WO2022089551A1

WO2022089551A1 - Server system and data processing method

Info

Publication number: WO2022089551A1
Application number: PCT/CN2021/127198
Authority: WO
Inventors: 于海鑫; 邹倩; 陈士阳
Original assignee: 华为云计算技术有限公司
Priority date: 2020-10-30
Filing date: 2021-10-28
Publication date: 2022-05-05
Also published as: CN114528032A

Abstract

Disclosed in the present application are a server system and a data processing method. By means of expansion using an offload card connecting to a heterogeneous card, the offload card can take charge of processing a management and control plane of a second virtual device of the heterogeneous card, and thus, resources of servers can be saved and implementation of lightweight servers can be facilitated. The server system in the present application comprises a server, an offload card, and a heterogeneous card, the offload card is connected to the server, and the heterogeneous card is connected to the offload card. The offload card is used to record a mapping relationship between a first virtual device of the offload card and a second virtual device of the heterogeneous card, and achieve communication between a cloud instance of the server and the second virtual device of the heterogeneous card according to the mapping relationship; the first virtual device of the offload card is directly connected to the cloud instance of the server, and the second virtual device of the heterogeneous card is directly connected to the offload card.

Description

A server system and data processing method

technical field

The present application relates to the field of computer technology, and in particular, to a server system and a data processing method.

Background technique

In a cloud scenario, a cloud instance, such as a virtual machine (virtual machine, VM), is deployed on the server. In order to improve the processing speed of input/output (I/O) services, some I/O services in the server can be offloaded to low-cost devices for execution, thereby releasing the central processing unit (central processing unit) of the server. CPU) resources. These devices for offloading I/O services are usually called offload cards, for example, application specific integrated circuits (ASICs) or field programmable gate arrays (FPGAs) and so on.

Under the special requirements of some cloud scenarios, additional heterogeneous cards need to be introduced, such as graphics processing unit (GPU), network interface controller (NIC), and so on. Generally, these heterogeneous cards can be directly connected to the server, so that the offload card and the heterogeneous card respectively occupy different high-speed serial computer expansion bus standard (peripheral component interconnect express, PCIE) interfaces of the server.

Heterogeneous cards usually support single-root I/O virtualization (SRIOV) function, so heterogeneous cards can create virtualized devices. Since these virtualized devices are directly connected to the cloud instance of the server, the heterogeneous card can communicate with the cloud instance of the server based on these virtualized devices.

In the above solution, the expansion of the server directly connected to the heterogeneous card will increase the complexity of the management plane. Therefore, the front-end server will occupy a lot of resources. These resources are used to process the control plane of the virtualization device of the heterogeneous card, which does not conform to the lightweight trend of the front-end server.

SUMMARY OF THE INVENTION

Embodiments of the present application provide a server system and a data processing method. Through the expansion of the connection between the offload card and the heterogeneous card, the offload card can be responsible for processing the management and control plane of the second virtualization device of the heterogeneous card, which can save server resources. , which is conducive to promoting the lightweight of the server.

A first aspect of the embodiments of the present application provides a server system, the server system includes: a server, an offload card and a heterogeneous card, the offload card is connected to the server, and the heterogeneous card is connected to the offload card. The offload card is used to record the mapping relationship between the first virtualization device of the offload card and the second virtualization device of the heterogeneous card, and realize the mapping relationship between the cloud instance of the server and the second virtualization device of the heterogeneous card according to the mapping relationship. communication, wherein the first virtualized device of the offload card is directly connected to the cloud instance of the server, and the second virtual device of the heterogeneous card is directly connected to the offload card.

In the above server system, the offload card is connected between the server and the heterogeneous card. Since the first virtualization device of the offload card is directly connected to the cloud instance of the server, the offload card can communicate with the cloud instance of the server based on the first virtualization device. Furthermore, since the second virtual device of the heterogeneous card is directly connected to the offload card, the heterogeneous card can communicate with the offload card based on the second virtual device. Therefore, the offload card can act as a bridge between the heterogeneous card and the cloud instance of the server, and realize the communication between the heterogeneous card and the cloud instance of the server. Since the offload card is directly connected to the expansion of the heterogeneous card, the offload card can be responsible for processing the management and control plane of the second virtualization device of the heterogeneous card, which can save the resources of the server and help to promote the lightweight of the server.

