CN112398817A - Data sending method and device - Google Patents

Abstract

Translated fromChinese

本发明实施例提供了一种数据发送的方法及设备，涉及通信技术领域，方法包括当第一虚拟机与第二虚拟机建立数据连接时，第一虚拟机根据第一虚拟机的虚拟地址，确定第一虚拟机对应的RDMA网络的全局身份标识GID，第一虚拟机通过GID对应的第一端口，向主机发送队列对QP属性设置信息，QP属性设置信息包括第一指令和第二指令，主机删除第一指令，将第二指令修改为第三指令，主机根据修改后的QP属性设置信息设置QP队列的属性，以使HCA卡基于设置后的QP队列向第二虚拟机发送数据。采用本发明即可以实现虚拟机的虚拟网络与物理网络隔离，保证虚拟机的安全，又可以兼容RoCE协议。

Embodiments of the present invention provide a method and device for sending data, which relate to the field of communications technologies. The method includes that when a first virtual machine establishes a data connection with a second virtual machine, the first virtual machine, Determine the global identity GID of the RDMA network corresponding to the first virtual machine, the first virtual machine sends the queue pair QP attribute setting information to the host through the first port corresponding to the GID, and the QP attribute setting information includes the first instruction and the second instruction, The host deletes the first instruction, modifies the second instruction into a third instruction, and sets the attributes of the QP queue according to the modified QP attribute setting information, so that the HCA card sends data to the second virtual machine based on the set QP queue. By adopting the invention, the virtual network of the virtual machine can be isolated from the physical network, the safety of the virtual machine can be ensured, and the RoCE protocol can be compatible.

Description

Data sending method and device

Technical Field

The present application relates to the field of communications technologies, and in particular, to a method and a device for data transmission.

Background

At present, a Remote Direct Access over Ethernet (RoCE) Protocol based on Converged Ethernet is widely applied to cloud computing, and the RoCE Protocol is adopted, so that on one hand, the problems of large network delay and high load of a Central Processing Unit (CPU) when a traditional Transmission Control Protocol/Internet Protocol (TCP/IP) Protocol transmits and receives data can be solved, and on the other hand, the Remote Direct Access over Ethernet (RoCE) Protocol can be compatible, and the requirements of low investment and low operation cost can be met. A plurality of virtual machines are generally arranged in a server in cloud computing, and meanwhile, a Host Channel Adapter (HCA) supporting a RoCE protocol is installed, the HCA card includes a plurality of physical ports, and for each physical port, the physical port may be abstracted into an ethernet interface supporting an ethernet protocol and an RDMA interface supporting a Remote Direct Memory Access (RDMA) protocol, and an administrator may configure a physical IP address corresponding to the virtual machine on the ethernet interface.

In the prior art, when the virtual machines in different servers need to communicate data, hardware virtualization functions, such as SR-IOV, need to be started on the HCA card. In this way, the virtual machine can establish connection with the other side according to the physical IP address configured by the administrator, establish a Queue Pair (QP) on the corresponding RDMA interface, and set the source address of the QP Queue as the physical IP address of the virtual machine itself and the destination address as the physical IP address of the other side. In the subsequent data transmission process, the HCA card directly reads the data in the memory corresponding to the application program, packages the data according to the source address and the destination address in the QP queue to obtain a data message, and transmits the data message through a physical network.

Therefore, in the prior art, network virtualization of the RoCE protocol cannot be realized through software, that is, the prior art cannot provide a data transmission method which can realize isolation of a virtual network of a virtual machine from a physical network, ensure safety of the virtual machine, and be compatible with the RoCE protocol.

Disclosure of Invention

The embodiment of the invention provides a data sending method and equipment, which can realize virtual network data isolation and reduce the security risk of a virtual machine. The technical scheme is as follows:

in a first aspect, a method for data transmission is provided, where an apparatus applied in the method includes a first virtual machine and a host, and the method includes: the host obtains a first instruction and a second instruction from the first virtual machine, wherein the first instruction is used for indicating that a source address of a queue pair QP queue is a virtual address of the first virtual machine, the second instruction is used for indicating that a destination address of the QP queue is a virtual address of a second virtual machine, and the QP queue is used for communication between the first virtual machine and the second virtual machine; the host deletes the first instruction and modifies the second instruction into a third instruction, wherein the third instruction is used for indicating that a destination address in the QP queue is a physical address of the second virtual machine; and the host sets the QP queue according to the third instruction so as to send data to the second virtual machine through the set QP queue.

In the embodiment of the invention, when a host receives a first instruction and a second instruction sent by a first virtual machine, the first instruction is used for indicating the host to set a source address in a QP queue as a virtual address of the first virtual machine, the second instruction is used for indicating the host to set a destination address in the QP queue as a virtual address of a second virtual machine, the host deletes the first instruction, modifies the second instruction into a third instruction, and the third instruction is used for indicating the host to set the destination address in the QP queue as a physical address of the second virtual machine; and then setting the attribute of the QP queue according to the modified instruction so as to send data to the second virtual machine based on the set QP queue. Based on the invention, the virtual network and the physical network of the virtual machine can be isolated, the safety of the virtual machine is ensured, and the RoCE protocol can be compatible.

In one possible implementation manner, the first virtual machine sends the first instruction and the second instruction to the host through a first port, where the first port is a port of a virtual remote direct access RDMA network card of the first virtual machine, and the first port corresponds to a global identity GID of the first virtual machine in an RDMA network. This global GID can uniquely identify the first virtual machine, thereby enabling the host to look up the relevant information.

In a possible implementation manner, the modifying the second instruction into a third instruction includes: the host determines a physical address of the second virtual machine according to a first mapping table stored in advance and a virtual address of the second virtual machine in the second instruction, wherein the first mapping table comprises a corresponding relation between the virtual address and the physical address; and the host modifies the virtual address of the second virtual machine in the second instruction into the physical address of the second virtual machine to obtain a third instruction.

In one possible implementation manner, the first virtual machine further includes a port of a virtual ethernet network card, and the method further includes: the first virtual machine acquires the identifier of the virtual RDMA network card and the identifier of the first port; the first virtual machine determines the identifier of a corresponding second port according to the identifier of the virtual RDMA network card, the identifier of the first port and a second mapping table prestored in the host, wherein the second mapping table comprises the corresponding relation among the identifier of the virtual RDMA network card, the identifier of the port of the virtual RDMA network card and the identifier of the port of the virtual Ethernet network card, and the second port is the port of the virtual Ethernet network card of the first virtual machine; the first virtual machine determines a first GID of the first port according to a preset GID generation algorithm and a virtual address corresponding to the identifier of the second port; the first virtual machine sets an address of the first port to the first GID.

In one possible implementation manner, the determining, by the first virtual machine, an identifier of a corresponding second port according to the identifier of the virtual RDMA network card, the identifier of the first port, and a second mapping table pre-stored in the host includes: the first virtual machine sends query request information to the host, wherein the query request information carries the identifier of the virtual RDMA network card and the identifier of the first port; the host determines the identifier of the corresponding second port according to the query request information and a pre-stored second mapping table; and the host sends query response information to the first virtual machine, wherein the query response information carries the identifier of the second port.

In one possible implementation, the method further includes: when the first virtual machine detects information carrying the identifier of the second port and used for changing the Ethernet address, determining the identifier of the corresponding virtual RDMA network card and the identifier of the first port according to the identifier of the second port and a prestored third mapping table, wherein the third mapping table comprises the corresponding relation among the identifier of the port of the virtual Ethernet network card, the identifier of the virtual RDMA network card and the identifier of the port of the virtual RDMA network card; and the first virtual machine updates the address of the first port according to the event type carried in the information for changing the Ethernet address.

