WO2023109891A1 - Multicast transmission method, apparatus and system - Google Patents

Abstract

The present application provides a multicast transmission method, apparatus and system. The method is applied to a first computing device, the first computing device comprising a first input/output (IO) device. The method comprises: the first IO device acquires a first request; the first IO device generates a first data packet according to first information and the first request, the first information comprising multicast information, the multicast information being used for identifying the link connections between multicast members of a multicast group and the first computing device, and the first data packet carrying the multicast information and data to be written corresponding to the first request; the first IO device sends the first data packet to the multicast members of the multicast group. The method can achieve reliable and efficient multicast data transmission.

Description

Multicast transmission method, device and system

This application claims the priority of the Chinese patent application with the application number 202111556558.4 and the title of the invention "Multicast transmission method, device and system" filed with the China Patent Office on December 17, 2021, the entire contents of which are incorporated by reference in this application middle.

technical field

The present application relates to the technical field of network communication, and more specifically, to a method, device and system for multicast transmission.

Background technique

Multicast (multicast) technology is a technology for network communication between a single sender and multiple receivers. The multicast technology uses the multicast group address as the destination address of the data packet, establishes a multicast tree, and uses the multicast tree to realize point-to-multi-point (P2MP) data forwarding, which is beneficial to reduce bandwidth consumption and Improve service quality. It is especially suitable for wide application in computing clusters.

The traditional multicast transmission method is realized based on the host (host) of the computing device. In the implementation process, the host needs to copy the data packet to be sent, and the operating system of the host needs to participate, which makes the method have high complexity and disadvantages. The problem of large delay and low transmission efficiency.

Therefore, there is an urgent need for a multicast transmission method to realize efficient multicast data transmission.

Contents of the invention

The present application provides a multicast transmission method, device and system, and the method can realize reliable and efficient multicast data transmission.

In a first aspect, a method for multicast transmission is provided, which is applied to a first computing device, and the first computing device includes a first input/output IO device. The method includes: the first IO device acquires a first request; the first The IO device generates a first data packet according to the first information and the first request, the first information includes multicast information, and the multicast information is used to identify the communication between the multicast member of the multicast group and the first computing device Link connection, the first data packet carries the multicast information and the data to be written corresponding to the first request; the first IO device sends the first data packet to the multicast members of the multicast group.

Wherein, a multicast group may include at least 2 multicast members. The multicast information is used to identify the link connections between the multicast members of the multicast group and the first computing device, that is, the multicast information is used to identify at least two link connections. The data to be written may be all data corresponding to the first request, that is, the data volume of the data to be written is the same as that of the data corresponding to the first request. Optionally, the data to be written may also be a part of the data corresponding to the first request, that is, the data volume of the data to be written is smaller than the data volume of the data corresponding to the first request.

In the above technical solution, the first IO device may perform data encapsulation according to the first request and the first information to generate the first data packet, and send the first data packet to the multicast members of the multicast group, instead of including the first data packet through the first host The processor encapsulates and generates data packets according to the work request, avoiding the problems of high complexity, large delay and low transmission efficiency in the traditional multicast transmission method. By setting the first data packet sent by the first IO device to carry the multicast information for identifying the link connection between the multicast members of the multicast group and the first computing device, reliable transmission of multicast data can be ensured. Therefore, the multicast transmission method provided by the present application can realize reliable and efficient multicast data transmission.

In a possible design, the multicast information includes a multicast identifier, and the multicast identifier is obtained when the first computing device establishes a link connection with the multicast group.

Wherein, the multicast group can include at least 2 multicast members (denoted as multicast member 1 and multicast member 2), and the multicast identifier can identify at least two link connections (denoted as link 1 and link 2), that is, link 1 is associated with multicast member 1 and the first computing device, and link 2 is associated with multicast member 2 and the first computing device.

In the above technical solution, the link connection between the multicast member of the multicast group and the first computing device can be identified through the multicast identifier included in the multicast information.

In another possible design, the first data packet further includes a port number, and the port number is used to indicate that the first data packet is transmitted in a multicast transmission manner.

Optionally, the port number may be the destination port number of the first data packet.

In another possible design, the first computing device further includes a first host, the first IO device communicates with the first host through an IO network, and the first request is to run in a processor included in the first host request sent by the application.

In another possible design, the first information further includes first window information, where the first window information is used to indicate the minimum amount of data that can be processed by the multicast members of the multicast group within the first time period.

In the above technical solution, when the first IO device generates the first data packet, the minimum amount of data that can be processed by the multicast members of the multicast group within the first time period is considered, and the amount of data carried by the first data packet can be controlled , to achieve network flow control.

Optionally, the first IO device may also update the window information. In an exemplary example, the first IO device may also perform the following operations: update the second window information to the first window information, and the second window information is that the multicast members of the multicast group before the first time period The minimum amount of data that can be processed within a period of time. The minimum amount of data that can be processed by the multicast members of the multicast group indicated by the second window information is different from the minimum amount of data that can be processed by the multicast members of the multicast group indicated by the first window information. In this implementation manner, by setting the first IO device to update the window information in a timely manner within a preset time period, the amount of data carried by the first data packet can be accurately controlled to realize network flow control.

In another possible design, the transmission protocol used by the first IO device to send the first data packet is TCP/IP.

In the above technical solution, a method of multicast transmission based on TCP/IP is provided, and this method avoids the problems of high complexity, large delay and low transmission efficiency in the traditional multicast transmission method, and can realize reliable and efficient multicast data transmission.

In another possible design, the data to be written is a part of the data corresponding to the first request, and the first information further includes indication information, and the indication information is used to indicate that the data to be written should be encapsulated , before the first IO device generates the first data packet according to the first information and the first request, the method further includes: the first IO device sends an IO write command to a multicast member of the multicast group, and the IO The write command is used to instruct to store the data corresponding to the first request into the first registration area MR, the data corresponding to the first request is located in the memory included in the first host included in the first computing device, and the first MR is The storage area in the memory included in the second host is registered to the storage area in the memory of the second IO device, the second computing device includes the second host and the second IO device, and the second computing device is a member of the multicast group A multicast member: the first IO device receives the indication information sent by the multicast member of the multicast group. Wherein, the data volume of the data to be written is smaller than the data volume of the data corresponding to the first request.

In the above technical solution, the first IO device may actively request to write data into the storage area of the multicast member of the multicast group, and encapsulate and generate the first data according to the received indication information sent by the multicast member of the multicast group packet, and send the first data packet to the multicast members of the multicast group, which can effectively reduce the probability of congestion on the side of the multicast members of the multicast group, so as to better meet the needs of the multicast members of the multicast group.

In another possible design, the IO write command includes a second key value and second position information, the second key value is used to identify the second MR, and the second position information is used to indicate the The location of the input data in the second MR.

In another possible design, the first information further includes a credit value, where the credit value is used to indicate the minimum number of requests that can be processed by the multicast members of the multicast group within the second time period, and the first data The basic transmission header BTH of the packet carries the credit value.

In the above technical solution, after the second IO device sends the credit value to the first IO device, when the first IO device generates the first data packet, the requests that the multicast members of the multicast group can process within the first time period are considered The minimum number is conducive to the realization of network flow control.

In another possible design, the first IO device obtaining the first request includes: the first IO device receiving the first request sent by the multicast member of the multicast group, the first request is used to indicate that the The data to be written in the second MR is stored in the storage area of the multicast members of the multicast group, and the second MR is registered in the storage area of the first IO device for the storage area in the memory included in the first host storage area, the first computing device also includes the first host.

In the above technical solution, the first IO device sends a first data packet to the multicast member of the multicast group according to the first request of the multicast member of the multicast group, and the first data packet carries the information corresponding to the first request. The data to be written and the multicast information of the multicast group can realize reliable and efficient multicast data transmission.

In another possible design, the first request includes a first key value, first location information and a preset field, the first key value is used to identify the second MR, and the first location information is used to Indicate the position of the data to be written in the second MR, and the value of the preset field is used to indicate the first request.

In another possible design, the transmission protocol used by the first IO device to send the first data packet is Ethernet-based Remote Direct Data Access (RDMA).

In the above technical solution, an RDMA-based multicast transmission method is provided, which can realize reliable and efficient multicast data transmission.

In another possible design, the first IO device sending the first data packet to the multicast members of the multicast group includes: the first IO device sending the first data packet to a forwarding device, and the forwarding device It is used to copy the first data packet, and forward the copied first data packet to the multicast members of the multicast group, and the link between the multicast members of the multicast group and the first computing device Road connections include the forwarding device.

In the above technical solution, the first IO device does not copy the first data packet, which is beneficial to reduce the resource overhead of the first IO device.

In another possible design, the multicast members of the multicast group include a second computing device, the second computing device is different from the first computing device, and the multicasting of the multicast group by the first IO device After the member sends the first data packet, the method further includes: the first IO device receives a second request sent by the second computing device, and the second request is used to request to obtain the first request carried in the first data packet Corresponding to the data to be written; the first IO device sends a second data packet to the second computing device, the second data packet carries the data to be written, and the port number included in the second data packet is used to indicate The transmission mode for transmitting the second data packet is a unicast transmission mode.

In the above technical solution, the first IO device sends the second data packet to the second computing device according to the second request, so that the multicast members who have not successfully received the first data packet receive the second data packet, which can ensure the integrity of the multicast data. Reliable transmission.

In another possible design, the multicast members of the multicast group only include the second computing device and the third computing device, the third computing device, any two of the second computing device and the first computing device Different computing devices, the method further includes: after the first IO device receives the first completion message and the second completion message, the first IO device sends a third completion message to the processor included in the first host, the The third completion message is used to indicate that the first request has been successfully executed, the first completion message is used to indicate that the second computing device has successfully executed the first request, and the second completion message is used to indicate that the third computing device has successfully executed The first request was executed successfully.

In the above technical solution, the first IO device will not feed back to the processor included in the first host until it receives the completion messages (that is, the first completion message and the second completion message) sent by all the multicast members of the multicast group. The processing result of the first request, instead of feeding back the processing result of the request corresponding to the completion message every time a completion message sent by a multicast member is received, this can improve the IO processing rate to achieve efficient multicast data transmission. .

The first IO device described in the first aspect above includes a network interface controller, an intelligent network interface controller, a host bus adapter, a host channel adapter, an accelerator, a data processor, an image processor, an artificial intelligence device, and a software-defined infrastructure at least one of . The IO network described in the first aspect above includes any one of the high-speed serial computer expansion bus standard PCIe, the computer fast link CXL, the cache coherent interconnection protocol CCIX, and the unified bus Ubus.

In a second aspect, a multicast transmission method is provided, which is applied to a second computing device, the second computing device is a multicast member of a multicast group, and the second computing device includes a second input and output IO device and a second host , the second IO device communicates with the second host through an IO network, the method includes: the second IO device receives a first data packet sent by the first IO device, the first data packet carries multicast information and a first request Corresponding to the data to be written, the multicast information is used to identify the link connection between the multicast member of the multicast group and the first computing device, the first computing device includes the first IO device, and the first computing device and The second computing device is different; the second IO device stores the data to be written into the memory of the second host according to the first data packet.

In the above technical solution, the second IO device receives the first data packet sent by the first IO device, the first data packet carries the multicast information and the data to be written corresponding to the first request, which can realize reliable and efficient multicast data transmission.

In a possible implementation manner, the method further includes: the second IO device sending a second request to the first IO device, where the second request is used to request to obtain the corresponding information of the first request carried in the first data packet. The data to be written.

In the above technical solution, when the second IO device fails to receive the first data packet, the second IO device will actively send a second request to the first IO device to obtain the unsuccessfully received data, which can realize reliable data transmission.

The second IO device described in the second aspect above includes a network interface controller, an intelligent network interface controller, a host bus adapter, a host channel adapter, an accelerator, a data processor, an image processor, an artificial intelligence device, and a software-defined infrastructure at least one of . The IO network described in the second aspect above includes any one of the high-speed serial computer expansion bus standard PCIe, the memory interconnection CXL, and the unified bus Ubus.

In a third aspect, a multicast transmission device is provided, the device includes a transceiver unit and a processing unit, the transceiver unit is used to acquire a first request; the processing unit is used to generate a request according to the first information and the first request The first data packet, the first information includes multicast information, the multicast information is used to identify the link connection between the multicast member of the multicast group and the multicast transmission device, and the first data packet carries the multicast The information and the data to be written corresponding to the first request; the transceiver unit is further configured to send the first data packet to the multicast members of the multicast group.

In a possible design, the multicast information includes a multicast identifier, and the multicast identifier is obtained when the multicast group establishes a link connection with the multicast transmission device.

In another possible design, the first data packet further includes a port number, and the port number is used to indicate that the first data packet is transmitted in a multicast transmission manner.

In another possible design, the first information further includes first window information, where the first window information is used to indicate the minimum amount of data that can be processed by the multicast members of the multicast group within the first time period.

Optionally, the processing unit is also used to update window information. In an example, the processing unit also uses the words: update the second window information to the first window information, the second window information is a time period before the first time period of the multicast members of the multicast group The minimum amount of data that can be processed. The minimum amount of data that can be processed by the multicast members of the multicast group indicated by the second window information is different from the minimum amount of data that can be processed by the multicast members of the multicast group indicated by the first window information.

