A system for sending completed computations ahead of time between serially connected electronic devices

By using the DMA engine and the accelerator in the computing system to send completion information in advance between the first electronic devices, the problem of communication delay between the series-connected electronic devices is solved, and the communication efficiency between the endpoint device and the host is improved.

CN113138950BActive Publication Date: 2024-11-05SAMSUNG ELECTRONICS CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202011284417.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2020-01-20
Filing Date
2020-11-17
Publication Date
2024-11-05
Estimated Expiration
2040-11-17

AI Technical Summary

Technical Problem

In computing systems, communication delays between electronic devices connected in series cause slower communication speeds between endpoint devices and hosts, especially in the case of high-speed electronic devices.

Method used

By introducing a first electronic device and a second electronic device in the computing system, the first electronic device includes a memory and an accelerator, and the data and command completion information are sent in advance using a direct memory access (DMA) engine to ensure that the accelerator is immediately after completing the calculation. Send complete information to reduce latency.

Benefits of technology

The command processing time and completion information reception time due to delay are effectively reduced, and the communication efficiency of the computing system is improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113138950B_ABST
    Figure CN113138950B_ABST
Patent Text Reader

Abstract

A computing system includes: a host; a first electronic device including a memory and an accelerator; and a second electronic device including a direct memory access (DMA) engine. Based on a command sent from the host through the first electronic device, the DMA engine sends data and completion information of the command to the first electronic device. The memory includes a data buffer for storing data and a completion queue buffer for storing completion information. The accelerator performs calculations on the data. The DMA engine sends the data to the first electronic device and then sends the completion information to the first electronic device.
Need to check novelty before this filing date? Find Prior Art

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the priority of Korean Patent Application No. 10-2020-0007182 filed on January 20, 2020, in the Korean Intellectual Property Office, the disclosure of which is incorporated herein by reference in its entirety. Technical Field

[0003] Embodiments of the present disclosure described herein relate to computing systems, and more particularly, to a computing system that sends completions ahead of time between serially connected electronic devices. Background Art

[0004] In a computing system, multiple electronic devices can be interconnected and communicate with each other. A bus of the computing system can interconnect multiple electronic devices of the computing system. Multiple electronic devices can be connected in series or sequentially. The serially connected electronic devices can communicate with a host of the computing system.

[0005] The electronic device corresponding to the endpoint device among the plurality of electronic devices may communicate with the host through at least one or more electronic devices. Due to the delay of at least one or more electronic devices placed between the endpoint device and the host, the communication speed between the endpoint device and the host may become slower. Specifically, in the case where the endpoint device is a high-speed electronic device, the delay may become more serious. Summary of the invention

[0006] Embodiments of the present disclosure provide a computing system that transmits a completion signal in advance between electronic devices connected in series.

[0007] According to an exemplary embodiment, a computing system includes: a host; a first electronic device including a memory and an accelerator; and a second electronic device including a direct memory access (DMA) engine. Based on a command sent from the host through the first electronic device, the DMA engine sends data and completion information of the command to the first electronic device. The memory includes a data buffer for storing data and a completion queue buffer for storing completion information. The accelerator performs calculations on the data. The DMA engine sends the data to the first electronic device and then sends the completion information to the first electronic device.

[0008] According to an exemplary embodiment, a computing system includes: a host; a first electronic device that communicates with the host; and a second electronic device that communicates with the host through the first electronic device. The first electronic device includes a memory that includes a data buffer that receives data from the second electronic device. The second electronic device receives a command from the host through the first electronic device. The first electronic device also includes: (1) a completion queue buffer that receives completion information about the command from the second electronic device; (2) an accelerator that performs calculations on data stored in the memory; and (3) an interface circuit that sends the completion information stored in the completion queue buffer to the host when the accelerator completely performs the calculation.

[0009] According to an exemplary embodiment, a computing system includes: a host that manages a submission queue of a host memory buffer in which a command is written and a completion queue of a host memory buffer in which completion information about the command is written. The computing system also includes a first electronic device that includes a memory and an accelerator. The memory includes a data buffer that stores data requested by the command and a completion queue buffer that stores completion information. The accelerator performs calculations on the data. The computing system also includes a second electronic device that obtains the command through the first electronic device and sends data and completion information to the first electronic device based on the command. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The above and other objects and features of the present disclosure will become apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings.

[0011] Figure 1 A block diagram of a computing system according to an exemplary embodiment of the present disclosure is shown.

[0012] Figure 2 Shown in detail Figure 1 Block diagram of a computing system.

[0013] Figure 3 shows Figure 1 Block diagram of a computing system.

[0014] FIG. 4 illustrates an operation timing diagram of the computing system of FIG. 3 .

[0015] Figure 5 Show Figure 1 Block diagram of a computing system.

[0016] Figure 6 Show Figure 5 The operation timing of the computing system.

[0017] Figure 7 Show Figure 5 A method of operating a computing system.

[0018] Figure 8 Show Figure 1Block diagram of a computing system.

[0019] Fig. 9 Show Figure 8 The operation timing of the computing system.

[0020] Fig.10 Show Figure 8 A method of operating a computing system.

[0021] Fig.11 3 to Fig.10 A block diagram of one of the FPGAs described.

[0022] Fig.12 3 to Fig.10 A block diagram of one of the storage devices described.

[0023] Fig.13 A block diagram showing a computing device according to an application example of the present disclosure. DETAILED DESCRIPTION

[0024] Hereinafter, embodiments of the present disclosure will be described in detail and clearly to such an extent that a person having ordinary skill in the art can easily implement the present disclosure.

[0025] Figure 1 A block diagram of a computing system according to an exemplary embodiment of the present disclosure is shown. The computing system 10 may include a host 11, a first electronic device 12, and a second electronic device 13. For example, the computing system 10 may be implemented on various electronic devices such as desktop computers, laptop computers, tablet computers, video game consoles, workstations, servers, computing devices, and mobile devices, or on a motherboard of an electronic device.

[0026] The host 11 may manage and process the overall operation of the computing system 10. For example, the host 11 may include a general-purpose processor, a dedicated processor, or an application processor. The host 11 may be the processor itself, or the host 11 may be an electronic device or system including a processor.

