JP6836087B1 - Information processing systems, platforms, and programs - Google Patents

Abstract

PROBLEM TO BE SOLVED: To suppress a decrease in communication speed due to release of a transmission buffer. An information processing system according to a first aspect of the present disclosure includes a plurality of platforms and a relay device. The first platform comprises a first bridge driver. The first bridge driver writes data to the transmission buffer of the first platform and causes the relay device to transfer the data to the second platform in response to the data transmission request to the second platform, and causes the second platform to transfer the data. In response to this, a data reception request is sent. Each time the first bridge driver writes data to the transmission buffer, the area in which the data is written in the transmission buffer is set as a lock area that cannot be overwritten, and transmission / reception of the written data is completed. In addition, when the reception completion notification from the second bridge driver is received a plurality of times, the lock area is set to an unlockable area that can be overwritten. [Selection diagram] Fig. 6

Description

This disclosure relates to information processing systems, platforms, and programs.

An information processing system including a plurality of processors (platforms) and a relay device having a bus for communicably connecting the plurality of processors is known. In such an information processing system, data transmission / reception between processors is generally performed via a transmission buffer of any device.

Japanese Unexamined Patent Publication No. 2008-041027

By the way, in the transmission / reception of data between processors (platforms) via the transmission buffer as described above, a lock area in which the data is written in the transmission buffer cannot be overwritten until the transmission / reception is completed. It is common to set to. In this case, when the transmission / reception of the written data is completed, the lock area is generally changed to an overwriteable unlock area, that is, the transmission buffer is generally released.

However, when communication between processors (platforms) occurs frequently, if the transmission buffer for which the transmission / reception of written data is completed is released every time the transmission / reception of data is completed, the communication becomes better. Since the resources allocated are reduced, it tends to lead to a decrease in communication speed.

The information processing system according to the first aspect of the present disclosure includes a plurality of platforms and a relay device. The relay device connects a plurality of platforms in a communicable manner via an expansion bus. The first platform of the plurality of platforms includes a first bridge driver. First bridge driver, the first platform of the plurality of platforms in response to a transmission request of data to the different second platform, writes the data to the transmission buffer of the first platform, to the second platform And send a data reception request. The second platform includes a second bridge driver. The second bridge driver acquires the data transferred from the transmission buffer to the predetermined storage area of the second platform via the relay device in response to the reception request transmitted by the first bridge driver, and transfers the data to the first platform. In response, a data reception completion notification is sent. Each time the first bridge driver writes data to the transmission buffer, the area in which the data is written in the transmission buffer is set as a lock area that cannot be overwritten, and transmission / reception of the written data is completed. In addition, when the reception completion notification from the second bridge driver is received two or more predetermined times, the lock area corresponding to the reception completion notification of the predetermined number of times is set as an overwriteable unlock area.

The platform according to the second aspect of the present disclosure is included in a plurality of platforms communicably connected via an expansion bus by a relay device. The platform includes a bridge driver. Bridge driver, the platform of the plurality of platforms in response to the transmission request of the data to different other platforms, writes the data to the transmission buffer of the platform, the received data to other platforms Send the request. Each time the bridge driver writes data to the transmission buffer, the data-written area of the transmission buffer is set as a non-overwritable lock area, and the transmission / reception of the written data is completed. , reception completion notification transmitted from other platforms to the platform by a predetermined the data transferred to the storage area of the other platforms platforms from the transmission buffer via the relay apparatus acquires in response to the received request is 2 When the above predetermined number of times of reception is received, the lock area corresponding to the reception completion notification of the predetermined number of times is set as an overwriteable unlock area.

The program according to the third aspect of the present disclosure is a computer as a platform included in a plurality of platforms connected by a relay device so as to be communicable via an extension bus, and another platform different from the platform among the plurality of platforms. in response to the transmission request of the data to, write the data to the transmission buffer of the platform, and transmitting the received request data to other platforms, each time data is written to the transmit buffer, set override non locking region data is written region in the transmission buffer, and transmitting and receiving of the written data is completed, and, for transmission via the relay device in response to receiving request when the reception completion notification transmitted from other platforms the platform is received more than a predetermined number of times by the data transferred from the buffer in a predetermined storage area of the other platforms other platforms acquires, the This is a program for setting a lock area corresponding to a predetermined number of reception completion notifications to an overwriteable unlock area and executing the operation.