In a fourth aspect, a multicast transmission device is provided, which is applied to a second computing device, the second computing device is a multicast member of a multicast group, and the second computing device includes a second input and output IO device and a second host , the second IO device communicates with the second host through an IO network, the device includes a transceiver unit and a processing unit, the transceiver unit is configured to receive a first data packet sent by the first IO device, and the first data packet carries a group broadcast information and the data to be written corresponding to the first request, the multicast information is used to identify the link connection between the multicast member of the multicast group and the first computing device, the first computing device includes the first IO device, The first computing device is different from the second computing device; the processing unit is configured to store the data to be written into the memory of the second host according to the first data packet.

In a possible design, the transceiver unit is further configured to send a second request to the first IO device, and the second request is used to request to obtain the to-be-written data.

In a fifth aspect, a first input/output IO device is provided, and the first IO device has a function of realizing the multicast transmission device described in the third aspect above. This function may be implemented based on hardware, or may be implemented by corresponding software based on hardware. The hardware or software includes one or more modules corresponding to the above functions.

In a possible implementation manner, the structure of the first IO device includes a processor, and the processor is configured to support the first IO device to perform corresponding functions in the foregoing method.

The first IO device may further include a memory, which is used to be coupled with the processor, and stores necessary program instructions and data of the first IO device.

In another possible implementation manner, the first IO device includes: a processor, a transmitter, a receiver, a random access memory, a read only memory, and a bus. Wherein, the processor is respectively coupled to the transmitter, the receiver, the random access memory and the read-only memory through the bus. Wherein, when the first IO device needs to be operated, the basic input/output system solidified in the read-only memory or the bootloader boot system in the embedded system is started to guide the first IO device into a normal operation state. After the first IO device enters the normal running state, run the application program and the operating system in the random access memory, so that the processor executes the method in the first aspect or any possible implementation manner of the first aspect.

A sixth aspect provides a second input/output IO device, the second IO device has the function of implementing the multicast transmission device described in the fourth aspect above. This function may be implemented based on hardware, or may be implemented by corresponding software based on hardware. The hardware or software includes one or more modules corresponding to the above functions.

In a possible implementation manner, the second IO device supports execution of corresponding functions in the foregoing method.

The second IO device may further include a memory, which is used to be coupled with the processor, and stores necessary program instructions and data of the second IO device.

In another possible implementation manner, the second IO device includes: a transmitter, a receiver, a random access memory, a read only memory, and a bus. Wherein, the processor is respectively coupled to the transmitter, the receiver, the random access memory and the read-only memory through the bus. Wherein, when the second IO device needs to be operated, the basic input/output system solidified in the read-only memory or the bootloader boot system in the embedded system is started to guide the second IO device into a normal operation state. After the second IO device enters the normal running state, run the application program and the operating system in the random access memory, so that the processor executes the method in the second aspect or any possible implementation manner of the second aspect.

In a seventh aspect, a computer program product is provided, and the computer program product includes: computer program code, when the computer program code is run on a computer, it causes the computer to execute the above-mentioned first aspect or the second aspect, and the above-mentioned first aspect Or any of the possible implementation methods of the second aspect.

In an eighth aspect, a computer-readable medium is provided, the computer-readable medium stores program code, and when the computer program code runs on a computer, the computer executes the above-mentioned first aspect or the second aspect, and the above-mentioned first aspect. Aspect or any method that may be implemented in the second aspect. These computer-readable storages include, but are not limited to, one or more of the following: read-only memory (read-only memory, ROM), programmable ROM (programmable ROM, PROM), erasable PROM (erasable PROM, EPROM), Flash memory, electrical EPROM (electrically EPROM, EEPROM) and hard disk drive (hard drive).

A ninth aspect provides a chip system, the chip system includes a processor and a data interface, wherein the processor reads the instructions stored in the memory through the data interface, so as to execute the first aspect or the second aspect above, and the first aspect above A method in any possible implementation of the first aspect or the second aspect. In the specific implementation process, the chip system can be based on a central processing unit (central processing unit, CPU), a microcontroller (micro controller unit, MCU), a microprocessor (micro processing unit, MPU), a digital signal processor (digital signal processor) signal processing, DSP), system on chip (system on chip, SoC), application-specific integrated circuit (application-specific integrated circuit, ASIC), field programmable gate array (field programmable gate array, FPGA) or programmable logic device (programmable logic device, PLD) in the form of realization.

A tenth aspect provides a multicast transmission system, which includes the multicast transmission device described in the third aspect and the multicast transmission device described in the fourth aspect.

On the basis of the implementation manners provided in the foregoing aspects, the present application may further be combined to provide more implementation manners.

Description of drawings

Fig. 1 is a schematic diagram of a computing cluster applicable to the embodiment of the present application.

FIG. 2 is a schematic diagram of computing devices included in the computing cluster in FIG. 1 .

FIG. 3 is a schematic flowchart of a multicast transmission method 300 provided by an embodiment of the present application.

Fig. 4 is a schematic diagram of a multicast transmission scenario provided by an embodiment of the present application.

FIG. 5 is a schematic flowchart of a multicast transmission method 500 provided by an embodiment of the present application.

FIG. 6 is a schematic diagram of a format of a data packet transmitted based on TCP/IP provided by an embodiment of the present application.

FIG. 7 is a schematic diagram of a window provided by an embodiment of the present application.

FIG. 8 is a schematic flowchart of a multicast transmission method 800 provided by an embodiment of the present application.

FIG. 9 is a schematic diagram of the format of data packets and messages transmitted based on the RoCE protocol provided by the embodiment of the present application.

FIG. 10 is a schematic flowchart of a multicast transmission method 1000 provided by an embodiment of the present application.

FIG. 11 is a schematic flowchart of a multicast transmission method 1100 provided by an embodiment of the present application.

FIG. 12 is a schematic structural diagram of a multicast transmission device 1200 provided by an embodiment of the present application.

FIG. 13 is a schematic diagram of a hardware structure of a multicast transmission device 1300 provided by an embodiment of the present application.

Detailed ways

The technical solutions in the embodiments of the present application will be described below with reference to the accompanying drawings.

The present application presents various aspects, embodiments or features in terms of a system comprising a number of devices, components, modules and the like. It is to be understood and appreciated that the various systems may include additional devices, components, modules, etc. and/or may not include all of the devices, components, modules etc. discussed in connection with the figures. Additionally, combinations of these schemes can also be used.

The network architecture and business scenarios described in the embodiments of the present application are for more clearly illustrating the technical solutions of the embodiments of the present application, and do not constitute limitations on the technical solutions provided by the embodiments of the present application. For the evolution of architecture and the emergence of new business scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

The chip referred to in the embodiment of the present application may be a system chip (system on chip, SoC), may also be a central processing unit (central processor unit, CPU), may also be a network processor (network processor, NP), may also be It is a digital signal processing circuit (digital signal processor, DSP), and it can also be an application processor (application processor, AP), or other integrated chips.

For ease of understanding, the following first introduces related terms and concepts that may be involved in the embodiments of the present application.

1. Multicast

Multicast is a one-to-many communication mode between hosts. Multicast is a technology that allows one or more multicast sources to send the same message to multiple receivers. That is, multicast technology can realize point to multipoint (point to multipoint) multi-point, P2MP) packet forwarding. The multicast source sends a message to a specific multicast address. Unlike a unicast address, a multicast address does not belong to a specific host, but belongs to a group of hosts. A multicast address represents a group, and all receivers who need to receive multicast packets join this group.

2. Unicast

Unicast is a one-to-one communication mode between hosts. The devices in the network select the transmission path according to the destination address contained in the network message, transmit the unicast message to the specified destination, and only forward the received data without will be copied. It can respond to each host in a timely manner.

3. Protocol stack

The protocol stack refers to the sum of all layers of protocols in the network, which vividly reflects the process of data transmission in a network: from the upper layer protocol to the lower layer protocol, and then from the lower layer protocol to the upper layer protocol. In simple terms, a protocol stack (for example, but not limited to, transmission control protocol/internet protocol (transmission control protocol/internet protocol, TCP/IP) stack) is an implementation of a protocol (for example, but not limited to TCP/IP).

4. Remote direct memory access (RDMA)

RDMA technology realizes the direct transmission of data buffer data between two nodes during the network transmission process. The local node can directly transmit the data to the memory of the remote node through the network, bypassing multiple memory copies in the operating system. Compared with traditional network transmission, RDMA does not require the intervention of the operating system and TCP/IP protocol, and can easily realize ultra-low-latency data processing and ultra-high-throughput transmission without the intervention of remote node CPU and other resources. The processing and migration of data consumes excessive resources.

The working process of RDMA is as follows:

1) When an application executes an RDMA read or write request, the RDMA request is sent from the application running in user space to the local network interface controller (NIC) without any kernel memory involvement. Wherein, the network card is also called a network interface controller.

2) The local NIC reads the buffered content and sends it over the network to the remote NIC.

3) The RDMA information transmitted on the network contains the target virtual address, the memory key and the data itself. Request completion can either be handled entirely in user space (by polling the user-level completion queue), or through kernel memory in the case where the application sleeps until the request completes. RDMA operations enable applications to read data from or write data to a remote application's memory.

4) The target NIC confirms the memory key and directly writes the data into the application cache. The remote virtual memory address used for the operation is included in the RDMA information.

5. Ethernet-based RDMA (RDMA over converged ethernet, RoCE)

RoCE supports the use of RDMA technology over standard Ethernet infrastructure. In the RoCE technology, the RDMA network card (RDMA-aware network interface controller, RNIC) offloads all the protocol stack (that is, the user datagram protocol (UDP)) to the ASIC chip of the RNIC for implementation. Wherein, offloading refers to handing over the processing work of the protocol stack on the host side from the CPU of the host to the network card for processing. Moreover, the user cache (buffer) to the network card cache on the host is also directly moved to the network card through direct memory access (DMA), and then the network card transmits the data to the peer through the network protocol UDP. After the peer end receives the data, it also directly receives the data on the network card, and directly DMAs it into the user cache. In this way, there is no CPU and memory copy involved in the whole process, thereby reducing the TCP/IP processing burden of the CPU and server I/O system, and eliminating the network bottleneck of the server.

6. Direct memory access (DMA)

DMA refers to a high-speed data transfer operation that allows direct reading and writing of data between IO devices and memory, neither through the CPU nor CPU intervention. In other words, DMA refers to an interface technology in which IO devices directly exchange data with system memory without going through the CPU. It can be understood that the DMA method is a working method in which the I/O exchange is completely performed by the hardware. The RDMA method is to directly transfer data to the storage area of the computer through the network, that is, to quickly move the data from one system to the remote system memory. That is to say, the RDMA method is a working method that combines software and hardware to perform I/O exchange.

Below, specifically introduce the relevant technical solutions of the present application:

A computing cluster (may be referred to simply as a cluster) is a computing system. Computing clusters connect a group of computing devices to work closely together to complete computing tasks. Individual computing devices in a computing cluster may be referred to as nodes. Fig. 1 is a schematic diagram of a computing cluster applicable to the embodiment of the present application. As shown in FIG. 1 , the computing cluster 100 includes but is not limited to multiple computing devices. FIG. 1 takes the computing cluster 100 including six computing devices as an example for illustration, and the six computing devices are computing device 111, computing device 112, computing device Device 113 , computing device 114 , computing device 115 , and computing device 116 . Wherein, any two computing devices (for example, computing device 111 and computing device 115 ) may communicate through the network 110 .

The computing device shown in FIG. 1 will be introduced below with reference to FIG. 2 . The computing device 100 shown in FIG. 2 may be any one of the computing devices 111 to 116 shown in FIG. 1 .

The computing device 200 shown in FIG. 2 includes a host 210 , an input-output (input output, IO) interconnection channel 220 and an IO device 230 . Wherein, the host 210 can be connected to the IO device 230 through the IO interconnection channel 220 .

The host 210 may be a computing core and a control core, and is configured to send a pending request to the IO device 230 and receive a processing result of the pending request sent by the IO device 230 . The host 210 includes a first processor 211 and a first memory 212, the first processor 211 may be a central processing unit (central processing unit, CPU), and the first processor 211 may also be other general-purpose processors, digital signal processing (digital signal processor, DSP), application specific integrated circuit (ASIC), field-programmable gate array (field-programmable gate array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general purpose processor may be a microprocessor or any conventional processor or the like. The first processor 211 may also be an on-chip (system of chip, SoC) or embedded processor. The first processor 211 has functions such as processing instructions, executing operations, and processing data. The first processor 211 can allocate independent memory resources for multiple processes, so as to run multiple processes. The address space addressable by the first processor 211 includes the first memory 212 . The first memory 212 may be implemented by a random access device (random access memory, RAM), a hard disk (for example, a solid state disk (solid state disk, SSD)) or other storage media. The first memory 212 can be used to store program codes of a plurality of processes.

The IO interconnection channel 220 is an interconnection mechanism between the host computer 210 and the IO device 230, for example, a high-speed serial computer expansion bus standard (peripheral component interconnect express, PCIe), a computer express link (compute express link, CXL), and a cache coherent interconnection Protocol (cache coherent interconnect for accelerators, CCIX), unified bus (unified bus, UB or Ubus) and so on.