[0027] The first electronic device 12 may be connected to the host 11 directly or in series (or sequentially). The second electronic device 13 may be connected to the first electronic device 12 directly or in series. For example, the host 11, the first electronic device 12, and the second electronic device 13 may be connected in a chain manner. The second electronic device 13 may be connected to the host 11 through the first electronic device 12. For example, the host 11 may communicate directly with the first electronic device 12, and may communicate with the second electronic device 13 through the first electronic device 12. For example, the second electronic device 13 may correspond to an endpoint device, and the first electronic device 12 may be an intermediate device that connects the host 11 and the endpoint device.

[0028] and Figure 1Different from the example shown, one or more electronic devices may be connected between the host 11 and the first electronic device 12, and one or more electronic devices may be connected between the first electronic device 12 and the second electronic device 13. For example, the first electronic device 12 and the second electronic device 13 may be implemented to be the same or different. For another example, the first electronic device 12 may correspond to a switch or a switch device that connects the host 11 and the second electronic device 13, and the second electronic device 13 may correspond to an endpoint device instead of a switch. For another example, although Figure 1 Not shown, the computing system 10 may further include a third electronic device capable of communicating with the host 11 through the first electronic device 12 and the second electronic device 13 .

[0029] For example, each of the host 11, the first electronic device 12, and the second electronic device 13 can be implemented using a system on a chip (SoC), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a storage device, a solid-state drive (SSD), a peripheral component interconnect express (PCIe) device, a non-volatile memory express (NVMe) device, a universal flash memory (UFS) device, a universal serial bus (USB) device, etc., but the present disclosure is not limited to the above examples.

[0030] Figure 2 Shown in detail Figure 1 The computing system 100 may correspond to Figure 1 The computing system 100 may include a processor 110, a root complex 120, a memory 130, and electronic devices 141, 142, 151 to 154, and 161 to 163. Here, the electronic devices may also be referred to as "input / output (I / O) devices".

[0031] The processor 110 may perform various types of arithmetic operations or logical operations. For example, the processor 110 may include at least one or more cores (e.g., homogeneous multi-cores or heterogeneous multi-cores) that perform the above operations and an internal cache memory. The processor 110 may execute program code, software, applications, firmware, etc. loaded on the memory 130.

[0032] The root complex 120 may mediate communication between the processor 110, the memory 130, and the electronic devices 141, 142, 151 to 154, and 161 to 163. For example, the root complex 120 may manage communication timing, communication sequence, communication environment, etc. between the processor 110, the memory 130, and the electronic devices 141, 142, 151 to 154, and 161 to 163. The root complex 120 may be implemented in a hardware manner, a software manner, or a combination thereof, and may be installed on a motherboard of the computing system 100. The root complex 120 may be a root that connects the processor 110 and the memory 130 to an I / O hierarchy including the electronic devices 141, 142, 151 to 154, and 161 to 163. The root complex 120 may include one or more downstream ports DP, the electronic devices 141 and 142 may be connected to the downstream ports DP of the root complex 120, the electronic devices 151 to 154 may be connected to the downstream ports DP of the root complex 120, and the electronic devices 161 to 163 may be connected to the downstream ports DP of the root complex 120. The number of downstream ports DP is not limited to Figure 2 The number of electronic devices connected to each downstream port DP may be one or more.

[0033] In an exemplary embodiment, communication between the root complex 120 and the electronic devices 141, 142, 151 to 154 and 161 to 163 can be performed in accordance with various interface protocols such as a Peripheral Component Interconnect Express (PCIe) protocol, a Mobile PCIe (M-PCIe) protocol, a Non-Volatile Memory Express (NVMe) protocol, a Universal Serial Bus (USB) protocol, a Small Computer System Interface (SCSI) protocol, an Advanced Technology Attachment (ATA) protocol, a Parallel ATA (PATA), a Serial ATA (SATA), a Serial Attached SCSI (SAS) protocol, an Integrated Drive Electronics (IDE) protocol, a Universal Flash Storage (UFS) protocol, and a Firewire protocol.

[0034] The memory 130 may temporarily store data used for the operation of the computing system 100. The memory 130 may store data processed by or to be processed by the processor 110. For example, the memory 130 may include a volatile memory such as a static random access memory (SRAM), a dynamic RAM (DRAM), a thyristor RAM (TRAM), or a non-volatile memory. An application program, a file system, or a device driver that can be executed by the processor 110 may be loaded onto the memory 130. The programs and software layers loaded onto the memory 130 are not limited to Figure 2 The example shown. The memory 130 may include a host memory buffer HMB. A portion of the entire storage area of ​​the memory 130 may be allocated to the host memory buffer HMB.

[0035] For example, processor 110 may correspond to Figure 1 For example, the processor 110 and the root complex 120 may correspond to Figure 1 For example, the processor 110, the root complex 120 and the memory 130 may all correspond to Figure 1 The processor 110, the root complex 120, and the memory 130 may be implemented using one chip to constitute the host 11. Alternatively, the processor 110, the root complex 120, and the memory 130 may be implemented using one or more chips to constitute the host 11.

[0036] Each of the electronic devices 142, 152, 154, and 163 may correspond to an endpoint device. Each of the electronic devices 142, 152, 154, and 163 may include an endpoint port EP. Each of the remaining electronic devices 141, 151, 153, 161, and 162 may correspond to an intermediate device connecting the root complex 120 and the endpoint device. Each of the electronic devices 141, 151, 153, 161, and 162 may include an upstream port UP and a downstream port DP. For example, the upstream port UP may be closer to the root complex 120 than the downstream port DP. The endpoint port EP of the endpoint device may be connected to the downstream port DP of the intermediate device or the root complex 120. The endpoint port EP may also be referred to as an “upstream port UP.” For example, one of the electronic devices 141, 151, 153, 161, and 162 may correspond to Figure 1 The first electronic device 12, one of the electronic devices 142, 152, 154 and 163 may correspond to Figure 1 The second electronic device 13.