In a possible implementation manner, the server includes a first interface, the uninstall card includes a second interface and a third interface, and the heterogeneous card includes a fourth interface; the first interface is connected to the second interface, and the third interface and the fourth interface connect. In the foregoing implementation manner, the unloading card itself has two interfaces, and the unloading card is connected to the server through one of the interfaces, and is connected to the heterogeneous card through the other interface, so that the unloading card is connected between the unloading card and the server.

In a possible implementation manner, the server system further includes an expansion card, the server includes a first interface, the expansion card includes a second interface, a third interface, a fourth interface and a fifth interface, and the uninstall card includes a sixth interface and a seventh interface Interface, the heterogeneous card includes an eighth interface; the first interface is connected to the second interface, the second interface is connected to the third interface, the fourth interface is connected to the fifth interface, the sixth interface is connected to the third interface, and the seventh interface is connected to The fourth interface is connected, and the eighth interface is connected with the fifth interface. In the foregoing implementation manner, the server, the unloading card and the heterogeneous card are respectively connected to different interfaces of the same expansion card. By setting the connection relationship between the interfaces of the expansion card, the unloading card is connected between the unloading card and the server.

In a possible implementation manner, the heterogeneous card supports the first protocol, the unloading card is loaded with the first driver corresponding to the first protocol, the server supports the second protocol, and the cloud instance of the server is loaded with the first driver corresponding to the second protocol the second driver; the unloading card is specifically used to receive the first operation request from the second virtualization device through the first driver, and the first operation request conforms to the format requirements of the first protocol; the unloading card is specifically used to convert the first operation request into The second operation request, the second operation request conforms to the format requirements of the second protocol; the uninstall card is specifically used to send the second operation request to the second driver through the first virtualization device, so that the cloud instance of the server is based on the second operation request. , to read the target data from the memory of the server's cloud instance or write the target data to the memory of the server's cloud instance. In the foregoing implementation manner, after receiving the first operation request from the second virtualization device of the heterogeneous card through the first driver, the uninstall card can convert the first operation request into a second operation request, and the first operation request conforms to the first protocol. The format requirements of the second operation request conform to the format requirements of the second protocol. Then, the uninstall card sends the first operation request to the second driver of the cloud instance of the server through the first virtualization device, so that the cloud instance of the server completes the corresponding operation based on the first operation request. Therefore, in the server system provided by the present application, even if the cloud instance of the server only loads the second driver, the communication between the cloud instance of the server and the heterogeneous card can be realized.

In a possible implementation manner, the unloading card includes an engine module supporting the second protocol; the unloading card is specifically used to load the first driver after it is determined that the heterogeneous card is started; the unloading card is specifically used to obtain the heterogeneous card through the first driver Device information of the card; the uninstall card is specifically used to configure the engine module as the first virtualization device based on the configuration information, and the configuration information is preset information; the uninstall card is specifically used to receive the data from the second driver through the first virtualization device The enumeration result is used to indicate that the cloud instance of the server has detected the first virtualization device; the uninstall card is specifically used to construct the second virtualization device of the heterogeneous card and the first virtual device of the uninstall card based on the device information and the enumeration result. The mapping relationship between virtualized devices. In the foregoing implementation manner, the offload card can configure the engine module as the first virtualized device based on the configuration information, and can construct the first virtualized device of the offload card based on the enumeration result of the cloud instance of the server and the device information of the heterogeneous card. The mapping relationship with the second virtualization device of the heterogeneous card. For the server side, the engine module can be presented as the first virtualized device, so the cloud instance of the server can send information to the first virtualized device according to its own requirements. After receiving the information through the first virtualization device, the uninstall card can send the information to the first virtualization device of the heterogeneous card based on the mapping relationship between the first virtualization device of the uninstall card and the second virtualization device of the heterogeneous card. Two virtualized devices, thereby realizing the communication between the cloud instance of the server and the heterogeneous card.