According to the information processing system, platform, and program of the present disclosure, it is possible to suppress a decrease in communication speed due to the release of the transmission buffer.

FIG. 1 is an exemplary and schematic block diagram showing the overall configuration of the information processing system according to the embodiment. FIG. 2 is an exemplary and schematic block diagram of a hardware configuration for the platform and relay device of the information processing system according to the embodiment. FIG. 3 is an exemplary and schematic block diagram showing the software configuration of the platform according to the embodiment. FIG. 4 is an exemplary and schematic diagram for explaining the communication processing executed by the information processing system according to the embodiment. FIG. 5 is an exemplary and schematic diagram for explaining the transmission and reception of data via the transmission buffer according to the embodiment. FIG. 6 is an exemplary and schematic sequence diagram showing a flow of processing executed when data is transmitted and received between platforms according to the embodiment.

Hereinafter, embodiments and modifications of the present disclosure will be described with reference to the drawings. The configurations of the embodiments and modifications described below, and the actions and effects brought about by the configurations are merely examples, and are not limited to the contents described below.

<Embodiment>
FIG. 1 is an exemplary and schematic block diagram showing the overall configuration of the information processing system 1 according to the embodiment.

As shown in FIG. 1, the information processing system 1 according to the embodiment includes a plurality of platforms 10-1 to 10-8, and a relay device 30. Platforms 10-1 to 10-8 are inserted into a plurality of slots on a board provided with, for example, a relay device 30. As a result, the platforms 10-1 to 10-8 are communicably connected to each other via the relay device 30.

In the following, when it is not necessary to distinguish from each other, any platform among the plurality of platforms 10-1 to 10-8 may be simply referred to as platform 10. Further, FIG. 1 illustrates a configuration in which the information processing system 1 includes eight platforms 10, but in the embodiment, the number of platforms 10 may be other than eight as long as it is plural. ..

Further, in the following, a case where communication between the platform 10 and the relay device 30 is executed according to a communication standard based on PCIe (Peripheral Component Interconnect-Express) (registered trademark) will be described, but in the embodiment, PCIe (registration) will be described. Communication according to a communication standard other than (trademark) may be executed.

Platforms 10-1 to 2-8 include a host device that functions as a control unit and GUI (Graphical User Interface) of the information processing system 1, and an arithmetic unit that executes AI (Artificial Intelligence) inference processing and image processing. Includes. In the following, as an example, it is assumed that the platform 10-1 functions as a host device and the platforms 10-2 to 10-8 function as an arithmetic unit. The platforms 10-2 to 10-8 as arithmetic units may have different functions, or at least a part of them may have a common function.

Platforms 10-1 to 2-8 are respectively provided with a route complex 11-1 to 11-8, and the relay device 30 is an endpoint 31-1 to corresponding to the route complex 11-1 to 11-8, respectively. It is equipped with 31-8. In the following, if it is not necessary to distinguish from each other, any root complex among the root complexes 11-11 to 11-8 is simply referred to as a root complex 11, and any of the endpoints 31-1 to 1-18. The endpoint may simply be referred to as endpoint 31.

Note that FIG. 1 illustrates a configuration in which the platform 10 and the root complex 11 have a one-to-one correspondence, but in the embodiment, the platform 10 and the root complex 11 always have a one-to-one correspondence. It doesn't have to be. For example, in an embodiment, one platform 10 may include two or more root complexes 11.

The platform 10 and the relay device 30 described above have a hardware configuration as shown in FIG. 2 below.

FIG. 2 is an exemplary and schematic block diagram showing the hardware configuration of the platform 10 and the relay device 30 of the information processing system 1 according to the embodiment.

First, the hardware configuration of the platform 10 will be described.

As shown in FIG. 2, platform 10-1 includes a root complex 11-1, a processor 12-1, a memory 13-1, and a storage unit 14-1. These hardwares (circuits) are communicably connected via buses (not shown).

Further, the other platforms 10-2 to 10-8 also have the same hardware configuration as the platform 10-1. That is, the platforms 10-2 to 10-8 have a root complex 11-2 to 11-8, a processor 12-2 to 12-8, a memory 13-2 to 13-8, and a storage unit 14-2, respectively. It is equipped with ~ 14-8.

