Multipath data access method, accelerator card, storage system, and readable storage medium
By introducing multi-path data access method and acceleration card in the SSD storage system, using soft-core processors and IP cores for data processing and transmission, the problems of poor performance and low efficiency caused by single-path storage are solved, and the more Efficient SSD access and system stability.
Patent Information
- Application Number
- CN202411110561.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-14
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2044-08-14
AI Technical Summary
The existing SSD storage system adopts a single-path storage method, resulting in poor performance and low access efficiency, and cannot effectively improve the access performance of solid-state drives.
Using multi-path data access method, the accelerator card is configured with soft core processor, on-chip memory, PCIE communication IP core and remote data access unit to realize local data access and one-way or multiple remote data access, and enhance the solid-state hard disk Access the host number and handle the IP core and the Ethernet protocol offload the IP core for data processing and transmission through the NVMEOF protocol.
It improves the access performance of the solid-state drive and ensures that the system does not cause the entire storage system to be paralyzed when any storage path fails, greatly improving security and stability.
Smart Images

Figure CN118646699B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of data processing technology, and in particular to a multi-path data access method, an acceleration card, a storage system, and a readable storage medium. Background Art
[0002] At present, the mainstream storage method has shifted from HDD (mechanical hard disk) to SSD (solid state drive). Compared with HDD, SSD has the following advantages: faster reading and writing speed, lower power consumption, strong shock and drop resistance, no noise, small appearance, easy to carry and install, etc. With the rapid development of SSD technology, the performance of SSD has soared, the bandwidth of the underlying flash memory has become wider and wider, and the media access delay has become lower and lower.
[0003] However, the current SSD storage system adopts a single-path storage method, which results in poor SSD performance and low SSD access efficiency. Summary of the invention
[0004] Based on this, it is necessary to provide a multi-path data access method, an acceleration card, a storage system and a readable storage medium to address the above technical problems, which can provide local data access and one or more remote data access for the solid-state drive, thereby expanding the number of hosts accessing the solid-state drive and improving the access performance of the solid-state drive.
[0005] A multi-path data access method is applied to an accelerator card. The accelerator card is configured with a soft-core processor, an on-chip memory, a PCIE communication IP core and a remote data access unit. The remote data access unit includes a physical layer IP core, an Ethernet protocol unloading IP core and an NVMEOF protocol processing IP core. A multi-path data access method includes: when the soft-core processor issues a local data processing instruction for a solid-state hard disk of the NVMEOF protocol, first command data is written into the on-chip memory and a doorbell notification is sent to the solid-state hard disk through the PCIE communication IP core. When the solid-state hard disk receives the doorbell notification, the first command data is read from the on-chip memory and processed; when the physical layer IP core receives a remote data processing instruction sent by a server, second command data sent by the server is received, and the second command data is sent to the server. To the Ethernet protocol unloading IP core, the Ethernet protocol unloading IP core deletes the protocol header in the second command data and then forwards it to the on-chip memory. If the second command data is the second management command data, the Ethernet protocol unloading IP core notifies the soft core processor in an interrupt manner, so that the soft core processor reads the second management command data from the on-chip memory and processes it. If the second command data is the second read / write command data, the Ethernet protocol unloading IP core sends a doorbell notification to the NVMEOF protocol processing IP core. The NVMEOF protocol processing IP core reads and stores the second read / write command data from the on-chip memory and sends a doorbell notification to the solid-state drive through the PCIE communication IP core. When the solid-state drive receives the doorbell notification, it reads and processes the second read / write command data from the NVMEOF protocol processing IP core.
[0006] In one embodiment, a solid-state drive processes first command data, including: when the first command data is first management command data, the solid-state drive parses the first management command data and obtains first management information, and manages the solid-state drive based on the first management information; when the first command data is first write command data, the solid-state drive parses the first write command data and obtains first write indication information, and based on the first write indication information, reads the first target data to be written from the on-chip memory through the PCIE communication IP core and writes it to the solid-state drive; when the first command data is first read command data, the solid-state drive parses the first read command data and obtains the first read indication information, and based on the first read indication information, reads the second target data from the solid-state drive and writes it to the on-chip memory through the PCIE communication IP core, so that the soft-core processor reads the second target data from the on-chip memory.
[0007] In one embodiment, the soft-core processor processes the second management command data, including: the soft-core processor identifies the corresponding solid-state hard disk corresponding to the second management command data, parses the second management information of the second management command data, and performs management of the corresponding solid-state hard disk based on the second management information.
[0008] In one of the embodiments, a multipath data access method also includes: if the second management command data is a data type that requires data return, the NVMEOF protocol processing IP core receives the interactive data returned by the soft-core processor and sends the interactive data to the on-chip memory; the Ethernet protocol offloading IP core reads the interactive data from the on-chip memory and sends the interactive data to the server through the physical layer IP; the NVMEOF protocol processing IP core receives the read and write completion signal sent by the soft-core processor and sends the read and write completion signal to the Ethernet protocol offloading IP core, and the Ethernet protocol offloading IP core sends the read and write completion signal to the server through the physical layer IP; if the second management command data is a data type that does not require data return, the NVMEOF protocol processing IP core receives the read and write completion signal sent by the soft-core processor and sends the read and write completion signal to the Ethernet protocol offloading IP core, and the Ethernet protocol offloading IP core sends the read and write completion signal to the server through the physical layer IP.
[0009] In one of the embodiments, the remote data access unit also includes a cache area, the NVMEOF protocol processing IP core reads and stores the second read / write command data from the on-chip memory and sends a doorbell notification to the solid-state hard disk through the PCIE communication IP core, including: if the second read / write command data is the second write command data, the NVMEOF protocol processing IP core reads the second write command data with the protocol header deleted from the on-chip memory and converts the second write command data with the protocol header deleted into a third read data command and sends it to the cache area; after receiving the third read data command, the cache area sends the third read data command to the Ethernet protocol unloading IP core and sends a doorbell signal, and the Ethernet After receiving the doorbell signal, the Ethernet protocol offloading IP core reads the third read data command from the cache area and sends a read request to the server; after receiving the response packet of the read request and the third target data fed back by the server, the Ethernet protocol offloading IP core converts the response packet of the read request into a response packet of the write request and sends it to the IP core for NVMEOF protocol processing, and stores the third target data in the IP core for NVMEOF protocol processing; the IP core for NVMEOF protocol processing sends a doorbell notification to the solid-state hard disk through the PCIE communication IP core, so that the solid-state hard disk reads and writes the third target data from the IP core for NVMEOF protocol processing.