In a possible implementation manner, the updating, by the first virtual machine, the address of the first port according to the event type carried in the ethernet address change information includes: when the event type is the network card starting, the first virtual machine acquires a virtual address corresponding to the second port identifier; and the first virtual machine determines a second GID of the first port according to a virtual address corresponding to the second port identifier and a preset GID generation algorithm, and sets the address of the first port as the second GID.

In a possible implementation manner, the updating, by the first virtual machine, the address of the first port according to the event type carried in the ethernet address change information includes: and when the event type is network card closing, the first virtual machine deletes the address of the first port.

In a second aspect, an apparatus is provided, which includes a host configured to obtain a first instruction and a second instruction from a first virtual machine, where the first instruction is configured to indicate that a source address of a queue pair QP queue is a virtual address of the first virtual machine, and the second instruction is configured to indicate that a destination address of the QP queue is a virtual address of a second virtual machine, and the QP queue is used by the first virtual machine to communicate with the second virtual machine; deleting the first instruction, and modifying the second instruction into a third instruction, wherein the third instruction is used for indicating that a destination address in the QP queue is a physical address of the second virtual machine; and the host sets the QP queue according to the third instruction so as to send data to the second virtual machine through the set QP queue.

In the embodiment of the invention, when a host receives a first instruction and a second instruction sent by a first virtual machine, the first instruction is used for indicating the host to set a source address in a QP queue as a virtual address of the first virtual machine, the second instruction is used for indicating the host to set a destination address in the QP queue as a virtual address of a second virtual machine, the host deletes the first instruction and modifies the second instruction into a third instruction, and the third instruction is used for indicating the host to set the destination address in the QP queue as a physical address of the second virtual machine; and then setting the attribute of the QP queue according to the modified third instruction so as to send data to the second virtual machine based on the set QP queue. Based on the invention, the virtual network and the physical network of the virtual machine can be isolated, the safety of the virtual machine is ensured, and the RoCE protocol can be compatible.

In one possible implementation, the apparatus further includes a first virtual machine, where the first virtual machine is configured to send the first instruction and the second instruction to the host through a first port, where the first port is a port of a virtual remote direct access RDMA network card of the first virtual machine, and the first port corresponds to a global identity GID of the first virtual machine in an RDMA network.

In a possible implementation manner, the host is specifically configured to: determining a physical address of the second virtual machine according to a first mapping table stored in advance and a virtual address of the second virtual machine in the second instruction, wherein the first mapping table comprises a corresponding relation between the virtual address and the physical address; and modifying the virtual address of the second virtual machine in the second instruction into the physical address of the second virtual machine to obtain a third instruction.

In a possible implementation manner, the first virtual machine further includes a port of a virtual ethernet network card, and the first virtual machine is further configured to: acquiring an identifier of the virtual RDMA network card and an identifier of the first port; determining a corresponding identifier of a second port according to the identifier of the virtual RDMA network card, the identifier of the first port and a second mapping table prestored in the host, wherein the second mapping table comprises a corresponding relation among the identifier of the virtual RDMA network card, the identifier of the port of the virtual RDMA network card and the identifier of the port of the virtual Ethernet network card, and the second port is the port of the virtual Ethernet network card of the first virtual machine; determining a first GID of the first port according to a preset GID generation algorithm and a virtual address corresponding to the identifier of the second port; setting an address of the first port to the first GID.

In a possible implementation manner, the first virtual machine is specifically configured to: sending query request information to the host, wherein the query request information carries the identifier of the virtual RDMA network card and the identifier of the first port; the host is specifically configured to: determining the identifier of a corresponding second port according to the query request information and a pre-stored second mapping table; and sending query response information to the first virtual machine, wherein the query response information carries the identifier of the second port.

In a possible implementation manner, the first virtual machine is further configured to: the first virtual machine is further configured to: when the first virtual machine detects information carrying the identifier of the second port and used for changing the Ethernet address, determining the identifier of the corresponding virtual RDMA network card and the identifier of the first port according to the identifier of the second port and a prestored third mapping table, wherein the third mapping table comprises the corresponding relation among the identifier of the port of the virtual Ethernet network card, the identifier of the virtual RDMA network card and the identifier of the port of the virtual RDMA network card; and the first virtual machine updates the address of the first port according to the event type carried in the information for changing the Ethernet address.

In a possible implementation manner, the first virtual machine is specifically configured to: when the event type is the starting of the network card, acquiring a virtual address corresponding to the second port identifier; and determining a second GID of the first port according to the virtual address and a preset GID generation algorithm, and setting the address of the first port as the second GID.

In a possible implementation manner, the first virtual machine is specifically configured to: and when the event type is the network card closing, deleting the address of the first port.

In a third aspect, an apparatus is provided, the apparatus comprising: the apparatus comprises: a memory, a processing circuit and a host channel adapter HCA card; wherein the memory is used for storing codes; the processing circuit is used for operating a first virtual machine and a host according to the codes stored in the memory; the processing circuit is further configured to implement the method of the first aspect by the first virtual machine and a host; the HCA card is used for sending data to the second virtual machine.

In a fourth aspect, there is provided a computer-readable storage medium comprising instructions which, when run on a computer, cause the computer to perform the method of the first aspect.

In a fifth aspect, there is provided a computer program product comprising instructions which, when run on a computer, cause the computer to perform the method of the first aspect.

In the embodiment of the invention, when a host receives a first instruction and a second instruction sent by a first virtual machine, the first instruction is used for indicating the host to set a source address in a QP queue as a virtual address of the first virtual machine, the second instruction is used for indicating the host to set a destination address in the QP queue as a virtual address of a second virtual machine, the host deletes the first instruction and modifies the second instruction into a third instruction, and the third instruction is used for indicating the host to set the destination address in the QP queue as a physical address of the second virtual machine; and then setting the attribute of the QP queue according to the modified third instruction so that the HCA card sends data to the second virtual machine based on the set QP queue. Based on the invention, the virtual network and the physical network of the virtual machine can be isolated, the safety of the virtual machine is ensured, and the RoCE protocol can be compatible.

Drawings

FIG. 1 is a flow chart of data transmission in a TCP/IP network according to an embodiment of the present invention;

FIG. 2 is a flow diagram of data transfer in an RDMA network according to an embodiment of the present invention;

fig. 3 is an architecture diagram of a RoCE protocol provided in an embodiment of the present invention;

FIG. 4 is a schematic structural diagram of an apparatus according to an embodiment of the present invention;

FIG. 5 is a schematic structural diagram of an apparatus according to an embodiment of the present invention;

fig. 6 is a flowchart of a method for associating a port of a virtual ethernet network card with a port of a virtual RDMA network card according to an embodiment of the present invention;

FIG. 7 is a flowchart illustrating an example of a method for associating a port of a virtual Ethernet network card with a port of a virtual RDMA network card according to the present invention;

fig. 8 is a flowchart of a QP attribute setting information management method according to an embodiment of the present invention;

fig. 9 is a schematic structural diagram of an apparatus according to an embodiment of the present invention.

Detailed Description

Fig. 1 is a flow chart of data transmission in a TCP/IP network according to an embodiment of the present invention, as shown in fig. 1, in a conventional TCP/IP network, first, a first server needs to copy data from a buffer of an application to a buffer of a TCP/IP protocol stack; then, packaging the data into a data packet according to a TCP/IP protocol, and copying the data packet to a buffer area driven by the network card; and finally, transmitting the data packet to a second server through the network card. Correspondingly, in the conventional TCP/IP network, the second server copies the data packet in the buffer area driven by the network card to the buffer area of the TCP/IP protocol stack, decapsulates the data packet according to the TCP/IP protocol to obtain data, and copies the data from the buffer area of the TCP/IP protocol stack to the buffer area of the application program.