In the following, if it is not necessary to distinguish from each other, any processor among processors 12-1 to 12-8 is simply referred to as processor 12, and any memory among memories 13-1 to 13-8 is simply memory. It may be described as 13, and any storage unit among the storage units 14-1 to 14-8 may be simply described as the storage unit 14.

The route complex 11 executes communication with the corresponding endpoint 31. In the embodiment, the root complex 11 and the endpoint 31 perform communication according to, for example, PCIe (Peripheral Component Interconnect Express) (registered trademark) as one of the communication standards. However, the platform 10 and the relay device 30 may execute communication according to a communication standard other than PCIe (registered trademark).

The processor 12 controls each part of the platform 10. The processor 12 includes, for example, a CPU (Central Processing Unit), an MPU (Micro Processing Unit), a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), an FPGA (Field Programmable Gate Array), or the like. Consists of. The processor 12 may be configured as a multi-core processor or may be configured as a combination of two or more processors.

The memory 13 stores data required for control by the processor 12, data generated as a result of control by the processor 12, and the like. The memory 13 is composed of, for example, a ROM (Read Only Memory) and a RAM (Random Access Memory). The ROM stores, for example, various software (computer programs) executed by the processor 12 and data for the software, and the RAM provides the processor 12 with a work area for executing the software stored in the ROM.

The storage unit 14 (nonvolatilely) stores various data including various software (computer programs) executed by the processor 12 and data transmitted / received to / from another platform 10. The storage unit 14 is composed of, for example, an HDD (Hard Disk Drive), an SSD (Solid State Drive), an SCM (Storage Class Memory), or the like.

The platform 10 realizes various functions for executing AI inference processing or image processing by executing a computer program stored in the memory 13 or the storage unit 14 by the processor 12.

The above computer program does not necessarily have to be stored in the memory 13 or the storage unit 14. For example, the computer programs described above may be in a format that can be installed on computer-readable recording media, such as various magnetic disks such as flexible disks (FDs) or various optical disks such as DVDs (Digital Versatile Disks). It may be provided as a computer program product recorded in an executable format.

In addition, the above computer program may be provided or distributed via a network such as the Internet. That is, the above computer program may be provided in the form of being stored on a computer connected to a network such as the Internet and being downloaded from the computer via the network.

Next, the hardware configuration of the relay device 30 that relays the communication between the platforms 10 will be described.

As shown in FIG. 2, the relay device 30 includes an endpoint 31, a processor 32, a memory 33, a storage unit 34, an internal bus 35, and an expansion bus 36.

The endpoint 31 is provided to correspond to the platform 10. The endpoint 31 executes data transmission (transfer) from the first platform of the platform 10 to a second platform different from the first platform via the expansion bus 36.

For example, when the route complex 11-5 transmits data to the platform 10-1 by DMA (Direct Memory Access) transfer, the relay device 30 starts from the endpoint 31-1 corresponding to the route complex 11-5 via the expansion bus 36. Data is received and the received data is transferred to the platform 10-1.

The processor 32 controls each part of the relay device 30. The processor 32 includes, for example, a CPU, an MPU, a DSP, an ASIC, a PLD, an FPGA, or the like. The processor 12 may be configured as a multi-core processor or may be configured as a combination of two or more processors.

The memory 33 stores data required for control by the processor 32, data generated as a result of control by the processor 32, and the like. The memory 33 is composed of, for example, a ROM and a RAM.

The storage unit 34 stores various data (nonvolatilely). The storage unit 34 is composed of, for example, an HDD, an SSD, an SCM, or the like.

The relay device 30 realizes various functions for relaying communication between the platforms 10 by executing a computer program stored in the memory 33 or the storage unit 34 by the processor 32.

The internal bus 35 communicatively connects the processor 32, the memory 33, the storage unit 34, and the expansion bus 36.

The expansion bus 36 connects a plurality of endpoints 31 and the internal bus 35 in a communicable manner. As a result, the expansion bus 36 enables data transfer between a plurality of endpoints 31, that is, data transfer between platforms 10. The expansion bus 36 is, for example, a bus compliant with PCIe (registered trademark).

Based on the hardware configuration as described above, the platform 10 according to the embodiment realizes the software configuration as shown in FIG. 3 below.

FIG. 3 is an exemplary and schematic diagram showing the software configuration of the platform 10 according to the embodiment.