[0010] In one of the embodiments, the NVMEOF protocol processing IP core reads and stores the second read / write command data from the on-chip memory and sends a doorbell notification to the solid-state hard disk through the PCIE communication IP core, and also includes: if the second read / write command data is the second read command data, the NVMEOF protocol processing IP core reads the second read command data of the protocol header deleted from the on-chip memory and converts the second read command data of the protocol header deleted from the on-chip memory into a third write data command and sends it to the cache area; the NVMEOF protocol processing IP core sends a doorbell notification to the solid-state hard disk and reads the second read command data of the protocol header deleted from the on-chip memory to the solid-state hard disk, so that the solid-state hard disk sends the fourth target data in the solid-state hard disk to the NVMEOF protocol processing IP core; the cache area sends the doorbell notification to the Ethernet protocol unloading IP core, the Ethernet protocol unloading IP core reads the third write data command from the cache area and reads the fourth target data from the NVMEOF protocol processing IP core based on the third write data command, and sends the fourth target data to the server through the physical layer IP.
[0011] In one of the embodiments, a filter and a direct memory access controller are also configured in the remote data access unit; before the step in which the physical layer IP core receives the remote data processing instruction sent by the server, it also includes: the physical layer IP core receives a ping data packet sent by the server and sends the ping data packet to the filter; the filter sends the ping data packet to the on-chip memory through the direct memory access controller; after the direct memory access controller generates an interrupt and reports it to the soft-core processor, the soft-core processor reads the ping data packet from the on-chip memory, swaps the source address and the destination address in the ping data packet and encapsulates it to obtain an encapsulated data packet, and sends the encapsulated data packet to the server through the direct memory access controller and the physical layer IP.
[0012] An accelerator card is configured with a soft-core processor, an on-chip memory, a PCIE communication IP core and a remote data access unit, wherein the remote data access unit comprises a physical layer IP core, an Ethernet protocol unloading IP core and an IP core for NVMEOF protocol processing; when the soft-core processor issues a local data processing instruction for a solid-state hard disk of the NVMEOF protocol, first command data is written into the on-chip memory and a doorbell notification is sent to the solid-state hard disk through the PCIE communication IP core, and when the solid-state hard disk receives the doorbell notification, the first command data is read from the on-chip memory and processed; when the physical layer IP core receives a remote data processing instruction sent by a server, second command data is received from the server, and the second command data is sent to the Ethernet protocol unloading IP core; The Ethernet protocol unloading IP core deletes the protocol header in the second command data and then forwards it to the on-chip memory. If the second command data is the second management command data, the Ethernet protocol unloading IP core notifies the soft-core processor in an interrupt manner, so that the soft-core processor reads the second management command data from the on-chip memory and processes it. If the second command data is the second read / write command data, the Ethernet protocol unloading IP core sends a doorbell notification to the NVMEOF protocol processing IP core. The NVMEOF protocol processing IP core reads and stores the second read / write command data from the on-chip memory and sends a doorbell notification to the solid-state drive through the PCIE communication IP core. When the solid-state drive receives the doorbell notification, it reads and processes the second read / write command data from the NVMEOF protocol processing IP core.
[0013] A storage system includes the acceleration card of any of the above embodiments and one or more solid-state hard disks.
[0014] A readable storage medium stores an embedded program, which implements the steps of any method in the above embodiments when executed by a processor.
[0015] The above-mentioned multi-path data access method, acceleration card, storage system and readable storage medium are applied to the acceleration card, which is configured with a soft-core processor, on-chip memory, PCIE communication IP core and remote data access unit, and the remote data access unit includes a physical layer IP core, an Ethernet protocol offloading IP core and an IP core for NVMEOF protocol processing; when the soft-core processor issues a local data processing instruction for a solid-state hard disk of the NVMEOF protocol, the first command data is written into the on-chip memory and a doorbell notification is sent to the solid-state hard disk through the PCIE communication IP core. When the solid-state hard disk receives the doorbell notification, it reads and processes the first command data from the on-chip memory; when the physical layer IP core receives the remote data processing instruction sent by the server, it receives the second command data sent by the server, and sends the second command data to the Ethernet protocol offloading IP core. The Ethernet protocol offloading IP core deletes the protocol header in the second command data and then forwards it to the on-chip memory. If the second command data is a second pipe The Ethernet protocol unloading IP core notifies the soft core processor in an interrupt mode, so that the soft core processor reads the second management command data from the on-chip memory and processes it. If the second command data is the second read / write command data, the Ethernet protocol unloading IP core sends a doorbell notification to the NVMEOF protocol processing IP core, and the NVMEOF protocol processing IP core reads and stores the second read / write command data from the on-chip memory and sends a doorbell notification to the solid-state hard disk through the PCIE communication IP core. When the solid-state hard disk receives the doorbell notification, it reads and processes the second read / write command data from the NVMEOF protocol processing IP core. Therefore, it is possible to provide local data access and one or more remote data access for the solid-state hard disk, thereby expanding the number of hosts accessing the solid-state hard disk and improving the access performance of the solid-state hard disk. In addition, by using one or more remote data access, when any storage path in the system fails, it will not cause the entire storage system to be paralyzed, which greatly improves the security and stability of the entire storage system. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 A schematic diagram of an application scenario of a multi-path data access method in one embodiment;
[0017] Figure 2 A schematic flow chart of a multi-path data access method in one embodiment;
[0018] Figure 3 A schematic diagram of information interaction of management command data in local data processing performed by a soft core processor in one embodiment;
[0019] Figure 4 A schematic diagram of information interaction of write command data in local data processing performed by a soft-core processor in one embodiment;
[0020] Figure 5A schematic diagram of information interaction of read command data when a soft-core processor performs local data processing in one embodiment;
[0021] Figure 6 A schematic diagram of information interaction of management command data in remote data processing performed by an accelerator card in one embodiment;
[0022] Figure 7 A schematic diagram of information interaction of writing command data when an accelerator card performs remote data processing in one embodiment;