[0037] In an exemplary embodiment, Figure 1 The electronic devices 141, 151, 153, 161 and 162 corresponding to the first electronic device 12 may be FPGA, PCIe switch, PCIe device, NVMe device, storage device, solid state drive (SSD), etc. Figure 1 The electronic devices 142, 152, 154 and 163 corresponding to the second electronic device 13 may also be FPGA, PCIe switch, PCIe device, NVMe device, storage device, SSD, etc. As described above, the computing system 100 may also include any other endpoint devices (not shown) connected to the electronic devices 142, 152, 154 and 163.

[0038] For example, an electronic device corresponding to an FPGA among the electronic devices 141, 151, 153, 161, and 162 may assist the operation or calculation of the processor 110. The electronic device as an FPGA may perform calculations on read data sent from one of the electronic devices 142, 152, 154, and 163, may send the calculation result to the root complex 120, and may assist the calculation of the processor 110. The electronic device as an FPGA may perform calculations on write data sent from the root complex 120, may send the calculation result to any other electronic device, and may assist the calculation of the processor 110.

[0039] Figure 3 shows Figure 1 FIG. 4 shows an operation sequence of the computing system of FIG. 3 . The computing system 200 may include a host 210, an FPGA 220, and a storage device 230. For example, the computing system 200, the host 210, the FPGA 220, and the storage device 230 may correspond to Figure 1 The computing system 100, the host 11, the first electronic device 12 and the second electronic device 13.

[0040] The host 210 may include a submission queue (SQ) 211, a completion queue (CQ) 212, and a data buffer 213. The submission queue 211, the completion queue 212, and the data buffer 213 may be placed or implemented in Figure 2 The host 210 may perform input / output operations on the storage device 230 through the FPGA 220 based on a pair of submission queues 211 and completion queues 212.

[0041] The FPGA 220 may be placed between the host 210 and the storage device 230, and may transmit an electrical signal from the host 210 (or the storage device 230) to the storage device 230 (or the host 210). The FPGA 220 may include a data buffer 221 and an accelerator 222. The data buffer 221 may store data exchanged between the host 210 and the storage device 230. The accelerator 222 may perform various types of arithmetic operations or calculations on the data stored on the data buffer 221.

[0042] The storage device 230 may receive a command (or request) from the host 210, may process the received command, and may send a processing result to the host 210. The storage device 230 may include a nonvolatile memory 239 and a controller 231 for controlling the nonvolatile memory 239. For example, the nonvolatile memory 239 may include a NAND flash memory, a NOR flash memory, a phase change RAM (PRAM), a magnetic RAM (MRAM), a resistive RAM (RRAM), a ferroelectric RAM (FeRAM), etc. The controller 231 may include a direct memory access (DMA) engine 232 configured to send a request or data directly to the FPGA 220 or the host 210.

[0043] The host 210 may input or write a command for the storage device 230 into an entry (or slot) of the submission queue 211, and may update a tail pointer of the submission queue 211, that is, a doorbell associated with the submission queue 211. For example, the doorbell may include an address indicating the submission queue 211. For example, a pair of the submission queue 211 and the completion queue 212 may be provided for each core of the processor 110, and the number of pairs may be one or more. The host 210 may submit or send the doorbell associated with the submission queue 211 to the storage device 230 through the FPGA 220. The host 210 may update the doorbell register of the controller 231 of the storage device 230 by sending the doorbell. The storage device 230 may read the doorbell stored in the doorbell register, and may determine that the host 210 inputs or writes a command in the submission queue 211. The storage device 230 may obtain or read the command of the submission queue 211 of the host 210 based on the doorbell. The storage device 230 may request a command of the submission queue 211 of the host 210, and may send the request to the host 210 through the FPGA 220. Next, the storage device 230 may read or receive the command of the submission queue 211 from the host 210 through the FPGA 220 (①).

[0044] When the switch delay of the FPGA 220 elapses, the DMA engine 232 of the storage device 230 may receive a command from the FPGA 220, and may store the received command in a memory associated with the controller 231 (see Fig.12631_2 or 638). Delay may also be referred to as "time". For example, the entire delay of FPGA 220 may be divided into a host-side switch delay and a storage device-side switch delay. In detail, FPGA 220 may include a transmission path including one or more transmission circuits for sending commands and a physical path in FPGA 220. For example, the time it takes for a command to pass through a portion of the transmission path of FPGA 220 may correspond to the host-side switch delay, and the time it takes for the command to pass through the rest of the transmission path of FPGA 220 may correspond to the storage device-side switch delay. As in the above-mentioned command transmission (①), switch delay may occur even in the case of sending a request for a command or sending a doorbell indicating that a command is written to the submission queue 211.

[0045] The controller 231 of the storage device 230 may receive a command sent from the host 210 through the FPGA 220 (①) and may process the received command. For example, the command may be a read command. The controller 231 may decode the command, and the DMA engine 232 may send data requested by the command to the FPGA 220 (②). The data requested by the command may correspond to data stored in the nonvolatile memory 239 or stored in any other memory (refer to Fig.12 631_2 or 638).

[0046] When the switch delay elapses after the read data is sent from the storage device 230 to the FPGA 220 (②), the read data may reach the data buffer 221, that is, may be stored in the data buffer 221. The accelerator 222 may perform calculations on the read data stored in the data buffer 221. When the accelerator 222 completely performs the calculations, the accelerator 222 may send one of the read data and the calculated data generated by the accelerator 222 to the host 210 (③). When the switch delay elapses after one of the calculated data and the read data is sent to the host 210, the calculated data and the read data may reach the host 210 (③). In addition, the accelerator 222 may send a signal (or notification signal) indicating that the accelerator 222 has completely performed the calculations to the storage device 230 (④). When the switch delay elapses after the signal indicating that the accelerator 222 has completely performed the calculations is sent, the signal may reach the DMA engine 232 of the storage device 230 (④). For example, as shown in FIG. 4 , the accelerator 222 may send one of the calculated data and the read data to the host 210 (③), and then may send a signal indicating that the accelerator 222 has completely performed the calculation to the storage device 230 (④). For another example, unlike the example shown in FIG. 4 , the accelerator 222 may send one of the calculated data and the read data to the host 210, and may send a signal indicating that the accelerator 222 has completely performed the calculation to the storage device 230. In any case, after the calculation of the accelerator 222 is completed, the FPGA 220 may send a signal indicating that the accelerator 222 has completely performed the calculation to the storage device 230.