The IO device 230 refers to hardware that can perform data transmission with the host 210 and/or other computing devices. Process the request. The IO device 230 is also configured to send the processing result of the processing request to the host 210 . In the embodiment of the present application, the IO device 230 has network capabilities and can realize the function of a protocol stack. Wherein the IO device 230 has a protocol stack (for example, TCP/IP protocol stack) function, which means that the IO device 230 has various protocols (for example, TCP/IP, UDP and Ethernet protocols, etc.) included in the protocol stack to realize pending requests The ability to handle all protocols transmitted in the network. The IO device 230 may be a network interface controller (network interface controller, NIC), a smart NIC (smart-NIC), an RDMA network card (RDMA-aware network interface controller, RNIC), a host bus adapter (host bus adapter, HBA), a host Channel adapter (host channel adapter, HCA), accelerator (accelerator), data processing unit (data processing unit, DPU), image processing unit (graphics processing unit, GPU), artificial intelligence (artificial intelligence, AI) equipment, software definition foundation At least one of the facilities (software defined infrastructure, SDI), etc. The IO device 230 may include a second processor 231 and a second memory 232 . The second storage 232 may be implemented by a random access device (random access memory, RAM), a hard disk (for example, SSD) or other storage media.

In the following, the multicast transmission method provided by the present application will be introduced in detail with reference to FIG. 3 .

FIG. 3 is a schematic flowchart of a multicast transmission method 300 provided by an embodiment of the present application. As shown in FIG. 3 , the method 300 includes step 310 to step 330 . It can be understood that, in the following, the method 300 is applied to a first computing device as an example. The first computing device includes a first input and output IO device and a first host. The first host includes a processor and a memory. The first IO device communicate with the first host through the IO network. Exemplarily, the first computing device in the method 300 may be any computing device in the computing cluster shown in FIG. 1 above, and the multicast group in the method 300 may include at least two multicast members, as shown in FIG. 1 For example, when the first computing device is computing device 111, the at least two multicast members may be computing device 112 and computing device 116, respectively. Exemplarily, the structure of the first computing device and the structure of any multicast member of the multicast group may be as shown in FIG. 2 above. Next, step 310 to step 330 will be described in detail.

Step 310, the first IO device obtains the first request.

Step 320, the first IO device generates a first data packet according to the first information and the first request, the first information includes multicast information, and the multicast information is used to identify the communication between the multicast members of the multicast group and the first computing device The link connection, the first data packet carries the multicast information and the data to be written corresponding to the first request.

The multicast information includes a multicast identifier, which is obtained when the first computing device establishes a link connection with the multicast group. In this embodiment of the present application, a multicast group may include at least 2 multicast members, that is, a multicast group may include 2 or more multicast group members. The multicast information is used to identify the link connections between the multicast members of the multicast group and the first computing device, that is, the multicast information is used to identify at least two link connections.

The first data packet further includes a port number, and the port number is used to indicate that the transmission mode of the first data packet is a multicast transmission mode. Optionally, the port number is the destination port number of the first data packet, that is, the value of the destination port number of the first data packet is used to indicate that the transmission mode of the first data packet is a multicast transmission mode.

In this embodiment of the present application, no specific limitation is imposed on the execution subject that sends the first request. In an example, the first request is a request sent by an application program running on a processor included in the first host. For ease of description, this scenario is referred to as scenario 1 hereinafter. It can be understood that the first request in Scenario 1 is a write request, and the write request may be used to request to store (write) the data stored in the memory included in the first host to the host corresponding to the multicast member of the multicast group. in memory. In another example, the first request is a request sent by an application program running in a processor included in the second host, the second computing device includes the second host and a second IO device, and the second computing device is a group of a multicast group Broadcasting members, for the convenience of description, this scenario will be recorded as scenario 2 in the following. It can be understood that the first request in Scenario 2 is a read request, and the read request may be used to request to store (read) the data stored in the memory included in the first host to the host corresponding to the multicast member of the multicast group. in memory. In the following, Scenario 1 and Scenario 2 will be introduced in detail.

scene one:

In Scenario 1, the first request is a request sent by an application running on a processor included in the first host. In the first scenario, according to different transmission protocols, it can be divided into two implementation modes. For the convenience of description, these two implementation modes will be recorded as implementation mode 1 and implementation mode 2 in the following. In the following, implementation manner 1 and implementation manner 2 will be introduced in detail.

Implementation method one:

In the first implementation, the transmission protocol used by the first IO device to send the first data packet is TCP/IP.

Optionally, the first information may further include first window information, where the first window information is used to indicate the minimum amount of data that can be processed by the multicast members of the multicast group within the first time period.

Optionally, the first IO device may also update the window information. In an exemplary example, the first IO device may also perform the following operations: update the second window information to the first window information, and the second window information is that the multicast members of the multicast group before the first time period The minimum amount of data that can be processed within a period of time. The minimum amount of data that can be processed by the multicast members of the multicast group indicated by the second window information is different from the minimum amount of data that can be processed by the multicast members of the multicast group indicated by the first window information. In this implementation manner, by setting the first IO device to update the window information in a timely manner within a preset time period, the amount of data carried by the first data packet can be accurately controlled to realize network flow control.

Optionally, when the data to be written carried by the first data packet is part of the data corresponding to the first request, the first IO device may also generate at least one data packet according to the first information and the first request, and the at least One data packet carries the multicast information and all data corresponding to the first request except the data to be written. When the data to be written carried in the first data packet is all data in the data corresponding to the first request, the first IO device may generate only one first data packet according to the first information and the first request. Optionally, when the multicast member of the multicast group fails to receive the first data packet, the first IO device may also generate a data packet, and the data packet carries all the data corresponding to the first request. Data and unicast information, where the unicast information is used to indicate the multicast members in the multicast group that have not successfully received the first data packet.

Implementation method two:

In the second implementation manner, the transmission protocol used by the first IO device to send the first data packet is Ethernet-based RDMA (for example, but not limited to, RoCEv2).

Optionally, the data to be written is a part of the data corresponding to the first request, and the first information further includes indication information, and the indication information is used to indicate to encapsulate the data to be written. In this implementation, before the first IO device generates the first data packet according to the first information and the first request, the method further includes: the first IO device sends an IO write command to the multicast members of the multicast group, and the IO write The command is used to instruct to store the data corresponding to the first request into a first registration area (memory registration, MR), the data corresponding to the first request is located in the memory included in the first host included in the first computing device, and the first MR is The storage area in the memory included in the second host is registered to the storage area in the memory of the second IO device, the second computing device includes the second host and the second IO device, and the second computing device is a multicast member of the multicast group; The first IO device receives the indication information sent by the multicast member of the multicast group. Wherein, the IO write command includes a second key value and second position information, the second key value is used to identify the second MR, and the second position information is used to indicate the position of the data to be written in the second MR.

Optionally, the first information further includes a credit value, and the credit value is used to indicate the minimum number of requests that the multicast member of the multicast group can handle within the second time period, and the basic transport header (base transport) of the first data packet header, BTH) carries the credit value.

Scene two:

In the second scenario, the first request is a read request, and the first request may be used to request to store (read) the data stored in the memory included in the first host into the memory of the host corresponding to the multicast member of the multicast group. In this scenario, the transmission protocol used by the first IO device to send the first data packet is Ethernet-based remote direct data access (RDMA).

Optionally, obtaining the first request by the first IO device includes: the first IO device receives the first request sent by the multicast member of the multicast group, and the first request is used to indicate that the data to be written in the second MR Stored in the storage area of the multicast members of the multicast group, the second MR is a storage area in the memory included in the first host and registered in the storage area in the memory of the first IO device, and the first computing device also includes the first host. Optionally, in this implementation, the window size of the first shared receive queue (shared receive queue, SRQ) is non-zero, and the non-zero window size of the first SRQ is used to indicate the receive queue elements (receive queue) available in the first SRQ. element.RQE) is non-zero, the available RQE in the first SRQ is used to process the data packet received by the first IO device, and the first SRQ is stored in the memory included in the first host.

Optionally, the first request includes a first key value, first location information and a preset field, the first key value is used to identify the second MR, and the first location information is used to indicate that the data to be written is in the second MR In the position, the value of the preset field is used to indicate the first request.

Step 330, the first IO device sends the first data packet to the multicast members of the multicast group.

The first IO device sends a first data packet to the multicast members of the multicast group. Optionally, the first data packet may be carried in the form of a message when transmitted in the network, and a message is generated according to the first data packet The method is not specifically limited. It can be understood that when the data corresponding to the first request is encapsulated into multiple data packets, and the multiple data packets include the first data packet, a message during transmission in the network may include the multiple data packets.

Optionally, the first IO device sending the first data packet to the multicast members of the multicast group includes: the first IO device sends the first data packet to the forwarding device, and the forwarding device is used to copy the first data packet, and The copied first data packet is forwarded to the multicast members of the multicast group, and the link connection between the multicast members of the multicast group and the first computing device includes a forwarding device. The number of forwarding devices included between the first computing device and the multicast members of the multicast group is not specifically limited, for example, 1, 2 or 5 forwarding devices may be included. It can be understood that the number of data packets obtained by duplicating the first data packet by the forwarding device is associated with the number of multicast members of the multicast group. Each first data packet after copying carries the data to be written corresponding to the first request, but the destination address carried by each first data packet after copying may be different, that is, each first data packet after copying The carried destination address is in one-to-one correspondence with a multicast member in the multicast group.

Optionally, the multicast members of the multicast group include a second computing device, and the second computing device is different from the first computing device. After the above step 330, the following operations may also be performed: the first IO device receives the information sent by the second computing device. The second request, the second request is used to request to obtain the data to be written corresponding to the first request carried by the first data packet; the first IO device sends the second data packet to the second computing device, and the second data packet carries the data to be written data, and the port number included in the second data packet is used to indicate that the transmission mode for transmitting the second data packet is a unicast transmission mode. Wherein, the port number included in the second data packet may be the destination port number included in the second data packet, that is, the value of the destination port number of the second data packet may be used to indicate that the transmission mode of the second data packet is unicast transmission Way.

Optionally, the multicast members of the multicast group only include the second computing device and the third computing device, and the third computing device is different from any two computing devices in the second computing device and the first computing device. In the above step 330 After that, the following operations can also be performed: after the first IO device receives the first completion message and the second completion message, the first IO device sends a third completion message to the processor included in the first host, and the third completion message is used to indicate The first request has been successfully executed, the first completion message is used to indicate that the second computing device has successfully executed the first request, and the second completion message is used to indicate that the third computing device has successfully executed the first request.

Optionally, in the implementation mode 1 of the above-mentioned scenario 1, after step 330, the following steps may also be included: when each multicast member of the multicast group receives the third message, the first IO device receives The confirmation data packet sent by each multicast member, the third message is used to indicate that the data packet carried in the third message includes the last data packet of the data to be written corresponding to the first request, and the confirmation data packet indicates Each multicast member successfully receives all data packets carrying the data to be written corresponding to the first request. This implementation of sending confirmation packets is also called cumulative acknowledgment (cumulative acknowledgment), that is, the receiving side will send an acknowledgment message to the sending end after receiving multiple messages, and the acknowledgment message indicates that the receiving side The plurality of messages is received. In this implementation manner, the first IO device can know the situation of the data packet corresponding to the first request received by the multicast member of the multicast group only by obtaining one confirmation data packet, which is beneficial to improve the efficiency of data transmission.

It should be understood that the method 300 shown in FIG. 3 above is only for illustration, and does not constitute any limitation to the multicast transmission method provided in the embodiment of the present application. For content that is not described in detail in the above method 300, refer to the relevant descriptions in the method embodiments corresponding to FIG. 4 to FIG. 11 below, and will not be described in detail here.

In this embodiment of the application, the first IO device may perform data encapsulation according to the first request and the first information to generate the first data packet, and send the first data packet to the multicast members of the multicast group instead of passing the first The processor included in the host machine encapsulates and generates data packets according to the work request, which avoids the problems of high complexity, large delay and low transmission efficiency in the traditional multicast transmission method. By setting the first data packet sent by the first IO device to carry the multicast information for identifying the link connection between the multicast members of the multicast group and the first computing device, reliable transmission of multicast data can be ensured. Therefore, the multicast transmission method provided by the embodiment of the present application can realize reliable and efficient multicast data transmission.

Fig. 4 is a schematic diagram of a multicast transmission scenario provided by an embodiment of the present application.

As shown in FIG. 4, suppose that the source computing device 1 (abbreviated as INI 1) in FIG. 4 is the computing device 111 in the computing cluster 100 shown in FIG. 1, and the target computing device 1 (abbreviated as TGT 1) The computing device 114 in the computing cluster 100, the target computing device 2 (abbreviated as TGT 2) is the computing device 115 in the computing cluster 100. In addition, for ease of description, it is assumed that at least one application program, called application program 1, exists in the host computer 10 below. As shown in Figure 4, under the situation that host computer 10 has at least one application program 1, create queue team (queue pairs, QP) 1 and complete queue (complete queue, CQ) 1 in the storage area of host computer 10. QP1 includes send queue (send queue, SQ) 1 and shared receive queue (shared receive queue, SRQ) 1. SRQ1 may be shared by multiple TGTs (ie, TGT 1 and TGT 2). SQ1 is used to store the work request (work request, WR) WR generated by the application program 1 in the host computer 10. CQ1 is used to store the processing result of the WR that the IO device 11 has completed. SRQ1 is associated with INI 1. Furthermore, QP2 and CQ2 are created in the storage area of the host 20, and QP2 includes RQ2 and SQ2. QP3 and CQ3 are created in memory included in host 30, QP3 including RQ3 and SQ3. The pending requests stored in the SQ may be referred to as send queue elements (send queue element, SQE), that is, one SQE is used for one pending request. The pending requests stored in the RQ can be called receive queue elements (receive queue element, RQE), that is, one RQE is used to store a received pending request. The processing result of the pending request stored in the CQ may be called a complete queue element (CQE), that is, a CQE is used to store the processing result of a pending request. Exemplarily, when the structure of any one of host 10, host 20, and host 30 is the structure of host 210 shown in FIG. 2 , the memory included in any one of host 10, host 20, and host 30 may be the first memory 212.