[0023] Figure 8 A schematic diagram of information interaction of reading command data when an accelerator card performs remote data processing in one embodiment;
[0024] Fig. 9 A schematic diagram of information interaction between an accelerator card and a server to establish a network communication connection in one embodiment;
[0025] Fig.10 A design block diagram of a dual-path NVMe storage system implemented based on FPGA in one embodiment. DETAILED DESCRIPTION
[0026] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0027] The present application provides a multi-path data access method. In one embodiment, a multi-path data access method is applied to Figure 1 In the application environment shown. Figure 1As shown, the acceleration card 10 is configured with a soft-core processor 100, an on-chip memory 101, a PCIE communication IP core 102 and a remote data access unit, and the remote data access unit includes a physical layer IP core 201, an Ethernet protocol offloading IP core 202 and an IP core 203 for NVMEOF protocol processing; when the soft-core processor 100 issues a local data processing instruction to the solid-state drive 20 of the NVMEOF protocol, the first command data is written into the on-chip memory 101 and a doorbell notification is sent to the solid-state drive 20 through the PCIE communication IP core 102. When the solid-state drive 20 receives the doorbell notification, it reads and processes the first command data from the on-chip memory 101; when the physical layer IP core 201 issues a remote data processing instruction to the server 30, it receives the second command data of the server 30 and sends the second command data to the Ethernet protocol offloading IP core 202. The Ethernet protocol offloading IP core 202 deletes the protocol header in the second command data and then forwards it to the on-chip memory 101. If the second command data is the second management command data, the Ethernet protocol offloading IP core 202 notifies the soft core processor 100 in an interrupt manner, so that the soft core processor 100 reads the second management command data from the on-chip memory 101 and processes it. If the second command data is the second read / write command data, the Ethernet protocol offloading IP core 202 sends a doorbell notification to the NVMEOF protocol processing IP core 204. The NVMEOF protocol processing IP core 204 reads and stores the second read / write command data from the on-chip memory 101 and sends a doorbell notification to the solid-state drive 20 through the PCIE communication IP core 102. When the solid-state drive 20 receives the doorbell notification, it reads and processes the second read / write command data from the NVMEOF protocol processing IP core 204.
[0028] In one embodiment, Figure 2 As shown, a multi-path data access method includes the following steps:
[0029] S202, when the soft core processor issues a local data processing instruction to the solid state drive of the NVMEOF protocol, the first command data is written into the on-chip memory and a doorbell notification is sent to the solid state drive through the PCIE communication IP core. When the solid state drive receives the doorbell notification, it reads and processes the first command data from the on-chip memory.
[0030] In this embodiment, the access to the solid state drive includes local access and remote access. In this step, the soft core processor directly accesses the solid state drive through the PCIE communication IP core. Figure 1 As shown, the soft core processor 100 writes the first command data into the on-chip memory 101 and notifies the solid state drive 20 through the PCIE communication IP core 102. When the solid state drive 20 receives the notification, it reads and processes the first command data from the on-chip memory 101 through the PCIE communication IP core 102.
[0031] Specifically, the solid state drive may be an NVMe (Non Volatile Memory Host Controller Interface Specification) solid state drive. The soft core processor may be a 32-bit soft core processor that is internally embedded with a Linux operating system and supports the NVMe Cli instruction set.
[0032] In one example, a solid-state hard disk processes first command data, including: when the first command data is first management command data, the solid-state hard disk parses the first management command data and obtains first management information, and manages the solid-state hard disk based on the first management information; when the first command data is first write command data, the solid-state hard disk parses the first write command data and obtains first write indication information, and based on the first write indication information, reads the first target data to be written from the on-chip memory through the PCIE communication IP core and writes it to the solid-state hard disk; when the first command data is first read command data, the solid-state hard disk parses the first read command data and obtains the first read indication information, and based on the first read indication information, reads the second target data from the solid-state hard disk and writes it to the on-chip memory through the PCIE communication IP core, so that the soft-core processor reads the second target data from the on-chip memory.
[0033] When the soft-core processor executes the management command data in the local data processing, the information interaction diagram corresponding to its processing process is shown in Figure 3 As shown. Specifically, the PCIE protocol can be used for data processing. Figure 3 As shown, MB refers to the soft-core processor, HBM1 refers to the on-chip memory, XDMA refers to the PCIE communication IP core, and SSD is the NVMe solid-state drive.
[0034] Specifically, a queue pair carrying NVMe commands is set, namely, an SQ queue and a CQ queue. The SQ queue indicates the queue where commands are submitted, and the CQ queue indicates the queue where commands are completed. NVMe commands include SQ commands and CQ commands. SQ commands refer to commands written into the SQ queue, and CQ commands refer to commands written into the CQ queue. See Figure 3 As shown in the figure, the processing flow of management command data in local data processing is as follows: MB writes the SQ command to HBM1, and then sends a doorbell to notify the SSD. After receiving the doorbell, the SSD reads the SQ command from HBM1 and executes the corresponding management command in the SQ command. After executing the corresponding management command, the SSD sends the completion CQ command to HBM1, and the SSD generates a PCIe interrupt and sends it to the MB. When the MB receives the PCIe interrupt, it determines that the corresponding management command processing is completed, and sends a CQ doorbell to the SSD to notify the SSD that it has received the information that the management command processing is completed. At the same time, the MB clears the PCIe interrupt so that the SSD can continue to apply for the PCIe interrupt next time.
[0035] When the soft-core processor executes the write command data in the local data processing, the information interaction diagram corresponding to its processing process is shown in Figure 4 When the soft-core processor executes the read command data in the local data processing, the information interaction diagram corresponding to its processing process is shown in Figure 5 shown.
[0036] Specifically, Figure 4 As shown, the MB writes the SQ command to HBM1, and then sends the SQ doorbell to notify the SSD. After receiving the SQ doorbell, the SSD reads the SQ command from HBM1, executes the corresponding write command in the SQ command, reads 4KB of data from HBM1 and writes it to the SSD. After executing the corresponding write command, the SSD sends a completed CQ command to the CQ queue of HBM1, and the SSD generates a PCIe interrupt and sends it to the MB. When the MB receives the PCIe interrupt, it determines that the corresponding write command has been processed, and sends a CQ doorbell to the SSD to notify the SSD that it has received the information that the write command has been processed. At the same time, the MB clears the PCIe interrupt so that the SSD can continue to apply for the PCIe interrupt next time.