[0047] The controller 231 may receive the signal (④), and may check through the signal that the accelerator 222 has completely calculated the read data and the command of the host 210 has been processed. The DMA engine 232 may send completion information (or completion) about the command to the host 210 through the FPGA 220 (⑤). For example, the completion information may include whether the command is completed, the result of processing the command, etc. When the switch delay elapses after the completion information is sent from the storage device 230 to the FPGA 220 (⑤), the completion information may arrive at the completion queue 212 of the host 210 (⑤). As in the above-mentioned command transmission, a switch delay may occur even when the completion information (⑤) is sent. Embodiments of the present disclosure may provide various computing systems 300, 400, and 1000 for improving the time or delay required to completely process a command (or the time or delay required to receive the completion information of the command) due to the switch delay caused by the FPGA 220 placed between the host 210 and the storage device 230.

[0048] Figure 5 Show Figure 1 Block diagram of a computing system. Figure 6 Show Figure 5 The operation timing of the computing system. Figure 7Show Figure 5 A method of operating a computing system.

[0049] The computing system 300 may include a host 310, an FPGA 320, and a storage device 330. The host 310 may include a submission queue 311, a completion queue 312, and a data buffer 313. The components 311 to 313 of the host 310 may be similar to the components 211 to 213 of the host 210, respectively. The FPGA 320 may include a data buffer 321 and an accelerator 322. The components 321 and 322 of the FPGA 320 may be similar to the components 211 and 212 of the FPGA 220, respectively. Compared with the FPGA 220, the FPGA 320 may also include a completion queue buffer 323. For example, the completion queue buffer 323 of the FPGA 320 may correspond to a copy or mirror of the completion queue 312 of the host 310. The completion queue buffer 323 may be similar to the completion queue 312. The storage device 330 may include a controller 331 and a non-volatile memory 339. The controller 331 may include a DMA engine 332. Components 331, 332, and 339 of storage device 330 may be similar to components 231, 232, and 239, respectively, of storage device 230. The description will focus on the differences between computing system 300 and computing system 200.

[0050] The host 310 may write or store a command associated with the storage device 330 in the submission queue 311. The host 310 may send a doorbell associated with the submission queue 311 to the FPGA 320, and the FPGA 320 may receive the doorbell. The FPGA 320 may send the doorbell to the storage device 330. Based on the doorbell, the storage device 330 may request a command of the submission queue 311 of the host 310 and may send the request to the FPGA 320. The FPGA 320 may send the request to the host 310. For example, the storage device 330 may access the submission queue 311 indicated by the address of the doorbell among one or more submission queues of the host 310.

[0051] In operation S305, the host 310 may receive the request and may send the command of the submission queue 311 to the FPGA 320 (①). When the switch delay elapses after the command of the submission queue 311 is sent from the host 310 to the FPGA 320, the FPGA 320 may receive the command of the submission queue 311 from the host 310 (①). In operation S310, the FPGA 320 may send the command of the submission queue 311 to the storage device 330. When the switch delay elapses after the command of the submission queue 311 is sent from the FPGA 320 to the storage device 330, the DMA engine 332 of the storage device 330 may receive the command of the submission queue 311 (①). That is, the storage device 330 may acquire the command of the submission queue 311. In operation S315, the controller 331 may determine whether the command of the submission queue 311 is a read command.

[0052] When the command of the submission queue 311 is a read command (Yes), in operation S320, the DMA engine 332 may send the data requested by the read command to the FPGA 320 based on the command in operation S310 (②). When the switch delay elapses after the data is sent from the storage device 330 to the FPGA 320, the FPGA 320 may store the data in the data buffer 321. In operation S325, the accelerator 322 may perform calculations on the data stored in the data buffer 321. For example, the host 310 may request in advance the type of calculation to be performed by the accelerator 322 of the FPGA 320. For another example, the type of calculation to be performed by the accelerator 322 may be predetermined and fixed.

[0053] When the command of the submission queue 311 is not a read command (No), in operation S330, the controller 331 may process the command. For example, when the command of the submission queue 311 is a write command, the DMA engine 332 may receive data stored in the data buffer 313 of the host 310 through the FPGA 320, and the controller 331 may store the data in the nonvolatile memory 339. As in the above description, the controller 331 may perform an operation corresponding to the command in operation S310.

[0054] In operation S320 or operation S330, the controller 331 may process the command communicated in operation S310. In operation S335, the DMA engine 332 may send completion information about the command to the FPGA 320 based on the command received in operation S310 (②). When the switch delay elapses after the completion information is sent from the storage device 330 to the FPGA 320, the FPGA 320 may store the completion information in the completion queue buffer 323 in operation S340. Unlike the above-mentioned computing system 200, the FPGA 320 of the computing system 300 may include or support the completion queue buffer 323. Therefore, after the command in operation S310 is completely processed in operation S320 or operation S330, the storage device 330 may send the completion information to the FPGA 320 (②) before the accelerator 322 completes the calculation in operation S335. For example, when the command in operation S310 is a read command, the DMA engine 332 may send data to the FPGA 320 in operation S320 (②), and then may send completion information to the FPGA 320 in operation S335 (②). The DMA engine 332 may send completion information to the FPGA 320 in operation S335 only according to whether the command in operation S310 is processed, regardless of the operation of the FPGA 320 (e.g., the calculation performed by the accelerator 322). Therefore, compared with the storage device 230, the storage device 330 may send the completion information in advance in operation S335 (②). Unlike the storage device 230, the storage device 330 may send the completion information to the FPGA 320 in operation S335 (②) without receiving a signal indicating that the calculation performed by the accelerator 322 is completed.

