deleted before and after existing packet data. At that time, since the size and the start position of the packet data are updated, they easily occur when they do not satisfy the above constraint.

Further, when the packet data is managed by the metadata (mbuf structure) and the packet buffer as in the case of the DPDK described above, the metadata also needs to be communicated by the bus. If the size or start position of the metadata does not satisfy the above constraint, the problem described above for the packet data will occur for the metadata as well.

Furthermore, the metadata contains the information needed to refer to the packet buffer and the packet data located in it, but if all the information in the metadata is needed by the receiver (software program, accelerator HW). Not necessarily. Although it is possible to reduce the communication volume by communicating only a part of it, there is a possibility that the above constraint may not be satisfied in that case, and as a result, the communication volume reduction effect may not be obtained or conversely increases. There is.

Patent Literature 1 discloses updating the data length when the aperiodic transfer data corresponding to the data length indicated by the second pointer cannot be transferred. However, Patent Document 1 does not take into consideration the restriction on the communication bus and the decrease in the utilization efficiency of the communication bus due to the restriction.

On the other hand, Patent Document 2 discloses an information processing device that communicates using the DPDK and mbuf formats. However, Patent Document 1 does not point out the restrictions associated with bus communication and the reduction in bus utilization efficiency.

An object of the present invention is made in view of such a problem, and an information processing apparatus capable of improving or suppressing a decrease in bus utilization efficiency when performing data communication between modules via a memory and a bus, To provide a method and a program.

According to the first aspect of the present invention, a bus that performs communication in a predetermined communication unit,
The communication device includes a memory connected to the bus, and a communication device that transmits and receives data to and from the memory via the bus, and the communication device has at least one of the data and metadata regarding an attribute of the communication unit of the bus. When the above does not match, the information processing apparatus is characterized in that the non-matching data is packed so as to match the communication unit of the bus for a plurality of data sets.

According to a second aspect of the present invention, the packing, the previous SL when at least one of data and metadata about the attributes is less than the communication units of the bus, and at least one of metadata about the data and its attributes The information processing apparatus according to the first aspect is obtained, which is executed when the communication unit is larger than the communication unit of the bus.

According to a third aspect of the present invention, the communication device, when at least one of the data and the metadata is less than a communication unit of the bus, outputs the fraction data that is a fraction of the communication unit. Information according to the first or second aspect, further comprising a transmission device that collects a plurality of data sets so as to be a communication unit and stores the data sets in a continuous area of the memory via the bus. A processing device is obtained.

According to a fourth aspect of the present invention, the communication device includes a receiving device for reading the fractional data of the data to be communicated from the continuous area. The information processing apparatus according to any one of the aspects is obtained.

According to a fifth aspect of the present invention, the communication device collects the data and metadata regarding an attribute of the data for a plurality of data sets, according to any one of the first to fourth aspects. The information processing device described in 1. can be obtained.

According to a sixth aspect of the present invention, in any one of the first to fifth aspects, the memory is a ring queue that uses a pointer as an entry, and packet data that can indirectly refer to the data by the pointer. An information processing apparatus according to a fifth aspect is obtained, which includes a packet buffer for storing, and an mbuf structure for storing the metadata regarding the packet data in an mbuf format.

According to the seventh aspect of the present invention, there is provided a bus that performs communication in a predetermined communication unit, a memory connected to the bus, and a communication device that transmits and receives data to and from the memory via the bus. In the information processing device, the communication device, when at least one of the data and the metadata regarding the attribute thereof does not match the communication unit of the bus, sets a plurality of non-matching data so as to match the communication unit of the bus. A communication method is obtained which is characterized in that packing is performed for collecting the data sets.

According to an eighth aspect of the present invention, when the communication device includes a transmitter that stores pointers relating to a plurality of the data in a continuous area (pointer data area) of the memory, the transmitter causes the data to be stored. Alternatively, when the start address and/or the size of data to be communicated, which is a part of the data, and/or both are less than the communication unit of the bus used for communication with the memory, the communication unit becomes a fractional part. Fractional data, which is the data to be communicated, is stored in a continuous area of the memory collectively for the plurality of data sets,
The communication method according to the seventh aspect is obtained in which the fractional data of the data to be communicated is read from the continuous area by the receiving device of the communication device.

According to a ninth aspect of the present invention, the pointer is a pointer to metadata including at least a pointer to the data, and the transmitting device collects a part of the metadata for the plurality of data sets. Storing in a continuous area and reading a part of the metadata from the continuous area by the receiving device,
The communication method according to the eighth aspect is obtained.

According to a tenth aspect of the present invention, there is provided a bus that performs communication in a predetermined communication unit, a memory connected to the bus, and a communication device that transmits and receives data to and from the memory via the bus. If at least one of the data and the metadata relating to the attribute does not match the communication unit of the bus, the computer that is provided with the computer may be provided with packing for collecting the mismatched data for a plurality of data sets so as to match the communication unit of the bus. A computer program characterized by being executed is obtained.

According to an eleventh aspect of the present invention, the transmitting device and the receiving device use the same memory area as the pointer data area and the continuous area,
The communication method according to any one of the seventh to tenth aspects is obtained.

With the information processing device, method, and program of the present invention, it is possible to improve the bus utilization efficiency when data communication is performed between modules via a memory and a bus, or to suppress the decrease.

It is a figure which shows the structural example of the general information processing apparatus provided with CPU, memory, and accelerator HW. It is a figure which shows an example of a ring queue and a mbuf format. It is a figure which shows an example of the restrictions in the bus connected to a memory. It is a block diagram explaining the principle of the present invention. It is a figure which shows the structural example of the information processing apparatus of the 1st Embodiment of this invention. It is a figure which shows an example of Packing data which the information processing apparatus of the 1st Embodiment of this invention uses. It is a figure which shows the structural example of the 1st transmission/reception processing part of the 1st Embodiment of this invention. It is a figure which shows the structural example of the 2nd transmission/reception processing part of the 1st Embodiment of this invention. It is a figure which shows the structural example of the ring queue which the information processing apparatus of the 1st Embodiment of this invention uses. It is a figure which shows the structural example of the mbuf structure which the information processing apparatus of the 1st Embodiment of this invention uses. It is a figure which shows the operation example of the information processing apparatus of the 1st Embodiment of this invention. It is a figure which shows the operation example of the information processing apparatus of the 1st Embodiment of this invention. It is a figure which shows the operation example of the information processing apparatus of the 1st Embodiment of this invention. It is a figure which shows the operation example of the information processing apparatus of the 1st Embodiment of this invention. It is a figure which shows the operation example of the information processing apparatus of the 1st Embodiment of this invention. It is a figure which shows the operation example of the information processing apparatus of the 1st Embodiment of this invention. It is a figure which shows the operation example of the information processing apparatus of the 1st Embodiment of this invention. It is a figure which shows the operation example of the information processing apparatus of the 1st Embodiment of this invention. It is a figure which shows the operation example of the information processing apparatus of the 1st Embodiment of this invention. It is a figure which shows the structural example of the information processing apparatus of the 2nd Embodiment of this invention. It is a figure which shows an example of Packing data which the information processing apparatus of the 2nd Embodiment of this invention uses. It is a figure which shows the structural example of the 1st transmission/reception processing part B of the 2nd Embodiment of this invention. It is a figure which shows the structural example of the 2nd transmission/reception processing part B of the 2nd Embodiment of this invention. It is a figure which shows the operation example of the information processing apparatus of the 2nd Embodiment of this invention. It is a figure which shows the operation example of the information processing apparatus of the 2nd Embodiment of this invention. It is a figure which shows the operation example of the information processing apparatus of the 2nd Embodiment of this invention. It is a figure which shows the operation example of the information processing apparatus of the 2nd Embodiment of this invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

[Configuration example of the first embodiment]
First, a configuration example of this embodiment will be described with reference to FIGS. In the present embodiment, a software program that processes a packet (first data processing unit 20 described below) and an accelerator HW (a second data processing unit 30 described below) that processes a packet mutually exchange packets via a memory. The following shows an example of communicating. They use ring queues and mbuf formats to communicate packets. However, the present invention is not limited to this.

First, FIG. 4A is a diagram for explaining the principle of the information processing apparatus according to the present invention. The illustrated information processing device 1 includes a memory unit 10, a bus unit 11, and a communication control unit 12. Here, the communication control unit 12 is connected to the memory unit 10 via the bus unit 11. The bus unit 11 can perform communication in a predetermined communication unit (for example, a unit of 64 bytes). The illustrated communication control unit 12 can send and receive at least one of data and metadata regarding the attribute of the data to and from the memory unit 10 via the bus unit 11.

Here, if at least one of the above-mentioned data and metadata does not match the communication unit of the bus unit 11, the communication control unit 12 performs processing that matches the communication unit of the bus unit 11. Specifically, for the communication unit, the communication control unit 12 performs a process of collecting fractional data in which at least one of data and metadata is a fractional data into a plurality of data sets. Fractional data, which is the data to be communicated, which is a fractional part of the communication unit, is collectively stored for a plurality of data in a continuous area of the memory unit 10 via the bus 11. This processing is executed not only when at least one of the data and the metadata is less than the communication unit, but also when it is greater than the communication unit.

As described above, the information processing apparatus according to the present invention includes the communication control unit 12 that processes data so as to match the communication unit in the bus unit 11. Hereinafter, the information processing device based on the principle of the present invention will be described more specifically.

FIG. 4B is a configuration diagram showing an example of the configuration of the information processing device 1 according to the first embodiment of the present invention. The information processing device 1 illustrated in FIG. 4B includes a memory unit 10, a bus unit 11, a first data processing unit 20, a second data processing unit 30, a first transmission/reception processing unit 40, and a second data processing unit 40. The transmission/reception processing unit 60 is included. The number of components and the connection relationship shown in FIG. 4B are examples.

The memory unit 10 is a memory that stores data, and stores a ring queue, an mbuf structure, packet data, and the like.