In a possible implementation manner, if the first operation request is the first read request, the uninstall card is specifically configured to receive, through the first virtualization device, a second sending request from the second driver, and the second sending request conforms to the first The format requirements of the second protocol, and the second sending request is used to indicate that the cloud instance of the server is about to send the target data; the uninstall card is specifically used to convert the second sending request into a first sending request, and the first sending request conforms to the format of the first protocol. requirements; the offload card is specifically configured to send the first sending request to the second virtualization device through the first driver, so that the heterogeneous card sends the first read request to the offload card through the second virtual device based on the first sending request. In the foregoing implementation manner, after receiving the second sending request of the second driver of the cloud instance of the server through the first virtualization device, the uninstall card can convert the second sending request into the first sending request, and the second sending request conforms to the second protocol. The format requirements of the first sending request conform to the format requirements of the first protocol. Then, the offload card sends the first sending request to the second virtualization device of the heterogeneous card through the first driver, so that the heterogeneous card can determine based on the first sending request that the cloud instance of the server is about to send the target data, thereby initiating the first read ask.

In a possible implementation manner, the uninstall card is specifically configured to receive target data from the second driver through the first virtualization device; the uninstall card is specifically configured to send the target data to the second virtualization device through the first driver. In the foregoing implementation manner, after reading the target data, the cloud instance of the server can send the target data to the first virtualization device of the offload card through its own second driver. Then, the offload card sends the target data to the second virtualization device of the heterogeneous card through the first driver, so that the heterogeneous card successfully reads the target data.

In a possible implementation manner, if the first operation request is a first read request and the second operation request is a second read request, the offload card is further configured to send the second read request to the second read request through the first virtualization device Second drive, so that the cloud instance of the server reads the target data from the memory of the cloud instance of the server based on the second read request, and transmits it to the first virtualization device; the offload card is also used to store the target data in the memory of the offload card The unloading card is also used to receive the first read request from the second virtualization device through the first driver; the unloading card is also used to read target data from the memory of the unloading card based on the first read request; the unloading card is also used for The target data is sent to the second virtualization device through the first driver. In the foregoing implementation manner, the offload card can send a second read request to the second driver of the cloud instance of the server through the first virtualization device, so as to obtain target data from the cloud instance of the server and store it in the memory of the offload card. When the offload card receives the first read request from the second virtualization device of the heterogeneous card through the first driver, the target data can be read from the memory and sent to the second virtualization device of the heterogeneous card, so that the The composition card successfully reads the target data.

In a possible implementation manner, if the first operation request is the first write request, the uninstall card is specifically configured to receive the first receive request from the second virtualization device through the first driver, and the first receive request conforms to the first receive request. The format requirements of the protocol, and the first receiving request is used to instruct the heterogeneous card to request to obtain the available memory space information of the cloud instance of the server; the unloading card is specifically used to convert the first receiving request into a second receiving request, and the second receiving request conforms to Format requirements of the second protocol; the uninstall card is specifically used to send the second reception request to the second driver through the first virtualization device; the uninstall card is specifically used to receive the available memory space information from the second driver through the first virtualization device The unloading card is specifically configured to send the available memory space information to the second virtualization device through the first driver, so that the heterogeneous card sends the first write request to the unloading card through the second virtualization device based on the available memory space information. In the foregoing implementation manner, after receiving the first reception request from the second virtualization device of the heterogeneous card through the first driver, the uninstall card can convert the first reception request into a second reception request, and the first reception request conforms to the first protocol. The format requirement of the second receiving request conforms to the format requirement of the second protocol. Then, the offload card sends the first reception request to the second driver of the cloud instance of the server through the first virtualization device, so that the cloud instance of the server sends available memory space information to the offload card through the second driver. Next, the offload card sends the available memory space information to the second virtualization device of the heterogeneous card through the first driver, so that the heterogeneous card determines which memory location of the cloud instance to write the target data to based on the available memory space information, thereby initiating a First write request.

A second aspect of the embodiments of the present application provides a data processing method, which is characterized in that the method is applied to a server system, and the server system includes: a server, an offload card, and a heterogeneous card, the offload card is connected to the server, and the heterogeneous card is connected to the server. The uninstall card is connected; the uninstall card records the mapping relationship between the first virtualization device of the uninstall card and the second virtualization device of the heterogeneous card; the uninstall card realizes the relationship between the cloud instance of the server and the second virtualization device of the heterogeneous card according to the mapping relationship. Communication between the two, wherein the first virtual device of the offload card is directly connected to the cloud instance of the server, and the second virtual device of the heterogeneous card is directly connected to the offload card.