This causes two problems, i.e., a large number of data copies and a large communication delay. On the other hand, the protocol stack needs a Central Processing Unit (CPU) to operate both for encapsulating and decapsulating data, which results in a high load on the CPU. For example, in a search application, each query results in several query requests to the back-end server, and these query requests have deadline requirements. Only query responses returned to the search application before the expiration time will have their results considered, while query responses that exceed the expiration time will be discarded by the search application. Thus, when the delay in a TCP/IP network is large, the proportion of the query response that is dropped will increase significantly. On the one hand, it will result in a reduced search quality of the search application; on the other hand, the large number of invalid query responses transmitted in the TCP/IP network further increases network delay. As another example, a higher CPU load may result in a reduction in the computing power that the server may provide. In cloud computing, a CPU is a core scheduling resource, and a high CPU load directly causes a decrease in the number of tenants that can be served, thereby reducing the profitability of the overall system.

In order to solve the problems of large delay and high CPU load in a TCP/IP network, the RDMA protocol is applied, and compared with the TCP/IP protocol, the RDMA protocol has the advantages of low delay and small CPU load. Fig. 2 is a flowchart of data transmission in an RDMA network according to an embodiment of the present invention, as shown in fig. 2, in the RDMA network, first, an application registers a memory in an operating system of a server before performing data transmission, and sends an address of the registered memory to an HCA card of the server, and then, the HCA card of the server can directly read and write the registered memory of the application. In this way, the transmission and reception of data no longer requires multiple copies between buffers, and the delay can be reduced considerably. Meanwhile, the HCA card also realizes the functions of an RDMA protocol stack, namely the HCA card completes the encapsulation and decapsulation processing of the data packet. Therefore, CPU resources are released, the load of the CPU is reduced, and the server can have more CPU resources allocated to the tenants for use.

The RDMA protocol is implemented based on an Infiniband (IB) network architecture. However, in the cloud computing scenario, the TCP/IP protocol remains the most dominant network protocol. Therefore, in order to be compatible with the TCP/IP protocol, the RoCE protocol is proposed in the industry. Currently, the RoCE technology mainly includes two versions, RoCEv1 and RoCEv 2. Fig. 3 is an architecture diagram of the RoCE protocol according to an embodiment of the present invention, and as shown in fig. 3, a link layer protocol of the TCP/IP protocol in the RoCEv1 version replaces a link layer protocol of the IB protocol. In the RoCEv2 version, the IB Protocol is replaced with a User Datagram Protocol (UDP) in the network layer Protocol of the TCP/IP Protocol and the transport layer Protocol. Specifically, the network layer protocol of the TCP/IP protocol is used for route learning and transmission, and the UDP protocol of the TCP/IP protocol can be used for equivalent routing (ECMP). Currently, the RoCEv2 version is commonly adopted in cloud computing networks.

In the prior art, when the virtual machines in different servers need to perform data communication, a hardware virtualization function, such as SR-IOV, needs to be started on the HCA card. In this way, the virtual machine can establish connection with the other side according to the physical IP address configured by the administrator, establish a Queue Pair (QP) on the corresponding RDMA interface, and set the source address of the QP Queue as the physical IP address of the virtual machine itself and the destination address as the physical IP address of the other side. In the subsequent data transmission process, the HCA card directly reads the data in the registered memory corresponding to the application program, packages the data according to the source address and the destination address in the QP queue to obtain a data message, and transmits the data message through a physical network.

In a conventional TCP/IP protocol, an administrator generally sets a virtual IP address in a virtual machine in order to isolate a virtual network of the virtual machine from a physical network and ensure the security of the virtual machine. However, based on the prior art, after the virtual machine sets the virtual IP address, when the virtual machine establishes a connection with the peer virtual machine, the virtual machine sets the source address in the QP queue as the virtual IP address of the virtual machine, and sets the destination address as the virtual IP address of the peer virtual machine, and when the subsequent HCA card encapsulates the data of the virtual machine according to the QP queue, the source address of the data packet is the virtual IP address of the virtual machine, and the destination address is the virtual IP address of the peer virtual machine. Since there is no routing table entry related to the virtual IP address in the network device of the physical network, the data packet cannot be transmitted.

The embodiment of the invention provides a method for sending data in a cloud computing scene, which is applied to equipment. In this scenario, as shown in fig. 4, aVirtual Machine 410, ahost 420, and anHCA card 430 supporting a RoCE or other similar protocols may be disposed in the device, where theVirtual Machine 410 is connected to thehost 420 and theHCA card 430, thehost 420 is connected to theHCA card 430, and thehost 420 may be a Virtual Machine with a management function or multiple implementation manners such as a Virtual Machine Monitor (VMM) or a Hypervisor, which is not limited in the present invention. Specifically, as shown in fig. 5, thevirtual machine 410 includes a virtualethernet network card 411, a virtualRDMA network card 412, a virtualethernet management module 413, and a virtualRDMA management module 414. Thehost 420 includes a virtualRDMA control module 421 and may also include a virtualethernet control module 422 and acontroller 423.

TheHCA card 430 includes a plurality of physical ports, and the administrator may virtualize theHCA card 430, and for each physical port on theHCA card 430, the administrator may virtualize the physical port into one or more groups of virtual RoCE network cards, where each group of virtual RoCE network cards may include onevirtual ethernet card 411 and one virtualRDMA network card 412. The invention is described by taking an example that one port is virtualized into a group of virtual RoCE network cards, and other situations are similar to the above. Thevirtual ethernet card 411 includes a plurality of ports of the virtual ethernet card, and thevirtual RDMA card 412 includes a plurality of ports of the virtual RDMA card.

Thevirtual machine 410 is connected to a virtualethernet network card 411 and a virtualRDMA network card 412, respectively, the virtualethernet network card 411 is connected to a port of thecorresponding HCA card 430 through a virtualethernet control module 422, the virtualRDMA network card 412 is connected to a port of thecorresponding HCA card 430 through a virtualRDMA control module 421, and meanwhile, the virtualRDMA control module 421 is connected to thecontroller 423.

A virtualRDMA management module 414 for registering and managing the virtualRDMA network card 412, a virtualethernet management module 413 for registering and managing the virtualethernet network card 411, a virtualethernet control module 422 for controlling the ethernet protocol packet sent by thevirtual machine 410, a virtualRDMA control module 421 for controlling the QP attribute setting information sent by thevirtual machine 410, and acontroller 423 for storing the correspondence among the identifier of the virtual RDMA network card, the identifier of the port of the virtual RDMA network card, and the identifier of the port of the virtual ethernet network card, and the correspondence between the virtual address and the physical address of the peer virtual machine. The virtual address of the virtual machine may be a virtual IP address, or a virtual Media Access Control (MAC) address, and similarly, the physical address may be a physical IP address, or a physical MAC address.

In the prior art, from the perspective of an operating system, an HCA network card is an HCA device, and an HCA device includes an ethernet device and an RDMA device. For an HCA device, the HCA driver first creates a base device in the operating system. And abstracts an ethernet device and an RDMA device on the base device. In the RDMA communication process, an IP address is set on the Ethernet equipment for establishing connection, and the RDMA equipment is used for reading and writing data. Since an application specifies only an IP address when performing an RDMA operation, the operating system must provide a method for determining an RDMA device by an IP address.

In the HCA card, because the ethernet device and the RDMA device are both mounted on the same basic device, the HCA driver can automatically determine the GID corresponding to the RDMA device according to the IP address of the ethernet device and a pre-stored GID generation algorithm, and set the address of the RDMA device as the GID. Meanwhile, when the IP address of the ethernet device changes, the HCA driver may automatically update the GID of the RDMA device according to the IP address of the ethernet device. In this way, the application program can determine the corresponding RDMA device according to its own IP address and the pre-stored GID generation algorithm, thereby implementing data transmission. However, for the virtual ethernet network card and the virtual RDMA network card, both are independent devices in the operating system, without any association. Therefore, for convenience of description, the embodiment of the present invention first introduces a method for providing a port of a virtual ethernet network card and an association method of a port of a virtual RDMA network card corresponding to a virtual machine, as shown in fig. 5 and fig. 6, the specific steps may be as follows:

instep 601, the firstvirtual machine 410 obtains the identifier of the preset virtualRDMA network card 412 and the identifier of the first port.