The bus unit 11 includes a memory unit 10, hardware (not shown) that realizes the first data processing unit and the first transmission/reception processing unit 40, and hardware (not shown) that realizes the second transmission/reception processing unit 60, for example. It is a bus that mediates data communication (read, write) between them. One or both of the following restrictions are imposed on the communication by the bus unit 11.
-The size of communication must be a multiple of a certain minimum unit size.
-The starting address for communication (start address of read or write) must be a multiple of a certain minimum unit size.

When the request from the read or write request source does not satisfy the constraint, the bus unit 11 adjusts the size and the start address so as to satisfy the constraint and performs actual communication. In the description of the present embodiment, it is assumed that the above-mentioned two restrictions are imposed on the communication by the bus unit 11 and that the minimum unit size is 64 bytes.

The first data processing unit 20 processes packet data. The first data processing unit 20 uses the mbuf format when storing packet data in the memory unit 10. The first data processing unit 20 uses the ring queue with the second data processing unit 30 to request the first transmission/reception processing unit 40 for transmission/reception when transmitting/receiving a packet. The first data processing unit 20 requests the first transmission/reception processing unit 40 to generate and release the ring queue (for the mbuf structure pointer). The first transmission/reception processing unit 40 assigns a queue ID as an identifier to the generated ring queue, and sets a set of a ring queue address and a queue ID in a second Enqueue processing unit 66 and a second Dequeue processing unit 68 described later.

The second data processing unit 30 processes packet data. The second data processing unit 30 uses the ring queue with the first data processing unit 20 to request the second transmission/reception processing unit 60 for transmission/reception when transmitting/receiving a packet.

The first transmission/reception processing unit 40 communicates data with the second transmission/reception processing unit 60 using the ring queue. The details will be described later.

The second transmission/reception processing unit 60 communicates data with the first transmission/reception processing unit 40 using the ring queue. The details will be described later.

The first transmission/reception processing unit 40 and the first data processing unit 20 may be realized, for example, as a software program executed by a CPU (not shown). Further, the second transmission/reception processing unit 60 and the second data processing unit 30 may be realized as, for example, an accelerator HW (not shown). As the accelerator HW, for example, any one of FPGA (Field-Programmable Gate Array), ASIC (Application Specific Integrated Circuit), coprocessor, and CPU may be used.

In the description of the present embodiment, the first transmission/reception processing unit 40 and the first data processing unit 20 are realized as a CPU and a software program, and the second transmission/reception processing unit 60 and the second data processing unit 30 are realized as an FPGA. I shall.

Here, the packing processing performed by the present embodiment will be described. In the present embodiment, regarding the packet data among the data communicated between the first data processing unit 20 and the second data processing unit 30 using the ring queue and the mbuf format, the communication is inefficient due to the restriction of the bus unit 11. The part to be copied is copied to another continuous area for communication. This is called packing processing or simply packing. The copied data is called Packing data. The first data processing unit 20 and the second data processing unit 30 include a program for packing, and by executing the program, packing is performed and Packing data is generated. The Packing data is a packet data fragment. That is, the Packing data is fractional data that is a fraction of the communication unit. Fractional data is generated not only when the packet data is less than the communication unit but also when it is larger than the communication packet data.

In the present embodiment, Packing data from a plurality of packet data is stored as a data set in a continuous area of the first and second data processing units 20 and 30, and communication is performed when the constraint of the bus unit 11 is satisfied. To do. This makes it possible to improve the ratio of useful data to be communicated among the data communicated by the bus unit 11, especially between the memory unit 10 and the second transmission/reception processing unit 60, and improve the bus utilization efficiency. You can

Packing will be described using the example of FIG. In the case of FIG. 3A, the restriction of the bus unit 11 is satisfied, and thus packing is unnecessary. In the case of FIG. 3B, since the packet data size is not a multiple of the minimum unit size, the 129th byte to the 136th byte are targets for packing and are copied as Packing data. In the case of FIG. 3C, the 1st byte to the 56th byte and the 121st byte to the 128th byte are objects to be packed and are copied as Packing data.

An example of generating Packing data will be described with reference to FIG. The first 1 byte of the Packing data of a packet is a flag indicating the presence/absence of Packing data (total 2 bits at the beginning and end of the packet data), and a data_len field (field length 6 Bit, value is in bytes) indicating the size of the Packing data. ). When the packing target exists only on either the head side or the tail side of the packet data, the target data is the second and subsequent bytes of the packing data. When the packing target exists on both the head side and the tail side of the packet data, the second byte of the Packing data is the data_len2 field indicating the data size of the tail side, and the third and subsequent bytes are the target data (head side, The order of the end side).

For example, the configuration of Packing data corresponding to FIG. 3B is as shown in FIG. 5B, and data_len is 8. In this case, the 129th byte to the 136th byte of the packet data are stored in data[0] to data[7]. The configuration of Packing data corresponding to FIG. 3C is that of FIG. 5C, and data_len is 56 and data_len2 is 8. In this case, the 1st to 56th bytes of the packet data are stored in data[0] to data[55], and the 121st to 128th bytes of the packet data are stored in data[56] to data[63]. It

Next, a configuration example of the first transmission/reception processing unit 40 in the present embodiment will be described using FIG. The first transmission/reception processing unit 40 illustrated in FIG. 6 includes a first IF (interface) unit 41, a first management unit 42, a first Packing transmission processing unit 43, a first Enqueue processing unit 44, and a first Packing. The reception processing unit 45 and the first Dequeue processing unit 46 are included.

The first IF unit 41 provides the first data processing unit 20 with an interface for operating the ring queue. The first IF unit 41 provides an IF for generating a ring queue, deleting the ring queue, enqueueing an mbuf structure pointer to the ring queue, and dequeueing an mbuf structure pointer from the ring queue.

The first management unit 42 manages the ring queue stored in the memory unit 10. When requested to generate a ring queue, the first management unit 42 allocates a ring queue memory area on the memory unit 10 and performs initialization. When the first management unit 42 is requested to delete the ring queue, the first management unit 42 releases the memory area allocated for the ring queue. The first management unit 42 creates and releases the ring queue for Packing data together with the ring queue for the mbuf structure pointer.

The first Packing transmission processing unit 43 receives the Enqueue request of the mbuf structure pointer, performs packing, and then requests the Enqueue of (mbuf structure pointer or Packing data) to the first Enqueue processing unit.

The first Enqueue processing unit 44 performs Enqueue processing of data (mbuf structure pointer and Packing data) to the ring queue.

The first Packing reception processing unit 45 receives the Dequeue request of the mbuf structure pointer, requests the first Dequeue processing unit 46 to Dequeue the data (mbuf structure pointer or Packing data), performs packing, and then dequeues. Mbuf structure pointer.

The first Dequeue processing unit 46 performs Dequeue processing of data (mbuf structure pointer and Packing data) from the ring queue.

Next, a configuration example of the second transmission/reception processing unit 60 in the present embodiment will be described using FIG. 7. The second transmission/reception processing unit 60 illustrated in FIG. 7 includes a second IF unit 61, a second Packing transmission processing unit 63, a second packet data transmission unit 64, a second mbuf transmission unit 65, and a second Enqueue process. The unit 66 is provided. The second transmission/reception processing unit 60 further includes a second Dequeue processing unit 68, a second mbuf reception unit 69, a second packet data reception unit 70, a second Packing reception processing unit 71, a memory access unit 73, It is configured to include. The second packet data transmission unit 64, the second mbuf transmission unit 65, the second mbuf reception unit 69, and the second packet data reception unit 70 also exchange with the memory access unit 73, but in order to avoid complication of the figure, Illustration is omitted.

The second IF unit 61 provides the second data processing unit 30 with an interface for transmitting and receiving packets to and from the first data processing unit 20. The second IF unit 61 receives, for example, the following transmission data from the second data processing unit 30 and transmits it to the first data processing unit 20.
・Queue ID
-Mbuf structure pointer-mbuf structure-send packet data

The second IF unit 61 delivers the following as the received data from the first data processing unit 20 to the second data processing unit 30.
・Queue ID
-Mbuf structure pointer-mbuf structure-received packet data

The second Packing transmission processing unit 63 receives the transmission data from the second IF unit 61, determines the necessity of packing, and writes the data to the second packet data transmission unit 64 and the second mbuf transmission unit 65 based on the result. Etc. send instructions. The instruction includes all or part of the transmission data received from the second IF unit 61. Further, the second Packing transmission processing unit 63 generates Packing data, and requests the second Enqueue processing unit 66 to Enqueue the Packing data generated in the second Enqueue processing unit 66 to the ring queue. The second Packing transmission processing unit 63 requests the second Enqueue processing unit 66 to perform Enqueue to the ring queue of the mbuf structure pointer to be transmitted.

The second packet data transmission unit 64 writes the packet data to the memory unit 10 via the memory access unit 73 based on the instruction from the second Packing transmission processing unit 63.

The second mbuf transmission unit 65 writes the mbuf structure in the memory unit 10 via the memory access unit 73 based on the instruction from the second Packing transmission processing unit 63.

The second Enqueue processing unit 66 performs Enqueue processing of data (mbuf structure pointer and Packing data) to the ring queue. The second Enqueue processing unit 66 refers to and updates the ring queue stored in the memory unit 10 via the memory access unit 73. The second Enqueue processing unit 66 performs Enqueue in units of multiples of the minimum unit size of the bus unit 11. The second Enqueue processing unit 66 has an internal buffer for temporarily storing data of less than the unit. The second Enqueue processing unit 66 holds a set of the address and the queue ID for each ring queue (ring queue table). The second Enqueue processing unit 66 provides the first data processing unit 20 with an interface (IF) for operating the ring queue table. The second Enqueue processing unit 66 refers to the p_ring_ptr field of the ring queue for the mbuf structure pointer, and stores the correspondence between the ring queue and the queue ID and the address of the Packing data ring queue.

The second Dequeue processing unit 68 performs Dequeue processing of data (mbuf structure pointer and Packing data) from the ring queue, and passes the Dequeued data to the second Packing reception processing unit 71. At that time, the second Dequeue processing unit 68 also passes the queue ID of the used ring queue. The second Dequeue processing unit 68 refers to and updates the ring queue stored in the memory unit 10 via the memory access unit 73. The second Dequeue processing unit 68 holds a set of its address and queue ID for each ring queue (ring queue table). The second Dequeue processing unit 68 provides the first data processing unit 20 with an IF for operating the ring queue table. The second Dequeue processing unit 68 refers to the p_ring_ptr field of the ring queue for the mbuf structure pointer and stores the correspondence between the ring queue and the queue ID and the address of the Packing data ring queue.