In the scenario shown in Figure 4, the multicast members of multicast group 1 include TGT 1 and TGT 2. TGT 1 includes a host 20 and an IO device 21, and TGT 2 includes a host 30 and an IO device 31. INI 1 can send information (for example, work requests or data packets) to multicast members of multicast group 1 through multicast, and receive information sent by multicast members of multicast group 1 through unicast. Any multicast member of multicast group 1 can also send information to INI 1 through unicast. In the embodiment of the present application, the link connection between the INI 1 shown in Figure 4 and all multicast members (that is, TGT1 and TGT 2) of the multicast group 1 shown in Figure 4 can be uniquely identified by the multicast identifier, for ease of description , hereinafter the multicast identifier is recorded as a replicat group (replicat group, RG) identifier. The INI can use an RG identifier to uniquely identify the multicast link connections between the INI and all multicast members of a multicast group. Any multicast member of the multicast group may use another RG identifier to uniquely identify the unicast link connection between the multicast member and the INI when the multicast member is a multicast member of the multicast group. The above-mentioned one RG identifier may be different from the above-mentioned other RG identifier. Optionally, the foregoing one RG identifier may also be the same as the foregoing another RG identifier.

For example, in Figure 4, INI 1 can use the ini_rg_id_1 identifier to uniquely identify the link connection between INI 1 and all multicast members of multicast group 1, the link connection includes link 1 and link 2, link 1 is the link from INI 1 to forwarding device 1, forwarding device 1 to TGT 1, and link 2 is the link from INI 1 to forwarding device 1, and forwarding device 1 to TGT 2. TGT 1 can use the tgt_rg_id_a identifier to uniquely identify the unicast link 1 between TGT 1 and INI 1 when TGT 1 is a member of multicast group 1. TGT 2 can use the tgt_rg_id_b identifier to uniquely identify the unicast link 2 between TGT 2 and INI 1 when TGT 2 is a member of multicast group 1. Any one of the ini_rg_id_1 identifier, tgt_rg_id_a identifier, and tgt_rg_id_b identifier may be different.

In the embodiment of this application, one INI is connected to the link between all multicast members of a multicast group, one INI is connected to the unicast link of each multicast member in the multicast group, and the RG , can be called a reliable point-to-multipoint connection (reliable point to multi-point, rP2M). Also take Figure 4 as an example, that is, one rP2MP includes: a unicast link connection (that is, a unicast link 1 and a unicast link 2), a multicast link connection (that is, a link 1 and a link 2 , link 1 is the link from INI 1 to forwarding device 1, forwarding device 1 to TGT 1, link 2 is the link from INI 1 to forwarding device 1, forwarding device 1 to TGT 2.), and RG (including INI 1 corresponds to the ini_rg_id_1 identifier, TGT 1 corresponds to the tgt_rg_id_a identifier and TGT2 corresponds to the tgt_rg_id_b identifier). An INI can also create multicast link connections with multiple multicast groups respectively. Correspondingly, multiple RG identifiers may be allocated to the one INI, and the multiple RG identifiers correspond to multiple multicast link connections one by one. A TGT can also establish multiple unicast link connections with multiple INIs respectively. Correspondingly, multiple RG identifiers may be allocated to the one TGT, and the multiple RG identifiers are in one-to-one correspondence with the multiple unicast link connections. A multicast group may include at least 2 multicast members, and optionally, the multicast group may also include more multicast members. For example, the multicast group 1 shown in FIG. 4 may also include 3, 4, 5, 9 or 20 multicast members.

It should be understood that the multicast transmission scenario shown in FIG. 4 above is only for illustration, and does not constitute any limitation to the multicast scenario applicable to the multicast transmission method provided in this application. For example, the multicast group in the multicast scenario shown in FIG. 4 may also include more multicast members, such as 3 or 5 members. In Figure 4, the unicast link 1 from INI 1 to TGT 1 does not include an intermediate forwarding device as an example. Optionally, in other implementation manners, the unicast link 1 from the INI 1 to the TGT 1 may also include one or more forwarding devices.

Taking the application scenario shown in FIG. 4 as an example, a specific implementation manner of the multicast transmission method provided by the embodiment of the present application will be described in detail below in combination with the embodiment in FIG. 5 .

FIG. 5 is a schematic flowchart of a multicast transmission method 500 provided by an embodiment of the present application. It should be understood that, for ease of description, only INI 1 and TGT 1 in FIG. 4 are taken as examples in FIG. 5 for illustration. The operations performed by TGT 2 in FIG. 4 are similar in principle to the operations performed by TGT1. For specific details, please refer to the operations corresponding to TGT 1 in method 500. As shown in FIG. 5 , the method 500 may include step 510 to step 592 , and step 510 to step 592 will be described in detail below.

Step 510, the host 10 obtains a work queue element (work queue element, WQE) 1, and puts WQE1 into SQ1, WQE1 is used to carry a write request 1, and the write request 1 carries the data 1 to be written and is located in the memory included in the host 10 source address.

Wherein, the acquisition of WQE1 by the host 10 may include the following steps: the application program 1 in the processor included in the host 10 generates WR1; the host 10 schedules the interface provided by the driver to convert WR1 into WQE1. WR1 and WQE1 carry the same information, but the format is different. It is understandable that after WQE1 is put into SQ1, WQE1 is also called SQE1. For ease of description, WQE1 obtained from SQ1 is collectively referred to as SQE1 hereinafter. For example, SQE1 may be used to indicate a write request 1, and the write request 1 is used to request that the data stored in the memory included in the host 10 with an address of 0x12345678 and a length of 10 bytes be written to the multicast members of the multicast group 1 in the memory of the corresponding computing device. Exemplarily, when the structure of the INI 1 in FIG. 4 is the structure of the computing device shown in FIG. 2, the application program 1 in the processor included in the host 10 may be the first processor included in the host 210 shown in FIG. 2 211, the memory included in the host 10 may be the first memory 212 shown in FIG. 2, the IO device 11 may be the IO device 230 shown in FIG. Two memory 232.

In step 520, the IO device 11 acquires SQE1 from SQ1.

Wherein, the format of SQE1 obtained by IO device 11 is a scatter gather list (SGL) format, that is, SQE1 includes the source address of data 1 to be written and the source address of SQE1, and the source address of data 1 to be written It may be the address of the data to be written 1 located in the memory included in the host 10 , and the source address of SQE1 may be the address of SQE1 located in the memory included in the host 10 .

In step 530, the IO device 11 encapsulates the data 1 to be written indicated by the SQE1 according to the SQE1, the multicast information and the window information, and generates a data packet 1.

Multicast information refers to the multicast information between INI 1 and the multicast members of multicast group 1, the multicast information includes but not limited to: the media access control (media access control, MAC) address of INI1, the single address of INI1 Broadcast IP address, MAC address of multicast group, IP address of multicast group, RG identifier of INI 1. The RG identifier of INI 1 indicates the multicast link connection between INI 1 and the multicast members of multicast group 1. The window information includes: TCP window information and congestion window information, the TCP window information is used to indicate the minimum TCP window size among all TCP window sizes corresponding to all members of the multicast group 1, and the congestion window information is used to indicate all members of the multicast group 1 The minimum congestion window size among all the congestion window sizes corresponding to the member. The algorithm for determining the congestion window in this embodiment of the present application is not specifically limited. For example, congestion control algorithms include, but are not limited to, slow start algorithms, congestion avoidance algorithms, fast retransmission algorithms, and fast recovery algorithms.

In the above step 530, the IO device 11 offloads the function of the TCP/IP protocol stack, that is, the IO device 11 can realize the function of the TCP/IP protocol stack. That is to say, the IO device 11 can use the TCP/IP protocol stack to encapsulate the data 1 to be written corresponding to the write request 1 into a data packet. Optionally, the IO device 11 may also parse the received data packet to obtain the content carried in the header of the data packet, the data carried in the payload of the data packet, and the like. Exemplarily, the format of the data packet 1 obtained after encapsulation by the IO device 11 may be shown in FIG. 6 , and the data packet 1 may sequentially include an Ethernet header, an IP header, a UDP header, a TCP header, a payload, and a frame check sequence. Wherein, the extension header in the TCP header may carry the RG identifier. Optionally, the extension header may also carry a message boundary, which indicates that the transmission protocol transmits the data as an independent message on the network. For the specific content carried in the header of the data packet shown in FIG. 6 , refer to Table 1 below.

Table 1

TGT i (i=1 or 2) in the above Table 1 is any multicast member in the multicast group 1 (that is, TGT 1 or TGT 2). In the above Table 1, the destination UDP port number is rp2m_port_x, which is used to indicate the multicast transmission mode. The destination UDP port number is rp2m_port_y, which is used to indicate the unicast transmission mode. It can be understood that the source UDP port number shown in Table 1 is src_port, and src_port can be a variable value, and the destination UDP port number shown in Table 1 is rp2m_port_x and rp2m_port_y is a fixed value. It should also be understood that, in this embodiment of the present application, the destination UDP port numbers for the same transmission mode (unicast transmission mode or multicast transmission mode) may be the same. For example, if rp2m_port_y is set equal to 1000, the destination UDP port number included in the data packets sent by INI1 to TGT 1 or TGT 2 through the unicast link is equal to 1000, and the packets sent by TGT 1 or TGT 2 to INI 1 through the unicast link The destination UDP port number included in the data packet is also equal to 1000. Optionally, the destination UDP port number may be selected from the Internet Corporation for Assigne Names an Numbers (ICANN) port range. Table 1 shows that the RG identifier assigned to INI 1 is ini_rg_id, and ini_rg_id is used to uniquely identify all link connections between INI 1 and multicast members of multicast group 1, and all link connections include link 1 and Link 2, link 1 is the link from INI 1 to forwarding device 1, forwarding device 1 to TGT 1, and link 2 is the link from INI 1 to forwarding device 1, and forwarding device 1 to TGT 2. Table 1 also shows that the RG identifier assigned to TGT i is tgt_rg_id[i], and when tgt_rg_id[i] can be used to uniquely identify the TGT i as a multicast member of multicast group 1, the TGT i to INI I unicast link connection.

In one implementation, in the above step 530, the IO device 11 encapsulates the data to be written 1 indicated by SQE1 according to SQE1, multicast information and window information, and generates a data packet 1, which may include the following steps: IO device 11 Acquire the data 1 to be written from the position where the data 1 to be written is stored in the host 10 according to SQE1; the IO device 11 determines the sending window size of the IO device 11 according to the window information, and the size of the TCP window indicated by the TCP window information is non-zero; the IO device 11 Encapsulate the data 1 to be written into the payload of the data packet 1, and encapsulate the multicast information into the data packet header of the data packet 1 to generate a data packet 1, the amount of data carried by the payload of the data packet 1 is equal to that of the IO device 11 Send window size. In the above implementation, when the TCP window size of any one of the multicast members of the above multicast group 1 is equal to zero, the IO device 11 will suspend encapsulating the data packet and sending the encapsulated data packet. At this time, the IO device 11 can use the unicast channel Detect and individually update the TCP window size of any multicast member of multicast group 1. Until the IO device 11 determines that the TCP window size of any multicast member of the multicast group 1 is non-zero, the IO device 11 performs data encapsulation according to the window information to generate a data packet and sends the data packet. Wherein, when the IO device 11 sends a data packet to any multicast member of the multicast group in a unicast manner, the parameters carried in the header of the data packet can refer to the content in the third column in Table 1 above. Wherein, the manner in which the INI 1 obtains the TCP window size and the congestion window size of each multicast member of the multicast group 1 is not specifically limited. In one example, when INI 1 establishes a TCP connection with each multicast member of multicast group 1, each multicast member of multicast group 1 will inform INI 1 of the TCP window size and congestion window of each multicast member size. In another example, the TGT (that is, any multicast member of the multicast group 1) can generate a regular TCP window size according to the state of its receiving buffer, and update the congestion window size according to the state of the network where it is located. INI 1 maintains the TCP window size and congestion window size for each multicast member of a given multicast group based on the individual window notifications of the TGT.

Take (1) in FIG. 7 as an example to illustrate the content carried by the above-mentioned data packet 1. The data to be written 1 may correspond to the content of the 10th byte to the 25th byte in (1) in FIG. 7 . In one implementation manner, the payload of data packet 1 carries all content from the 10th byte to the 25th byte, and the parameters carried in the header of data packet 1 can refer to the content in the second column in Table 1 above. In this implementation, the sending window size of IO device 11 is 16 bytes, the TCP window size of TGT 1 and the TCP window size of TGT2 can be equal to 16 bytes, the congestion window size of TGT 1 and the congestion window of TGT 2 Size can also be equal to 16 bytes.