[0037] Specifically, Figure 5 As shown, the MB writes the SQ command to HBM1, and then sends the SQ doorbell to notify the SSD. After receiving the SQ doorbell, the SSD reads the SQ command from HBM1, executes the corresponding read command in the SQ command, and writes the 4KB data in the SSD to HBM1. After executing the corresponding read command, the SSD sends the completion CQ command to the CQ queue of HBM1, and the SSD generates a PCIe interrupt and sends it to the MB. When the MB receives the PCIe interrupt, it determines that the corresponding read command processing is completed, and sends a CQ doorbell to the SSD to notify the SSD that it has received the information that the write command processing is completed. At the same time, the MB clears the PCIe interrupt so that the SSD can continue to apply for the PCIe interrupt next time.
[0038] S204, when the physical layer IP core receives the remote data processing instruction sent by the server, it receives the second command data sent by the server, and sends the second command data to the Ethernet protocol offloading IP core. The Ethernet protocol offloading IP core deletes the protocol header in the second command data and then forwards it to the on-chip memory. If the second command data is the second management command data, the Ethernet protocol offloading IP core notifies the soft core processor in an interrupt manner, so that the soft core processor reads the second management command data from the on-chip memory and processes it. If the second command data is the second read / write command data, the Ethernet protocol offloading IP core sends a doorbell notification to the NVMEOF protocol processing IP core. The NVMEOF protocol processing IP core reads and stores the second read / write command data from the on-chip memory and sends a doorbell notification to the solid-state drive through the PCIE communication IP core. When the solid-state drive receives the doorbell notification, it reads and processes the second read / write command data from the NVMEOF protocol processing IP core.
[0039] In this embodiment, a remote data access unit is configured in the acceleration card 10. The remote data access unit can be multiple, and the devices configured inside each remote data access unit and the circuits formed by each device are the same. Each remote data access unit 200 is used to implement a remote access mode. Here, the remote data access unit 200 is taken as an example for detailed description:
[0040] like Figure 1 As shown, in the remote data access unit 200, the physical layer IP core 201 receives the second command data sent by the server 30, and sends the second command data to the Ethernet protocol unloading IP core 202. Among them, the Ethernet protocol unloading IP core 202 is an RDMA (Remote Direct Memory Access) protocol unloading IP core. Specifically, the second command data can be transferred to the Ethernet protocol unloading IP core 202 after passing through the remote direct data access filter 205. The Ethernet protocol unloading IP core 202 deletes the protocol header in the second command data and then forwards it to the on-chip memory 101. If the second command data is the second management command data, the Ethernet protocol unloading IP core 202 notifies the soft core processor 100 so that the soft core processor 100 reads and processes the second management command data from the on-chip memory 101. Among them, the on-chip memory 101 is divided into two left and right stacks.
[0041] If the second command data is the second read / write command data, the Ethernet protocol offloading IP core 202 sends a doorbell notification to the NVMEOF protocol processing IP core 204. NVMEOF protocol: NVMe over Fabrics, Fabrics can be understood as a high-speed network. NVMe over Fabrics: The local NVMe protocol is extended to the NVMe of the network, and the original motherboard bus PCle becomes a high-speed network such as RDMA, which extends the original NVME protocol. NVME protocol: NVM Express, non-volatile memory host controller interface specification. The NVMEOF protocol processing IP core 204 reads and stores the second read / write command data from the on-chip memory 101 and sends a doorbell notification to the solid-state drive 20 through the PCIE communication IP core 102. When the solid-state drive 20 receives the doorbell notification, it reads and processes the second read / write command data from the NVMEOF protocol processing IP core 204.
[0042] The above-mentioned multi-path data access method is applied to an accelerator card. The accelerator card is configured with a soft-core processor, an on-chip memory, a PCIE communication IP core and a remote data access unit. The remote data access unit includes a physical layer IP core, an Ethernet protocol offloading IP core and an IP core for NVMEOF protocol processing; when the soft-core processor issues a local data processing instruction for a solid-state hard disk of the NVMEOF protocol, the first command data is written into the on-chip memory and a doorbell notification is sent to the solid-state hard disk through the PCIE communication IP core. When the solid-state hard disk receives the doorbell notification, it reads and processes the first command data from the on-chip memory; when the physical layer IP core receives the remote data processing instruction sent by the server, it receives the second command data sent by the server, and sends the second command data to the Ethernet protocol offloading IP core. The Ethernet protocol offloading IP core deletes the protocol header in the second command data and then forwards it to the on-chip memory. If the second command data is a second pipe The Ethernet protocol unloading IP core notifies the soft core processor in an interrupt mode, so that the soft core processor reads the second management command data from the on-chip memory and processes it. If the second command data is the second read / write command data, the Ethernet protocol unloading IP core sends a doorbell notification to the NVMEOF protocol processing IP core, and the NVMEOF protocol processing IP core reads and stores the second read / write command data from the on-chip memory and sends a doorbell notification to the solid-state hard disk through the PCIE communication IP core. When the solid-state hard disk receives the doorbell notification, it reads and processes the second read / write command data from the NVMEOF protocol processing IP core. Therefore, it is possible to provide local data access and one or more remote data access for the solid-state hard disk, thereby expanding the number of hosts accessing the solid-state hard disk and improving the access performance of the solid-state hard disk. In addition, by using one or more remote data access, when any storage path in the system fails, it will not cause the entire storage system to be paralyzed, which greatly improves the security and stability of the entire storage system.
[0043] In one example, after the step in which the soft-core processor feeds back a message indicating that the second command data has been completed to the network interface card, the step also includes: after the network interface card receives a response signal from the server based on the process data feedback of the second command data, the second interrupt signal is sent to the soft-core processor; after the soft-core processor writes the process data of processing the second command data into the on-chip memory and notifies the network interface card, the step also includes: after the network interface card receives a confirmation signal from the server based on the process data feedback of the second command data, the third interrupt signal is sent to the soft-core processor.
[0044] In one embodiment, the soft core processor processes the second management command data, including: the soft core processor identifies the corresponding solid state drive corresponding to the second management command data, parses the second management information of the second management command data and performs management of the corresponding solid state drive based on the second management information.