The second Packing reception processing unit 71 receives the mbuf structure pointer Packing data from the second Dequeue processing unit 68, and processes the packet. The second Packing reception processing unit 71 determines whether or not the packing is used, and based on the result, sends an instruction such as data reading to the second mbuf receiving unit 69 and the second packet data receiving unit 70. Further, the second Packing reception processing unit 71 restores the packet data using the Packing data. The second Packing reception processing unit 71 generates reception data including the queue ID, the mbuf structure pointer, the mbuf structure, and the packet data, and notifies the second IF unit 61 of the reception data.

The second mbuf reception unit 69 reads the mbuf structure from the memory unit 10 via the memory access unit 73 using the designated mbuf structure pointer based on the instruction from the second Packing reception processing unit 71.

The second packet data receiving unit 70 reads the packet data stored in the memory unit 10 via the memory access unit 73 based on the instruction from the second Packing reception processing unit 71.

The memory access unit 73 performs reading and writing from the memory unit 10 via the bus unit 11. The memory access unit 73 provides an IF for that purpose.

The connection between the second data processing unit 30 and the second transmission/reception processing unit 60, and between the respective units in the second transmission/reception processing unit 60 described above is, for example, an HW (eg, FPGA) that realizes the second transmission/reception processing unit 60. ) May be performed by the FIFO using the memory. In that case, for example, a FIFO is provided between the second Packing reception processing unit 71 and the second packet data receiving unit 70, and the second Packing reception processing unit 71 writes the instruction data to the FIFO to receive the second packet data reception. The unit 70 is instructed, and the second packet data receiving unit 70 receives the instruction by reading from the FIFO. Further, the second packet data receiving unit 70 writes the read packet data in the FIFO, and the second Packing reception processing unit 71 receives the packet data by reading from the FIFO.

Further, the second Dequeue processing unit 68 may perform the Dequeue from the ring queue in advance before the Dequeue request from the second Packing reception processing unit 71. In that case, the dequeued data is stored in the FIFO provided between the Dequeued data and the second Packing reception processing unit 71. In this case, the fact that the second Packing reception processing unit 71 makes a Dequeue request to the second Dequeue processing unit 68 means that the second Packing reception processing unit 71 reads data from the FIFO.

FIG. 8 shows the structure of the ring queue used in this embodiment, and FIG. 9 shows the structure of the mbuf structure. These are a partial extension of the ring queue and mbuf structure used by DPDK for the sake of simplicity.

The ring queue used in the present embodiment shown in FIG. 8 is mainly composed of three parts. The first is the first 64 bytes, which is updated by the Producer that mainly enqueues the ring queue. The second is the portion corresponding to the next 64 bytes, which is mainly updated by the Consumer that performs Dequeue from the ring queue. The third part is the part from the beginning to the 128th byte onward, which is a FIFO array that stores data (8 bytes) to be enqueued and dequeued in the ring queue (ring[0], ring[1] in FIG. 8). etc). The array for FIFO is updated by the Producer and referenced by the Consumer. In the following, p is attached to fields related to Producer and c is attached to fields related to Consumer.

The first 16 bytes of the ring queue is a name field that stores the name of the ring queue. The p_size field and the c_size field are the size of the FIFO array (8-byte unit, that is, the number of entries). In p_ring_ptr and c_ring_ptr, when the ring queue is for the mbuf structure pointer, the address of the related ring queue for Packing data is set (does not exist in DPDK). p_flag and c_flag are fields that store flags related to the use of the ring queue. The values of these are set when the ring queue is created and are not updated thereafter.

The p_index field is a producer index and indicates the array index in which the data (8 bytes) to be enqueued next in the FIFO array should be stored. The initial value is 0, and it is updated every time data is enqueued by the producer.

The c_index field is a consumer index and indicates an array index from which data (8 bytes) to be dequeued next in the array for FIFO is to be taken out. The initial value is 0, and it is updated every time data is dequeued by the consumer.

The part shown in gray in FIG. 8 is padding.

The first 16 bytes of the mbuf structure used in the present embodiment shown in FIG. 9 are composed of a buf_addr field and a buf_physaddr field which hold a logical address and a physical address of a packet buffer which makes a pair with the mbuf structure.

The buf_len field indicates the size (byte unit) of the packet buffer. The data_off field indicates a packet data start position (byte unit) in the packet buffer. The data_len field indicates the size (byte unit) of the packet data in the packet buffer.

The mbuf structure can compose a Linked List. For example, when storing a large packet that cannot be stored in one packet buffer, a plurality of mbuf structures and packet buffers are used, and they are linked by a Linked List to store the packet. The pkt_len field indicates the total size (in bytes) of such a packet. The next field indicates a pointer to the next mbuf structure in the Linked List.

[Operation example of the first embodiment]
Next, the following operations (A), (B), (C), (D), (E), and (in the information processing apparatus 1 illustrated in FIGS. 4 to 9 as a configuration example of the present embodiment: Regarding F), one example thereof will be described in detail in order with reference to the drawings.
(A) Ring queue generation process (B) Packet transmission process by the first data processing unit 20 (C) Packet reception process by the second data processing unit 30 (D) Packet transmission process by the second data processing unit 30 (E) Packet reception process (F) Ring queue release process by one data processing unit 20

<(A) Ring queue generation process>
The ring queue generation process of the information processing device 1 (FIG. 4) is described with reference to the flowchart of FIG. This ring queue generation process is a process of generating a ring queue. The processing is executed, for example, when the operation of the first data processing unit 20 is started. The first data processing unit 20 executes the processing for each ring queue for the mbuf structure pointer used with the second data processing unit 30 to generate a ring queue.

The first data processing unit 20 requests the first IF unit 41 of the first transmission/reception processing unit 40 shown in FIG. 6 to generate a ring queue (step S100). At that time, the name and size of the ring queue are passed as parameters. The request is transferred to the first management unit 42.

The first management unit 42 allocates the ring queue memory area to be generated on the memory unit 10 (FIG. 4) and initializes each field as follows (step S101). This ring queue is for the mbuf structure pointer.
・Name: Name given by parameter ・p_size and c_size: Size given by parameter ・Other fields: 0

The first management unit 42 allocates a ring queue memory area for Packing data on the memory unit 10 and initializes each field (step S102). The first management unit 42 automatically determines the size of the ring queue for Packing data, for example.

In this case, the first management unit 42 may use a value obtained by multiplying the size given in step S100 by a predetermined numerical value (for example, the minimum unit size of the bus unit 11). Alternatively, the first management unit 42 may receive the size of the Packing data ring queue from the data processing unit 20 via the first IF unit 41. The first management unit 42 automatically determines the name of the ring queue for Packing data, for example. The first management unit 42 may use the name of the Packing data ring queue as a name by adding a predetermined character string (for example, “_P”) to the name given in step S100. The first management unit 42 sets the address of the ring queue assigned in this step in the p_ring_ptr and c_ring_ptr fields of the ring queue for the mbuf structure pointer initialized in step S101.

The first management unit 42 notifies the address of the ring queue generated in step S101 to the first data processing unit 20 via the first IF unit 41 (step S103).

The first data processing unit 20 shown in FIG. 4 is the second Enqueue processing unit 66 of the second transmission/reception processing unit 60 shown in FIG. (When receiving) or the second Dequeue processing unit 68 (when the first data processing unit 20 transmits and the second data processing unit 30 receives), the address of the ring queue obtained in step S103 is set (step S104). ).

<(B) Packet Transmission Process by First Data Processing Unit 20>
A process in which the first data processing unit 20 of the information processing device 1 transmits a packet to the second data processing unit 30 will be described with reference to the flowcharts of FIGS. 11 and 12. The process is executed, for example, when the first data processing unit 20 offloads a part of the packet processing to the second data processing unit 30.

The first data processing unit 20 requests Enqueue from the first IF unit 41 using the address of the ring queue (for mbuf structure pointer) and the mbuf structure pointer as parameters (step S110).

The request is transferred by the first IF unit 41 (FIG. 6) to the first Packing transmission processing unit 43, and the first Packing transmission processing unit 43 determines whether or not packing is required for the packet (step S111). The first Packing transmission processing unit 43 refers to buf_addr, data_off, and data_len of the mbuf structure based on the mbuf structure pointer, and confirms the position and size of the packet data in the packet buffer. The first Packing transmission processing unit 43 determines that packing is necessary when the position and size do not satisfy the constraints of the bus unit 11, and determines that packing is unnecessary when the position and size satisfy the constraint. If it is determined that packing is unnecessary, the process proceeds to step S114.

The first Packing transmission processing unit 43 generates Packing data according to the packing described above (step S112).

The first Packing transmission processing unit 43 acquires the address of the ring queue for Packing data by referring to the p_ring_ptr field of the ring queue, and uses the address and the Packing data generated in step S112 as parameters to set the first Enqueue processing unit 44. To request the Enqueue process (step S113). Details of the Enqueue process by the first Enqueue processing unit 44 will be described later. Enqueue to the ring queue is performed in 8-byte units. Therefore, the first Packing transmission processing unit 43, when Packing data of less than 8B occurs during Enqueue to the ring queue, suspends the Enqueue of the data and puts it together with the Packing data regarding the packet to be processed next or thereafter. Enqueue is performed when the number of bytes is exceeded.

The first Packing transmission processing unit 43 requests Enqueue processing from the first Enqueue processing unit 44 using the address of the ring queue for the mbuf structure pointer and the mbuf structure pointer as parameters (step S114).

Next, the process in which the first Enqueue processing unit 44 enqueues data in the ring queue will be described with reference to FIG. The first Enqueue processing unit 44 receives the address of the ring queue, the data to be enqueued (size of multiple of 8 bytes), and the size of the data (8 byte unit) as parameters.