In an implementation, the administrator may pre-store the identification of the virtualRDMA network card 412 in the virtualRDMA management module 414 of the virtual machine in order to manage the virtualRDMA network card 412 through the virtualRDMA management module 414. When the virtualRDMA management module 414 detects that a certain virtualRDMA network card 412 is started, the virtualRDMA management module 414 may obtain a pre-stored identifier of the virtualRDMA network card 412, and then send a registration message to an RDMA network card driver module (such as an ib-core module), where the registration message carries the identifier of the virtualRDMA network card 412, so as to register the virtualRDMA network card 412. Meanwhile, the registration message may also carry a link layer type, where the link layer type may be RDMA, and after receiving the registration message, the RDMA network card driver module may stop managing the GIDs of the ports of the virtual RDMA network cards on the virtualRDMA network card 412. For example, the virtualRDMA management module 414 calls the ib _ register _ device () interface to register the virtualRDMA network card 412 with the ib _ core module. Wherein, the virtualRDMA management module 414 specifies the parameter link layer type in the IB _ register _ device () interface as "IB" so that the IB _ core stops managing the gib _ cache of the first port of the virtualRDMA network card 412.

After the virtualRDMA management module 414 completes the registration of the virtualRDMA network card 412, the virtualRDMA management module 414 may scan in the bus, and query the identification of the first port on the virtualRDMA network card 412, where the identification of the first port may be a port number of the first port.

Instep 602, the firstvirtual machine 410 determines the identifier of the corresponding second port according to the identifier of the virtualRDMA network card 412, the identifier of the first port, and a second mapping table pre-stored in thehost 420.

The second mapping table includes a correspondence relationship among an identifier of the virtualRDMA network card 412, an identifier of a port of the virtual RDMA network card, and an identifier of a port of the virtual ethernet network card, and the second port is the port of the virtual ethernet network card of the first virtual machine.

In an implementation, thehost 420 may be provided with a virtualRDMA control module 421, and the virtualRDMA control module 421 may be stored with a second mapping table in advance. The second mapping table may be preset by an administrator, and when the administrator virtualizes the physical ports of the HCA into the virtualethernet network card 411 and the virtualRDMA network card 412, the administrator may store the correspondence between the identifier of the virtualRDMA network card 412, the identifiers of the ports of the virtual RDMA network cards on the virtualRDMA network card 412, and the identifiers of the ports of the corresponding virtual ethernet network cards in the second mapping table. The identifier of the port of the virtual ethernet network card may be a Media Access Control (MAC) address of the port of the virtual ethernet network card.

After acquiring the identifier of the virtualRDMA network card 412 and the identifier of the first port, the virtualRDMA management module 414 may query the identifier of the second port corresponding to the first port through the virtualRDMA control module 421. The specific steps can be as follows:

in step one, the firstvirtual machine 410 sends query request information to thehost 420.

The query request may carry an identifier of the virtualRDMA network card 412 and an identifier of the first port.

In step two, thehost 420 determines the identifier of the corresponding second port according to the query request information and a second mapping table stored in advance.

In implementation, after receiving the query request information, the virtualRDMA control module 421 analyzes the query request information to obtain the identifier of the virtualRDMA network card 412 and the identifier of the first port, which are carried in the query request information, and then queries the identifier of the corresponding second port (i.e., the identifier of the second port corresponding to the first port) in the second mapping table according to the identifier of the virtualRDMA network card 412 and the identifier of the first port.

Step three, thehost 420 sends the query response information to the firstvirtual machine 410.

In implementation, after querying the identifier of the second port corresponding to the first port, the virtualRDMA control module 421 may send query response information to the virtualRDMA management module 414, where the query response information carries the identifier of the second port.

It should be noted that, in order to improve the query efficiency, the administrator may only cache the entry of the second mapping table that has been queried in the virtualRDMA control module 421, and store the complete second mapping table in thecontroller 423 of thehost 420. The virtualRDMA control module 421 receives the query request information (for convenience of distinction, hereinafter referred to as first query request information) sent by the virtualRDMA management module 414, obtains the identifier of the virtualRDMA network card 412 and the identifier of the first port, and then locally queries whether the identifier of the second port corresponding to the first port exists. If yes, send a query response message to the virtualRDMA management module 414, where the query response message carries the identifier of the second port (for convenience of differentiation, hereinafter referred to as a first query response message); otherwise, the virtualRDMA control module 421 sends query request information (hereinafter referred to as second query request information for convenience of distinction) to thecontroller 423. Thecontroller 423 receives the second query request information sent by the virtualRDMA control module 421, obtains the identifier of the virtualRDMA network card 412 and the identifier of the first port, then queries whether the identifier of the second port corresponding to the first port exists in a second mapping table stored in advance, and if so, sends a query response information to the virtualRDMA control module 421, where the query response information carries the identifier of the second port (for convenience of distinguishing, hereinafter referred to as second query response information); otherwise, query failure information is returned to the virtualRDMA control module 421. When the virtualRDMA control module 421 receives the second query response message, the identifier of the second port is obtained, and the first query response message is sent to the virtualRDMA management module 414. Meanwhile, the virtualRDMA management module 414 stores the correspondence between the identifier of the virtualRDMA network card 412, the identifier of the first port, and the identifier of the second port locally, so as to provide a fast query in the following. When the virtualRDMA control module 421 receives the query failure information, the query failure information is returned to the virtualRDMA management module 414, and after receiving the query failure information, the virtualRDMA management module 414 may notify the administrator in a display or a short message manner, so that the administrator may modify the error in time.

Step 603, the firstvirtual machine 410 determines the first GID of the first port according to a preset GID generation algorithm and a virtual IP address corresponding to the identifier of the second port.

In implementation, the virtualRDMA management module 414 stores a GID generation algorithm in advance. The GID generation algorithm can adopt an industry standard GID conversion method. For example, if the address length of the GID is 128 bits, if the IPv4 address is converted into the GID, since the address length of the IPv4 is 32 bits, the IPv4 address can be used as the lower 32 bits of the GID, and the upper 96 bits of the GID are all complemented by 0; if the IPv6 address is converted into the GID, since the address length of the IPv6 is 128 bits, the IPv6 address can be directly used as the GID.

After receiving the query response message, the virtualRDMA management module 414 parses the query response message to obtain the identifier of the second port corresponding to the first port. The virtualRDMA management module 414 may scan in the bus according to the identifier of the second port, query the corresponding port of the virtual ethernet network card and the virtual IP address pre-configured on the port of the virtual ethernet network card, and obtain the first GID corresponding to the first port based on the virtual IP address and the pre-stored GID generation algorithm.

Instep 604, the firstvirtual machine 410 sets the address of the first port to the first GID.

In an implementation, after the virtualRDMA management module 414 determines the first GID of the first port, the address of the first port may be set to the first GID.

In this embodiment of the present invention, when the administrator modifies the virtual IP address of the second port, the virtualRDMA management module 414 may further perform corresponding configuration on the GID of the first port, and the specific steps may be as follows:

step one, when the firstvirtual machine 410 detects the ethernet address change information carrying the identifier of the second port, the identifier of the corresponding virtualRDMA network card 412 and the identifier of the first port are determined according to the identifier of the second port and a pre-stored third mapping table.

The third mapping table includes a correspondence relationship among the identifier of the port of the virtualethernet network card 411, the identifier of the virtualRDMA network card 412, and the identifier of the port of the virtualRDMA network card 412.