[0055] When it is determined in operation S345 that the calculation performed by the accelerator 322 is not completed (No), the FPGA 320 may wait until the calculation performed by the accelerator 322 is completed. When it is determined in operation S345 that the calculation performed by the accelerator 322 is completed (Yes), in operation S350, the FPGA 320 may send data to the host 310 (③). When the switch delay elapses after the data is sent from the FPGA 320 to the host 310, the host 310 may store the data in the data buffer 313. Here, the data may be calculated data generated by the accelerator 322 or read data read from the storage device 330. For another example, the FPGA 320 may send both the calculated data generated by the accelerator 322 and the read data read from the storage device 330 to the host 310.

[0056] The accelerator 322 may complete the calculation and may control the completion queue buffer 323 (④). The accelerator 322 may access the completion queue buffer 323 with reference to the address of the completion queue buffer 323 placed on the on-chip memory and the identifier ID of the command. The completion information sent from the storage device 330 may include the identifier of the command. For example, when the command is submitted to the submission queue 311, the identifier of the command may be assigned by the host 310. The FPGA 320 may not send a signal indicating that the calculation of the accelerator 322 is completed to the storage device 330. In operation S355, the FPGA 320 may send the completion information stored in the completion queue buffer 323 to the host 310 (⑤). For example, when the calculation performed by the accelerator 322 is completed, the FPGA 320 may send the completion information stored in the completion queue buffer 323 to the host 310. When the switch delay elapses after the completion information is sent from the FPGA 320 to the host 310, the host 310 may receive the completion information and may store the completion information in the completion queue 312.

[0057] In an exemplary embodiment, the DMA engine 332 may perform operation S320 or operation S330, and then may perform operation S335. As in the above description, the FPGA 320 may perform operation S350, and then may perform operation S355. The order of performing operation S325, operation S335, and operation S340 is not limited to Figure 7 For example, while the accelerator 322 performs calculations on the data, the DMA engine 332 may send completion information about the command to the FPGA 320. While the accelerator 322 performs calculations on the data, the completion queue buffer 323 may store completion information about the command.

[0058] In an exemplary embodiment, components 321 to 323 of FPGA 320 may be implemented in hardware, software, or a combination thereof. In the case of using a hardware approach, components 321 to 323 may be implemented differently using registers, latches, flip-flops, logic circuits, logic gates, etc. Data buffer 321 and completion queue buffer 323 may correspond to an area allocated on an on-chip memory included in FPGA 320. In an exemplary embodiment, controller 331 may be implemented in hardware, software, or a combination thereof.

[0059] Figure 8 Show Figure 1 Block diagram of a computing system. Fig. 9 Show Figure 8 The operation timing of the computing system. Fig.10 Show Figure 8 A method of operating a computing system.

[0060] The computing system 400 may include a host 410, an FPGA 420, and a storage device 430. The host 410 may include a submission queue 411, a completion queue 412, and a data buffer 413. The components 411 to 413 of the host 410 may be similar to the components 311 to 313 of the host 310, respectively. The FPGA 420 may include a data buffer 421, an accelerator 422, and a completion queue buffer 423. The components 421 to 423 of the FPGA 420 may be similar to the components 321 to 323 of the FPGA 320, respectively. The storage device 430 may include a controller 431 and a nonvolatile memory 439. The controller 431 may include a DMA engine 432. The components 431, 432, and 439 of the storage device 430 may be similar to the components 331, 332, and 339 of the storage device 330, respectively. The description will focus on the differences between the computing system 400 and the computing system 300.

[0061] The host 410 may write or store a command associated with the storage device 430 in the submission queue 411. The host 410 may send a doorbell associated with the submission queue 411 to the FPGA 420, and the FPGA 420 may receive the doorbell. The FPGA 420 may send the doorbell to the storage device 430. Based on the doorbell, the storage device 430 may request a command of the submission queue 411 of the host 410 and may send the request to the FPGA 420. The FPGA 420 may send the request to the host 410. For example, the storage device 430 may access the submission queue 411 indicated by the address of the doorbell among one or more submission queues of the host 410.

[0062] In operation S405, the host 410 may receive the request and may send the command of the submission queue 411 to the FPGA 420 (①). When the switch delay elapses after the command of the submission queue 411 is sent from the host 410 to the FPGA 420, the FPGA 420 may receive the command of the submission queue 411 from the host 410 (①). In operation S410, the FPGA 420 may send the command of the submission queue 411 to the storage device 430. When the switch delay elapses after the command of the submission queue 411 is sent from the FPGA 420 to the storage device 430, the DMA engine 432 of the storage device 430 may receive the command of the submission queue 411 (①). In operation S415, the controller 431 may determine whether the command of the submission queue 411 is a write command.

[0063] When the command of the submission queue 411 is a write command (Yes), in operation S420, the DMA engine 432 may send a data request to the FPGA 420 based on the write command (②). When the switch delay elapses after the data request is sent from the storage device 430 to the FPGA 420, the FPGA 420 may receive the data request (②). In operation S425, the FPGA 420 may send the data request to the host 410 (②). When the switch delay elapses after the data request is sent from the FPGA 420 to the host 410, the host 410 may receive the data request (②). In operation S430, the host 410 may send the write data of the data buffer 413 to the FPGA 420 based on the data request (③). When the switch delay elapses after the write data is sent from the host 410 to the FPGA 420, the FPGA 420 may receive the write data and may store the write data in the data buffer 421.

[0064] In operation S435, the accelerator 422 may perform calculations on the write data stored in the data buffer 421. For example, the host 410 may pre-request the type of calculation to be performed by the accelerator 422 to the FPGA 420. As another example, the type of calculation to be performed by the accelerator 422 may be predetermined and fixed. The accelerator 422 may complete the calculation and may control the completion queue buffer 423. The accelerator 422 may access the completion queue buffer 423 with reference to the address of the completion queue buffer 423 placed on the on-chip memory and the identifier ID of the command. The completion information sent from the storage device 430 may include the identifier of the command. The identifier of the command may be assigned by the host 410 when the command is submitted to the submission queue 411. When it is determined in operation S440 that the calculation performed by the accelerator 422 is not completed (No), the FPGA 420 may wait until the calculation performed by the accelerator 422 is completed. When it is determined in operation S440 that the calculation performed by the accelerator 422 is completed (Yes), in operation S445, the FPGA 420 may transmit one of the calculated data and the write data generated by the accelerator 422 to the storage device 430 (④). When the switch delay elapses after one of the calculated data and the write data generated by the accelerator 422 is transmitted from the FPGA 420 to the storage device 430, the DMA engine 432 may receive one of the calculated data and the write data (④).