The first Enqueue processing unit 44 determines whether the Enqueue destination ring queue is Full (step S120). The first Enqueue processing unit 44 determines that the ring queue is Full when the following conditions are satisfied. When determining that the ring queue is Full, the first Enqueue processing unit 44 ends the processing and returns a failure response. Alternatively, the first Enqueue processing unit 44 may repeatedly execute step S120 until the ring queue becomes a state other than Full.

p_index-c_index + size of data to be enqueued (in units of 8 bytes) ≤ p_size (when p_index is greater than or equal to c_index).

p_index + 2^32-c_index + Enqueue data size (8 bytes unit) ≤ p_size (when p_index is less than c_index).

The first Enqueue processing unit 44 stores the data to be enqueued at the position (and after) of the array index p_index of ring which is the array for FIFO (step S121).

The first Enqueue processing unit 44 adds the size of the data to be enqueued (in units of 8 bytes) to p_index (step S122).

<(C) Packet reception process by the second data processing unit 30>
A process in which the second data processing unit 30 (FIG. 4) of the information processing device 1 receives a packet from the first data processing unit 20 will be described with reference to the flowcharts of FIGS. 13 and 14. The process is executed, for example, when the first data processing unit 20 offloads a part of the packet processing to the second data processing unit 30.

The second Packing reception processing unit 71 (FIG. 7) requests Dequeue of the mbuf structure pointer to the second Dequeue processing unit 68, and the second Dequeue processing unit 68 returns the dequeued mbuf structure pointer together with the queue ID (step S130). The operation of the second Dequeue processing unit 68 at this time will be described later. As described above, the Dequeue request by the second Packing reception processing unit 71 may be realized by a read request to the FIFO provided between the second Packing reception processing unit 71 and the second Dequeue processing unit 68. The second Dequeue processing unit 68 may continuously perform the Dequeue processing and write the dequeued data in the FIFO.

The second Packing reception processing unit 71 instructs the second mbuf receiving unit 69 to read the mbuf structure based on the mbuf structure pointer obtained in step S130, and the second mbuf receiving unit 69 causes the memory access unit 73 to read. The mbuf structure is read from the memory unit 10 via (step S131).

The second Packing reception processing unit 71 determines whether packing is used for the packet (step S132). The second Packing reception processing unit 71 makes a determination by the same method as the first Packing transmission processing unit 43 in step S111. That is, the second Packing reception processing unit 71 refers to buf_addr, data_off, and data_len of the mbuf structure to confirm the position and size of the packet data in the packet buffer. The second Packing reception processing unit 71 determines that the packing is not used when the position and the size satisfy the constraint of the bus unit 11, and determines that the packing is used when the position and the size do not satisfy the constraint. When the packing is used, the second Packing reception processing unit 71 instructs the second packet data receiving unit 70 to read the packet data (only the portion satisfying the constraint of the bus unit 11) (shift to step S133). When the packing is not used, the second Packing reception processing unit 71 instructs the second packet data receiving unit 70 to read the entire packet data (shift to step S136).

The second packet data receiving unit 70 reads the designated portion of the packet data from the memory unit 10 via the memory access unit 73 (steps S133 and S136).

The second Packing reception processing unit 71 requests the second Dequeue processing unit 68 to Dequeue the Packing data, and acquires the Packing data (step S134). Dequeue of Packing data is performed in units of multiples of the minimum unit size of the bus unit 11. The second Packing reception processing unit 71 internally stores the dequeued Packing data, refers to it whenever necessary, and consumes only the necessary amount. In this step, the second Packing reception processing unit 71 requests the second Dequeue processing unit 68 to Dequeue the Packing data when the internally stored Packing data becomes insufficient.

The second Packing reception processing unit 71 uses the packet data read in Step S133 (only the portion that satisfies the constraint of the bus unit 11) to include part of the packet data included in the Packing data obtained in Step S134 (start side, end side). Side, one or both) and packet data is restored (step S135).

The second packet data receiving unit 70 creates reception data including a queue ID, an mbuf structure pointer, an mbuf structure, and packet data, and processes the reception data via the second IF unit 61 to the second data processing unit 30 ( (Fig. 4) is notified (step S137).

Next, the process by which the second Dequeue processing unit 68 shown in FIG. 7 dequeues data from the ring queue will be described with reference to FIG. The second Dequeue processing unit 68 performs the following operation for each of the ring queues (for mbuf structure pointers) registered in the ring queue table and the Packing data ring queues associated with them.

The second Dequeue processing unit 68 determines whether or not there is data in the ring queue of the Dequeue source (step S140). The second Dequeue processing unit 68 determines that there is data when p_index and c_index of the ring queue are not equal, and determines that there is no data (the queue is empty) when they are equal. If there is no data in the ring queue, the second Dequeue processing unit 68 ends the processing and returns a failure response. Alternatively, the second Dequeue processing unit 68 may repeatedly execute step S140 until the ring queue is not empty.

The second Dequeue processing unit 68 confirms the size of the data stored in the FIFO array and determines whether the size is equal to or larger than the minimum unit size of the bus unit 11 (step S141). The size (in bytes) of the data stored in the FIFO array is calculated as follows.
(P_index-c_index) *8 (when p_index is greater than or equal to c_index)
(C_index-p_index) *8 (when p_index is less than c_index)

The second Dequeue processing unit 68 ends the process and returns a failure response when the size of the stored data is less than the minimum unit size of the bus unit 11. Alternatively, the second Dequeue processing unit 68 may repeatedly execute step S141 until the condition is satisfied.

The second Dequeue processing unit 68 transfers the data (an integer multiple of the minimum unit size of the bus unit 11) stored at the position of the array index c_index of ring, which is the FIFO array, via the memory access unit 73. Is read from (step S142).

The second Dequeue processing unit 68 adds a value obtained by dividing the size of the extracted data by 8 to c_index and writes the value in the ring queue on the memory unit 10 via the memory access unit 73 (step S143).

<(D) Packet transmission process by second data processing unit 30>
A process in which the second data processing unit 30 (FIG. 4) of the information processing device 1 transmits a packet to the first data processing unit 20 will be described with reference to the flowcharts of FIGS. 15 and 16. The process is executed, for example, when the second data processing unit 30 transmits the packet part after the offload process to the first data processing unit 20.

The second data processing unit 30 requests the second IF unit 61 to transmit a packet (step S150). At that time, the second data processing unit 30 passes the queue ID, the mbuf structure pointer, the mbuf structure, and the packet data as transmission data.

The second IF unit 61 passes the transmission data to the second Packing transmission processing unit 63, and the second Packing transmission processing unit 63 determines whether packing is required for the packet (step S151). The second Packing transmission processing unit 63 refers to buf_addr, data_off and data_len of the mbuf structure to confirm the position and size of the packet data in the packet buffer. The second Packing transmission processing unit 63 determines that packing is necessary when the position and size do not satisfy the constraint of the bus unit 11, and determines that packing is unnecessary when the position and size satisfy the constraint. When it is determined that the packing is unnecessary, the second Packing transmission processing unit 63 instructs the second packet data transmission unit 64 to write the packet data (shift to step S155). At that time, the second Packing transmission processing unit 63 specifies the entire packet data as the write target area.

The second Packing transmission processing unit 63 generates Packing data according to the packing described above (step S152).

The second Packing transmission processing unit 63 requests the second Enqueue processing unit 66 for Enqueue of the generated Packing data, and the second Enqueue processing unit 66 performs Enqueue (step S153). Enqueue processing by the second Enqueue processing unit 66 will be described later. After that, the second Packing transmission processing unit 63 instructs the second packet data transmission unit 64 to write the portion of the packet data that satisfies the constraint of the bus unit 11.

The second packet data transmission unit 64 writes the designated portion of the packet data in the memory unit 10 via the memory access unit 73 (steps S154 and S155).

The second Packing transmission processing unit 63 instructs the second mbuf transmission unit 65 to write the mbuf structure, and the second mbuf transmission unit 65 writes the mbuf structure to the memory unit 10 via the memory access unit 73 ( Step S156).

The second Packing transmission processing unit 63 requests the second Enqueue processing unit 66 to perform Enqueue to the ring queue of the mbuf structure pointer, and the second Enqueue processing unit 66 performs Enqueue processing (step S157).

Next, the process in which the second Enqueue processing unit 66 enqueues data in the ring queue will be described with reference to FIG. The second Enqueue processing unit 66 receives the queue ID of the ring queue, the data to be enqueued (size of multiple of 8 bytes), and the size of the data (8 byte unit) as parameters.

The second Enqueue processing unit 66 adds the input data to the internal buffer (step S160).

The second Enqueue processing unit 66 confirms whether the size of the data stored in the internal buffer is the minimum unit size of the bus unit 11 (step S161). If the size is less than the minimum unit size, the second Enqueue processing unit 66 ends this processing.

The second Enqueue processing unit 66 determines whether or not the Enqueue destination ring queue has a free space equal to or larger than the minimum unit size of the bus unit 11 (step S162). The second Enqueue processing unit 66 determines that there is a free space equal to or larger than the minimum unit size of the bus unit 11 when the following conditions are satisfied. MAX (A, B) means the larger of A and B.

p_index-c_index + MAX (size of enqueued data (data in internal buffer) (8-byte unit), minimum unit size (byte unit)/8) ≤ p_size (when p_index is greater than or equal to c_index).

p_index + 2^32-c_index + MAX (size of enqueued data (data in internal buffer) (8 bytes unit), minimum unit size (byte unit)/8) ≤ p_size (when p_index is less than c_index).

When the second Enqueue processing unit 66 determines that the ring queue has no space, it ends the process.

The second Enqueue processing unit 66 writes the data to be enqueued (data in the internal buffer) at the position (and after) of the array index p_index of the ring that is the FIFO array via the memory access unit 73 (step S163). The second Enqueue processing unit 66 writes data in units of multiples of the minimum unit size of the bus unit 11.

The second Enqueue processing unit 66 adds the size (8-byte unit) of the data enqueued in step S162 to p_index and writes it in the p_index field of the ring queue stored in the memory unit 10 via the memory access unit 73 (step S164). ).