In an implementation, a third mapping table may be stored in the virtualRDMA management module 414, and the third mapping table may be set by the virtualRDMA management module 414. The virtualRDMA management module 414 receives the query response information sent by the virtualRDMA control module 421, analyzes the query response information, and obtains the identifier of the second port, and the virtualRDMA management module 414 may store the correspondence between the identifier of the virtualRDMA network card 412, the identifier of the first port, and the identifier of the second port in the third mapping table.

For real-time monitoring of IP address change of the second port, the virtualRDMA management module 414 may register an ethernet address change event with the operating system of the virtual machine, so that when the administrator modifies the virtual IP address on the second port, the virtualRDMA management module 414 may configure the GID of the first port accordingly. For example, virtualRDMA management module 414 calls the register _ inetaddr _ notifier () interface to register a callback function for an ethernet address change event in the operating system.

When the operating system of the virtual machine detects a management instruction corresponding to a virtual IP address, the ethernet address change event is triggered, where the management instruction may be a delete instruction or a configuration instruction.

When the operating system detects a delete instruction for deleting the virtual IP address of the second port by the administrator, the operating system may determine that the type of the ethernet address change event is network card shutdown. When the operating system detects a configuration instruction for configuring the virtual IP address of the port of the virtual ethernet network card by the administrator, the operating system may determine that the type of the ethernet address change event is network card startup. The operating system may send ethernet address change event information to the virtualRDMA management module 414, where the ethernet address change event information carries an identifier of the determined type of the ethernet address change event.

The os sends ethernet address change information to theRDMA management module 414, where the ethernet address change information carries the identifier of the port of the virtual ethernet network card and the type of the ethernet address change event.

The virtualRDMA management module 414 receives the ethernet address change information, parses the ethernet address change information to obtain the identifier of the port of the virtual ethernet network card, and then queries whether the identifier of the virtualRDMA network card 412 and the identifier of the port of the virtual RDMA network card corresponding to the identifier of the port of the virtual ethernet network card exist in the third mapping table. If yes, executing step two, otherwise, not executing other processing.

Step two, the firstvirtual machine 410 configures the GID of the first port according to the event type carried in the ethernet address change information.

In implementation, after obtaining the event type carried in the ethernet address change information, the virtualRDMA management module 414 may correspondingly update the GID of the first port according to the event type.

Optionally, the virtualRDMA management module 414, according to the type of the ethernet address change event, specifically divides the configuration update of the GID of the first port into the following two cases:

in the first case: when the event type is network card shutdown, the firstvirtual machine 410 deletes the address of the first port.

In the second case: when the event type is network card starting, the firstvirtual machine 410 acquires a virtual IP address corresponding to the second port identifier, and the firstvirtual machine 410 determines the second GID of the first port according to the virtual IP address and a preset GID generation algorithm, and sets the first port address as the second GID.

In implementation, after receiving the ethernet address change event information, the virtualRDMA management module 414 obtains an identifier of an event type, and if the identifier is used to indicate that the network card is closed, deletes the address of the first port corresponding to the port of the virtual ethernet network card.

If the identifier is used to indicate that the network card is started, the virtualRDMA management module 414 may query, in the bus, the virtual IP address configured on the second port according to the identifier of the second port, and then determine, according to the virtual IP address and a pre-stored GID generation algorithm, the second GID of the first port corresponding to the port of the virtual ethernet network card. After the virtualRDMA management module 414 determines the second GID for the first port, the address of the first port may be set to the second GID.

The present invention further provides an example of a method for associating a port of a virtual ethernet network card with a port of a virtual RDMA network card, as shown in fig. 5 and 7, the specific processing procedures are as follows:

instep 701, after the virtualRDMA network card 412 is started, the virtualRDMA management module 414 sends a registration message to the RDMA network card driver module to register the virtualRDMA network card 412, where the registration message carries the identifier of the virtualRDMA network card 412.

Instep 702, the RDMA network card driver module registers the virtualRDMA network card 412 and stops managing the GIDs of the ports of the virtual RDMA network cards on the virtualRDMA network card 412.

Instep 703, the virtualRDMA management module 414 obtains the identifier of the virtualRDMA network card 412 and the identifier of the first port.

Instep 704, the virtualRDMA management module 414 sends a query request message to the virtualRDMA control module 421, where the query request message carries the identifier of the virtualRDMA network card 412 and the identifier of the first port.

Step 705, the virtualRDMA control module 421 returns a query response message to the virtualRDMA management module 414, where the query response message carries the identifier of the second port.

Instep 706, the virtualRDMA management module 414 queries whether the second port is configured with a virtual IP address, if so, then step 707 and 708 are executed, otherwise,step 709 is executed.

Instep 707, the virtualRDMA management module 414 determines the first GID of the first port according to the preset GID generation algorithm and the virtual IP address corresponding to the identifier of the second port.

Atstep 708, virtualRDMA management module 414 sets the address of the first port to the first GID.

Instep 709, virtualRDMA management module 414 registers the Ethernet Address Change event with the operating System.

Step 710, when the virtualRDMA management module 414 detects the ethernet address change information, determining the identifier of the corresponding virtualRDMA network card 412 and the identifier of the first port according to the identifier of the second port and a third mapping table stored in advance.

Instep 711, the virtualRDMA management module 414 obtains the event type carried in the ethernet address change information, if the event type is network card start,step 712 and 713 are executed, and if the event type is network card close,step 714 is executed.

Instep 712, the virtualRDMA management module 414 obtains the virtual IP address corresponding to the second port identifier, and determines the second GID of the first port according to the virtual IP address and the preset GID generation algorithm.

Instep 713, virtualRDMA management module 414 sets the first port address to the second GID.

Instep 714, virtualRDMA management module 414 deletes the address of the first port.

The specific processing procedures fromstep 701 to step 714 can refer tosteps 601 to 604, which are not described in detail herein.

In this embodiment of the present invention, after the virtualRDMA management module 414 determines the GID of the first port of the virtualRDMA network card 412, the virtualRDMA control module 421 may manage the QP attribute setting information sent by the virtual machine, as shown in fig. 5 and 8, the steps may be as follows:

instep 801, the firstvirtual machine 410 determines the global identity GID of the RDMA network corresponding to the firstvirtual machine 410 according to the virtual address of the firstvirtual machine 410.

In implementation, when an application in the firstvirtual machine 410 needs to perform data transmission with an application in the second virtual machine, the application in the firstvirtual machine 410 may determine the GID of the corresponding first port according to the local virtual IP address and the pre-stored GID generation algorithm.

Step 802, the firstvirtual machine 410 sends QP attribute setting information to thehost 420 through a first port corresponding to the GID, where the first port is a port of the virtual RDMA network card of the first virtual machine, the QP attribute setting information includes a first instruction and a second instruction, the first instruction is used to instruct thehost 420 to set a source address in a QP queue as a virtual address of the firstvirtual machine 410, the second instruction is used to instruct thehost 420 to set a destination address in the QP queue as a virtual address of the second virtual machine, and the QP queue is a QP queue used for communication between the first virtual machine and the second virtual machine in the QP queue corresponding to the HCA card.

In implementation, the application in the firstvirtual machine 410 may send QP attribute setting information to the virtualRDMA control module 421 through the first port corresponding to the GID. Wherein the QP attribute setting information includes a first instruction and a second instruction, wherein the first instruction may be used to indicate that the source address is set to the virtual IP address of the firstvirtual machine 410, and the second instruction may be used to indicate that the destination address is set to the virtual IP address of the second virtual machine. The virtual IP address of the second virtual machine may be set in advance by an administrator in the application of the firstvirtual machine 410. For example, the firstvirtual machine 410 sends QP attribute setting information to thehost 420 through an ib _ modify _ QP (QP _ attr) interface, where QP _ attr specifies a source address as the virtual IP address of the firstvirtual machine 410 and a destination address as the virtual IP address of the second virtual machine.