In the above method, the offload card is connected between the server and the heterogeneous card. Since the first virtualization device of the offload card is directly connected to the cloud instance of the server, the offload card can communicate with the cloud instance of the server based on the first virtualization device. Furthermore, since the second virtual device of the heterogeneous card is directly connected to the offload card, the heterogeneous card can communicate with the offload card based on the second virtual device. Therefore, the offload card can act as a bridge between the heterogeneous card and the cloud instance of the server, and realize the communication between the heterogeneous card and the cloud instance of the server. Since the offload card is directly connected to the expansion of the heterogeneous card, the offload card can be responsible for processing the management and control plane of the second virtualization device of the heterogeneous card, which can save the resources of the server and help to promote the lightweight of the server.

In a possible implementation manner, the heterogeneous card supports the first protocol, the unloading card is loaded with the first driver corresponding to the first protocol, the server supports the second protocol, and the cloud instance of the server is loaded with the first driver corresponding to the second protocol For the second driver, the offload card implements the communication between the cloud instance of the server and the second virtualization device of the heterogeneous card according to the mapping relationship, including: the offload card receives the first operation request from the second virtualization device through the first driver, An operation request meets the format requirements of the first protocol; the uninstall card converts the first operation request into a second operation request, and the second operation request meets the format requirements of the second protocol; the uninstall card converts the second operation request through the first virtualization device The data is sent to the second driver, so that the cloud instance of the server reads the target data from the memory of the cloud instance of the server or writes the target data to the memory of the cloud instance of the server based on the second operation request.

In a possible implementation manner, the offload card includes an engine module supporting the second protocol, and the offload card recording the mapping relationship between the first virtualization device of the offload card and the second virtualization device of the heterogeneous card includes: After the heterogeneous card is started, the first driver is loaded; the unloading card obtains the device information of the heterogeneous card through the first driver; the unloading card configures the engine module as the first virtualized device based on the configuration information, and the configuration information is preset information; The card receives the enumeration result from the second driver through the first virtualization device, and the enumeration result is used to indicate that the cloud instance of the server has detected the first virtualization device; the uninstall card constructs the heterogeneous card based on the device information and the enumeration result. The mapping relationship between the second virtualization device and the first virtualization device of the offload card.

In a possible implementation manner, if the first operation request is the first read request, before the uninstall card receives the first operation request from the second virtualization device through the first driver, the method further includes: A virtualized device that receives a second sending request from the second driver, the second sending request conforms to the format requirements of the second protocol, and the second sending request is used to indicate that the cloud instance of the server is about to send target data; The sending request is converted into a first sending request, and the first sending request conforms to the format requirements of the first protocol; the offload card sends the first sending request to the second virtualization device through the first driver, so that the heterogeneous card is based on the first sending request , sending a first read request to the offload card through the second virtualization device.

In a possible implementation manner, the method further includes: the offload card receives the target data from the second driver through the first virtualization device; the offload card sends the target data to the second virtualization device through the first driver.

In a possible implementation manner, if the first operation request is a first read request and the second operation request is a second read request, after the uninstall card sends the first sending request to the second virtualization device through the first driver , the method further includes: the unloading card sends the second read request to the second driver through the first virtualization device, so that the cloud instance of the server reads the target data from the memory of the cloud instance of the server based on the second read request, and transmission to the first virtualization device; the unloading card stores the target data in the memory of the unloading card; the unloading card receives the first read request from the second virtualization device through the first driver; The target data is read in the memory of the device; the unloading card sends the target data to the second virtualization device through the first driver.