As shown in FIG. 2, the platform 10-1 as a host device includes a BIOS (Basic Input Output System) 202, an OS (Operating System) 203, a driver 204, a service 205, and a virtual LAN (Local Area Network). ) The driver 206, the distributed control unit 207, and the common software 208 are realized. Based on these software configurations, platform 10-1 executes application 209 and controls a PC (Personal Computer) platform 1001 as a hardware resource provided in platform 10-1.

The BIOS 202 reads the OS 203 when the platform 10-1 is started, and executes basic input / output control for the platform 10-1. OS203 is, for example, Windows (registered trademark), but may be another OS.

OS203 accesses the route complex 11-1 of platform 10-1 by reading various drivers 204 including the bridge driver 204A for controlling the expansion bus, and communicates with other platforms 10. Execute. Further, the OS 203 executes various processes by reading the service 205 for performing various controls.

The virtual LAN driver 206 and the distributed control unit 207 are realized in the upper layers of the driver 204 and the service 205. The common software 208 is realized in the upper layer of the virtual LAN driver 206 and the distributed control unit 207 and in the lower layer of the application 209. As a result, the application 209 accesses the virtual LAN driver 206 via the common software 208, assuming that the other platform 10 exists on the virtual LAN, and communicates with the other platform 10. To execute.

On the other hand, platforms 10-2 and 10-3 as arithmetic units realize Bootloader 212, OS 213, driver 214, virtual LAN driver 215, distributed control unit 216, and common software 217. Based on these software configurations, platforms 10-2 and 10-3 execute distributed processing A and distributed processing B, respectively, and are hardware platforms as hardware resources provided in platforms 10-2 and 10-3, respectively. Control 1011.

In FIG. 3, for simplification, only the software configurations of platforms 10-2 and 10-3 are shown, but in the embodiment, other platforms 10-4 to 10-8 as arithmetic units are also shown. Software configurations similar to platforms 10-2 and 10-3 can be achieved.

The Bootloader 212 is started when the power of the corresponding platform 10 is turned on, and the OS 213 is started. OS203 is, for example, Linux®, but may be another OS.

OS 213 accesses the corresponding route complex 11 by reading various drivers 214 including the bridge driver 214A for controlling the expansion bus, and executes communication with other platforms 10. ..

The virtual LAN driver 215 and the distributed control unit 216 are realized on the upper layer of the driver 214. The common software 217 is realized in the upper layer of the virtual LAN driver 215 and the distributed control unit 216 and in the lower layer of the corresponding distributed processing. As a result, the distributed processing A and the distributed processing B access the virtual LAN driver 215 via the common software 217, assuming that the other platform 10 exists on the virtual LAN, and the other platform 10 Execute communication with.

Based on the above software configuration, the information processing system 1 according to the embodiment executes the communication process as shown in FIG. 4 below when communicating between the platforms 10. Although FIG. 4 illustrates the communication processing between the platform 10-1 as the host device and the platform 10-5 as the arithmetic unit, the communication processing between the other platforms 10 can be executed in the same manner.

FIG. 4 is an exemplary and schematic diagram for explaining the communication processing executed by the information processing system 1 according to the embodiment.

As shown in FIG. 4, the information processing system 1 has a layer structure including a plurality of predetermined layers. The information processing system 1 executes communication between the platforms 10 via a plurality of layers.

For example, a case where the software of the platform 10-1 as a host device transmits (transfers) data to the platform 10-5 as an arithmetic unit will be described. In this case, the software of platform 10-1 is transmitted to the physical layer of the relay device 30, more specifically platform 10-1, via the transaction layer, data link layer, and physical layer (PHY) of platform 10-1. Data is transferred to the physical layer corresponding to the connected endpoint 31-1.

When data is transferred to the physical layer corresponding to the endpoint 31-1, the relay device 30 passes the data to the transaction layer via the data link layer. Then, the relay device 30 transfers data by tunneling to the endpoint 31-5 to which the platform 10-5 is connected in the transaction layer. Then, the relay device 300 transfers the data transferred by tunneling to the physical layer of the platform 10-5 via the data link layer and the physical layer corresponding to the endpoint 31-5.

The platform 10-5 passes the data transferred from the relay device 30 to the software via the physical layer, the data link layer, and the transaction layer.