In another implementation manner, the amount of data 1 to be written is relatively large, and it is difficult for the buffers of the multicast members of the multicast group 1 to receive the data 1 to be written at one time. Based on this, the IO device 11 encapsulates the data to be written 1 indicated by SQE1 according to SQE1, multicast information and window information, and generates a data packet 1, which may include the following steps: the IO device 11 stores the data to be written from the host 10 according to SQE1. Enter the position of data 1 to obtain data 1 to be written; IO device 11 determines the size of the sending window according to the window information, the size of the TCP window indicated by the TCP window information is non-zero, and the size of the sending window is less than the amount of data to be written into data 1; The IO device 11 obtains the data to be encapsulated 1 from the data to be written 1 according to the size of the sending window, the data to be written 1 includes the data to be encapsulated 1; the IO device 11 encapsulates the data to be encapsulated 1 into the payload of the data packet 1, and The multicast information is encapsulated into the data packet header of the data packet 1 to generate the data packet 1, and the payload of the data packet 1 only carries part of the data to be written into the data 1. In this implementation, after the above step 530, the following steps may also be included: the IO device 11 will also send at least one data packet 2 to the members of the multicast group 1 through the multicast transmission mode, and the content carried in the header of the data packet 2 The same as the content carried in the header of data packet 1, the size of the content carried in the payload of data packet 2 is determined according to the TCP window information and the data to be encapsulated 2, the data to be encapsulated 2 is the data to be written in the data 1 except the data to be encapsulated 1 The data. In the above implementation, when the size of the TCP window indicated by the TCP window information is equal to zero, the IO device 11 suspends encapsulating the message and sending the encapsulated message. At this time, the IO device 11 can use a unicast channel to detect and update the window indicated by the TCP window information separately. TCP window size. Until the IO device 11 determines that the size of the TCP window indicated by the TCP window information is not zero, the IO device 11 performs data encapsulation according to the window information to generate a message and sends the message. Wherein, when the IO device 11 sends a message to any member of the multicast group through unicast, the parameters carried in the header of the data packet included in the message can refer to the content in the third column in Table 1 above.

Take (2) in Figure 7 as an example to illustrate the above steps. Assume that the IO device 11 sends data packet 1 and data packet 2 to the multicast group 1 through multicast transmission, and then the data to be written 1 can be transmitted to the multicast group. The purpose of the multicast membership of group 1. When INI 1 establishes a TCP connection with the multicast member of multicast group 1, the TCP window size of TGT 1 is 7 bytes, and the TCP window size of TGT 1 is 6 bytes. Based on this, the data packet 1 generated by the IO device 11 carries the content of the data to be encapsulated 1 shown in (2) in FIG. 7 . Thereafter, INI 1 receives the TCP window size update of TGT 1 to 10 bytes, and receives the TCP window size update of TGT 1 to 15 bytes. Based on this, the data packet 2 generated by IO device 11 carries the data shown in Figure 7. The content in the data to be packaged 2 shown in (2) in. That is to say, the IO device 11 sends two data packets (that is, the data packet 1 and the data packet 2 ) to realize the purpose of transmitting the data 1 to be written to the multicast members of the multicast group 1 .

In step 540, the IO device 11 sends the data packet 1 to the multicast members of the multicast group 1.

The multicast members of multicast group 1 include TGT 1 and TGT 2. Correspondingly, both the IO device 21 and the IO device 31 will receive the data packet 1 . The IO device 31 is an IO device included in the TGT 2. It can be understood that the TGT 2 is not shown in FIG. 5 .

In some implementations, the IO device 11 encapsulates the data to be written 1 to obtain only one data packet 1, after which the IO device 11 can send the data packet 1 through multicast, that is, the information carried in the header of the data packet 1 can be as described above shown in the second column of Table 1.

In other implementation manners, the IO device 11 encapsulates the data to be written 1 to obtain the data packet 1 and the data packet 2, and then the IO device 11 sends the data packet 1 and the data packet 2 through multicast. In this implementation manner, the timing of sending the data packet 1 and the data packet 2 by the IO device 11 is not specifically limited. In an example, the IO device 11 may send the data packet 2 to the multicast group 1 through multicast transmission after receiving the acknowledgment (ACK) information of the data packet 1 sent by each member of the multicast group 1 . In another example, after IO device 11 sends data packet 1 to multicast group 1 through multicast transmission, it may send data packet 2 to multicast group 1 through multicast transmission. In this implementation, IO device 11 It is not necessary to send data packet 2 after receiving the ACK information of data packet 1 sent by each multicast member of multicast group 1, and each multicast member of multicast group 1 does not need to receive a message every time Just send an ACK message afterwards, each multicast member (for example, TGT 1) of this multicast group 1 can send another ACK message after receiving a plurality of messages, and this ACK message can represent this multicast group 1 Each multicast member receives the plurality of messages. This method of sending confirmation information is also called the cumulative confirmation method, that is, the receiving side will send a confirmation message to the sending terminal after receiving multiple messages, and the confirmation message indicates that the receiving side has received the multiple messages. information. In yet another example, an acknowledgment sequence number (ACK number) field of the TCP header of the message may also be selected to be set, and each message may carry a unique identifier to help message aggregation at the receiving side.

Optionally, after the above-mentioned step 540, the following operations can also be performed: a timer (timer) is respectively set for each multicast member of the multicast group 1 on the INI 1 side, that is, timer 1 corresponds to TGT 1, and timer 1 2 corresponds to TGT 2 to ensure reliable data transmission. After the IO device 11 sends a data packet (for example, data packet 1) to the multicast member of the multicast group 1, the timing of timer 1 and timer 2 is started to determine the round trip time (round trip time, RTT). When the IO device 11 determines that the RTT of a TGT exceeds a retransmission timeout (retransmission time out, RTO), the IO device 11 may resend the message to the TGT using a unicast link. At this time, the parameters encapsulated in the data packet sent by the IO device 11 can refer to the content in the third column in the above Table 1.

For the convenience of description, in the following, INI 1 sends a data packet to the multicast members of multicast group 1 through multicast transmission as an example. That is to say, any multicast member of the multicast group 1 receives the data packet, that is, receives all the contents of the data 1 to be written.

Step 550, the IO device 21 parses the data packet 1, obtains the data to be written 1, and generates rqe1, which carries the processing result 1 of SQE1, and the processing result 1 of SQE1 indicates that the data to be written in the memory of the IO device 21 1 is written into the memory of the host computer 20.

The step of parsing the data packet 1 by the IO device 21 in the above step 550 and the step of generating the data packet 1 by the IO device 11 in the above step 530 are reciprocal. Taking Fig. 2 as an example, when the structure of the TGT 1 is the structure of the computing device shown in Fig. 2, the memory of the host computer 20 can be the first memory 212 shown in Fig. 2, and the memory of the IO device 21 can be the first memory 212 shown in Fig. 2 . The second memory 232.

Optionally, when CQ1 is shared by multiple QPs on the INI 1 side, rqe1 also needs to carry the index of SQ1 (identity, ID) and SQE1, and the index of SQE1 indicates the position of SQE1 in SQ1.

Step 560 , the IO device 21 sends rqel to the host 20 . Correspondingly, the host 20 stores the received rqel into RQ2.

Wherein, the format of rqel is SGL format, that is, rqel includes the source address of the data to be written 1 located in the memory of the IO device 21 and the source address of rqel located in the memory of the IO device 21 .

Step 570 , the host 20 acquires rqel from RQ2 , reads the data 1 to be written from the memory of the IO device 21 according to rqel , and stores the data 1 to be written into the memory of the host 20 .

In some implementation manners, RQ2 is associated with the memory in which the data 1 to be written is stored in the host 20 , that is, RQ2 is used to indicate the memory of the host 20 .

In step 580, the IO device 21 sends an acknowledgment ACK message 1 to the IO device 11 in a unicast manner, and the ACK message 1 indicates that the IO device 11 has successfully received the data 1 carried in the data packet 1 to be written.

In the above step 580, the parameters carried in the header of the ACK message 1 can refer to the content in the fourth column in the above Table 1. Optionally, the ACK message 1 may also carry the TCP window size and the congestion window size. An ACK message 1 needs to consume a shared receive queue element (SRQE) in SRQ1. Optionally, in some implementations, after the IO device 11 receives the ACK message 1, it is found that all SRQEs in the SRQ1 of the IO device 11 are occupied (that is, the receiving window of the IO device 11 is equal to zero), and at this time the IO device 11 may also discard the ACK message 1, and send the receiving window of the IO device 11 equal to zero to the multicast members of the multicast group 1, so that the multicast members of the multicast group 1 suspend sending ACK messages. Until there is an unused SRQE in SRQ1 of the IO device 11 , the IO device 11 sends the updated receiving window of the IO device 11 to the multicast members of the multicast group 1 . Correspondingly, after the multicast members of the multicast group 1 know that the receiving window of the IO device 11 is not zero, the multicast members of the multicast group 1 will continue to send ACK messages to the IO device 11 .

The IO device 21 executes the above step 580, and correspondingly, the IO device 11 receives the confirmation ACK message 1 . Wherein the IO device 11 receives the ACK messages from the multicast group members of the multicast group 1 in strict order, and the IO device 11 can determine whether the out-of-order phenomenon of the ACK message occurs according to the TCP sequence number (sequence number) carried in the ACK message. When the IO device 11 determines that the currently received ACK message is not the ACK message that the IO device 11 is currently expecting to receive, the IO device 11 may discard the ACK message and instruct the corresponding multicast member of the multicast group to regenerate the ACK message.

Step 590 , the host 20 sends SQE2 to the IO device 21 .

Wherein, the SQE2 format is the SGL format, that is, SQE2 carries the location of SQE2 in the memory of the host 20 and the location of the data to be written 1 in the memory of the host 20 . SQE2 indicates that the data request corresponding to SQE1 has been executed.

Correspondingly, the IO device 21 receives the SQE2, and the IO device 21 analyzes the SQE2 to determine that the host 20 has successfully received the data 1 to be written indicated by the SQE1.

In step 591, the IO device 21 sends an IO completion message 1 to the IO device 11, and the IO completion message 1 indicates that the execution of the SQE1 is completed.

Before step 591, after receiving the SQE2 sent by the host 20, the IO device analyzes the SQE2, determines that the data 1 to be written has been successfully written into the memory of the host 20, and generates an IO completion message 1.

In step 592, the IO device 11 sends RQE1 to the host 10, and RQE1 indicates that the execution of SQE1 is completed.

Correspondingly, the host 10 receives RQE1, and stores the received RQE1 into SRQ1, and the RQE1 stored in SRQ1 is also called SRQE1.

In the above step 510 to step 592, the transmission processing flow between INI 1 and TGT 1 is taken as an example for introduction. It can be understood that the transmission processing flow between INI 1 and TGT 2 is similar to the transmission processing flow between INI 1 and TGT 1. It can be understood that, the above steps are all described by taking successful reception as an example. Optionally, the success of reception in the above steps can also be replaced by failure of reception. In this implementation, INI 1 needs to resend the unsuccessfully received message through multicast or unicast. It can be understood that, in the above step 510 to step 592, INI 1 sends a data packet 1 to TGT 1 as an example for description. Optionally, the above method is also applicable when INI 1 needs to send a plurality of data packets to TGT 1. In the example of implementing a data request corresponding to WQE. It can be understood that, in the above steps, TGT 1 and TGT 2 are described as multicast members of multicast group 1 as an example. Optionally, multicast group 1 may also include more multicast members.

It can be understood that, when the above-mentioned IO device 11 sends the data packet 1 to the multicast members of the multicast group 1 in the network, the data packet 1 may be carried in the form of a message (message).

In the embodiment of the present application, a TCP/IP-based multicast transmission method is provided. During concrete implementation, the IO equipment 11 that INI 1 comprises can generate data packet 1 according to write request 1 and multicast information, and send this data packet 1 to the multicast member of multicast group 1, rather than by the main frame 10 that INI 1 comprises The processor in the device performs data encapsulation to generate data packets, which avoids the problems of high complexity, large delay and low transmission efficiency in the prior art when implementing multicast transmission based on TCP/IP. By setting the data packet 1 sent by the IO device 11 included in the INI 1 to carry the multicast information used to identify the link connection between the multicast member of the multicast group 1 and the INI 1, the reliable transmission of the multicast data can be ensured.

Taking the application scenario shown in FIG. 4 as an example below, another specific implementation manner of the multicast transmission method provided by the embodiment of the present application will be described in detail in combination with the embodiment in FIG. 8 .

FIG. 8 is a schematic flowchart of a multicast transmission method 800 provided by an embodiment of the present application. As shown in FIG. 8 , the method 800 includes step 810 to step 880 . Next, step 810 to step 880 will be described in detail. In the embodiment of the present application, any one of the IO devices in the IO device 11, the IO device 21, and the IO device 31 supports RoCE.

Step 810, the host 10 obtains SQE1, and stores SQE1 into SQ1.

SQE1 is used to carry the write request 1, and the write request 1 carries the source address of the data 1 to be written in the memory included in the host 10. The acquisition of SQE1 by the host 10 may include the following steps: the application program 1 in the host 10 generates WR1; the host 10 schedules the interface provided by the driver to convert WR1 into WQE1. WR1 and WQE1 carry the same information, but the format is different. It is understandable that after WQE1 is put into SQ1, WQE1 is also called SQE1. Hereinafter, the WQE1 obtained from SQ1 is collectively referred to as SQE1.

In step 811, the IO device 11 acquires SQE1 from SQ1.