<(E) Packet Reception Processing by First Data Processing Unit 20>
A process in which the first data processing unit 20 (FIG. 4) of the information processing device 1 receives a packet from the second data processing unit 30 will be described with reference to the flowchart of FIG. The process is executed, for example, when the first data processing unit 20 receives the packet after the offload process from the second data processing unit 30.

The first data processing unit 20 requests Dequeue from the first IF unit 41 (FIG. 6) using the address of the ring queue (for the mbuf structure pointer) as a parameter (step S170).

The request is transferred to the first Packing reception processing unit 45 by the first IF unit 41, and the first Packing reception processing unit 45 requests the first Dequeue processing unit 46 to Dequeue the mbuf structure pointer, and performs the first Dequeue processing. The unit 46 dequeues the mbuf structure pointer from the ring queue (step S171). Dequeue processing by the first Dequeue processing unit 46 will be described later.

The first Packing reception processing unit 45 determines whether packing is used for the packet (step S172). The first Packing reception processing unit 45 makes a determination by the same method as the second Packing transmission processing unit 63 in step S151. That is, the first Packing reception processing unit 45 refers to buf_addr, data_off, and data_len of the mbuf structure based on the mbuf structure pointer, and confirms the position and size of the packet data in the packet buffer. The first Packing reception processing unit 45 determines that the packing is used when the position and size do not satisfy the constraint of the bus unit 11, and determines that the packing is not used when the position and size do not satisfy the constraint. If packing is not used, the process proceeds to step S177.

The first Packing reception processing unit 45 requests the first Dequeue processing unit 46 to Dequeue the Packing data (step S173).

The first Dequeue processing unit 46 dequeues the Packing data from the ring queue and notifies the first Packing reception processing unit 45 (step S174). Dequeue processing by the first Dequeue processing unit 46 will be described later.

The first Packing reception processing unit 45 processes the dequeued Packing data (step S175). Specifically, when the Packing data includes data on the packet data head side, the first Packing reception processing unit 45 adds the data to the address obtained by adding the values of the buf_addr field and the data_off field of the mbuf structure. Write. In addition, when the Packing data includes the data on the end side of the packet data, the first Packing reception processing unit 45 writes the data in the address obtained as follows.
buf_addr field value of mbuf structure + data_off field value + data_len field value-size of the data

The first Packing reception processing unit 45 notifies the mbuf structure pointer obtained in step S171 to the first data processing unit 20 via the first IF unit 41 (step S176).

The Dequeue process from the ring queue by the first Dequeue processing unit 46 in steps S171 and S174 is basically the same as the Dequeue process by the second Dequeue processing unit 68 described with reference to FIG. However, the first Dequeue processing unit 46 acquires the ring queue address as a parameter from the Dequeue requester. Further, the first Dequeue processing unit 46 does not use the memory access unit 73 (FIG. 7) for referring to and updating the ring queue on the memory unit 10. Further, the first Dequeue processing unit 46 may omit step S141 shown in FIG. 14, and may extract data of a size smaller than the minimum unit size of the bus unit 11 in steps S142 and S143.

<(F) Ring queue release processing>
A process in which the information processing device 1 releases the ring queue will be described with reference to the flowchart of FIG. The process is executed, for example, when the operation of the first data processing unit 20 ends. The first data processing unit 20 executes the processing for each ring queue for the mbuf structure pointer used with the second data processing unit 30.

The first data processing unit 20 includes a second Enqueue processing unit 66 (FIG. 7) (when the second data processing unit 30 transmits and the first data processing unit 20 receives) or a second Dequeue processing unit 68 (first data). When the processing unit 20 transmits and the second data processing unit 30 receives), it requests the deletion of the address of the ring queue (step S180).

The first data processing unit 20 requests the first IF unit 41 to release the ring queue (step S180). At that time, the address of the ring queue is passed as a parameter. The request is transferred to the first management unit 42.

The first management unit 42 (FIG. 6) acquires the address of the ring queue for Packing data by referring to the p_ring_ptr field of the ring queue given by the parameter, and releases the memory area for the ring queue (step S182). ).

The first management unit 42 releases the memory area for the ring queue given the parameter (step S183).

As described in detail above, the following effects can be obtained in the present embodiment.

First, a case will be described in which the first data processing unit 20 illustrated in FIG. 4 uses the ring queue to transmit the packet stored in the memory unit 10 in the mbuf format to the second data processing unit 30. According to the present embodiment, when generating the ring queue for the mbuf structure pointer used by the first data processing unit 20, the first transmission/reception processing unit 40 also generates the ring queue for Packing data. It is assumed that the first data processing unit 20 enqueues the mbuf structure pointer in the mbuf structure pointer ring queue. In this case, the first transmission/reception processing unit 40 generates Packing data in which the portions of the packet data of the packet that do not satisfy the constraint of the bus unit 11 are collected and enqueues the Packing data in the Ring queue for Packing data. ..

Further, it is assumed that the second data processing unit 30 receives the packet from the first data processing unit 20 using the ring queue. In this case, the second transmission/reception processing unit 60 dequeues the mbuf structure pointer from the ring queue and reads the mbuf structure. The second transmission/reception processing unit 6 reads, from the packet buffer, a portion of the packet data of the packet that matches the constraint of the bus unit 11 (a portion that matches the communication unit of the bus unit 11). On the other hand, the second transmission/reception processing unit 60, from the packet data ring queue, packs data (fractional data) for a part of the packet data that does not match the constraint of the bus unit 11 (a part that does not match the communication unit of the bus unit 11). Read as. At that time, the second transmission/reception processing unit 60 dequeues the Packing data in units of multiples of the minimum unit size of the bus unit 11. The second transmission/reception processing unit 60 updates the packet data read from the packet buffer with Packing data as necessary, and then notifies the second data processing unit 30 together with the mbuf structure and the like.

Here, the case where the second data processing unit 30 uses the ring queue to transmit the packet stored in the memory unit 10 in the mbuf format to the first data processing unit 20 will be described. In this case, according to the present embodiment, when the mbuf structure pointer ring queue used by the first data processing unit 20 is generated, the first transmission/reception processing unit 40 also generates the Packing data ring queue. Here, it is assumed that the second data processing unit 30 enqueues the mbuf structure pointer in the ring queue for the mbuf structure pointer. In this case, the second transmission/reception processing unit 60 generates Packing data in which the packet data of the packet that does not satisfy the constraint of the bus unit 11 are collected and enqueues the Packing data in the ring queue for Packing data. .. At that time, the second transmission/reception processing unit 60 enqueues the Packing data in units of multiples of the minimum unit size of the bus unit 11. Further, the second transmission/reception processing unit 60 writes the mbuf structure in the memory area indicated by the mbuf structure pointer, and also writes the portion of the packet data that matches the constraints of the bus unit 11 in the packet buffer.

A case where the first data processing unit 20 receives a packet from the second data processing unit 30 using the ring queue will be described. In this case, the first transmission/reception processing unit 40 dequeues the mbuf structure pointer from the ring queue, dequeues the Packing data from the ring queue for Packing data as necessary, and dequeues the Packing data of the bus unit 11 of the packet data of the packet. The part that does not match the constraint is updated using the packing data. After that, the first transmission/reception processing unit 40 notifies the first data processing unit 20 of the dequeued mbuf structure pointer.

As a result, the packet data of the portion that does not satisfy the constraint of the bus unit 11 is communicated as Packing data in a state that the constraint of the bus unit 11 is satisfied, and it is possible to suppress a decrease in bus utilization efficiency.

Therefore, according to the present embodiment, it is possible to obtain the effect of suppressing a decrease in bus utilization efficiency when performing data communication between modules via a memory and a bus.

In the description of the present embodiment, an example in which the information processing apparatus 1 indicates the existence of the partial packet data on the head side and the tail side of the packet data by using the head 2 bits of the Packing data has been described. It is not limited to this. As another example, the information processing device 1 replaces the presence of the partial packet data with the first 2 bits of the Packing data, or in addition to the presence of the data (8 bytes) to be enqueued and dequeued in the ring queue for the mbuf structure pointer. May be represented as part. For example, the first 2 bits of the data (8 bytes) may be used as the 2 bits indicating the existence of the partial packet data, and the remaining 62 bits may be used as the mbuf structure pointer. In that case, when the first transmission/reception processing unit 40 and the second transmission/reception processing unit 60 pass the mbuf structure pointer dequeued from the ring queue to the first data processing unit 20 and the second data processing unit 30, the dequeued data The first 2 bits are set to 0 and passed. With such a configuration, it is possible to obtain the effect of reducing the frequency of access to the Packing data ring queue.

In addition, in the description of the present embodiment, an example in which the information processing device 1 generates a Packing data ring queue separately from the mbuf structure pointer ring queue and uses it for Packing data communication has been described. It is not limited to this. As another example, the information processing device 1 may communicate the Packing data using the ring queue for the mbuf structure pointer. In that case, the first transmission/reception processing unit 40 and the second transmission/reception processing unit 60 perform Enqueue of Packing data not on the ring queue for Packing data but on the ring queue for mbuf structure. In addition, the first transmission/reception processing unit 40 and the second transmission/reception processing unit 60 perform Dequeue of Packing data from the ring queue for the mbuf structure, not from the ring queue for Packing data.

The first transmission/reception processing unit 40 and the second transmission/reception processing unit 60 determine whether the data to be enqueued and dequeued in the ring queue for the mbuf structure pointer is the mbuf structure pointer or the packing data, for example, at the beginning. Perform using 1 Bit. In that case, when the first transmission/reception processing unit 40 and the second transmission/reception processing unit 60 pass the mbuf structure pointer dequeued from the ring queue to the first data processing unit 20 and the second data processing unit 30, the dequeued data The first 1 Bit is set to 0 and passed. If the size of the Packing data is not a multiple of 8 bytes, the first transmission/reception processing unit 40 and the second transmission/reception processing unit 60 add padding to make it a multiple of 8 bytes, and then use the ring for the mbuf structure pointer. You may Enqueue and Dequeue to the queue. The first transmission/reception processing unit 40 does not need to generate the ring queue for Packing data when generating the ring queue in response to the request from the first data processing unit 20. Further, when generating the ring queue, the first transmission/reception processing unit 40 may generate a ring queue of a size different from the size given by the first data processing unit 20. For example, the first transmission/reception processing unit 40 may generate a ring queue having a size obtained by multiplying the given size by the minimum unit size of the bus unit 10. With such a configuration, it is possible to obtain the effects of reducing the size of the memory area used and reducing the frequency of access to the memory unit 10.
Further, when the Packing data is communicated using the ring queue for the mbuf structure pointer, the information processing device 1 does not communicate the mbuf structure pointer and the Packing data alternately using the ring queue, but instead uses a predetermined number. The mbuf structure pointer may be continuously communicated for each packet of, and then the Packing data relating to them may be continuously communicated. With such a configuration, when the first Packing reception processing unit 45 dequeues the mbuf structure pointer (step S171), the prefetch instruction (prefetch instruction) is executed for the mbuf structure pointer. , CPU cache miss can be suppressed.