[0065] When the command of the submission queue 411 is not a write command (No), in operation S450, the controller 431 may process the command without operations S420 to S445. The controller 431 may perform an operation corresponding to the command received in operation S410. As described above, when the command of the submission queue 411 is a write command, the DMA engine 432 may receive the data stored in the data buffer 413 of the host 410 through the FPGA 420, and the controller 431 may store the data in a storage space (e.g., on-chip memory 631_2 (refer to Fig.12 ), memory buffer 638 (reference Fig.12 ) or in non-volatile memory 439).

[0066] In operation S450, the controller 431 may process the command received in operation S410. In operation S455, the DMA engine 432 may send completion information about the command to the FPGA 420 (⑤). When the switch delay elapses after the completion information is sent from the storage device 430 to the FPGA 420, the FPGA 420 may receive the completion information and may store the completion information in the completion queue buffer 423 (⑤). In operation S460, the FPGA 420 may send the completion information stored in the completion queue buffer 423 to the host 410. When the switch delay elapses after the completion information is sent from the FPGA 420 to the host 410, the host 410 may receive the completion information and may store the completion information in the completion queue 412 (⑤).

[0067] In an exemplary embodiment, components 421 to 423 of FPGA 420 may be implemented in hardware, software, or a combination thereof. In the case of using hardware, components 421 to 423 may be implemented differently using registers, latches, flip-flops, logic circuits, logic gates, etc. Data buffer 421 and completion queue buffer 423 may correspond to an area allocated on an on-chip memory included in FPGA 420. In an exemplary embodiment, controller 431 may be implemented in hardware, software, or a combination thereof.

[0068] Fig.11 3 to Fig.10 FPGA 520 may be one of FPGAs 220 to 420. FPGA 520 may include on-chip memory 521, accelerator 522, and interface circuit 524.

[0069] The on-chip memory 521 may include latches, registers, SRAM, DRAM, TRAM, tightly coupled memory (TCM), etc. The on-chip memory 521 may include the above-mentioned data buffer 221, 321, or 421 and the completion queue buffer 323 or 423. Areas of the data buffer 221, 321, or 421 and the completion queue buffer 323 or 423 may be allocated to the on-chip memory 521.

[0070] The accelerator 522 may perform calculations on data stored in a data buffer of the on-chip memory 521. The accelerator 522 may correspond to the above-described accelerator 222, 322, or 422. The accelerator 522 may include an arithmetic logic unit (ALU), a floating point unit (FPU), registers, latches, flip-flops, logic circuits, logic gates, etc., and may perform various calculation operations.

[0071] The interface circuit 524 can communicate with the outside (eg, the host 11, 210, 310, or 410, the root complex 120, the electronic devices 141, 151, 153, 161, and 162, and the storage device 230, 330, or 430: see Figures 2 to 10 ) communication. For example, the interface circuit 524 can communicate with the reference Figure 2 The downstream port DP and the upstream port UP described are connected. For example, the interface protocol may be USB, SCSI, PCIe, M-PCIe, NVMe, ATA, PATA, SATA, SAS, IDE, UFS, Firewire, etc. The interface circuit 524 may receive data from the host 210, 310, or 410 and may send the received data or the calculated data generated by the accelerator 522 to the storage device 230, 330, or 430. The interface circuit 524 may receive data from the storage device 230, 330, or 430 and may send the received data and one of the calculated data generated by the accelerator 522 to the host 210, 310, or 410. In detail, the interface circuit 524 may receive data from the DMA engine 332, and then may receive completion information from the DMA engine 432, and may provide the data and the completion information to the data buffer and the completion queue buffer of the on-chip memory 521, respectively. For example, the interface circuit 524 may send one of the data generated by the accelerator 522 and the data received from the DMA engine 432 to the host 310, and then may send the completion information stored in the completion queue buffer of the on-chip memory 521 to the host 410 (refer to Figure 7 In detail, when the calculation performed by the accelerator 522 is completed, the interface circuit 524 may send the completion information stored in the completion queue buffer of the on-chip memory 521 to the host 310.

[0072] Fig.123 to Fig.10 The storage device 630 may be one of the storage devices 230, 330, or 430. The storage device 630 may include a controller 631, a memory buffer 638, and a non-volatile memory 639.

[0073] The controller 631 may be implemented using SoC, ASIC, FPGA, etc. The controller 631 may include a processor 631_1, an on-chip memory 631_2, a non-volatile memory interface circuit 631_3, an external interface circuit 631_4, a DMA engine 631_5, and a buffer interface circuit 631_6. The processor 631_1 may control components 631_2 to 631_6 of the controller 631. The processor 631_1 may include at least one or more cores (e.g., homogeneous multi-cores or heterogeneous multi-cores) and an internal cache memory. The processor 631_1 may execute program codes, software, applications, etc. loaded onto the on-chip memory 631_2 or the memory buffer 638.

[0074] The on-chip memory 631_2 may include latches, registers, SRAM, DRAM, TRAM, TCM, etc. Firmware or software such as a host interface layer (HIL), a flash translation layer (FTL), and a flash interface layer (FIL) may be loaded onto the on-chip memory 631_2. The HIL may manage input / output (I / O) commands from the outside. The FTL may manage the mapping between the logical address provided from the outside and the physical address of the non-volatile memory 639. The FTL may also perform garbage collection, wear leveling, etc., as well as the above-mentioned address mapping. The FTL may manage the read and write operations of the non-volatile memory 639.