Instep 803, thehost 420 deletes the first instruction, and modifies the second instruction into a third instruction, where the third instruction is used to instruct thehost 420 to set the destination address in the QP queue as the physical address of the second virtual machine.

In implementation, if the source address is the virtual IP address of the firstvirtual machine 410 and the destination address is the virtual IP address of the second virtual machine in the attributes of the QP queue, when thesubsequent HCA card 430 performs protocol encapsulation on data sent by the application program of the firstvirtual machine 410 to the application program of the second virtual machine, the virtual IP address of the firstvirtual machine 410 is used as the source address and the virtual IP address of the second virtual machine is used as the destination address, and then the data is transmitted through the physical network between the firstvirtual machine 410 and the second virtual machine. Since the physical network does not have the routing entry of the virtual IP address corresponding to the firstvirtual machine 410 and the second virtual machine, but only has the routing entry of the physical IP address corresponding to the firstvirtual machine 410 and the second virtual machine, after receiving the QP attribute setting information, the virtualRDMA control module 421 needs to delete the first instruction, and at the same time, modify the virtual IP address corresponding to the second virtual machine in the second instruction to the physical IP address corresponding to the second virtual machine, so as to obtain the third instruction. The third instruction is to instruct to set the destination address to a physical IP address of the second virtual machine.

Optionally, the second instruction is modified into a third instruction, and the steps may be as follows:

in step one, thehost 420 determines a physical IP address of the second virtual machine according to the pre-stored virtual IP address of the second virtual machine in the first mapping table and the second instruction.

The first mapping table comprises a corresponding relation between a virtual IP address and a physical IP address.

In implementation, the virtualRDMA control module 421 may store a first mapping table in advance, and the first mapping table may be set by an administrator in advance.

After receiving the QP attribute setting information, the virtualRDMA control module 421 parses the QP attribute setting information to obtain the virtual IP address of the second virtual machine in the second instruction, and then queries the physical IP address corresponding to the virtual IP address of the second virtual machine in the first mapping table.

It should be noted that, in order to improve the query efficiency, the administrator may only cache the entry of the first mapping table that has been queried in the virtualRDMA control module 421, and store the complete first mapping table in thecontroller 423 of thehost 420. The virtualRDMA control module 421 receives the QP attribute setting information, obtains the virtual IP address of the second virtual machine in the second instruction, and then locally queries whether a physical IP address corresponding to the virtual IP address of the second virtual machine exists. If yes, executing the step two; otherwise, the virtualRDMA control module 421 sends an address query request message to thecontroller 423, where the address query request message carries the virtual IP address of the second virtual machine. Thecontroller 423 receives the address query request information sent by the virtualRDMA control module 421 to obtain the virtual IP address of the second virtual machine, and then thecontroller 423 queries whether a physical IP address corresponding to the virtual IP address of the second virtual machine exists in a pre-stored first mapping table, if so, sends address query response information to the virtualRDMA control module 421, where the address query response information carries the physical IP address; otherwise, address query failure information is returned to the virtualRDMA control module 421. When the virtualRDMA control module 421 receives the address query response message, the physical IP address is obtained, then step two is executed, and the corresponding relationship between the virtual IP address of the second virtual machine and the physical IP address is stored locally, so as to provide fast query in the following. When the virtualRDMA control module 421 receives the address query failure information, it may notify the administrator through a display or a short message, so that the administrator can modify the error in time.

Step two, thehost 420 modifies the virtual IP address of the second virtual machine in the second instruction into the physical IP address of the second virtual machine, and obtains a third instruction.

In implementation, after the virtualRDMA control module 421 determines the physical IP address corresponding to the virtual IP address of the second virtual machine, the virtual IP address of the second virtual machine in the second instruction may be deleted, and then the destination address in the second instruction is set as the physical IP address of the second virtual machine, so as to obtain the third instruction.

Instep 804, thehost 420 sets the attribute of the QP queue according to the modified QP attribute setting information, so that theHCA card 430 sends data to the second virtual machine based on the set QP queue.

In implementation, the virtualRDMA control module 421 sets the destination address in the QP queue to the physical IP address of the second virtual machine according to the modified QP attribute setting information. Meanwhile, the virtualRDMA control module 421 scans the bus, queries the physical IP address configured on the corresponding ethernet interface, and then sets the source address as the physical IP address of the ethernet interface. Subsequently, when the application program in the firstvirtual machine 410 sends data to the application program in the second virtual machine, the RDMA interface may read data from the memory corresponding to the application program of the firstvirtual machine 410, encapsulate the data with the physical IP address of the ethernet interface as a source address and the physical IP address of the second virtual machine as a destination address, obtain a data packet, and then send the data packet to the port of theHCA card 430 corresponding to the second virtual machine through the physical network between the firstvirtual machine 410 and the second virtual machine.

In the embodiment of the present invention, when thehost 420 receives QP attribute setting information sent by the firstvirtual machine 410, the QP attribute setting information includes a first instruction and a second instruction, the first instruction is used to instruct thehost 420 to set a source address in a QP queue as a virtual address of the firstvirtual machine 410, the second instruction is used to instruct thehost 420 to set a destination address in the QP queue as a virtual address of the second virtual machine, thehost 420 deletes the first instruction and modifies the second instruction into a third instruction, and the third instruction is used to instruct thehost 420 to set the destination address in the QP queue as a physical address of the second virtual machine; and then setting the attribute of the QP queue according to the modified QP attribute setting information, so that theHCA card 430 sends data to the second virtual machine based on the set QP queue. Based on the invention, the virtual network and the physical network of the virtual machine can be isolated, the safety of the virtual machine is ensured, and the RoCE protocol can be compatible.

Based on the same technical concept, an embodiment of the present invention further provides an apparatus, as shown in fig. 5, the apparatus includes a first virtual machine 410, a host 420, and a host channel adapter HCA card 430, where the first virtual machine 410 includes a virtual ethernet card 411 and a virtual RDMA card 412, the virtual ethernet card 411 includes a port of the virtual ethernet card, and the virtual RDMA card 412 includes a port of the virtual RDMA card; wherein: the first virtual machine 410 is configured to determine, according to a virtual internet protocol address of the first virtual machine 410, a global identity GID of an RDMA network corresponding to the first virtual machine 410 when a data connection needs to be established with a second virtual machine; the first virtual machine 410 is further configured to send QP attribute setting information to the host 420 through the first port corresponding to the GID, where the QP attribute setting information includes a first instruction and a second instruction, the first instruction is used to instruct the host 420 to set a source address in a QP queue as a virtual address of the first virtual machine 410, and the second instruction is used to instruct the host 420 to set a destination address in the QP queue as a virtual address of the second virtual machine; the host 420 is used for deleting the first instruction and modifying the second instruction into a third instruction, wherein the third instruction is used for indicating the host 420 to set a destination address in the QP queue as a physical address of the second virtual machine; and the host 420 is configured to set an attribute of the QP queue according to the modified QP attribute setting information, so that the HCA card 430 sends data to the second virtual machine based on the set QP queue.

In a possible implementation manner, thehost 420 is specifically configured to: determining a physical address of a second virtual machine according to a first mapping table stored in advance and the virtual address of the second virtual machine in a second instruction, wherein the first mapping table comprises a corresponding relation between the virtual address and the physical address; and modifying the virtual address of the second virtual machine in the second instruction into the physical address of the second virtual machine to obtain a third instruction.