In addition, in the description of the present embodiment, an example in which the information processing device 1 determines whether or not the packing is necessary only based on whether or not the position or size of the packet data satisfies the constraint of the bus unit 11 has been described. The invention is not limited to this. As another example, the information processing device 1 may additionally use a condition that the packing is used only when the packet size is smaller than a predetermined value. With such a configuration, the effect of reducing the load due to packing can be obtained. Further, as another example, the information processing device 1 receives, for each packet, information on whether or not to use packing for the packet from the first data processing unit 20 or the second data processing unit 30, and based on the information. The use/non-use of packing may be switched.

A case will be described in which the use/nonuse of packing is switched using information other than the position and size of packet data. In this case, the first transmission/reception processing unit 40 and the second transmission/reception processing unit 60 provide, for each packet to be processed, a flag indicating whether or not to use packing for the packet, and refer to the flag to perform packing on the packet. Switch between doing and not doing. The first transmission/reception processing unit 40 and the second transmission/reception processing unit 60, for example, record the flag in any place of the mbuf structure pointer or the mbuf structure.

In the description of the present embodiment, the information processing apparatus 1 uses the ring queue and the mbuf format for packet communication, but the present invention is not limited to this. As another example, the information processing device 1 may use, for example, Virtio used in virtualization technology for packet communication. In that case, the information processing device 1 uses the Virtio queue instead of the ring queue for the mbuf structure pointer. Further, the information processing device 1 uses a virtio ring descriptor (vring_deSc structure) instead of the mbuf structure as a data structure for managing the position and size of the packet buffer. The information processing device 1 uses a vring_avail structure and a vring_used structure instead of the data structure of the ring queue shown in FIG. The information processing device 1 uses an index to the vring_desc structure array indicated by the desc field of the vring structure instead of the mbuf structure pointer. The information processing device 1 uses a ring queue for communication of Packing data.
Further, in the present embodiment, for example, the first transmission/reception processing unit 40 and the first data processing unit 20 realized as a software program executed by the CPU, and the second transmission/reception processing unit 60 and the first data transmission processing unit 60 realized by, for example, an FPGA. Although the two data processing units 30 communicate with each other, the present invention is not limited to this. For example, as another example, a first transmission/reception processing unit 40 and a first data processing unit 20, which are realized as a software program executed by a CPU, and another first transmission/reception processing unit 40 and a first data processing unit 20. , May communicate. Similarly, for example, the second transmission/reception processing unit 60 and the second data processing unit 30 which are realized by the FPGA and another second transmission/reception processing unit 60 and the second data processing unit 30 may communicate with each other.

[Description of Second Embodiment]
In the first embodiment, an example in which the portion of the packet data that does not match the constraint of the bus unit 11 is used as the Packing data has been described, but the present invention is not limited to this. As another example, a part of the mbuf structure may be included in the Packing data, and the second transmission/reception processing unit 60 may not directly refer to or update the mbuf structure stored in the memory unit 10.

Hereinafter, the second embodiment of the present invention will be described in detail with reference to the drawings. In each of the drawings referred to in the description of the present embodiment, the same configurations as those in the first embodiment of the present invention and steps operating in the same manner are designated by the same reference numerals, and detailed description of the present embodiment will be given. The description is omitted.

[Example of Configuration of Second Embodiment]
A configuration example of this embodiment will be described with reference to FIGS.

FIG. 19 is a configuration diagram showing an example of the configuration of the information processing device 2 according to the second embodiment of the present invention. The information processing device 2 illustrated in FIG. 19 includes a memory unit 10, a bus unit 11, a first data processing unit 20, a second data processing unit 30, a first transmission/reception processing unit 40B, and a second transmission/reception processing unit. And 60B. That is, the configurations of the first transmission/reception processing unit 40B and the second transmission/reception processing unit 60B are different from those of the first transmission/reception processing unit 40 and the second transmission/reception processing unit 60 shown in FIG. Note that the numbers of components and connection relationships shown in FIG. 19 are examples.

The first transmission/reception processing unit 40B uses the ring queue to communicate data with the second transmission/reception processing unit 60B. The details will be described later.

The second transmission/reception processing unit 60B uses the ring queue to communicate data with the first transmission/reception processing unit 40B. The details will be described later.

Here, the packing performed by the present embodiment will be described. In this embodiment, in addition to the packing described in the first embodiment, the mbuf structure is also handled by packing. In the packing in the present embodiment, Packing data including only the information required by the second data processing unit 30 and the second transmission/reception processing unit 60 among the fields of the mbuf structure is generated. Information required by the second data processing unit 30 and the second transmission/reception processing unit 60B is shown in FIG. That is, FIG. 20 is a configuration example of Packing data relating to the mbuf structure in the case of the buf_physaddr field, data_off field, and data_len field of the mbuf structure. Considering a general packet size, a field length of the data_off field and the data_len field of 15 bits or less is sufficient. In the present embodiment, it is assumed that the first 2 bits of the data_off field of Packing data are always 0. As a result, it is possible to distinguish the Packing data regarding the packet data shown in FIG. 5 from the Packing data regarding the mbuf structure. In the example of FIG. 20, a data_off field having a 15-bit length and a data_len field having a 15-bit length follow the 2 bits.

Next, a configuration example of the first transmission/reception processing unit 40B in the present embodiment will be described using FIG. The first transmission/reception processing unit 40B illustrated in FIG. 21 includes a first IF unit 41, a first management unit 42, a first Packing transmission processing unit 43B, a first Enqueue processing unit 44, and a first Packing reception processing unit. 45B and the first Dequeue processing unit 46.

The first Packing transmission processing unit 43B receives the Enqueue request for the mbuf structure pointer, performs the above-described packing, and then requests the first Enqueue processing unit for the Enqueue of the mbuf structure pointer and Packing data.

The first Packing reception processing unit 45B receives the Dequeue request for the mbuf structure pointer, requests the Dequeue processing unit 46 for the mbuf structure pointer and the Dequeue of the Packing data, and the mbuf structure and the packet data based on the packing described above. After executing the update, returns the dequeued mbuf structure pointer.

Next, a configuration example of the second transmission/reception processing unit 60B in the present embodiment will be described using FIG. The second transmission/reception processing unit 60B illustrated in FIG. 22 includes a second IF unit 61, a second Packing transmission processing unit 63B, a second packet data transmission unit 64, a second mbuf transmission unit 65, and a second Enqueue process. It is configured to include a unit 66, a second Dequeue processing unit 68, a second mbuf reception unit 69, a second packet data reception unit 70, a second Packing reception processing unit 71B, and a memory access unit 73. ..

The second Packing transmission processing unit 63B receives the transmission data from the second IF unit 61, determines the necessity of packing, and sends an instruction to the second packet data transmission unit 64 and the second mbuf transmission unit 65 based on the result. .. The instruction includes all or part of the transmission data received from the second IF unit 61. Further, the second Packing transmission processing unit 63B generates Packing data and requests the second Enqueue processing unit 66 to Enqueue the generated Packing data to the ring queue. In addition, the second Packing transmission processing unit 63B requests the second Enqueue processing unit 66 to enqueue the mbuf structure pointer to the ring queue.

The second Packing reception processing unit 71B receives the dequeued mbuf structure pointer and the Packing data from the second Dequeue processing unit 68 together with the queue ID, and determines their processing methods. The second Packing reception processing unit 71B instructs the second mbuf receiving unit 69 to read the mbuf structure, if necessary. When the Packing data includes a part of the mbuf structure, the second Packing reception processing unit 71B generates the mbuf structure using the data. The second Packing reception processing unit 71B instructs the second packet data receiving unit 70 to read the packet data. The second Packing reception processing unit 71B generates reception data including the mbuf structure pointer, queue ID (ring queue for mbuf structure pointer), mbuf structure pointer, mbuf structure, and Packing data, and the second IF unit Notify 61.

[Operation example of the second embodiment]
Next, an example of each of the following operations (B), (C), (D), and (E) in the information processing device 2 illustrated in FIGS. 19 to 22 as a configuration example of the present embodiment will be described. A detailed description will be given with reference to the drawings.
(B) Packet transmission process by the first data processing unit 20 (C) Packet reception process by the second data processing unit 30 (D) Packet transmission process by the second data processing unit 30 (E) Packet by the first data processing unit 20 Reception processing

<(B) Packet Transmission Process by First Data Processing Unit 20>
A process in which the first data processing unit 20 of the information processing device 2 transmits a packet to the second data processing unit 30 will be described with reference to the flowchart of FIG.

The first data processing unit 20 requests Enqueue from the first IF unit 41 using the address of the ring queue (for the mbuf structure pointer) and the mbuf structure pointer as parameters (step S110).

The request is transferred to the first Packing transmission processing unit 43B by the first IF unit 41, and the first Packing transmission processing unit 43B determines whether packing is required for the packet (step S111B). In the present embodiment, it is always determined that packing is necessary.

The first Packing transmission processing unit 43B generates Packing data for the mbuf structure and Packing data for the packet data according to the packing described above (step S112B). However, when the packet data satisfies the constraint of the bus unit 11, the first Packing transmission processing unit 43B does not have to generate the Packing data regarding the packet data.