In a possible implementation manner, if the first operation request is the first write request, before the uninstall card receives the first operation request from the second virtualization device through the first driver, the method further includes: a driver receives a first reception request from the second virtualization device, the first reception request conforms to the format requirements of the first protocol, and the first reception request is used to instruct the heterogeneous card to request to obtain the available memory space information of the cloud instance of the server; The unloading card converts the first receiving request into a second receiving request, and the second receiving request conforms to the format requirements of the second protocol; the unloading card sends the second receiving request to the second driver through the first virtualization device; the unloading card passes the first The virtualization device receives the available memory space information from the second driver; the uninstall card sends the available memory space information to the second virtualization device through the first driver, so that the heterogeneous card passes the second virtualization device based on the available memory space information. Send the first write request to the offload card.

A third aspect of the embodiments of the present application provides a server system, the server system includes: a server, an unloading card, and a heterogeneous card, the heterogeneous card supports a first protocol, and both the server and the unloading card are loaded with the corresponding first protocol. The first driver, one end of the uninstall card is connected to the server, and the other end of the uninstall card is connected to the heterogeneous card; the uninstall card is used to receive the operation request from the heterogeneous card through the first driver of the uninstall card, and the operation request conforms to the first protocol. Format requirements; the offload card is also used to send the operation request to the first driver of the cloud instance of the server through the pass-through channel, so that the cloud instance of the server reads the target data from the memory of the cloud instance of the server or sends the target data to the cloud instance of the server based on the operation request. The memory of the cloud instance writes the target data.

In a possible implementation manner, the server includes a first interface, the uninstall card includes a second interface and a third interface, and the heterogeneous card includes a fourth interface; the first interface is connected to the second interface, and the third interface and the fourth interface connect.

In a possible implementation manner, the server system further includes an expansion card, the server includes a first interface, the expansion card includes a second interface, a third interface, a fourth interface and a fifth interface, and the uninstall card includes a sixth interface and a seventh interface Interface, the heterogeneous card includes an eighth interface; the first interface is connected to the second interface, the second interface is connected to the third interface, the fourth interface is connected to the fifth interface, the sixth interface is connected to the third interface, and the seventh interface is connected to The fourth interface is connected, and the eighth interface is connected with the fifth interface.

In a possible implementation manner, the uninstall card is further configured to receive an initialization request from the first driver of the cloud instance of the server through the pass-through channel; the uninstall card is further configured to send the initialization request to the heterogeneous through the first driver of the uninstall card card, so that the heterogeneous card sends the device information of the heterogeneous card to the first driver of the uninstall card based on the initialization request; the uninstall card is also used for receiving the device information of the heterogeneous card through the first driver of the uninstall card, The channel sends the device information of the heterogeneous card to the first driver of the cloud instance of the server.

In a possible implementation manner, if the operation request is a read request, the offload card is further configured to receive a sending request from the first driver of the cloud instance of the server through the pass-through channel, and the sending request conforms to the format requirements of the first protocol, And the sending request is used to indicate that the cloud instance of the server is about to send the target data; the unloading card is also used to send the sending request to the heterogeneous card through the first driver of the unloading card, so that the heterogeneous card can send the data to the first driver of the unloading card based on the sending request. A driver sends a read request.

In a possible implementation manner, the offload card is further configured to receive target data from the first driver of the cloud instance of the server through the pass-through channel; the offload card is further configured to send the target data to the heterogeneous through the first driver of the offload card Card.

In a possible implementation manner, the offload card is further configured to send a read request to the first driver of the cloud instance of the server through the pass-through channel, so that the cloud instance of the server can delete the memory of the cloud instance of the server from the memory of the cloud instance of the server based on the read request of the offload card. Read the target data in the server, and send the target data to the pass-through channel through the first driver of the cloud instance of the server; the offload card is also used to store the target data in the memory of the offload card; the offload card is also used to pass the first drive of the offload card. A driver receives the read request from the heterogeneous card; the unloading card is also used to read target data from the memory of the unloading card based on the read request of the heterogeneous card; the unloading card is also used to send the target data through the first driver of the unloading card to heterogeneous cards.

In a possible implementation manner, if the operation request is a write request, the unloading card is further configured to receive the receiving request from the heterogeneous card through the first driver of the unloading card, the receiving request conforms to the format requirements of the first protocol, and the receiving The request is used to instruct the heterogeneous card to request to obtain the available memory space information of the cloud instance of the server; the offload card is also used to send the receiving request to the first driver of the cloud instance of the server through the pass-through channel; the offload card is also used to receive through the pass-through channel Available memory space information from the first driver of the cloud instance of the server; the uninstall card is also used to send the available memory space information to the heterogeneous card through the first driver of the uninstall card, so that the heterogeneous card can send the information to the heterogeneous card based on the available memory space information. The offload card sends a write request.