[0045] Among them, if the second management command data is a data type that requires data return, the NVMEOF protocol processing IP core receives the interactive data returned by the soft-core processor, and sends the interactive data to the on-chip memory; the Ethernet protocol offloading IP core reads the interactive data from the on-chip memory and sends the interactive data to the server through the physical layer IP; the NVMEOF protocol processing IP core receives the read and write completion signal sent by the soft-core processor and sends the read and write completion signal to the Ethernet protocol offloading IP core, and the Ethernet protocol offloading IP core sends the read and write completion signal to the server through the physical layer IP; if the second management command data is a data type that does not require data return, the NVMEOF protocol processing IP core receives the read and write completion signal sent by the soft-core processor and sends the read and write completion signal to the Ethernet protocol offloading IP core, and the Ethernet protocol offloading IP core sends the read and write completion signal to the server through the physical layer IP.
[0046] Specifically, the second management command data is remote management command data. Remote management command: refers to the management command of the server to the NVME SSD. Remote management commands are divided into two cases: data needs to be returned to the server and data does not need to be returned to the server. Among them, the remote management command includes a command for obtaining basic information and basic properties of the SSD disk. The second management command data sent by the server is command data in the ROCEV2 data format.
[0047] Specifically, combined Figure 6 As shown in the figure, RHOST is the server, CMAC is the physical layer IP core, ERNIC is the Ethernet protocol offload IP core, and HBM2 is the on-chip memory. The processing flow of the second management command data:
[0048] RHOST sends the second management command data SQ of NVME SSD to ERNIC through CMAC. ERNIC sends the second management command data SQ after removing the ROCEV2 protocol header to HBM2. After receiving the data, HBM2 sends an interrupt signal to notify MB. After receiving the interrupt signal, MB takes out the second management command data SQ from HBM2 and executes the second management command data.
[0049] If the second command data is the second management command data that requires data to be returned, MB writes the returned read-write interactive data into HBM2, and ERNIC takes the read-write interactive data out of HBM2 and sends it to RHOST. After receiving the read-write interactive data sent by ERNIC, RHOST generates a completed response signal ACK to ERNIC. After receiving the response signal ACK, ERNIC generates an interrupt signal and sends it to MB. At this point, a RHOST management command interaction that requires data to be returned is completed.
[0050] If the second command data is the second management command data that does not require data return, then after receiving the second management command data, MB generates an RDMA (Remote Direct Memory Access) completion signal to ERNIC. ERNIC sends the RDMA completion signal to RHOST. After receiving the RDMA completion command, RHOST generates an ACK signal to ERNIC, and then ERNIC sends an interrupt signal to MB. At this point, a RHOST management command interaction without data return is completed.
[0051] In one embodiment, the remote data access unit further includes a cache area, the NVMEOF protocol processing IP core reads and stores the second read / write command data from the on-chip memory and sends a doorbell notification to the solid-state hard disk through the PCIE communication IP core, including: if the second read / write command data is the second write command data, the NVMEOF protocol processing IP core reads the second write command data with the protocol header deleted from the on-chip memory and converts the second write command data with the protocol header deleted into a third read data command and then sends it to the cache area; after receiving the third read data command, the cache area sends the third read data command to the Ethernet protocol unloading IP core and sends a doorbell signal, and the Ethernet After receiving the doorbell signal, the protocol offloading IP core reads the third read data command from the buffer area and sends a read request to the server; after receiving the response packet of the read request and the third target data fed back by the server, the Ethernet protocol offloading IP core converts the response packet of the read request into a response packet of the write request and sends it to the IP core for NVMEOF protocol processing, and stores the third target data in the IP core for NVMEOF protocol processing; the IP core for NVMEOF protocol processing sends a doorbell notification to the solid-state hard disk through the PCIE communication IP core, so that the solid-state hard disk reads and writes the third target data from the IP core for NVMEOF protocol processing.
[0052] In one example, the NVMEOF protocol processing IP core reads and stores the second read / write command data from the on-chip memory and sends a doorbell notification to the solid-state drive through the PCIE communication IP core, and also includes: if the second read / write command data is the second read command data, the NVMEOF protocol processing IP core reads the second read command data of the protocol header deleted from the on-chip memory and converts the second read command data of the protocol header deleted from the on-chip memory into a third write data command and then sends it to the cache area; the NVMEOF protocol processing IP core sends a doorbell notification to the solid-state drive and reads the second read command data of the protocol header deleted from the on-chip memory to the solid-state drive, so that the solid-state drive sends the fourth target data in the solid-state drive to the NVMEOF protocol processing IP core; the cache area sends the doorbell notification to the Ethernet protocol unloading IP core, the Ethernet protocol unloading IP core reads the third write data command from the cache area and reads the fourth target data from the NVMEOF protocol processing IP core based on the third write data command, and sends the fourth target data to the server through the physical layer IP.
[0053] like Figure 7 As shown in the figure, RHOST writes 4KB of data to NVME SSD. This command is an IO command, which is a command packet in RDMA format. IO command refers to the command for reading and writing operations on NVME SSD. The brief process of RHOST write command: RHOST write data command-RDMA read data command-ERNCI reads RHOST data and writes it to NVME SSD. In addition, Figure 7As shown, WQECache represents the cache area, and NVMoF represents the IP core for NVMEOF protocol processing.