The first Packing transmission processing unit 43B acquires the address of the ring queue for Packing data by referring to the p_ring_ptr field of the ring queue, and uses the address and the Packing data generated in step S112B as parameters to set the first Enqueue processing unit 44B. Request the Enqueue process (in order of mbuf structure and packet data) (step S113B).

The first Packing transmission processing unit 43B requests Enqueue processing from the first Enqueue processing unit 44 using the address of the ring queue for the mbuf structure pointer and the mbuf structure pointer as parameters (step S114B).

<(C) Packet reception process by the second data processing unit 30>
A process in which the second data processing unit 30 of the information processing device 2 receives a packet from the first data processing unit 20 will be described with reference to the flowchart of FIG.

The second Packing reception processing unit 71B (FIG. 22) requests the Dequeue of the mbuf structure pointer to the second Dequeue processing unit 68, and the second Dequeue processing unit 68 returns the dequeued mbuf structure pointer together with the queue ID (step S200).

The second Packing reception processing unit 71B requests the second Dequeue processing unit 68 to Dequeue the Packing data, and the second Dequeue processing unit 68 returns the Dequeued Packing data together with the queue ID (step S201). Dequeue of Packing data is performed in units of multiples of the minimum unit size of the bus unit 11. The second Packing reception processing unit 71B internally stores the dequeued Packing data, refers to it whenever necessary, and consumes only the necessary amount. In this step, the second Packing reception processing unit 71B requests the second Dequeue processing unit 68 to Dequeue the Packing data when the internally stored Packing data becomes insufficient.

The second Packing reception processing unit 71B determines whether the Dequeued Packing data relates to the mbuf structure (step 202). The second Packing reception processing unit 71B determines that the Packing data is related to the mbuf structure when the first 2 Bits are all 0.

When the Packing data is related to the mbuf structure, the second Packing reception processing unit 71B generates the mbuf structure from the Packing data (step S203). Specifically, the second Packing reception processing unit 71B obtains and sets the value of the field included in the Packing data from the Packing data among the fields of the mbuf structure, and sets the other fields to 0, for example. Set to.

In step S202, when the Packing data is not related to the mbuf structure, the second Packing reception processing unit 71B passes the mbuf structure pointer to the second mbuf receiving unit 69 to instruct the reading, and the second mbuf receiving unit 69. Reads the mbuf structure from the memory unit 10 via the memory access unit 73 based on the pointer (step S204).

The second Packing reception processing unit 71B determines whether packing is used for the packet data of the packet (step S205). The second Packing reception processing unit 71B makes a determination by the same method as the first Packing transmission processing unit 43 in step S111. When the packing is used, the second Packing reception processing unit 71B instructs the second packet data receiving unit 70 to read the packet data (only the portion satisfying the constraint of the bus unit 11) (shift to step S206). When the packing is not used, the second Packing reception processing unit 71B instructs the second packet data receiving unit 70 to read the entire packet data (shift to step S209).

The second packet data receiving unit 70 reads the designated portion of the packet data from the memory unit 10 via the memory access unit 73 (steps S206 and S209).

The second Packing reception processing unit 71B dequeues the Packing data relating to the packet data (step S207). It should be noted that this step is performed only when the packing data dequeued in step S201 relates to the mbuf structure.

The second Packing reception processing unit 71B uses the packet data read in Step S206 (only the portion satisfying the constraint of the bus unit 11) to include a part of the packet data included in the Packing data obtained in Step S207 (start side, end side). Side, one or both), and packet data is restored (step S208).

The second Packing reception processing unit 71B generates reception data including a queue ID (for mbuf structure pointer), mbuf structure pointer, mbuf structure, and packet data, and receives the reception data via the second IF unit 61. The second data processing unit 30 is notified (step S210).

<(D) Packet transmission process by second data processing unit 30>
A process in which the second data processing unit 30 of the information processing device 2 transmits a packet to the first data processing unit 20 will be described with reference to the flowchart of FIG.

The second data processing unit 30 requests the second IF unit 61 to transmit a packet (step S220). At that time, the second data processing unit 30 passes the queue ID, the mbuf structure pointer, the mbuf structure, and the packet data as transmission data.

The second IF unit 61 passes the transmission data to the second Packing transmission processing unit 63B, and the second Packing transmission processing unit 63B determines whether or not each of the mbuf structure of the packet and the packet data needs to be packed ( Step S221). In the present embodiment, the second Packing transmission processing unit 63B determines that the mbuf structure is always necessary. Regarding packing of packet data, the second Packing transmission processing unit 63B refers to buf_physaddr, data_off, data_len of the mbuf structure to confirm the position and size of the packet data in the packet buffer. The second Packing transmission processing unit 63B determines that packing is necessary when the position and size do not satisfy the constraint of the bus unit 11, and determines that packing is unnecessary when the position and size satisfy the constraint. When it is determined that the packing is unnecessary, the second Packing transmission processing unit 63B instructs the second packet data transmission unit 64 to write the packet data (shift to step S224). At that time, the second Packing transmission processing unit 63 specifies the entire packet data as the write target area.

The second Packing transmission processing unit 63B generates Packing data for the mbuf structure and the packet data as needed according to the packing described above (step S222).

The second Packing transmission processing unit 63B requests the Enqueue of the generated Packing data from the second Enqueue processing unit 66 (mbuf structure, packet data order), and the second Enqueue processing unit 66 performs Enqueue (step S223). .. After that, the second Packing transmission processing unit 63B instructs the second packet data transmission unit 64 to write the packet data (only the portion that satisfies the constraint of the bus unit 11).

The second packet data transmission unit 64 writes the designated packet data in the memory unit 10 via the memory access unit 73 (step S224).

In step S221, when the mbuf structure is not packed, the second Packing transmission processing unit 63B determines that the mbuf structure needs to be written, and the second mbuf transmission unit 65 informs the second mbuf structure unit 65 of the mbuf structure pointer and the mbuf structure. The body is passed over to instruct writing (step S225).

The second mbuf transmission unit 65 writes the mbuf structure in the memory unit 10 via the memory access unit 73 (step S226).

The second Packing transmission processing unit 63B requests the second Enqueue processing unit 66 to Enqueue the mbuf structure pointer to the ring queue, and the second Enqueue processing unit 66 performs Enqueue processing (step S227).

<(E) Packet Reception Processing by First Data Processing Unit 20>
A process in which the first data processing unit 20 of the information processing device 2 receives a packet from the second data processing unit 30 will be described with reference to the flowchart of FIG.

The first data processing unit 20 requests the first IF unit 41 for Dequeue using the ring queue address (for the mbuf structure pointer) as a parameter (step S231).

The request is transferred to the first Packing reception processing unit 45B by the first IF unit 41, and the first Packing reception processing unit 45B requests the first Dequeue processing unit 46 to Dequeue the mbuf structure pointer, and performs the first Dequeue processing. The unit 46 dequeues the mbuf structure pointer from the ring queue (step S232).

The first Packing reception processing unit 45B requests the first Dequeue processing unit 46 to dequeue the Packing data, and the first Dequeue processing unit 46 dequeues the Packing data from the ring queue (step S233). The first Packing reception processing unit 45B acquires the address of the Packing data ring queue from the p_ring_ptr field of the mbuf structure pointer ring queue.

The first Packing reception processing unit 45B determines whether or not the Dequeued Packing data relates to the mbuf structure (step 223). The first Packing reception processing unit 45B determines that the Packing data is related to the mbuf structure when the first 2 Bits are all 0.

When the Packing data is related to the mbuf structure, the first Packing reception processing unit 45B updates the mbuf structure from the Packing data (step S235). Specifically, the first Packing reception processing unit 45B updates the fields included in the Packing data among the fields of the mbuf structure with the values included in the Packing data.

The first Packing reception processing unit 45B determines whether packing is used for the packet data (step S236). The first Packing reception processing unit 45B makes a determination by the same method as the second Packing transmission processing unit 63 in step S151. When packing is not used for the packet data, this process is terminated, and the mbuf structure pointer is notified to the first data processing unit 20 via the first IF unit 41 in step S232.

When packing is used for the packet data, the first Packing reception processing unit 45B requests the first Dequeue processing unit 46 to dequeue the Packing data related to the packet data (step S237). Note that this step is performed only when the Packing data dequeued in step S233 is related to the mbuf structure.

The first Packing reception processing unit 45B updates the packet data using the dequeued Packing data (step S238). Specifically, when the Packing data includes data on the packet data head side, the first Packing reception processing unit 45 adds the data to the address obtained by adding the values of the buf_addr field and the data_off field of the mbuf structure. Write. In addition, when the Packing data includes the data on the end side of the packet data, the first Packing reception processing unit 45 writes the data in the address obtained as follows.
buf_addr field value of mbuf structure + data_off field value + data_len field value-size of the data

The first Packing reception processing unit 45 ends this processing and notifies the first data processing unit 20 of the mbuf structure pointer obtained in step S232 via the first IF unit 41.

As described in detail above, the following effects can be obtained in the present embodiment.

According to the present embodiment, in addition to the first embodiment, the mbuf structure is also included in the Packing data. The first transmission/reception processing unit 40B generates Packing data including a part of the mbuf structure and sends it to the second transmission/reception processing unit 60B via the ring queue. Further, the first transmission/reception processing unit 40B updates the mbuf structure stored in the memory unit 10 based on the information included in the Packing data received from the second transmission/reception processing unit 60B via the ring queue. The second transmission/reception processing unit 60B generates an mbuf structure from the information included in the Packing data received from the first transmission/reception processing unit 40B via the ring queue, and directly refers to the mbuf structure stored in the memory unit 10. do not do. The second transmission/reception processing unit 60B also generates Packing data including a part of the mbuf structure and sends it to the first transmission/reception processing unit 40B via the ring queue.

As a result, the amount of data to be communicated on the bus unit 11 can be reduced, and the packet data and the part of the mbuf structure that does not satisfy the constraints of the bus unit 11 can be used as Packing data while satisfying the constraints of the bus unit 11. As a result of communication, the bus utilization efficiency can be improved or its degradation can be suppressed.

Therefore, according to the present embodiment, it is possible to obtain the effect of improving the bus utilization efficiency when suppressing data communication between the modules via the memory and the bus or suppressing the decrease.