As described above, in the embodiment, the data transfer from the platform 10-1 to the platform 10-5 is realized by tunneling the data in the transaction layer of the relay device 30. Further, the transfer of data from the platform 10-5 to the platform 10-1 is realized by moving the data in each layer in the reverse order of the above.

In the embodiment, when the transfer of data from another platform 10 is not concentrated on a specific platform 10, it is also possible to execute data transfer in parallel between any plurality of different sets of platforms 10. is there.

That is, when a plurality of communications from the other platform 10 are concentrated on the specific platform 10, the relay device 30 serially processes the plurality of communications from the other platform 10. However, when different platforms 10 communicate with each other and communication is not concentrated on a specific platform 10, the relay device 30 can process a plurality of communications between the platforms 10 in parallel.

Here, in the embodiment, the transmission / reception of data between the platforms 10 is executed via the transmission buffer 502 provided in each platform 10 in the form shown in FIG. 5 below. Although the relay device 30 is not shown in FIG. 5, as described above with reference to FIG. 4, data transmission / reception by DMA (Direct Memory Access) between the platforms 10 is relayed by the relay device 30. Will be done.

FIG. 5 is an exemplary and schematic diagram for explaining the transmission and reception of data via the transmission buffer 502 according to the embodiment.

As shown in FIG. 5, each platform 10 has a transmission buffer 502 as a storage area for writing data to be transmitted to each platform 10. Here, the transmission buffers 502 to 502-8 of the platforms 10-1 to 10-8 are collectively referred to as the transmission buffer 502 (however, the one shown in FIG. 5 is the transmission buffer 502). Credit buffers 502-1 and 502-5 only). The transmission buffer 502 is realized, for example, on the memory 13.

Note that the example shown in FIG. 5 assumes a case where data is transmitted from the platform 10-5 to the platform 10-1 as in the example shown in FIG. 4 described above. Therefore, in the following, the transmission and reception of data via the transmission buffer 502-5 of the platform 10-5 will be mainly described.

In the example shown in FIG. 5, first, in step S501, for example, the virtual LAN driver 206 of the platform 10-5 sends a data transmission request to the platform 10-1 to the transmission buffer 502-5 of the platform 10-5. Issued to the area indicated by Slot # 0.

Here, the area represented by Slot # 0 is an area for transmitting data to the platform 10-1. The transmission buffer 502-5 has a plurality of areas for transmitting data to another platform 10 in addition to the address range indicated by Slot # 0.

Then, in step S502, the bridge driver 214A of the platform 10-5 writes the data to be transmitted to the platform 10-1 in the area indicated by Slоt # 0 in response to the above transmission request.

Then, in step S503, the relay device 30 (not shown in FIG. 5) corresponds to the endpoints 31-5 and the platform 10-1 corresponding to the platform 10-5 by processing such as EPtoEP (Endpoint to Endpoint). It connects to the endpoint 31-1 and transfers data from platform 10-5 to platform 10-1.

By the way, when the data writing in the above step S502 is completed, the bridge driver 214A of the platform 10-5 requests the platform 10-1 to receive the data via the relay device 30.

Then, in step S504, the bridge driver 204A of the platform 10-1 acquires the data transferred from the platform 10-5 via the relay device 30 in response to the reception request from the platform 10-5, and the acquired data. Is stored in the storage unit 1401 via a buffer provided in the memory 13-1.

In this way, data transmission from platform 10-5 to platform 10-1 is executed via the relay device 30.

As shown in FIG. 5, the platform 10-1 is also provided with a transmission buffer 502-1 similar to the transmission buffer 502-5 of the platform 10-5. Platform 10-1 can use the transmission buffer 502-1 to execute data transmission to another platform 10. Further, although not shown in FIG. 5, other platforms 10 also have a similar transmit buffer 502. Therefore, in the embodiment, it is possible to realize the transmission / reception of data between arbitrary platforms 10 by the same processing as described above.

By the way, in the transmission / reception of data via the transmission buffer 502 as described above, the area in which the data is written in the transmission buffer 502 is set as a lock area that cannot be overwritten until the transmission / reception is completed. Is common. In this case, when the transmission / reception of the written data is completed, the lock area is generally changed to an overwritable unlock area, that is, the transmission buffer 502 is released.

However, in a configuration in which a plurality of platforms 10 are provided, such as the information processing system 1 according to the embodiment, communication between the platforms 10 may occur frequently. In this case, if the transmission buffer 502 for which the transmission / reception of the written data is completed is released every time the transmission / reception of data is completed, the resources devoted to the communication are reduced, so that the communication speed is reduced. Likely to happen.