In one possible implementation, the firstvirtual machine 410 is further configured to: acquiring an identifier of a preset virtualRDMA network card 412 and an identifier of a first port; determining the corresponding identifier of the second port according to the identifier of the virtualRDMA network card 412, the identifier of the first port and a second mapping table pre-stored in thehost 420, wherein the second mapping table comprises the corresponding relationship among the identifier of the virtualRDMA network card 412, the identifier of the port of the virtual RDMA network card and the identifier of the port of the virtual Ethernet network card; determining a first GID of the first port according to a preset GID generation algorithm and a virtual address corresponding to the identifier of the second port;

the address of the first port is set to the first GID.

In a possible implementation manner, the firstvirtual machine 410 is specifically configured to: sending query request information to thehost 420, where the query request carries the identifier of the virtualRDMA network card 412 and the identifier of the first port; thehost 420 is specifically configured to: determining the identifier of the corresponding second port according to the query request information and a pre-stored second mapping table; and sending query response information to the firstvirtual machine 410, wherein the query response information carries the identifier of the second port.

In one possible implementation, the firstvirtual machine 410 is further configured to: when detecting the change information of the ethernet address carrying the identifier of the second port, determining the identifier of the corresponding virtualRDMA network card 412 and the identifier of the first port according to the identifier of the second port and a pre-stored third mapping table, where the third mapping table includes the correspondence between the identifier of the port of the virtualethernet network card 411, the identifier of the virtualRDMA network card 412 and the identifier of the port of the virtualRDMA network card 412; and updating the address of the first port according to the event type carried in the Ethernet address change information.

In a possible implementation manner, the firstvirtual machine 410 is specifically configured to: when the event type is the network card starting, acquiring a virtual address corresponding to the second port identification; and determining a second GID of the first port according to the virtual address and a preset GID generation algorithm, and setting the address of the first port as the second GID.

In a possible implementation manner, the firstvirtual machine 410 is specifically configured to: and when the event type is the network card closing, deleting the address of the first port.

In the embodiment of the present invention, when thehost 420 receives QP attribute setting information sent by the firstvirtual machine 410, the QP attribute setting information includes a first instruction and a second instruction, the first instruction is used to instruct thehost 420 to set a source address in a QP queue as a virtual address of the firstvirtual machine 410, the second instruction is used to instruct thehost 420 to set a destination address in the QP queue as a virtual address of the second virtual machine, thehost 420 deletes the first instruction and modifies the second instruction into a third instruction, and the third instruction is used to instruct thehost 420 to set the destination address in the QP queue as a physical address of the second virtual machine; and then setting the attribute of the QP queue according to the modified QP attribute setting information, so that theHCA card 430 sends data to the second virtual machine based on the set QP queue. Based on the invention, the virtual network and the physical network of the virtual machine can be isolated, the safety of the virtual machine is ensured, and the RoCE protocol can be compatible.

Based on the same technical concept, an embodiment of the present invention further provides an apparatus, as shown in fig. 9, where theapparatus 900 includes:memory 910,processing circuitry 920, and host channeladapter HCA card 930. Theapparatus 900 can perform any one of the above methods described in the present application, and the system architecture, noun explanation, beneficial effect, etc. involved in any one of the above methods, please refer to the related paragraphs, which are not described herein again.

Thememory 910 is used for storing codes.

Theprocessing circuit 920 is configured to run a first virtual machine and a host according to the code stored in thememory 910.

TheHCA card 930 is configured to send data to the second virtual machine.

Theprocessing circuit 920 is further configured to implement, by the first virtual machine and the host, the following method:

the first virtual machine includes a port of a virtual remote direct data access, RDMA, network card, the method comprising:

the first virtual machine determines a Global Identity (GID) of an RDMA network corresponding to the first virtual machine according to the virtual address of the first virtual machine;

the first virtual machine sends queue pair QP attribute setting information to the host through a first port corresponding to the GID, wherein the first port is a port of the virtual RDMA network card of the first virtual machine, the QP attribute setting information comprises a first instruction and a second instruction, the first instruction is used for indicating the host to set a source address in a QP queue as a virtual address of the first virtual machine, the second instruction is used for indicating the host to set a destination address in the QP queue as a virtual address of the second virtual machine, the QP queue is a QP queue used for communication between the first virtual machine and the second virtual machine, and the destination address in the QP queue is a virtual address of the second virtual machine;

the host deletes the first instruction and modifies the second instruction into a third instruction, wherein the third instruction is used for indicating the host to set a destination address in a QP queue as a physical address of the second virtual machine;

and the host sets the attribute of the QP queue according to the modified QP attribute setting information so that the HCA card sends data to the second virtual machine based on the set QP queue.

In a possible implementation manner, the modifying the second instruction into a third instruction includes:

the host determines a physical address of the second virtual machine according to a first mapping table stored in advance and a virtual address of the second virtual machine in the second instruction, wherein the first mapping table comprises a corresponding relation between the virtual address and the physical address;

and the host modifies the virtual address of the second virtual machine in the second instruction into the physical address of the second virtual machine to obtain a third instruction.

In one possible implementation manner, the first virtual machine further includes a port of a virtual ethernet network card, and the method further includes:

the first virtual machine acquires a preset identifier of the virtual RDMA network card and an identifier of the first port;

the first virtual machine determines the identifier of a corresponding second port according to the identifier of the virtual RDMA network card, the identifier of the first port and a second mapping table prestored in the host, wherein the second mapping table comprises the corresponding relation among the identifier of the virtual RDMA network card, the identifier of the port of the virtual RDMA network card and the identifier of the port of the virtual Ethernet network card, and the second port is the port of the virtual Ethernet network card of the first virtual machine;

the first virtual machine determines a first GID of the first port according to a preset GID generation algorithm and a virtual address corresponding to the identifier of the second port;

the first virtual machine sets an address of the first port to the first GID.

In one possible implementation manner, the determining, by the first virtual machine, an identifier of a corresponding second port according to the identifier of the virtual RDMA network card, the identifier of the first port, and a second mapping table pre-stored in the host includes:

the first virtual machine sends query request information to the host, wherein the query request carries the identifier of the virtual RDMA network card and the identifier of the first port;

the host determines the identifier of the corresponding second port according to the query request information and a pre-stored second mapping table;

and the host sends query response information to the first virtual machine, wherein the query response information carries the identifier of the second port.

In one possible implementation, the method further includes:

when the first virtual machine detects Ethernet address change information carrying the identifier of the second port, determining the identifier of the corresponding virtual RDMA network card and the identifier of the first port according to the identifier of the second port and a prestored third mapping table, wherein the third mapping table comprises the corresponding relation among the identifier of the port of the virtual Ethernet network card, the identifier of the virtual RDMA network card and the identifier of the port of the virtual RDMA network card;

and the first virtual machine updates the address of the first port according to the event type carried in the Ethernet address change information.

In a possible implementation manner, the updating, by the first virtual machine, the address of the first port according to the event type carried in the ethernet address change information includes:

when the event type is the network card starting, the first virtual machine acquires a virtual address corresponding to the second port identifier;

and the first virtual machine determines a second GID of the first port according to the virtual address and a preset GID generation algorithm, and sets the address of the first port as the second GID.

In a possible implementation manner, the updating, by the first virtual machine, the address of the first port according to the event type carried in the ethernet address change information includes:

and when the event type is network card closing, the first virtual machine deletes the address of the first port.

In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded and executed on a computer, cause the processes or functions described in accordance with the embodiments of the invention to occur, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, from one website site, computer, server, or data center to another website site, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that incorporates one or more of the available media. The usable medium may be a magnetic medium (e.g., floppy Disk, hard Disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., Solid State Disk (SSD)), among others.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims (19)

1. A method for data transmission, wherein a device applied by the method comprises a first virtual machine and a host, and the method comprises the following steps:

the host obtains a first instruction and a second instruction from the first virtual machine, wherein the first instruction is used for indicating that a source address of a queue pair QP queue is a virtual address of the first virtual machine, the second instruction is used for indicating that a destination address of the QP queue is a virtual address of a second virtual machine, and the QP queue is used for communication between the first virtual machine and the second virtual machine;

the host deletes the first instruction and modifies the second instruction into a third instruction, wherein the third instruction is used for indicating that a destination address in the QP queue is a physical address of the second virtual machine;

and the host sets the QP queue according to the third instruction so as to send data to the second virtual machine through the set QP queue.