Further, in the description of the present embodiment, an example has been shown in which the information processing device 2 always performs packing regarding the mbuf structure, but the present invention is not limited to this. As another example, the information processing device 2 may use the packing related to the mbuf structure only when the packet size is smaller than a predetermined value. Further, as still another example, the information processing device 2 may use the packing related to the mbuf structure only when packing related to the packet data. With such a configuration, the effect of reducing the load due to packing can be obtained. In that case, the first transmission/reception processing unit 40B and the second transmission/reception processing unit 60B provide, for each packet to be processed, a flag indicating whether or not to use packing regarding the mbuf structure for the packet. It suffices to switch whether packing is performed or not for the packet with reference to the flag. The first transmission/reception processing unit 40B and the second transmission/reception processing unit 60B, for example, record the flag in any place of the mbuf structure pointer or the mbuf structure.
Further, in the description of the present embodiment, an example in which the information processing device 2 generates a Packing data ring queue separately from the mbuf structure pointer ring queue and uses it for Packing data communication has been described. It is not limited to this. As another example, the information processing device 2 may communicate the Packing data using the ring queue for the mbuf structure pointer. In that case, the first transmission/reception processing unit 40B and the second transmission/reception processing unit 60B perform Enqueue of Packing data not on the Ring queue for Packing data but on the ring queue for mbuf structure. Further, the first transmission/reception processing unit 40B and the second transmission/reception processing unit 60B perform Dequeue of Packing data from the ring queue for the mbuf structure instead of the ring queue for Packing data.
Further, when communicating the Packing data by using the ring queue for the mbuf structure pointer, the information processing device 2 does not communicate by alternately using the ring queue for the mbuf structure pointer and the Packing data, but a predetermined number. The mbuf structure pointer may be continuously communicated for each packet. In this case, thereafter, Packing data (a part of the mbuf structure) regarding them may be continuously communicated, and then Packing data (packet data) regarding them may be continuously communicated. With such a configuration, when the first Packing reception processing unit 45B dequeues the mbuf structure pointer (step S232), a prefetch instruction (prefetch instruction) is executed for the mbuf structure pointer, and When the one Packing reception processing unit 45B dequeues the Packing data (a part of the mbuf structure) (step S233), the prefetch instruction (prefetch instruction) is executed for the packet data. This can suppress CPU cache misses.

The whole or part of the exemplary embodiments disclosed above can be described as, but not limited to, the following supplementary notes.
[Appendix 1]
A bus that performs communication in a predetermined communication unit,
A memory connected to the bus,
A communication device that sends and receives data to and from the memory via the bus;
The communication device packs unmatched data into a plurality of data sets so as to match the communication unit of the bus when at least one of the data and the metadata regarding the attribute does not match the communication unit of the bus. An information processing apparatus characterized by performing.
In this case, the packing is performed when at least one of the metadata regarding the data and its attribute is less than the communication unit of the bus, and when at least one of the metadata regarding the data and its attribute is larger than the communication unit of the bus. To be executed.

[Appendix 2]
The packing is performed when at least one of the metadata regarding the data and its attribute is less than the communication unit of the bus, and when at least one of the metadata regarding the data and its attribute is larger than the communication unit of the bus. The information processing apparatus according to appendix 1, characterized in that:

[Appendix 3]
When at least one of the data and the metadata is less than the communication unit of the bus, the communication device divides the fractional data, which is fractional with respect to the communication unit, into a plurality of data sets so as to be the communication unit. The information processing device according to appendix 1 or 2, further comprising a transmission device that stores the data in a continuous area of the memory via the bus.

[Appendix 4]
The information processing device according to any one of appendices 1 to 3, wherein the communication device includes a receiving device that reads the fractional data of the data to be communicated from the continuous area.

[Appendix 5]
5. The information processing apparatus according to any one of appendices 1 to 4, wherein the communication device collects the data and metadata regarding an attribute of the data for a plurality of data sets.

[Appendix 6]
In any one of appendices 1 to 5, the memory is a ring queue that uses a pointer as an entry, a packet buffer that stores packet data that can indirectly refer to the data by the pointer, and the metadata related to the packet data. The information processing apparatus according to claim 5, further comprising an mbuf structure that stores the mbuf format.

[Appendix 7]
In an information processing device including a bus that performs communication in a predetermined communication unit, a memory connected to the bus, and a communication device that transmits and receives data to and from the memory via the bus,
When the communication device does not match at least one of the data and the metadata regarding the attribute to the communication unit of the bus, packing that does not match the plurality of data sets so as to match the communication unit of the bus. A communication method characterized by performing.

[Appendix 8]
When the communication device includes a transmission device that stores pointers relating to a plurality of the data in a continuous area (pointer data area) of the memory,
When the starting address and/or the size and/or both of the data to be communicated, which is the data or a part of the data, are less than the communication unit of the bus used for communication with the memory by the transmitting device, Fractional data, which is the data to be communicated that is fractional with respect to the communication unit, is collectively stored in the continuous area of the memory for the plurality of data sets,
The communication method according to claim 7, wherein the reception device of the communication device reads the fractional data of the data to be communicated from the continuous area.

[Appendix 9]
The pointer is a pointer to metadata including at least a pointer to the data, the transmission device collectively stores a portion of the metadata in the continuous area for the plurality of data sets,
By the receiving device,
Reading a part of the metadata from the continuous area,
The communication method according to appendix 8, characterized in that

[Appendix 10]
The transmitting device stores either or both of the fractional data and a part of the metadata in the continuous area only for data satisfying a predetermined condition,
The communication method according to appendix 7, characterized in that

[Appendix 11]
11. The communication method according to appendix 10, wherein the predetermined condition is that the size of the data to be communicated is smaller than a predetermined value.

[Appendix 12]
The transmitting device and the receiving device use the same memory area as the pointer data area and the continuous area,
12. The communication method according to any one of appendices 7 to 11, characterized in that.

[Appendix 13]
A computer equipped with a bus that performs communication in a predetermined communication unit, a memory connected to the bus, and a communication device that transmits and receives data to and from the memory via a bus, and A computer program, wherein when at least one of the data does not match the communication unit of the bus, packing is performed to collect the unmatched data for a plurality of data sets so as to match the communication unit of the bus.

INDUSTRIAL APPLICABILITY The present invention can improve utilization efficiency of a bus or a medium in a communication system including a plurality of CPUs and a bus or medium in which communication units are restricted.

1 Information processing apparatus 2 Information processing apparatus 10 Memory section 11 Bus section 20 First data processing section 30 Second data processing section 40 First transmission/reception processing section 40B First transmission/reception processing section 41 First IF section 42 First management section 43th One Packing transmission processing unit 43B First Packing transmission processing unit 44 First Enqueue processing unit 45 First Packing reception processing unit 45B First Packing reception processing unit 46 First Dequeue processing unit 60 Second transmission/reception processing unit 60B Second transmission/reception processing unit 61 Second IF Unit 63 Second Packing Transmission Processor 63B Second Packing Transmission Processor 64 Second Packet Data Transmitter 65 Second mbuf Transmitter 66 Second Enqueue Processor 68 Second Dequeue Processor 69 Second mbuf Receiver 70 Second Packet Data Receiving Section 71 Second Packing Reception Processing Section 71B Second Packing Reception Processing Section 73 Memory Access Section

Claims (10)

A bus that performs communication in a predetermined communication unit,
A memory connected to the bus,
A communication device that sends and receives data to and from the memory via the bus;
The communication device packs unmatched data into a plurality of data sets so as to match the communication unit of the bus when at least one of the data and the metadata regarding the attribute does not match the communication unit of the bus. An information processing apparatus characterized by performing. The packing is performed when at least one of the metadata regarding the data and its attribute is less than the communication unit of the bus, and when at least one of the metadata regarding the data and its attribute is larger than the communication unit of the bus. The information processing apparatus according to claim 1, wherein: When at least one of the data and the metadata is less than the communication unit of the bus, the communication device divides the fractional data, which is fractional with respect to the communication unit, into a plurality of data sets so as to be the communication unit. The information processing apparatus according to claim 1 or 2, further comprising: a transmission device that stores the data in a continuous area of the memory via the bus. The information processing apparatus according to any one of claims 1 to 3, wherein the communication apparatus includes a receiving apparatus that reads the fractional data of the data to be communicated from the continuous area. The information processing apparatus according to claim 1, wherein the communication apparatus collects the data and metadata regarding attributes of the data for a plurality of data sets. 6. The ring queue according to claim 1, wherein the memory uses a pointer for an entry, a packet buffer that stores packet data that can indirectly refer to the data by the pointer, and the meta data related to the packet data. The information processing apparatus according to claim 5, further comprising an mbuf structure that stores data in the mbuf format. In an information processing device including a bus that performs communication in a predetermined communication unit, a memory connected to the bus, and a communication device that transmits and receives data to and from the memory via the bus,
When the communication device does not match at least one of the data and the metadata regarding the attribute to the communication unit of the bus, packing that does not match the plurality of data sets so as to match the communication unit of the bus. A communication method characterized by performing. When the communication device includes a transmission device that stores pointers relating to a plurality of the data in a continuous area (pointer data area) of the memory,
When the starting address and/or the size and/or both of the data to be communicated, which is the data or a part of the data, are less than the communication unit of the bus used for communication with the memory by the transmitting device, Fractional data, which is the data to be communicated that is fractional with respect to the communication unit, is collectively stored in the continuous area of the memory for the plurality of data sets,
The communication method according to claim 7, wherein the reception device of the communication device reads the fractional data of the data to be communicated from the continuous area. The pointer is a pointer to metadata including at least a pointer to the data, the transmission device collectively stores a portion of the metadata in the continuous area for the plurality of data sets,
By the receiving device,
Reading a part of the metadata from the continuous area,
9. The communication method according to claim 8, wherein A computer equipped with a bus that performs communication in a predetermined communication unit, a memory connected to the bus, and a communication device that transmits and receives data to and from the memory via a bus, and A computer program, wherein when at least one of the data does not match the communication unit of the bus, packing is performed to collect the unmatched data for a plurality of data sets so as to match the communication unit of the bus.

Images