Therefore, in the embodiment, the release of the transmission buffer 502 is not performed every time the data transmission / reception is completed, but at a location where the transmission / reception of the written data is completed according to the data transmission / reception being completed a plurality of times. It is configured to run. According to such a configuration, the frequency of releasing the transmission buffer 502 is reduced, so that more resources can be devoted to communication. Therefore, it is possible to suppress a decrease in communication speed due to the release of the transmission buffer 502.

Further, as shown in FIG. 6, the embodiment depends on the free information of the transmission buffer 502, that is, the size of the unlock area writable for data transmission in the transmission buffer 502. , The frequency of releasing the transmission buffer 502 can be changed dynamically.

FIG. 6 is an exemplary and schematic sequence diagram showing a flow of processing executed when data is transmitted / received between platforms 10 according to the embodiment.

Here, as an example, the transmission (transfer) of data from the platform 10-5 to the platform 10-1 will be described. In the example shown in FIG. 6, the virtual LAN driver 215 and the bridge driver 214A of platform 10-5 are illustrated as the "sender virtual LAN driver" and the "sender bridge driver", respectively, and the bridge driver 204A of platform 10-1 is It is illustrated as a "receiver bridge driver".

As shown in FIG. 6, in the embodiment, first, in step S610, the virtual LAN driver 215 as the transmitting side virtual LAN driver is referred to the platform 10-1 with respect to the bridge driver 214A as the transmitting side bridge driver. Execute a data transmission request.

Then, in step S620, the bridge driver 214A responds to the transmission request from the virtual LAN driver 215 to the area for transmitting data to the platform 10-1 in the transmission buffers 502-5 of the platform 10-5. The data to be transmitted is written, and the relay device 30 is made to execute the transfer of the data to the platform 10-1. Then, the bridge driver 214A updates the free information regarding the free space of the transmission buffer 502-5. The free information can be realized as a table for managing, for example, a flag indicating whether each area of the transmission buffer 502-5 is set in the lock area or the unlock area.

Then, in step S630, the bridge driver 214A requests the bridge driver 204A as the receiving side bridge driver to receive data via the relay device 30.

Then, in step S640, the bridge driver 204A acquires the data transferred by the relay device 30 from the transmission buffer 502-5 in response to the reception request from the bridge driver 214A. At this time, the transmission / reception completion flag is set at the location (area) where the data acquisition is completed in the transmission buffer 502-5.

Then, in step S650, the bridge driver 204A transmits a reception completion notification to the bridge driver 214A via the relay device 30 in response to the completion of data acquisition.

Then, in step S660, the bridge driver 214A confirms the free information of the transmission buffer 502 of the platform 10-5 in response to the reception completion notification from the bridge driver 204A. Then, the bridge driver 214A executes a process as described below according to the free information.

For example, if the free information indicates that the transmission buffer 502-5 is free, that is, the transmit buffer 502-5 has zero writable unlock area for data transmission, the process of step S671 is executed. Will be done. In this case, it can be considered that there is a shortage of free space in the transmission buffer 502-5 because the frequency of releasing the transmission buffer 502-5 is low.

Therefore, in this case, in step S671, the bridge driver 214A increases the frequency of releasing the transmission buffer 502. For example, if the latest setting is to release the part of the transmission buffer 502-5 where the transmission / reception of the written data is completed when the reception completion notification is received m times, the bridge The driver 214A of the transmission buffer 502-5 so as to release the portion of the transmission buffer 502-5 where the transmission / reception of the written data is completed when the reception completion notification is received m-1 times. Reduce the number of times the reception completion notification is received, which is the criterion for whether or not to execute the release. Whether or not data transmission / reception is completed is determined based on the transmission / reception completion flag or the like described above.

On the other hand, when the free information indicates that the free space in the transmission buffer 502-5 is equal to or greater than the threshold value, the process of step S672 is executed. The threshold is set as a value sufficiently larger than a value indicating the amount of data to be transmitted. In this case, even if the frequency of releasing the transmission buffer 502-5 is reduced, it can be considered that a shortage of free space in the transmission buffer 502-5 is unlikely to occur.