[0054] See also Figure 7 As shown: First, RHOST sends a write command SQ to write 4K data to ERNIC through CMAC. ERNIC removes the ROCEV2 protocol header in the write command SQ and sends the write command SQ to HBM2. ERNIC sends the doorbell RQ_DB to NVMEOF. After receiving the doorbell, NVMEOF takes the write command SQ from HBM2, and then NVMEOF converts the write command SQ into an RDMA_READ command and sends it to WQE CACHE. After receiving the written RDMA_READ command, WQE CACHE sends a doorbell to ERNIC. After receiving the doorbell from WQE CACHE, ERNIC takes the RDMA_READ command from WQE CACHE and then sends an RDMA read data request to RHOST. After receiving the RDMA read data request sent by ERNIC, RHOST sends a read request response packet, which carries 4K data and sends it to ERNIC. After that, ERNIC removes the ROCEV2 protocol header from the read request response packet and sends the data to NVMEOF. At the same time, ERNIC generates a completion CQ command and sends it to WQE CACHE, and sends the doorbell CQ_DB to NVMEOF. After receiving the response data, NVMEOF sends an SQ command to the SSD, and sends the doorbell SQ_Tail_DB to the SSD. After receiving the doorbell, the SSD reads out the 4KB data, and after completion, the SSD writes the completion command CQ to NEMOF. NVMEOF writes the completion command CQ to WQE CACHE, and after writing, WQE CACHE sends the doorbell QP_SQ_PIDB to ERNIC. After receiving the doorbell from WQE CACHE, ERNIC takes out the completion command CQ from WQE CACHE, and after taking it out, ERNIC continues to send the completion command CQ to RHOST, and then RHOST generates a response signal ACK and sends it to ERNIC. After receiving the response signal ACK from RHOST, ERNIC writes the completion command CQ to WQE CACHE, and ERNIC sends the doorbell CQ_DB to NVMEOF. NVMEOF sends the doorbell CQ_Head_DB to the NVME SSD. So far, RHOST has completed a NVME SSD data writing process.
[0055] In this embodiment, the brief flow of the RHOST read command is: RHOST read data command-RDMA write data command-ERNCI writes NVME SSD data to RHOST. RHOST read 4KB data flow, this command is an IO command, similar to the RHOST write 4KB data flow.
[0056] See also Figure 8 As shown: First, RHOST sends a read command SQ to read 4KB data to ERNIC through CMAC. After receiving the read command SQ of the read 4K data command sent by RHOST, ERNIC sends an ACK command to RHOST. After receiving the read command SQ, ERNIC removes the ROCEV2 protocol header and stores the read command SQ in HBM2. ERNIC also sends the doorbell RQ_DB to NVMEOF. After receiving the doorbell, NVMEOF takes out the read command SQ from HBM2, and then NVMEOF sends the doorbell signal SQ_Tail_DB to NVME SSD. After receiving the doorbell, NVME SSD takes out the read command SQ from NVMEOF, and then NVMESSD writes 4KB data to NVMEOF. At the same time, NVMEOF parses the read command SQ and converts it into an RDMA write command and writes it into WQE CACHE. WQE CACHE generates a doorbell QP_SQ_PIDB to ERNIC to notify ERNIC to get the RDME write command from WQE CACHE. After ERNIC takes out the RDMA write command, it reads 4K data from NVMEOF, and then ERNIC sends this part of 4K data to RHOST. After receiving this part of 4K data, RHOST generates an ACK response signal and sends it to ERNIC. After receiving the ACK response signal, ERNIC writes the ACK response signal to WQE CACHE, and ERNIC sends the doorbell CQ_DB to NVMEOF. NVMEOF further sends the NVMoF doorbell CQ_Head_DB to NVME SSD. At this point, RHOST has completed an NVME SSD data read process.
[0057] In one embodiment, a filter and a direct memory access controller are also configured in the remote data access unit; before the step of the physical layer IP core receiving the remote data processing instruction sent by the server, it also includes: the physical layer IP core receives the ping data packet sent by the server and sends the ping data packet to the filter; the filter sends the ping data packet to the on-chip memory through the direct memory access controller; after the direct memory access controller generates an interrupt and reports it to the soft-core processor, the soft-core processor reads the ping data packet from the on-chip memory, swaps the source address and the destination address in the ping data packet and encapsulates it to obtain an encapsulated data packet, and sends the encapsulated data packet to the server through the direct memory access controller and the physical layer IP.
[0058] In this embodiment, before the remote data access unit responds to the remote data processing instruction of the server, it needs to establish a network communication connection with the server. Specifically, the server sends a data packet to the external acceleration card. If it can ping through, it means that the acceleration card feeds back data to the server, and the server and the acceleration card exchange data. The ping data packet sent by the server is a Non-RoCE packet. In order to establish network communication with the acceleration card.
[0059] For the data processing flow, see Fig. 9 As shown. Fig. 9 As shown, RDMA_Pkt_Filter is a filter for remote direct data access, and axi_DMA is a controller for direct memory access. For details, see Fig. 9 As shown, RHOST sends a ping data packet, i.e., a Non-RoCE packet. The Non-RoCE packet reaches RDMA_Pkt_Filter through CMAC. The RDMA_Pkt_Filter is mainly used to filter RDMA's ROCEV2 data packets and non-RDMA's non-ROCEV2 data packets. Among them, the ping data packet is a non-RDMA data packet, and after being filtered by RDMA_Pkt_Filter, the data packet is sent to axi_DMA. axi_DMA writes the filtered data packet into HBM2 and sends a DMA interrupt to MB at the same time. After receiving the DMA interrupt, MB takes out the ping data packet from HBM2, then swaps the source data with the destination data, and encapsulates it to obtain a reply packet, writes the reply packet into HBM2, and then MB controls axi_DMA to take out the encapsulated reply packet from HBM2 and send it to CMAC, and then sends the reply packet to the RHOST end.
[0060] Based on the above-mentioned multi-path data access method, the present application also provides an acceleration card corresponding to its implementation. The acceleration card is configured with a soft-core processor, on-chip memory, PCIE communication IP core and remote data access unit, and the remote data access unit includes a physical layer IP core, an Ethernet protocol offloading IP core and an IP core for NVMEOF protocol processing; when the soft-core processor issues a local data processing instruction for the solid-state hard disk of the NVMEOF protocol, the first command data is written into the on-chip memory and a doorbell notification is sent to the solid-state hard disk through the PCIE communication IP core. When the solid-state hard disk receives the doorbell notification, it reads and processes the first command data from the on-chip memory; when the physical layer IP core receives the remote data processing instruction sent by the server, it receives the second command data sent by the server, and sends the second command data to the Ethernet protocol offloading IP core, and the Ethernet protocol The unloading IP core deletes the protocol header in the second command data and then forwards it to the on-chip memory. If the second command data is the second management command data, the Ethernet protocol unloading IP core notifies the soft core processor in an interrupt manner, so that the soft core processor reads the second management command data from the on-chip memory and processes it. If the second command data is the second read / write command data, the Ethernet protocol unloading IP core sends a doorbell notification to the NVMEOF protocol processing IP core. The NVMEOF protocol processing IP core reads and stores the second read / write command data from the on-chip memory and sends a doorbell notification to the solid-state drive through the PCIE communication IP core. When the solid-state drive receives the doorbell notification, it reads and processes the second read / write command data from the NVMEOF protocol processing IP core.