2. The method of claim 1, further comprising:

the first virtual machine sends the first instruction and the second instruction to the host through a first port, wherein the first port is a port of a virtual remote direct access (RDMA) network card of the first virtual machine, and the first port corresponds to a Global Identity (GID) of the first virtual machine in an RDMA network.

3. The method of claim 1, wherein modifying the second instruction into a third instruction comprises:

the host determines a physical address of the second virtual machine according to a first mapping table stored in advance and a virtual address of the second virtual machine in the second instruction, wherein the first mapping table comprises a corresponding relation between the virtual address and the physical address;

and the host modifies the virtual address of the second virtual machine in the second instruction into the physical address of the second virtual machine to obtain a third instruction.

4. The method of claim 2, wherein the first virtual machine further comprises a port of a virtual ethernet network card, the method further comprising:

the first virtual machine acquires the identifier of the virtual RDMA network card and the identifier of the first port;

the first virtual machine determines the identifier of a corresponding second port according to the identifier of the virtual RDMA network card, the identifier of the first port and a second mapping table prestored in the host, wherein the second mapping table comprises the corresponding relation among the identifier of the virtual RDMA network card, the identifier of the port of the virtual RDMA network card and the identifier of the port of the virtual Ethernet network card, and the second port is the port of the virtual Ethernet network card of the first virtual machine;

the first virtual machine determines a first GID of the first port according to a preset GID generation algorithm and a virtual address corresponding to the identifier of the second port;

the first virtual machine sets an address of the first port to the first GID.

5. The method of claim 4, wherein the first virtual machine determining the identity of the corresponding second port from the identity of the virtual RDMA network card, the identity of the first port, and a second mapping table pre-stored in the host comprises:

the first virtual machine sends query request information to the host, wherein the query request information carries the identifier of the virtual RDMA network card and the identifier of the first port;

the host determines the identifier of the corresponding second port according to the query request information and a pre-stored second mapping table;

and the host sends query response information to the first virtual machine, wherein the query response information carries the identifier of the second port.

6. The method of claim 4, further comprising:

when the first virtual machine detects information carrying the identifier of the second port and used for changing the Ethernet address, determining the identifier of the corresponding virtual RDMA network card and the identifier of the first port according to the identifier of the second port and a prestored third mapping table, wherein the third mapping table comprises the corresponding relation among the identifier of the port of the virtual Ethernet network card, the identifier of the virtual RDMA network card and the identifier of the port of the virtual RDMA network card;

and the first virtual machine updates the address of the first port according to the event type carried in the information for changing the Ethernet address.

7. The method according to claim 6, wherein the updating, by the first virtual machine, the address of the first port according to the event type carried in the information for changing the ethernet address includes:

when the event type is the network card starting, the first virtual machine acquires a virtual address corresponding to the second port identifier;

and the first virtual machine determines a second GID of the first port according to a virtual address corresponding to the second port identifier and a preset GID generation algorithm, and sets the address of the first port as the second GID.

8. The method according to claim 6, wherein the updating, by the first virtual machine, the address of the first port according to the event type carried in the ethernet address change information includes:

and when the event type is network card closing, the first virtual machine deletes the address of the first port.

9. A data sending device, characterized in that the device comprises a host, wherein the host is used for obtaining a first instruction and a second instruction from a first virtual machine, the first instruction is used for indicating that a source address of a queue pair QP queue is a virtual address of the first virtual machine, the second instruction is used for indicating that a destination address of the QP queue is a virtual address of a second virtual machine, and the QP queue is used by the first virtual machine for communicating with the second virtual machine; deleting the first instruction, and modifying the second instruction into a third instruction, wherein the third instruction is used for indicating that a destination address in the QP queue is a physical address of the second virtual machine; and the host sets the QP queue according to the third instruction so as to send data to the second virtual machine through the set QP queue.

10. The apparatus of claim 9, further comprising a first virtual machine,

the first virtual machine is used for sending the first instruction and the second instruction to the host through a first port, the first port is a port of a virtual remote direct access (RDMA) network card of the first virtual machine, and the first port corresponds to a Global Identity (GID) of the first virtual machine in an RDMA network.

11. The data sending device of claim 9, wherein the host is specifically configured to:

determining a physical address of the second virtual machine according to a first mapping table stored in advance and a virtual address of the second virtual machine in the second instruction, wherein the first mapping table comprises a corresponding relation between the virtual address and the physical address;

and modifying the virtual address of the second virtual machine in the second instruction into the physical address of the second virtual machine to obtain a third instruction.

12. The data sending device of claim 9, wherein the first virtual machine further comprises a port of a virtual ethernet network card, and the first virtual machine is further configured to:

acquiring an identifier of the virtual RDMA network card and an identifier of the first port; determining a corresponding identifier of a second port according to the identifier of the virtual RDMA network card, the identifier of the first port and a second mapping table prestored in the host, wherein the second mapping table comprises a corresponding relation among the identifier of the virtual RDMA network card, the identifier of the port of the virtual RDMA network card and the identifier of the port of the virtual Ethernet network card, and the second port is the port of the virtual Ethernet network card of the first virtual machine; determining a first GID of the first port according to a preset GID generation algorithm and a virtual address corresponding to the identifier of the second port; setting an address of the first port to the first GID.

13. The data sending device of claim 9, wherein the first virtual machine is specifically configured to:

sending query request information to the host, wherein the query request information carries the identifier of the virtual RDMA network card and the identifier of the first port;

the host is specifically configured to: determining the identifier of a corresponding second port according to the query request information and a pre-stored second mapping table; and sending query response information to the first virtual machine, wherein the query response information carries the identifier of the second port.

14. The data sending device of claim 11, wherein the first virtual machine is further configured to:

when the first virtual machine detects information carrying the identifier of the second port and used for changing the Ethernet address, determining the identifier of the corresponding virtual RDMA network card and the identifier of the first port according to the identifier of the second port and a prestored third mapping table, wherein the third mapping table comprises the corresponding relation among the identifier of the port of the virtual Ethernet network card, the identifier of the virtual RDMA network card and the identifier of the port of the virtual RDMA network card;

and the first virtual machine updates the address of the first port according to the event type carried in the information for changing the Ethernet address.

15. The data sending device of claim 14, wherein the first virtual machine is specifically configured to:

when the event type is the starting of the network card, acquiring a virtual address corresponding to the second port identifier;

and determining a second GID of the first port according to the virtual address corresponding to the second port identifier and a preset GID generation algorithm, and setting the address of the first port as the second GID.

16. The data sending device of claim 14, wherein the first virtual machine is specifically configured to:

and when the event type is the network card closing, deleting the address of the first port.

17. A data transmission apparatus, characterized in that the apparatus comprises: a memory, a processing circuit and a host channel adapter HCA card;

wherein the memory is used for storing codes;

the processing circuit is used for operating a first virtual machine and a host according to the codes stored in the memory;

the processing circuitry, further to implement the method of any of claims 1 to 8 by the first virtual machine and the host;

and the HCA card is used for sending data to the second virtual machine.

18. A computer-readable storage medium comprising instructions that, when executed on a computer, cause the computer to perform the method of any of claims 1 to 8.

19. A computer program product comprising computer program instructions which, when executed by processing circuitry, uses the processing circuitry to perform the method of any of claims 1 to 8.

Images