Therefore, in this case, in step S672, the bridge driver 214A reduces the frequency of releasing the transmission buffer 502-5. For example, if the latest setting is to release the part of the transmission buffer 502-5 where the transmission / reception of the written data is completed when the reception completion notification is received m times, the bridge The driver 214A releases the transmission buffer 502-5 so as to release the portion of the transmission buffer 502-5 where the transmission / reception of the written data is completed when the reception completion notification is received m + 1 times. Increase the number of times the reception completion notification is received, which is the criterion for whether or not to execute.

If the free information indicates that the free space in the transmission buffer 502-5 is larger than zero and smaller than the threshold value, neither of the processes of steps S671 and S672 is executed. That is, in this case, the bridge driver 214A keeps the frequency of releasing the transmission buffer 502-5, that is, the number of receptions of the reception completion notification which is a criterion for whether or not to execute the release of the transmission buffer 502-5, as it is. maintain.

Then, the bridge driver 214A releases the transmission buffer 502-5 at a frequency adjusted (or maintained) as described above.

For example, if the current setting is to execute the release of the transmission buffer 502-5 when the reception completion notification is received n times, in other words, the current setting of the release of the transmission buffer 502-5. Consider the case where the frequency of is set to 1 / n. In this case, the bridge driver 214A was written out of the transmission buffers 502-5 in step S680 only when the transmission buffer 502-5 was not released in the latest n-1 reception completion notification. Release the part where data transmission / reception is completed. However, when the transmission buffer 502-5 is released in response to any of the latest n-1 reception completion notifications, the bridge driver 214A releases the transmission buffer 502-5. do not do.

In this way, in the embodiment, the frequency of release of the transmission buffer 502-5 is dynamically adjusted. The frequency of adjustment is considered to converge to the optimum frequency.

The example shown in FIG. 6 is just an example. In the embodiment, the bridge driver 204A or 214A of the platform 10 other than the platform 10-5 becomes the transmitting bridge driver, and the bridge driver 214A of the platform 10 other than the platform 10-1 becomes the receiving bridge driver. Is also possible.

As described above, the transmitting side bridge driver according to the embodiment sets the area in the transmission buffer 502 to which the data is written as a lock area that cannot be overwritten each time data is written to the transmission buffer 502. , When the transmission / reception of the written data is completed and the reception completion notification from the receiving bridge driver is received multiple times, the lock area is set to the unlockable area that can be overwritten. According to such a configuration, for example, the frequency of releasing the transmission buffer 502 is lower than that in the case where the transmission buffer 502 is always released every time data transmission / reception is completed. More resources can be devoted to it. Therefore, it is possible to suppress a decrease in communication speed due to the release of the transmission buffer 502.

More specifically, the transmitting bridge driver according to the embodiment receives the reception completion notification, which is a criterion for whether or not to set the lock area as the unlock area, based on the free information regarding the free space of the transmission buffer 502. Dynamically change. According to such a configuration, the frequency of releasing the transmission buffer 502 can be appropriately adjusted based on the free information.

For example, when the above-mentioned free information indicates that the free space of the transmission buffer 502 is equal to or more than a predetermined value, the transmitting bridge driver according to the embodiment receives the lock area so that the lock area is set to the unlock area less frequently. Change the number of times. According to such a configuration, it is possible to suppress a reduction in resources devoted to communication due to the release of the transmission buffer 502 even when there is a margin in the free space.

<Modification example>
In the above-described embodiment, a configuration in which the frequency of releasing the transmission buffer 502 is dynamically adjusted is exemplified. However, as a modification, a configuration in which the number of receptions of the reception completion notification, which is a criterion for whether or not to release the transmission buffer 502, is fixedly set as a value of 2 or more can be considered. Even with such a configuration, for example, the transmission buffer 502 has a transmission buffer 502 as compared with the case where the transmission / reception of the written data is completed every time the transmission / reception of data is completed. Since the frequency of release is reduced, it is possible to suppress a decrease in communication speed.

Further, in the above-described embodiment, as an example, a configuration in which the number of receptions of the reception completion notification, which is a criterion for whether or not to release the transmission buffer 502, is increased or decreased by one is illustrated. However, as a modification, a configuration in which the number of receptions of the reception completion notification, which is a criterion for whether or not to release the transmission buffer 502, is increased or decreased by 2 or more can be considered. In this case, the number to be increased or decreased may be determined according to the free information.