Wherein, the format of SQE1 is SGL format, that is, SQE1 includes the source address of the data 1 to be written and the source address of SQE1, and the source address of the data 1 to be written is that the data 1 to be written is located in the memory included in the host computer 10. Address, the source address of SQE1 is the location of SQE1 in the memory included in the host computer 10 .

Step 812 , the IO device 11 obtains the data 1 to be written from the memory of the host 10 according to the SQE1 , and encapsulates the data 1 to be written to generate a data packet 1 .

The format of data packet 1 can be as shown in (1) among Fig. 9, and this data packet 1 comprises Ethernet head, IP head, UDP head, BTH, load, invariant cyclic redundancy check (invariant cyclic redundancy check) successively check, ICRC) and frame check sequence (frame check sequence, FCS) namely. For the format of the BTH, refer to (2) in FIG. 9 . In (2) in FIG. 9 , the operation code (opration code, Opcode) is used to indicate the type (type) of the packet or the higher layer protocol type in the payload (payload). S is the abbreviation of Solicited Event, indicating that the responder should generate an event. M is the abbreviation of MigReq, which is generally used for migrating status. Pad indicates how many extra bytes are padded into IB PayLoad. TVer is the abbreviation of Transport Header Version, indicating the version number of the package. Partition Key is used to represent the logical memory partition associated with this Packet. rsvd is an abbreviation of reserved, and this field is reserved. Destination QP indicates the sequence number of the destination Queue Pair. A is Acknowledge Request, indicating that the response of this packet can be scheduled by the responder. PSN is the sequence number (packet sequence number, PSN) of the data packet, which is used on the receiving side to determine whether the data packet is out of order. Table 2 below defines the OpCode list corresponding to the RDMA operation. The BTH also includes an extended header, and the extended header may include at least one of the following fields: AETH field, RETH field, ImmDt field and SynETH field. The AETH field is used to indicate the number of available SQEs in SQ1, the RETH field is used to indicate RDMA read operations or RDMA write operations, the ImmDt field is used to indicate carrying immediate data, and the value of the SynETH field is used to uniquely indicate a read request. The read request is a request sent to INI 1 by the multicast member of multicast group 1, that is, the read request is used to request to store the data to be written stored in INI 1 to the storage area corresponding to the multicast member of the multicast group 1 middle.

Table 2

It can be understood that, in this embodiment of the application, the ImmDt field included in the header of the data packet 1 is used to indicate that no immediate value is carried, the RETH field included in the header of the data packet 1 is used to indicate the RDMA write operation, and the SynETH field The value is empty, that is, it is not used to uniquely indicate a read request. In addition, other parameter information included in the header of the data packet 1 can refer to the content in the second column in Table 1 above.

In step 820, the IO device 11 sends a message 1 to the multicast members of the multicast group 1, where the message 1 includes a data packet 1. Correspondingly, the members of the multicast group 1 (ie, the IO device 21 and the IO device 31 ) respectively receive the message 1 .

Exemplarily, (3) in FIG. 9 shows the format of the data packet 1 included in the message 1 . Optionally, in other implementation manners, the IO device 11 may also generate multiple data packets after encapsulating the data 1 to be written, and the multiple data packets jointly carry the data 1 to be written. In this implementation manner, message 1 may also include the multiple data packets. Exemplarily, (3) in FIG. 9 shows a format in which a message includes 2 data packets.

In the above implementation, when the IO device 21 or the IO device 31 receives the message 1, and after checking the message 1, the IO device 21 or the IO device 31 can send an ACK message to the INI 1, and the ACK message indicates that the IO device 21 Or the IO device 31 has correctly received message 1.

Step 830 , the host 20 acquires rqel from RQ2 of the IO device 21 , rqel is used to indicate to write the data 1 to be written into the location 1 in the memory of the host 20 .

The format of rqel is SGL format, that is, rqel carries the source address of rqel in the memory of the IO device 21 and the source address of the data to be written 1 in the memory of the IO device 21 .

Before step 830, the IO device 21 is further configured to perform the following operations: parse the message 1, obtain the data 1 to be written, and store the generated rqel into RQ2.

Step 831 , the host 30 acquires rqe2 from RQ3 of the IO device 31 , and rqe2 is used to indicate to write the data 1 to be written into the location 2 in the memory of the host 30 .

The format of rqe2 is the SGL format, that is, rqe2 carries the source address of rqe2 located in the memory of the IO device 31 and the source address of the data to be written 1 located in the memory of the IO device 31 .

Before step 831, the IO device 31 is further configured to perform the following operations: parse the message 1, obtain the data 1 to be written, and store the generated rqe2 into RQ3.

Step 840 , the host 20 reads the data 1 to be written from the memory of the IO device 21 according to rqel, and stores the data 1 to be written into the location 1 in the memory of the host 20 .

Step 841, the host 30 reads the data 1 to be written from the storage area of the IO device 31 according to rqe2, and stores the data 1 to be written to the location 2 in the memory of the host 30.

In step 850, the IO device 11 sends CQE1 to the host 10, and the CQE1 includes completion information of the SQE1.

The above step 850 is executed after the IO device 11 receives the ACK message sent by the IO device 21 and the IO device 31, and the ACK message is used to indicate that the message 1 is successfully received.

Step 860, the host 20 sends SQE2 to the IO device 21, and the SQE2 includes the completion information of rqel.

Step 861, the IO device 21 sends an IO completion message 1 to the INI 1, and the IO completion message 1 indicates that the IO device 21 successfully executes the write request 1 corresponding to the SQE1. Correspondingly, the IO device 11 of the INI 1 receives the IO completion message 1, and sends an ACK message to the TGT 1.

In step 862, the IO device 11 sends a CQE2 to the host 10 of the host 10, and the CQE1 includes the completion information of the SQE2.

Step 870, the host 30 sends SQE3 to the IO device 31, and SQE2 indicates the completion information of rqe2.

Step 871, the IO device 31 sends an IO completion message 2 to the INI 1, and the IO completion message 2 indicates that the IO device 31 successfully executes the write request 1 corresponding to the SQE1. Correspondingly, the IO device 11 of the INI 1 receives the IO completion message 1, and sends an ACK message to the TGT 2.

In step 870, the IO device 11 sends a CQE3 to the host 10, and the CQE3 includes the completion information of the SQE2.

In step 880, the host 10 sends an IO completion message 3 to the application program 1, and the IO completion message 3 indicates that the write request 1 corresponding to WR1 has been executed.

Before step 880, the following step may also be included: the host 10 generates an IO completion message 3 according to CQE2 and CQE3.

It can be understood that, the execution sequence of the above step 810 to step 880 is only for illustration and does not constitute any limitation. For example, step 841 may also be performed first and then step 840 is performed.

The above implementation manner provides an RDMA-based multicast transmission method, which can realize reliable and efficient data transmission.

Taking the application scenario shown in FIG. 4 as an example below, another specific implementation manner of the multicast transmission method provided by the embodiment of the present application is described in detail in combination with the embodiment in FIG. 10 .

FIG. 10 is a schematic flowchart of a multicast transmission method 1000 provided by an embodiment of the present application. As shown in FIG. 10 , the method 1000 includes step 1010 to step 1080 . Next, step 1010 to step 1080 will be described in detail.

In the embodiment of the present application, any one of the IO devices in the IO device 11, the IO device 21, and the IO device 31 supports RoCE. When transmitting packets based on RoCE between two IO devices, memory registration (MR) is required. After registering an MR, the MR has the following attributes: RDMA operation context (context), MR registered cache address (address, abbreviated as addr), MR registered cache length (length), MR registered local key ( local key, lkey) and MR registered remote key (remote key, rkey). Taking INI 1 as an example, the storage area in the memory of the host computer 10 included in INI 1 can be registered in the memory of the IO device 11 included in INI 1, and then the IO device 11 can directly implement the storage in the memory of the host computer 10. An operation on a storage area (such as a read operation or a write operation).

Optionally, before step 1010, INI 1 interacts with TGT 1 to obtain the address and remote key value rkey1 of area 1 in the storage area of host 30 included in TGT 1, that is, INI 1 has obtained the read and write of area 1 After that, INI 1 can directly access area 1 like accessing its own memory to achieve read or write operations. Similarly, the interaction between INI 1 and TGT 2 can obtain the address of area 2 and the remote key value rkey2 in the memory of the host computer 20 included in TGT 2, that is, INI 1 has obtained the read and write authority of this area 2, after which INI 1 can send Access area 2 directly as accessing its own memory for read or write operations. The key value of rkey1 is the same as the key value of rkey2. For the convenience of description, the key value of rkey1 and the key value of rkey2 are collectively referred to as the key value of rkey in the following. It can be understood that, for TGT 1, the key value of rkey or the key value of rkey1 is used to indicate area 1. For TGT 2, either the key value of rkey or the key value of rkey2 is used to indicate area 2.

Step 1010, the host 10 obtains SQE1, and stores SQE1 into SQ1.

SQE1 is used to carry the write request 1, and the write request 1 carries the source address of the data 1 to be written in the memory included in the host 10. The acquisition of SQE1 by the host 10 may include the following steps: the application program 1 in the host 10 generates WR1; the host 10 schedules the interface provided by the driver to convert WR1 into WQE1. WR1 and WQE1 carry the same information, but the format is different. It is understandable that after WQE1 is put into SQ1, WQE1 is also called SQE1. Hereinafter, the WQE1 obtained from SQ1 is collectively referred to as SQE1.

In step 1011, the IO device 11 acquires SQE1 from SQ1.

Wherein, the format of SQE1 is SGL format, that is, SQE1 includes the source address of the data 1 to be written and the source address of SQE1, and the source address of the data 1 to be written is that the data 1 to be written is located in the memory included in the host computer 10. Address, the source address of SQE1 is the location of SQE1 in the memory included in the host computer 10 .

Step 1012 , the IO device 11 encapsulates the data 1 to be written indicated by the SQE1 according to the SQE1 , the multicast link information and the rkey key value to generate a message 1 , the message 1 includes the data packet 1 .

Optionally, the following step is also included before step 1012: the IO device 11 obtains the data 1 to be written from the memory of the host 10 according to SQE1.

Wherein, data packet 1 carries data 1 to be written. The format of the data packet 1 may be as shown in (1) in FIG. 9 .

In step 1020, the IO device 11 sends message 1 to the multicast members of the multicast group 1.

The multicast members of multicast group 1 include TGT 1 and TGT 2. Correspondingly, the IO device 21 and the IO device 31 will receive the message 1 . For the parameters encapsulated in message 1, refer to the second column in Table 1 above.

Step 1030, the IO device 21 parses the message 1, obtains the data 1 to be written and the key value of the rkey, and stores the data 1 to be written into the area 1 indicated by the key value of the rkey.

Step 1040, the IO device 31 parses the message 1, obtains the data 1 to be written and the key value of the rkey, and stores the data 1 to be written into the area 2 indicated by the key value of the rkey.

Step 1050, the IO device 11 sends CQE1 to the host 10, and the CQE1 includes the completion information of the SQE1. Correspondingly, the host 10 stores the received CQE1 into CQ1.

Step 1060, the IO device 21 sends an IO completion message 1 to the INI 1. Correspondingly, the IO device 11 receives the IO completion message 1 . Wherein, the IO completion message 1 is used to indicate that the TGT 1 successfully executes the write request 1 corresponding to the SQE1.

Optionally, after step 1060, the IO device 11 may also send an ACK message of the IO completion message 1 to the IO device 21.

Step 1061 , the IO device 11 sends CQE2 to the host 10 . Correspondingly, the host 10 stores the received CQE2 into CQ1.

Step 1070, the IO device 31 sends an IO completion message 2 to the INI 1.

Wherein, the IO completion message 2 is used to indicate that the TGT 2 successfully executes the write request 1 corresponding to the SQE1. Correspondingly, the IO device 31 receives the IO completion message 1. Wherein, the IO completion message 3 is used to indicate that the TGT 2 successfully executes the write request 1 corresponding to the SQE1.

Optionally, after step 1070, the IO device 11 may also send an ACK message of the IO completion message 2 to the IO device 31.

Step 1071 , the IO device 11 sends CQE3 to the host 10 . Correspondingly, the host 10 stores the received CQE3 into CQ1.

Step 1080, the IO device 11 sends an IO completion message 3 to the application program 2, and the IO completion message 3 indicates that the write request 1 corresponding to WR1 has been successfully executed.

Before step 1080, the following step may also be included: the host 10 generates an IO completion message 3 according to CQE2 and CQE3.

It can be understood that, the execution order of the above steps 1010 to 1080 is only for illustration and does not constitute any limitation. For example, step 1040 may also be performed first and then step 1030 is performed.

The above implementation manner provides an RDMA-based multicast transmission method, which can realize reliable and efficient data transmission.

Taking the application scenario shown in FIG. 4 as an example below, another specific implementation manner of the multicast transmission method provided by the embodiment of the present application is described in detail in combination with the embodiment in FIG. 11 .

FIG. 11 is a schematic flowchart of a multicast transmission method 1100 provided by an embodiment of the present application. As shown in FIG. 11 , the method 1100 includes step 1110 to step 1130 . Next, step 1110 to step 1130 will be described in detail.