[0075] The nonvolatile memory interface circuit 631_3 may communicate with the nonvolatile memory 639 in accordance with an interface protocol such as a switching double data rate (DDR). The nonvolatile memory interface circuit 631_3 may communicate with one or more nonvolatile memories 639 through a channel CH1, and may communicate with one or more nonvolatile memories 639 through a channel CHn (n is an integer of 2 or more). The number of channels CH1 to CHn between the controller 631 and the nonvolatile memory 639 may be one or more, the number of nonvolatile memories allocated to one channel may be one or more, and each nonvolatile memory 639 may be a plurality of channels CH1 to CHn. Fig.10 Under the control of the processor 631_1, the non-volatile memory interface circuit 631_3 can transmit data from the outside (for example, the host 11, the root complex 120, the electronic devices 141, 151, 153, 161 and 162, and the FPGA 220, 320 or 420: see Figures 2 to 10) is sent to the nonvolatile memory 639, and the write data may be stored in the nonvolatile memory 639. Under the control of the processor 631_1, the nonvolatile memory interface circuit 631_3 may receive read data sent from the nonvolatile memory 639.

[0076] The external interface circuit 631_4 can communicate with the outside (eg, the host 11, the root complex 120, the electronic devices 141, 151, 153, 161, and 162, and the FPGA 220, 320, or 420: see Figures 2 to 10 ) communication. For example, the interface protocol may be USB, SCSI, PCIe, M-PCIe, NVMe, ATA, PATA, SATA, SAS, IDE, UFS, Firewire, etc. For example, the external interface circuit 631_4 may communicate with the reference Figure 2 Describes the endpoint port EP is connected to.

[0077] Under the control of the processor 631_1, the DMA engine 631_5 can directly access the outside (eg, the host 11, the root complex 120, the electronic devices 141, 151, 153, 161 and 162, and the FPGA 220, 320 or 420: see Figures 2 to 10 ) various memory buffers 130, HMB, 211 to 213, 311 to 313, 321, 323, 411 to 413, 421, and 423. For example, the DMA engine 631_5 can access one of the above memory buffers, can receive a command, can receive write data, and can send read data of the storage device 630. The DMA engine 631_5 can access the on-chip memory 631_2 and the memory buffer 638 of the storage device 630 and can exchange data therewith.

[0078] The buffer interface circuit 631_6 may communicate with the memory buffer 638 in accordance with an interface protocol such as the DDR standard. Under the control of the processor 631_1, the buffer interface circuit 631_6 may exchange data with the memory buffer 638. The memory buffer 638 may include latches, registers, SRAM, DRAM, TRAM, TCM, etc. For example, the memory buffer 638 may be placed outside the controller 631, or may be placed inside the storage device 630. As another example, the memory buffer 638 may not be included in the storage device 630. In the case where the memory buffer 638 is included in the storage device 630, the processor 631_1 may use the memory buffer 638 and the on-chip memory 631_2 as a cache memory. In an exemplary embodiment, the controller 431 may perform operations with reference to Figure 7 and Fig.10The described storage devices 330 and 430 are associated with operations S315, S320, S330, S335, S415, S420, S450, and S455.

[0079] Fig.13 A block diagram of a computing device according to an application example of the present disclosure is shown. Figures 1 to 10 In the described computing systems 10 and 100 to 400, various embodiments of the present disclosure may be applied to a computing device 1000. The computing device 1000 may include a main processor 1100, a memory 1200, a user interface 1300, a storage device 1400, a communication block 1500, and a graphics processor 1600. For example, the computing device 1000 may also be referred to as a "mobile device".

[0080] The main processor 1100 may control the overall operation of the computing device 1000. The main processor 1100 may be configured to process various types of arithmetic operations or logic operations. The main processor 1100 may be implemented using, but not limited to, a dedicated logic circuit, FPGA, ASIC, or SoC including one or more processor cores. The main processor 1100 may be implemented using a general-purpose processor, a dedicated processor, or an application processor. For example, each of the hosts 11 and 210 to 410 and the processor 110 may correspond to the main processor 1100.

[0081] The memory 1200 may temporarily store data used for the operation of the computing device 1000. The memory 1200 may store data processed by or to be processed by the main processor 1100. For example, the memory 1200 may correspond to the above-mentioned memory 130.

[0082] The user interface 1300 may perform communication mediation between the user and the computing device 1000 under the control of the main processor 1100. For example, the user interface 1300 may process input from a keyboard, a mouse, a keypad, a button, a touch panel, a touch screen, a touch pad, a touch ball, a camera, a gyro sensor, a vibration sensor, etc. In addition, the user interface 1300 may process output to be provided to a display device, a speaker, a motor, etc.

[0083] The storage device 1400 may include a storage medium capable of storing data regardless of whether it is powered. Figures 1 to 10 One of the described electronic devices 12, 13, 141, 142, 151 to 154 and 161 to 163, FPGAs 220 to 420 or storage devices 230 to 430. The storage device 1400 may be an intermediate device, and another intermediate device and another endpoint device connected to the storage device 1400 may be further included in the computing device 1000.

[0084] The communication block 1500 may communicate with an external device / system of the computing device 1000 under the control of the main processor 1100. For example, the communication block 1500 may communicate with an external device / system of the computing device 1000 based on at least one of various wired communication protocols such as Ethernet, Transmission Control Protocol / Internet Protocol (TCP / IP), Universal Serial Bus (USB), and Firewire, and / or at least one of various wireless communication protocols such as Long Term Evolution (LTE), Worldwide Interoperability for Microwave Access (WiMax), Global System for Mobile Communications (GSM), Code Division Multiple Access (CDMA), Bluetooth, Near Field Communication (NFC), Wireless Fidelity (Wi-Fi), and Radio Frequency Identification (RFID).

[0085] The graphics processor 1600 may include a plurality of processor cores (eg, graphics processing units). The processor cores included in the graphics processor 1600 may process graphics data in parallel quickly. For example, the graphics processor 1600 may process various graphics operations such as pixel shaders, super sampling, and color space conversion by using the processor cores.