Further, in the above-described embodiment, PCIe (registered trademark) is exemplified as the I / O interface of each part, but the I / O interface is not limited to PCIe (registered trademark). For example, the I / O interface of each part may be a technology capable of transferring data between a device (peripheral controller) and a processor by a data transfer bus. The data transfer bus may be a general-purpose bus capable of transferring data at high speed in a local environment (for example, one system or one device) provided in one housing or the like. The I / O interface may be either a parallel interface or a serial interface.

The I / O interface may be configured so that point-to-point connection is possible and data can be serially transferred on a packet basis. The I / O interface may have a plurality of lanes in the case of serial transfer. The layer structure of the I / O interface may include a transaction layer that generates and decodes packets, a data link layer that performs error detection, and a physical layer that converts serial and parallel. The I / O interface also includes a root complex at the top of the hierarchy with one or more ports, endpoints that are I / O devices, switches for increasing ports, and bridges that translate protocols. It's fine. The I / O interface may multiplex the data to be transmitted and the clock signal by a multiplexer and transmit them. In this case, the receiving side may separate the data and the clock signal by a demultiplexer.

1 Information processing system 10, 10-1 to 10-8 platforms (1st platform, 2nd platform)
30 Relay device 36 Expansion bus 204A, 214A Bridge driver (1st bridge driver, 2nd bridge driver)
502 Transmission buffer

Claims (5)

With multiple platforms
A relay device that connects the multiple platforms so that they can communicate via an expansion bus,
With
The first platform among the plurality of platforms is
Wherein the plurality of the first platform of the platform in response to a transmission request of data to the different second platform, writes the data to the transmission buffer of the first platform relative to said second platform A first bridge driver for transmitting the data reception request is provided.
The second platform is
In response to the reception request transmitted by the first bridge driver, the data transferred from the transmission buffer to a predetermined storage area of the second platform is acquired via the relay device, and the first platform A second bridge driver for transmitting the data reception completion notification to the user is provided.
Each time the data is written to the transmission buffer, the first bridge driver sets the area of the transmission buffer in which the data is written as a lock area that cannot be overwritten, and transmits / receives the written data. Is completed, and when the reception completion notification from the second bridge driver is received two or more predetermined times , the lock area corresponding to the reception completion notification of the predetermined number of times can be overwritten. Set in the lock area,
Information processing system. The first bridge driver dynamically changes the number of times of receiving the reception completion notification, which is a criterion for whether or not to set the lock area in the unlock area, based on the free information regarding the freeness of the transmission buffer. To do
The information processing system according to claim 1. When the free information indicates that the transmission buffer has more than a predetermined amount of free space, the first bridge driver changes the number of receptions so that the frequency of setting the lock area to the unlock area decreases. To do
The information processing system according to claim 2. A platform included in multiple platforms that are communicably connected via an expansion bus by a relay device.
Wherein the plurality of the platform of the platform in response to the transmission request of the data to different other platforms, writes the data to the transmission buffer of the platform, the reception of the data to the other platforms Equipped with a bridge driver to send requests
Each time the bridge driver writes the data to the transmission buffer, the bridge driver sets the area of the transmission buffer in which the data is written to a lock area that cannot be overwritten, and the transmission / reception of the written data is completed. and and and wherein by said data transferred from said transmitting buffer via the relay device in a predetermined storage area of the other platforms in response to the received request the other platform to get when the reception completion notification transmitted from other platforms to the platform has been received more than a predetermined number, to set the lock region corresponding to a reception completion notice of the predetermined number of times overwritable unlocking region ,
platform. To a computer as a platform included in multiple platforms connected to be communicable via an expansion bus by a relay device
Wherein the plurality of the platform of the platform in response to the transmission request of the data to different other platforms, writes the data to the transmission buffer of the platform, the reception of the data to the other platforms Sending a request and
Each time the data is written to the transmission buffer, the area of the transmission buffer to which the data is written is set as a lock area that cannot be overwritten, and the transmission / reception of the written data is completed. the from the other platforms by the data transferred from the transmission buffer through the relay device in a predetermined storage area of the other platforms in response to the received request the other platform to get When the reception completion notification sent to the platform is received two or more predetermined times , the lock area corresponding to the predetermined number of reception completion notifications is set as an overwriteable unlock area.
A program to execute.

Images