In this embodiment of the application, the data 1 to be written can be stored in the area 1 of the memory of the host 10, and before step 1110, after the INI 1 interacts with the TGT 1, the TGT 1 can obtain the address VA1 of the area 1 and the remote key The value is rkey1, that is, TGT1 has obtained the read and write permission of this area 1, and then TGT1 1 can directly access area 1 to perform read or write operations like accessing its own memory. Similarly, after INI 1 interacts with TGT 2, TGT 2 can also obtain the address VA1 of area 1 and the remote key value rkey1, that is, TGT 2 has obtained the read and write permission of this area 1, and then TGT1 2 can access its own memory The same direct access to area 1 for read or write operations. Wherein, the procedure for TGT 1 and TGT 2 to obtain the data stored in area 1 is as described in steps 1110 to 1120 below.

Step 1110, the host 10 obtains SQE1, and stores SQE1 in SQ1.

Before step 1110 , the following steps are also included: the host 10 creates an IO command message 1 in the memory of the host 10 . Wherein, SQE1 includes an IO command message 1, and the IO command message 1 is used to instruct to store the data 1 to be written into the memory of the hosts corresponding to the multicast members of the multicast group 1 (ie, the host 20 and the host 30). Exemplarily, the data 1 to be written may be, but not limited to, 10 bytes of data.

In step 1111, the IO device 11 acquires SQE1 from SQ1.

The format of SQE1 is SGL format, that is, SQE1 includes the information that the data 1 to be written corresponding to the IO command message 1 is located in area 1 of the memory of the host 10, and SQE1 includes VA1 and rkey1.

In step 1112, the IO device 11 sends an IO command message 1 to the multicast members of the multicast group 1.

The multicast members of the multicast group 1 include TGT 1 and TGT 2. Correspondingly, the IO device 21 and the IO device 31 will receive the IO command message 1. The IO command message 1 may include at least one RoCE data packet, and the parameters carried in the header of the at least one RoCE data packet may refer to the content in the second column in Table 1 above.

Optionally, before step 1112, the IO device 11 may also perform the following operations: generate the at least one RoCE data packet according to the multicast link information and the rkey key value.

It can be understood that the above IO command message 1 carries VA1 and rkey1, but does not carry the data 1 to be written.

Step 1113 , the IO device 21 sends rqel to the host 20 . Correspondingly, the host 20 receives rqel.

Wherein, the format of rqel is the SGL format, that is, rqe1 includes the source address of the IO command message 1 located in the memory of the IO device 21, and the address (that is, VA1) of the data to be written 1 located in the memory of the host computer 10.

Optionally, before step 1113, the IO device 21 is also configured to perform the following operations: parse the IO command message 1 to obtain rq1, and store the generated rqe1 in RQ2, where rqe1 is used to indicate that the data to be written 1 Stored in area 1 of the memory of the host computer 20 .

Optionally, after step 1113, the IO device 21 will also send an ACK message to the IO device 11, where the ACK message is used to indicate that the IO device 21 has successfully received the IO command message 1.

Step 1114 , the IO device 31 sends rqe2 to the host 30 . Correspondingly, the host 30 receives rqe2.

Before step 1114, the IO device 31 is also used to perform the following operations: parse the IO command message 1 to obtain rq2, and store the generated rqe2 in RQ2, and rqe2 is used to indicate that the data 1 to be written is stored in the host 30 area 2 of the memory.

The format of rqe2 is SGL format, that is, rqe2 includes the source address of the IO command message 1 located in the memory of the IO device 31 and the address of the data to be written 1 located in the memory of the host 10 .

Optionally, after step 1114, the IO device 31 will also send an ACK message to the IO device 11, where the ACK message is used to indicate that the IO device 31 has successfully received the IO command message 1.

Optionally, after step 1114, INI 1 may use time 1 as the start time of the aggregation time, and time 1 is the time when IO device 11 receives the last ACK message in ACK message 1 and ACK message 2.

Step 1115 , the IO device 11 sends CQE1 to the host 10 .

Wherein, CQE1 is used to indicate that the IO command message 1 included in SQE1 has been executed. In some implementation manners, after the IO device 11 receives the ACK message of the IO command message 1 sent by all the multicast members of the multicast group 1, the IO device 11 performs the above step 6.

Step 1116, the host 20 sends SQE2 to the IO device 21.

Wherein, the format of SQE2 may be SGL format, that is, SQE2 includes the address of SQE2 in the memory of the host 20 and the address of the data 1 to be written in area 1 of the memory of the host 20 . That is to say, SQE2 can carry address information 1, address information 1 includes VA1 and lkey1, lkey1 is used to indicate the registration area MR2 in IO device 21, VA1 is used to indicate an area 2 in MR2, and this area 2 is used to store Data 1 to be written. The SQE2 may also carry read information 1 , which is used to indicate that the data 1 to be written is acquired from the area 1 in the memory of the host 10 . Read information 1 may include (rkey1, size, offset), where rkey1 is used to indicate area 1, size is used to indicate the size of data in area 1, and offset is used to indicate the address of the area to be read relative to the start of area 1 The offset of the address. In an example, read information 1 may be specifically used to indicate that all data to be written in data 1 is obtained from area 1, that is, read information 1 may include (rkey1, size, offset), and at this time, the size of size is equal to The size of data 1 to be written, and the value of offset is used to indicate the offset from the end address of area 1 to the start address of area 1. In another example, the read information 1 can be specifically used to indicate that part of the data to be written into the data 1 is obtained from the area 1, that is, the read information 1 can include (rkey1, size, offset), and at this time, the size of the size It is equal to the size of the part of data to be written into data 1, and the value of offset is used to indicate the offset of the address of the part of data in area 1. It can be understood that, in this implementation manner, TGT 1 needs to send multiple (for example, at least two) read messages to obtain all the contents of the data 1 to be written.

Step 1117, the host 30 sends SQE3 to the IO device 31.

Wherein, the format of SQE3 may be SGL format, that is, SQE3 includes the address of SQE3 located in the memory of the host 30 and the address of area 1 of the data 1 to be written located in the memory of the host 30 . That is to say, SQE3 can carry address information 2, address information 2 includes VA2 and lkey2, lkey2 is used to indicate the registration area MR3 in IO device 31, VA2 is used to indicate an area 3 in MR3, and this area 3 is used to store Data 1 to be written. The SQE3 may also carry read information 2 , and the read information 2 is used to indicate that the data 1 to be written is obtained from the memory of the host 10 . Read information 2 may include (rkey1, size, offset), where rkey1 is used to indicate area 1, size is used to indicate the size of the data in area 1 to be read, and offset is used to indicate that the address of the area to be read is relative to area 1 The offset of the starting address of . In an example, read information 2 may be specifically used to indicate that all data to be written in data 1 is obtained from area 1, that is, read information 2 may include (rkey1, size, offset), and at this time, the size of size is equal to The size of data 1 to be written, and the value of offset is used to indicate the offset from the end address of area 1 to the start address of area 1. In another example, the read information 2 may be specifically used to indicate that part of the data to be written into the data 1 is obtained from the area 1, that is, the read information 2 may include (rkey1, size, offset), and at this time, the size of the size It is equal to the size of the part of data to be written into data 1, and the value of offset is used to indicate the offset of the address of the part of data in area 1. It can be understood that, in this implementation manner, TGT 2 needs to send multiple (for example, at least two) read messages to obtain all the contents of the data 1 to be written.

In step 1118, the IO device 21 sends a read request message 1 to the IO device 11. Correspondingly, the IO device 11 will receive the read request message 1 .

Optionally, before step 1118, the IO device 21 also needs to generate a read request message 1 according to SQE2 and the contents of the fourth column shown in Table 1 above, that is, the read request message 1 carries address information 1 and read information 1 . The format of the read request message 1 can be referred to as shown in (4) in FIG. The value of the SynETH field 1 is used to uniquely indicate a synchronous read operation request 1 , and the read operation request 1 is used to request to read data 1 to be written in area 1 of the memory of the host 10 . The value of RETH field 1 is used to indicate a read operation. Specifically, the value of SynETH field 1 is equal to tag1, that is, tag1 is used to uniquely indicate synchronous read operation request 1.

Wherein, the IO device 21 sends a read request message 1 to the IO device 11 in a unicast manner, and the address information encapsulated in the read request message 1 can be referred to in the fourth column in Table 1 above.

In step 1119, the IO device 31 sends a read request message 2 to the IO device 11. Correspondingly, the IO device 11 receives the read request message 2 .

Optionally, before step 1119, the IO device 31 also needs to generate a read request message 2 according to SQE3 and the contents of the fourth column shown in Table 1 above, that is, the read request message 2 carries address information 2 and read information 2 . The format of the read request message 2 can be referred to as shown in (4) in FIG. The value of the SynETH field 2 is used to uniquely indicate the synchronous read operation request 1, and the read operation request 1 is used to request to read the data 1 to be written in the area 1 of the INI 1. The value of RETH field 1 is used to indicate a read operation. Specifically, the value of the SynETH field 2 is equal to tag1, that is, tag1 is used to uniquely indicate the synchronous read operation request 1.

Wherein, the IO device 31 sends a read request message 2 to the IO device 11 in a unicast manner, and the address information encapsulated in the read request message 2 can be referred to in the fourth column in Table 1 above.

In step 1120, the IO device 11 sends a synchronous read request response message 1 to the members of the multicast group 1. Correspondingly, the members of the multicast group 1 receive the synchronous read request response message 1 . The multicast members of multicast group 1 include TGT 1 and TGT 2.

Optionally, before step 1120, the IO device 11 may also perform the following steps: process the read request message 1 and the read request message 2, generate a synchronous read request response message 1, and the synchronous read request response message 1 carries data to be written 1 and the information of multicast group member 1. Wherein, the IO device 11 processes the read request message 1 and the read request message 2 to generate a synchronous read request response message 1, which may include the following steps: the IO device 11 parses the read request message 1 to obtain the SynETH field 1 and the read information 1 (that is, (rkey1, size, offset)), and read request message 2 is parsed to obtain SynETH field 2 and read information 2 (that is, (rkey1, size, offset)); IO device 11 according to SynETH field 1 and the value of SynETH field 2 to determine that the read request task corresponding to read request message 1 and the read request task corresponding to read request message 2 are the same task; Input data 1, and encapsulate according to the link information of multicast group 1 and data 1 to be written, and generate synchronous read request response message 1. For the content of the link information of multicast group 1, refer to the content in the second column in Table 1 above.

In step 1121, the IO device 21 parses the synchronous read request response message 1, obtains the data 1 to be written, and stores the data 1 to be written into the area 1 indicated by the key value of rkey2.

Optionally, after step 1121, the IO device 21 may also send an ACK message 3 to the IO device 11, where the ACK message 3 indicates that the IO device 21 has successfully received the synchronous read request response message 1.

In step 1122, the IO device 31 parses the synchronous read request response message 1, acquires the data 1 to be written, and stores the data 1 to be written in the area 2 indicated by the key value of rkey3.

Optionally, after step 1122, the IO device 31 may also send an ACK message 4 to the IO device 11, where the ACK message 4 indicates that the IO device 31 has successfully received the synchronous read request response message 1.

Step 1123 , the IO device 11 sends CQE2 to the host 10 , and the CQE2 includes the completion information of sending the synchronous read request response message 1 to the multicast group 1 .

Step 1124, the host 20 sends SQE4 to the IO device 11, and the SQE4 includes the completion information of TGT 1 executing the synchronous read request response message 1.

Wherein, the format of SQE4 is the SGL format, that is, SQE4 includes the address information of the area 1 storing the data 1 to be written in the TGT 1.

Step 1125, the host 30 sends SQE5 to the IO device 11, and the SQE5 includes the completion information of the TGT 2 executing the synchronous read request response message 1.

Wherein, the format of SQE5 is the SGL format, that is, SQE5 includes the address information of the area 2 storing the data 1 to be written in the TGT 2.

Step 1126, the IO device 21 sends an IO completion message 1 to the IO device 11, and the IO completion message 1 is used to indicate that the TGT 1 successfully executes the task corresponding to the synchronous read request response message 1. Correspondingly, the IO device 11 receives the IO completion message 1 .

Optionally, after step 1126, the IO device 11 may also send an ACK message of the IO completion message 1 to the IO device 21, indicating that the IO device 11 has received the IO completion message 1.

In step 1127, the IO device 11 sends CQE3 to the host 10, and the CQE3 includes the completion information that TGT 1 successfully executes the synchronous read request response message 1.

Step 1128, the IO device 31 sends an IO completion message 2 to the IO device 11, and the IO completion message 2 is used to indicate that the TGT 2 successfully executes the response corresponding to the synchronous read request response message 1. Correspondingly, the IO device 11 receives the IO completion message 1 .

Optionally, after step 1128, the IO device 11 may also send an ACK message of the IO completion message 2 to the IO device 31, indicating that the IO device 11 has received the IO completion message 2.

Optionally, after step 1128, the IO device 11 may also send an ACK message of the IO completion message 2 to the IO device 31.

In step 1129, the IO device 11 sends a CQE4 to the host 10, and the CQE4 includes the completion information that the TGT 2 successfully executes the synchronous read request response message 1.

Step 1130, the host 10 sends an IO completion message 3 to the application program 1, and the IO completion message 3 indicates that the IO command message 1 corresponding to the SQE1 has been executed.

Before step 1130, the following step may also be included: processor 1 generates IO completion message 3 according to CQE3 and CQE4.