[0061] In this embodiment, each component in the acceleration card is specifically used to implement the operations mentioned in each embodiment of the above multi-path data access method. Among them, the remote data access unit can be two-way.
[0062] Specifically, Fig.10 As shown, this embodiment provides a dual-path data access method for an NVMe storage system implemented using FPGA. Unlike the previous single-path NVMe storage system, two hosts are supported to access the NVMe SSD separately, and the read and write bandwidth of each host is consistent with the single path. Therefore, for the NVMe SSD under the same storage system, the number of accessible hosts is increased. In addition, since the NVMe storage system adopts a dual-path redundant design, when any storage path in the system fails, it will not cause the entire NVMe storage system to be paralyzed, greatly improving the security and stability of the entire NVMe storage system.
[0063] In a dual-path data access method for an NVMe storage system implemented using FPGA, the design block diagram of the dual-path NVMe storage system is as follows: Fig.10As shown in the figure, the accelerator card is implemented by configuring the FPGA. There are two main ways to access the NVMe SSD inside the FPGA: local access and remote access.
[0064] The local access method is: the MB (MicroBlaze) inside the FPGA directly accesses the NVMe SSD. MicroBlaze is a 32-bit soft-core processor inside the FPGA. The soft-core processor is embedded with the Linux operating system and supports the NVMeCli instruction set. However, since the main frequency of MicroBlaze is 200M, and the data interaction with the NVMe SSD is achieved by directly operating the XDMAIP core, without high-speed IP cores such as ERNIC and NVMeOF, the access bandwidth to the NVMe SSD is limited.
[0065] Remote access mode: The server accesses the NVMe SSD. The Admin commands and IO commands sent by the server are processed by the high-speed IP cores such as CMAC, ERNIC, and NVMeOF inside the FPGA and then forwarded to the NVMe SSD. The internal controller of the NVMe SSD completes the data interaction with the server according to the corresponding instructions. Since the bandwidth of data processing by high-speed IP cores such as CMAC, ERNIC, and NVMeOF can support up to 100G, the remote access mode greatly improves the read and write performance of the NVMe SSD.
[0066] The present application also provides a storage system, which includes the acceleration card in any of the above embodiments and one or more solid-state hard disks.
[0067] In a specific embodiment, Fig.10 As shown, the acceleration card is implemented through FPGA, and the solid-state drive is NVMe SSD. Fig.10 An embodiment of an NVMe SSD is given in FIG. 1 . In other implementations, an FPGA may also control data access to multiple NVMe SSDs.
[0068] It should be understood that, although the steps in the flowchart are shown in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.
[0069] In one embodiment, a readable storage medium is provided, on which an embedded program is stored. When the embedded program is executed by a processor, a multi-path data access method described in any of the above embodiments is implemented.
[0070] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing related hardware through an embedded program, and the embedded program can be stored in a non-volatile computer-readable storage medium. When the embedded program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
[0071] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0072] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.
Claims
1. A multi-path data access method, characterized in that: Applied to an accelerator card, the accelerator card is equipped with a soft core processor, an on-chip memory, a PCIE communication IP core and a remote data access unit, the remote data access unit includes a physical layer IP core, an Ethernet protocol offloading IP core and an NVMEOF protocol processing IP core; the method includes: When the soft-core processor issues a local data processing instruction to the solid-state drive of the NVMEOF protocol, the first command data is written into the on-chip memory and a doorbell notification is sent to the solid-state drive through the PCIE communication IP core. When the solid-state drive receives the doorbell notification, it reads and processes the first command data from the on-chip memory; When the physical layer IP core receives the remote data processing instruction sent by the server, it receives the second command data sent by the server, and sends the second command data to the Ethernet protocol unloading IP core, and the Ethernet protocol unloading IP core deletes the protocol header in the second command data and then forwards it to the on-chip memory. If the second command data is the second management command data, the Ethernet protocol unloading IP core notifies the soft core processor in an interrupt mode, so that the soft core processor reads the second management command data from the on-chip memory and processes it. If the second command data is the second read / write command data, the Ethernet protocol unloading IP core sends a doorbell notification to the NVMEOF protocol processing IP core, and the NVMEOF protocol processing IP core reads and stores the second read / write command data from the on-chip memory and sends a doorbell notification to the solid-state drive through the PCIE communication IP core. When the solid-state drive receives the doorbell notification, it reads and processes the second read / write command data from the NVMEOF protocol processing IP core. The remote data access unit also includes a cache area, the NVMEOF protocol processing IP core reads and stores the second read / write command data from the on-chip memory and sends a doorbell notification to the solid-state hard disk through the PCIE communication IP core, including: If the second read / write command data is the second write command data, the NVMEOF protocol processing IP core reads the second write command data of the protocol header deleted from the on-chip memory and converts the second write command data of the protocol header deleted into a third read data command and then sends it to the cache area; after receiving the third read data command, the cache area sends the third read data command to the Ethernet protocol unloading IP core and sends a doorbell signal, and after receiving the doorbell signal, the Ethernet protocol unloading IP core reads the third read data command from the cache area and then sends a read request to the server; after receiving the response packet of the read request and the third target data fed back by the server, the Ethernet protocol unloading IP core converts the response packet of the read request into a response packet of the write request and then sends it to the NVMEOF protocol processing IP core and stores the third target data in the NVMEOF protocol processing IP core; the NVMEOF protocol processing IP core sends a doorbell notification to the solid state drive through the PC IE communication IP core, so that the solid state drive reads and writes the third target data from the NVMEOF protocol processing IP core; If the second read / write command data is the second read command data, the NVMEOF protocol processing IP core reads the second read command data of the protocol header from the on-chip memory and converts the second read command data of the protocol header into a third write data command and sends it to the cache area; the NVMEOF protocol processing IP core sends a doorbell notification to the solid-state drive and reads the second read command data of the protocol header from the on-chip memory to the solid-state drive, so that the solid-state drive sends the fourth target data in the solid-state drive to the NVMEOF protocol processing IP core; the cache area sends a doorbell notification to the Ethernet protocol unloading IP core, the Ethernet protocol unloading IP core reads the third write data command from the cache area and reads the fourth target data from the NVMEOF protocol processing IP core based on the third write data command, and sends the fourth target data to the service end through the physical layer IP.