[0086] Each of the main processor 1100, the memory 1200, the user interface 1300, the storage device 1400, the communication block 1500, and the graphic processor 1600 may be implemented using a circuit-level, chip-level, and / or package-level device to be installed in the computing device 1000. Alternatively, each of the main processor 1100, the memory 1200, the user interface 1300, the storage device 1400, the communication block 1500, and the graphic processor 1600 may be implemented using an independent electronic device to be assembled in the computing device 1000. The installed or assembled components may be connected to each other via the bus 1700.

[0087] The bus 1700 may provide a communication path between components of the computing device 1000. The components of the computing device 1000 may exchange data with each other based on a bus format of the bus 1700. For example, the bus format may include one or more of various protocols such as PCIe, NVMe, SCSI, ATA, PATA, SATA, SAS, and UFS.

[0088] According to an exemplary embodiment of the present disclosure, since the FPGA includes a completion queue buffer as a copy of the completion queue of the host, the storage device can send completion information in advance compared to the case where the FPGA does not include the completion queue buffer. Therefore, the delay required to complete the command issued by the host can be improved.

[0089] As is conventional in the art, embodiments may be described and illustrated by blocks that perform the described functions or multiple functions. These blocks (herein referred to as units or modules, etc.) are physically implemented by analog and / or digital circuits such as logic gates, integrated circuits, microprocessors, microcontrollers, memory circuits, passive electronic components, active electronic components, optical components, hard-wired circuits, etc., and may optionally be driven by firmware and / or software. For example, the circuit may be embodied in one or more semiconductor chips or on substrate supports such as printed circuit boards, etc. The circuits constituting the blocks may be implemented by dedicated hardware or by a processor (e.g., one or more programmed microprocessors and associated circuits) or by a combination of dedicated hardware that performs some functions of the blocks and processors that perform other functions of the blocks. Without departing from the scope of the present disclosure, the various blocks of the embodiments may be physically separated into two or more interacting and discrete blocks. Similarly, without departing from the scope of the present disclosure, the blocks of the embodiments may be physically combined into more complex blocks. One aspect of the embodiments may be implemented by instructions stored in a non-transitory storage medium and executed by a processor.

[0090] While the present disclosure has been described with reference to exemplary embodiments thereof, it will be apparent to those skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope of the disclosure as set forth in the following claims.

Claims

1. A computing system, include: Host; A first electronic device including a memory and an accelerator; as well as a second electronic device comprising a direct memory access engine, wherein based on a command sent from the host through the first electronic device, the direct memory access engine sends data and completion information of the command to the first electronic device, wherein: The memory includes a data buffer for storing the data and a completion queue buffer for storing the completion information received from the second electronic device, The accelerator performs computations on the data stored in the memory, and The direct memory access engine sends the data to the first electronic device and then sends the completion information to the first electronic device instead of to the host.

2. The computing system according to claim 1, in: The first electronic device is a field programmable gate array, and The second electronic device is a high-speed non-volatile memory device.

3. The computing system according to claim 1, in, The direct memory access engine sends the completion information to the first electronic device before the accelerator completely performs the calculation.

4. The computing system according to claim 1, in, The first electronic device further includes an interface circuit configured to transmit one of the data calculated and the data generated by the accelerator to the host, and then transmit the completion information stored in the completion queue buffer to the host.

5. The computing system according to claim 4, in, The interface circuit receives the data from the second electronic device, then receives the completion information from the second electronic device, and provides the data and the completion information to the data buffer and the completion queue buffer, respectively.

6. The computing system according to claim 1, in, The accelerator accesses the completion queue buffer with reference to an address of the completion queue buffer placed on the memory and an identifier of the command.

7. The computing system according to claim 6, in, The completion information includes the identifier of the command.

8. The computing system according to claim 1, in, The completion queue buffer stores the completion information about the command while the accelerator performs computations on the data.

9. A computing system, include: Host; a first electronic device configured to communicate with the host; as well as a second electronic device configured to communicate with the host through the first electronic device, send data to the first electronic device, and send completion information to the first electronic device instead of to the host, wherein: The first electronic device comprises: a memory including a data buffer for receiving the data from the second electronic device and a completion queue buffer for receiving completion information about a command from the second electronic device, wherein the second electronic device receives the command from the host through the first electronic device; an accelerator configured to perform computations on the data stored in the memory; and An interface circuit is configured to send the completion information stored in the completion queue buffer to the host when the calculation is completely performed by the accelerator.

10. The computing system according to claim 9, in, The second electronic device includes a direct memory access engine configured to obtain the command and send the data and the completion information to the first electronic device based on the command.

11. The computing system according to claim 10, in, The accelerator accesses the completion queue buffer with reference to an address of the completion queue buffer placed on the memory and an identifier of the command.

12. The computing system according to claim 11, in, The completion information includes the identifier of the command.

13. The computing system according to claim 10, in, The direct memory access engine sends the data to the first electronic device and then sends the completion information to the first electronic device.

14. The computing system according to claim 10, in, The direct memory access engine sends the completion information to the first electronic device before the computation is completely performed by the accelerator.

15. The computing system of claim 9, in, The interface circuit transmits one of the data and the data calculated by the accelerator to the host, and then transmits the completion information stored in the completion queue buffer to the host.

16. A computing system, include: a host configured to manage a submission queue of a host memory buffer having commands written therein and a completion queue of the host memory buffer having completion information about the commands written therein; A first electronic device comprising a memory and an accelerator, wherein: The memory includes a data buffer storing data requested by the command and a completion queue buffer storing the completion information received from the second electronic device, and The accelerator performs calculations on the data stored in the memory; as well as The second electronic device is configured to obtain the command through the first electronic device and send the data and the completion information to the first electronic device instead of to the host based on the command.

17. The computing system of claim 16, in, When the calculation is completely performed by the accelerator, the first electronic device sends the completion information stored in the completion queue buffer to the host.

18. The computing system of claim 17, in, The first electronic device transmits one of the calculated data and the data generated by the accelerator to the host, and then transmits the completion information stored in the completion queue buffer to the host.

19. The computing system of claim 16, in, The second electronic device transmits the data to the first electronic device, and then transmits the completion information to the first electronic device.

20. The computing system of claim 19, in, Before the calculation is completely performed by the accelerator, the second electronic device sends the completion information to the first electronic device.