It can be understood that, the execution order of the above steps 1110 to 1130 is only for illustration and does not constitute any limitation. For example, step 1114 may also be performed first and then step 1113 is performed. For example, step 1117 may also be performed first and then step 1116 is performed.

The above implementation manner provides another RDMA-based multicast transmission method, which can realize reliable and efficient data transmission.

When the computing device (for example, INI 1, TGT 1 or TGT 2) in each of the above method embodiments is realized by a virtual machine, the above-mentioned host and IO device correspond to the host and the IO device in the virtual machine respectively, and the host in the virtual machine and IO devices are realized through physical hosts and physical IO devices that carry their virtual functions. Its implementation is similar to the above implementation and will not be repeated here.

The multicast transmission method described above is only for illustration, and does not constitute any limitation to the multicast transmission method provided by the embodiment of the present application. The multicast transmission method provided by the embodiment of the present application is described in detail above with reference to FIG. 3 to FIG. 11 , and the embodiment of the device of the present application will be described in detail below in conjunction with FIG. 12 and FIG. 13 . The descriptions of the method embodiments correspond to the descriptions of the device embodiments. Therefore, for parts not described in detail, reference may be made to the foregoing method embodiments.

FIG. 12 is a schematic structural diagram of a multicast transmission device 1200 provided by an embodiment of the present application. The multicast transmission apparatus 1200 shown in FIG. 12 may execute corresponding steps of the multicast transmission method in the foregoing embodiments. As shown in FIG. 12 , the multicast transmission device 1200 includes: a transceiver unit 1210 and a processing unit 1220 .

In some implementations, the apparatus 1200 is applied to the first IO device, and the transceiver unit 1210 is configured to receive the above-mentioned step 520, step 540, step 580, step 591, step 592, step 811, step 820, step 850, step 861, Step 862, Step 871, Step 872, Step 1101, Step 1020, Step 1050, Step 1061, Step 1070, Step 1071, Step 1111, Step 1112, Step 1115, Step 1118, Step 1119, Step 1120, Step 1123, Step 1126 , step 1127 , step 1128 , step 1129 , step 310 and step 330 . The processing unit 1220 is configured to execute the above step 530 , step 812 , step 1012 , and step 320 . For details of the above steps, refer to the relevant descriptions in the above method embodiments, and details are not repeated here.

It should be understood that the apparatus 1200 in the embodiment of the present application may be used to implement the multicast transmission method in the foregoing embodiments. Specifically, when the apparatus 1200 is hardware, the apparatus 1200 may be the IO device itself, or may also be some modules in the IO device. When the apparatus 1200 is software, the apparatus 1200 may be a software system deployed in an IO device.

In other implementations, the apparatus 1200 is applied to the second IO device, and the transceiver unit 1210 is used to perform the above steps 540, 560, 830, 831, 860, 861, 870, 871, and 1060 , step 1070, step 1112, step 1114, step 1116, step 1117, step 1118, step 1119, step 1126, step 1128. The processing unit 1220 is configured to execute the above step 550 , step 580 , step 590 , step 591 , step 840 , step 841 , step 1030 , step 1040 , step 1121 , and step 1122 . For details of the above steps, refer to the relevant descriptions in the above method embodiments, and details are not repeated here.

FIG. 13 is a schematic diagram of a hardware structure of a multicast transmission device 1300 provided by an embodiment of the present application. The multicast transmission apparatus 1300 shown in FIG. 13 can execute the operation steps of the method implemented by the IO device in the multicast transmission method of the above embodiment.

As shown in FIG. 13 , the device 1300 includes a processor 1301 , a memory 1302 , a communication interface 1303 and a data transmission line 1304 . Wherein, the processor 1301, the memory 1302, and the communication interface 1303 communicate through the data transmission line 1304, and the communication may also be realized through other means such as wireless transmission. The memory 1302 is used to store instructions, and the processor 1301 is used to execute the computer instructions or program codes stored in the memory 1302 .

In this embodiment of the present application, the processor 1301 may be a processor in a network card or an iNIC. The memory 1302 may include computer instructions or program codes, and the computer instructions or program codes may be used to realize the functions of the transceiver unit 1210 in FIG. 12 and/or the functions of the processing unit 1220 in FIG. 12 . For the functions of the transceiver unit 1210 and the functions of the processing unit 1220, reference may be made to the above description, and details will not be repeated here.

The memory 1302 may include read-only memory and random-access memory, and provides instructions and data to the processor 1301 . Memory 1302 may also include non-volatile random access memory. For example, memory 1302 may also store device type information.

The memory 1302 can be volatile memory or nonvolatile memory, or can include both volatile and nonvolatile memory. Among them, the non-volatile memory can be read-only memory (read-only memory, ROM), programmable read-only memory (programmable ROM, PROM), erasable programmable read-only memory (erasable PROM, EPROM), electrically programmable Erases programmable read-only memory (electrically EPROM, EEPROM) or flash memory. Volatile memory can be random access memory (RAM), which acts as external cache memory. By way of illustration and not limitation, many forms of RAM are available such as static random access memory (static RAM, SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (synchronous DRAM, SDRAM), Double data rate synchronous dynamic random access memory (double data date SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (enhanced SDRAM, ESDRAM), synchronous connection dynamic random access memory (synchlink DRAM, SLDRAM) and direct Memory bus random access memory (direct rambus RAM, DR RAM).

The data transmission line 1304 is used to connect the processor 1301 , the memory 1302 and the communication interface 1303 .

The embodiment of the present application also provides a computer-readable medium, the computer-readable medium stores program codes, and when the computer program codes run on the computer, the computer executes the above-mentioned first IO device or the second IO device. method. These computer-readable storages include, but are not limited to, one or more of the following: read-only memory (read-only memory, ROM), programmable ROM (programmable ROM, PROM), erasable PROM (erasable PROM, EPROM), Flash memory, electrical EPROM (electrically EPROM, EEPROM) and hard drive (hard drive).

An embodiment of the present application also provides a computing device, which includes the aforementioned host and an IO device.

An embodiment of the present application also provides a computing cluster, including a plurality of computing devices described above, and each computing device in the computing devices includes the aforementioned IO device and the aforementioned host.

Those skilled in the art can appreciate that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed by hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may use different methods to implement the described functions for each specific application, but such implementation should not be regarded as exceeding the scope of the present application.

Those skilled in the art can clearly understand that for the convenience and brevity of the description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the foregoing method embodiment, which will not be repeated here.

In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods may be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or May be integrated into another system, or some features may be ignored, or not implemented. In another point, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be in electrical, mechanical or other forms.

The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit.

If the functions described above are realized in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including Several instructions are used to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disc and other media that can store program codes. .

The above is only a specific implementation of the application, but the scope of protection of the application is not limited thereto. Anyone familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the application. Should be covered within the protection scope of this application. Therefore, the protection scope of the present application should be determined by the protection scope of the claims.

Claims (23)

1. A multicast transmission method, characterized in that it is applied to a first computing device, the first computing device includes a first input and output IO device, and the method includes:

   The first IO device acquires a first request;

   The first IO device generates a first data packet according to the first information and the first request, the first information includes multicast information, and the multicast information is used to identify the multicast members of the multicast group and the A link connection between the first computing devices, the first data packet carrying the multicast information and the data to be written corresponding to the first request;

   The first IO device sends the first data packet to the multicast members of the multicast group.
2. The method according to claim 1, characterized in that,

   The multicast information includes a multicast identifier, and the multicast identifier is obtained when the first computing device establishes a link connection with the multicast group.
3. The method according to claim 1 or 2, characterized in that,

   The first data packet further includes a port number, and the port number is used to indicate that the transmission mode of the first data packet is a multicast transmission mode.
4. The method according to any one of claims 1 to 3, characterized in that,

   The first computing device further includes a first host, the first IO device communicates with the first host through an IO network, and the first request is sent by an application program running in a processor included in the first host request.
5. The method according to claim 4, characterized in that,

   The first information further includes first window information, and the first window information is used to indicate the minimum amount of data that can be processed by the multicast members of the multicast group within a first time period.
6. The method according to claim 5, characterized in that,

   The transmission protocol used by the first IO device to send the first data packet is Transmission Control Protocol/Internet Protocol TCP/IP.
7. The method according to claim 4, wherein the data to be written is a part of the data corresponding to the first request, and the first information further includes indication information, and the indication information is used to indicate Encapsulating the data to be written, before the first IO device generates a first data packet according to the first information and the first request, the method further includes:

   The first IO device sends an IO write command to the multicast members of the multicast group, the IO write command is used to instruct to store the data corresponding to the first request in the first registration area MR, and the first The data corresponding to a request is located in the memory included in the first host included in the first computing device, and the first MR is registered as a storage area in the memory included in the second host to a storage area in the memory of the second IO device , the second computing device includes the second host and the second IO device, and the second computing device is a multicast member of the multicast group;

   The first IO device receives the indication information sent by the multicast member of the multicast group.
8. The method according to claim 7, characterized in that,

   The IO write command includes a second key value and second location information, the second key value is used to identify the second MR, and the second location information is used to indicate that the data to be written is in the The location in the second MR described above.
9. The method according to claim 7 or 8, characterized in that,

   The first information also includes a credit value, the credit value is used to indicate the minimum number of requests that can be processed by the multicast members of the multicast group within the second time period, and the basic transmission header of the first data packet Part BTH carries the credit value.
10. The method according to any one of claims 1 to 3, wherein said first IO device obtaining a first request comprises:

    The first IO device receives the first request sent by the multicast member of the multicast group, and the first request is used to indicate that the data to be written in the second MR is stored in the group The storage area of the multicast member of the broadcast group, the second MR registers the storage area in the memory included in the first host to the storage area in the memory of the first IO device, and the first computing device also includes the Describe the first host.
11. The method according to claim 10, characterized in that,

    The first request includes a first key value, first location information and a preset field, the first key value is used to identify the second MR, and the first location information is used to indicate the to-be-written The position of the input data in the second MR, and the value of the preset field is used to indicate the first request.
12. The method according to any one of claims 8 to 11, characterized in that,

    The transmission protocol used by the first IO device to send the first data packet is Ethernet-based Remote Direct Data Access (RDMA).
13. The method according to any one of claims 1 to 12, wherein the first IO device sends the first data packet to the multicast members of the multicast group, comprising:

    The first IO device sends the first data packet to a forwarding device, and the forwarding device is configured to copy the first data packet and forward the copied first data packet to the multicast A multicast member of a group, the link connection between the multicast member of the multicast group and the first computing device includes the forwarding device.
14. The method according to any one of claims 1 to 13, wherein the multicast members of the multicast group include a second computing device,

    After the first IO device sends the first data packet to the multicast members of the multicast group, the method further includes:

    The first IO device receives a second request sent by the second computing device, and the second request is used to request to obtain the data to be written corresponding to the first request carried in the first data packet;

    The first IO device sends a second data packet to the second computing device, the second data packet carries the data to be written, and the port number included in the second data packet is used to indicate the transmission of the The transmission mode of the second data packet is a unicast transmission mode.
15. The method according to any one of claims 1 to 14, wherein the multicast members of the multicast group include a second computing device and a third computing device,

    The method also includes:

    After the first IO device receives the first completion message and the second completion message, the first IO device sends a third completion message to the processor included in the first host, and the third completion message is used for Indicating that the first request has been successfully executed, the first completion message is used to indicate that the second computing device has successfully executed the first request, and the second completion message is used to indicate that the third computing device has successfully executed The first request was successfully executed.
16. A multicast transmission device, characterized in that it includes a transceiver unit and a processing unit,

    The transceiver unit is configured to obtain a first request;

    The processing unit is configured to generate a first data packet according to the first information and the first request, the first information includes multicast information, and the multicast information is used to identify the multicast members of the multicast group and the The link connection between the multicast transmission devices, the first data packet carries the multicast information and the data to be written corresponding to the first request;

    The transceiver unit is further configured to send the first data packet to the multicast members of the multicast group.
17. The device according to claim 16, characterized in that,

    The multicast information includes a multicast identifier, and the multicast identifier is obtained when the multicast transmission device establishes a link connection with the multicast group.
18. Apparatus according to claim 16 or 17, characterized in that,

    The first data packet further includes a port number, and the port number is used to indicate that the transmission mode of the first data packet is a multicast transmission mode.
19. Apparatus according to any one of claims 16 to 18, characterized in that

    The transmission protocol used by the processing unit to send the first data packet is Transmission Control Protocol/Internet Protocol TCP/IP, or Remote Direct Data Access RDMA based on Ethernet.
20. An input and output IO device, characterized in that the IO device includes at least one processor and a communication interface, and the at least one processor is used to execute computer programs or instructions, so that the IO device performs the process described in claim 1. The method described in any one of to 15.
21. A computing device, characterized in that the computing device includes a host and an input/output IO device, and the host and the IO device communicate through an IO network;

    The host is used to run an application, and send a request generated by the application to the IO device;

    The IO device is used to execute the method according to any one of claims 1-15.
22. A computer-readable storage medium, characterized by comprising a computer program, which causes the computer to execute the method according to any one of claims 1 to 15 when the computer program is run on the computer.
23. A multicast transmission system, characterized in that the system includes a first computing device and a multicast group, and the multicast group is the multicast group in the method according to any one of claims 1 to 15, wherein The first computing device includes a first input and output IO device, and the first IO device is used to execute the operation steps of the method described in any one of claims 1 to 15.

Images