2. The method according to claim 1, characterized in that: The solid state drive processes the first command data, including: When the first command data is first management command data, the solid state drive parses the first management command data and obtains first management information, and manages the solid state drive based on the first management information; When the first command data is first write command data, the solid state drive parses the first write command data and obtains first write indication information, and reads first target data to be written from the on-chip memory through the PCIE communication IP core based on the first write indication information and writes the first target data to be written into the solid state drive; When the first command data is the first read command data, the solid-state drive parses the first read command data and obtains first read indication information, reads second target data from the solid-state drive based on the first read indication information, and writes the second target data into the on-chip memory through the PCIE communication IP core, so that the soft-core processor reads the second target data from the on-chip memory.
3. The method according to claim 1, characterized in that The soft core processor processes the second management command data, including: The soft-core processor identifies the corresponding solid-state hard disk corresponding to the second management command data, parses the second management information of the second management command data, and performs management of the corresponding solid-state hard disk based on the second management information.
4. The method according to claim 3, characterized in that: The method further comprises: If the second management command data is a data type that requires data to be returned, the NVMEOF protocol processing IP core receives the interactive data returned by the soft core processor, and sends the interactive data to the on-chip memory; The Ethernet protocol offloading IP core reads the interaction data from the on-chip memory and sends the interaction data to the server through the physical layer IP; The NVMEOF protocol processing IP core receives the read / write completion signal sent by the soft core processor and sends the read / write completion signal to the Ethernet protocol unloading IP core, and the Ethernet protocol unloading IP core sends the read / write completion signal to the server through the physical layer IP; If the second management command data is a data type that does not require data return, the NVMEOF protocol processing IP core receives the read and write completion signal sent by the soft core processor and sends the read and write completion signal to the Ethernet protocol unloading IP core, and the Ethernet protocol unloading IP core sends the read and write completion signal to the server through the physical layer IP.
5. The method according to claim 1, characterized in that The remote data access unit is also configured with a filter and a direct memory access controller; before the physical layer IP core receives the remote data processing instruction sent by the server, it also includes: The physical layer IP core receives the ping data packet sent by the server and sends the ping data packet to the filter; The filter sends the ping packet to the on-chip memory via the direct memory access controller; After the direct memory access controller generates an interrupt and reports it to the soft-core processor, the soft-core processor reads the ping data packet from the on-chip memory, swaps the source address and the destination address in the ping data packet and encapsulates it to obtain an encapsulated data packet, and sends the encapsulated data packet to the server through the direct memory access controller and the physical layer IP.
6. An accelerator card, characterized in that: The accelerator card is equipped with a soft-core processor, on-chip memory, a PCIE communication IP core and a remote data access unit, wherein the remote data access unit includes a physical layer IP core, an Ethernet protocol offloading IP core and an NVMEOF protocol processing IP core; When the soft-core processor issues a local data processing instruction to the solid-state drive of the NVMEOF protocol, the first command data is written into the on-chip memory and a doorbell notification is sent to the solid-state drive through the PCIE communication IP core. When the solid-state drive receives the doorbell notification, it reads and processes the first command data from the on-chip memory; When the physical layer IP core receives the remote data processing instruction sent by the server, it receives the second command data sent by the server, and sends the second command data to the Ethernet protocol unloading IP core, and the Ethernet protocol unloading IP core deletes the protocol header in the second command data and then forwards it to the on-chip memory. If the second command data is the second management command data, the Ethernet protocol unloading IP core notifies the soft core processor in an interrupt mode, so that the soft core processor reads the second management command data from the on-chip memory and processes it. If the second command data is the second read / write command data, the Ethernet protocol unloading IP core sends a doorbell notification to the NVMEOF protocol processing IP core, and the NVMEOF protocol processing IP core reads and stores the second read / write command data from the on-chip memory and sends a doorbell notification to the solid-state drive through the PCIE communication IP core. When the solid-state drive receives the doorbell notification, it reads and processes the second read / write command data from the NVMEOF protocol processing IP core. The remote data access unit also includes a cache area, the NVMEOF protocol processing IP core reads and stores the second read / write command data from the on-chip memory and sends a doorbell notification to the solid-state hard disk through the PCIE communication IP core, including: If the second read / write command data is the second write command data, the NVMEOF protocol processing IP core reads the second write command data of the protocol header deleted from the on-chip memory and converts the second write command data of the protocol header deleted into a third read data command and then sends it to the cache area; after receiving the third read data command, the cache area sends the third read data command to the Ethernet protocol unloading IP core and sends a doorbell signal, and after receiving the doorbell signal, the Ethernet protocol unloading IP core reads the third read data command from the cache area and then sends a read request to the server; after receiving the response packet of the read request and the third target data fed back by the server, the Ethernet protocol unloading IP core converts the response packet of the read request into a response packet of the write request and then sends it to the NVMEOF protocol processing IP core and stores the third target data in the NVMEOF protocol processing IP core; the NVMEOF protocol processing IP core sends a doorbell notification to the solid-state drive through the PCIE communication IP core, so that the solid-state drive reads and writes the third target data from the NVMEOF protocol processing IP core; If the second read / write command data is the second read command data, the NVMEOF protocol processing IP core reads the second read command data of the protocol header from the on-chip memory and converts the second read command data of the protocol header into a third write data command and sends it to the cache area; the NVMEOF protocol processing IP core sends a doorbell notification to the solid-state drive and reads the second read command data of the protocol header from the on-chip memory to the solid-state drive, so that the solid-state drive sends the fourth target data in the solid-state drive to the NVMEOF protocol processing IP core; the cache area sends a doorbell notification to the Ethernet protocol unloading IP core, the Ethernet protocol unloading IP core reads the third write data command from the cache area and reads the fourth target data from the NVMEOF protocol processing IP core based on the third write data command, and sends the fourth target data to the service end through the physical layer IP.
7. A storage system, characterized in that: The storage system includes the acceleration card described in claim 6, and also includes one or more solid-state hard drives with NVMEOF protocol.
8. A readable storage medium having an embedded program stored thereon, characterized in that: When the embedded program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.