A fourth aspect of the embodiments of the present application provides a data processing method. The method is applied to a server system. The server system includes a server, an offload card, and a heterogeneous card. The heterogeneous card supports a first protocol. The server and the offload card Both are loaded with a first driver corresponding to the first protocol, one end of the unloading card is connected to the server, and the other end of the unloading card is connected to the heterogeneous card; the method includes: the unloading card receives data from the heterogeneous card through the first driver of the unloading card The operation request conforms to the format requirements of the first protocol; the uninstall card sends the operation request to the first driver of the cloud instance of the server, so that the cloud instance of the server reads the target from the memory of the cloud instance of the server based on the operation request data or write target data to the memory of the server's cloud instance.

In a possible implementation manner, the method further includes: the uninstall card receives an initialization request from the first driver of the cloud instance of the server through the pass-through channel; the uninstall card sends the initialization request to the heterogeneous card through the first driver of the uninstall card , so that the heterogeneous card sends the device information of the heterogeneous card to the first driver of the unloading card based on the initialization request; after the unloading card receives the device information of the heterogeneous card through the first driver of the unloading card, the The device information of the card is sent to the first driver of the cloud instance of the server.

In a possible implementation manner, if the operation request is a read request, before the unloading card receives the operation request from the heterogeneous card through the first driver of the unloading card, the method further includes: the unloading card receives the operation request from the server through the pass-through channel The sending request of the first driver of the cloud instance, the sending request conforms to the format requirements of the first protocol, and the sending request is used to indicate that the cloud instance of the server is about to send the target data; the unloading card sends the sending request to the first driver of the unloading card to Heterogeneous card, so that based on the sending request, the heterogeneous card sends a read request to the first driver that unloads the card.

In a possible implementation manner, the method further includes: the offload card receives target data from the first driver of the cloud instance of the server through the pass-through channel; the offload card sends the target data to the heterogeneous card through the first driver of the offload card .

In a possible implementation manner, after the offload card sends the sending request to the heterogeneous card through the first driver of the offload card, the method further includes: the offload card sends the read request to the first driver of the cloud instance of the server through the pass-through channel drive, so that the cloud instance of the server reads the target data from the memory of the cloud instance of the server based on the read request of the offload card, and sends the target data to the pass-through channel through the first driver of the cloud instance of the server; the offload card sends the target data Stored in the memory of the unloading card; the unloading card receives the read request from the heterogeneous card through the first driver of the unloading card; the unloading card reads the target data from the memory of the unloading card based on the read request of the heterogeneous card; The first driver of the card sends the target data to the heterogeneous card.

In a possible implementation manner, if the operation request is a write request, before the unloading card receives the operation request from the heterogeneous card through the first driver of the unloading card, the method further includes: the unloading card passes the first driver of the unloading card Receive the receiving request from the heterogeneous card, the receiving request conforms to the format requirements of the first protocol, and the receiving request is used to instruct the heterogeneous card to request to obtain the available memory space information of the cloud instance of the server; the unloading card sends the receiving request to the The first driver of the cloud instance of the server; the unloading card receives the available memory space information from the first driver of the cloud instance of the server through the pass-through channel; the unloading card sends the available memory space information to the heterogeneous card through the first driver of the unloading card, So that the heterogeneous card sends a write request to the offload card based on the available memory space information.

A fifth aspect of the embodiments of the present application provides an unloading card. One end of the unloading card is connected to the server, and the other end of the unloading card is connected to the heterogeneous card. The unloading card includes: a recording module for recording the first virtual data of the unloading card. The mapping relationship between the virtualization device and the second virtualization device of the heterogeneous card, the communication module is used to realize the communication between the cloud instance of the server and the second virtualization device of the heterogeneous card according to the mapping relationship, wherein, the first virtualization device of the uninstall card is A virtualized device is directly connected to the cloud instance of the server, and the second virtualized device of the heterogeneous card is directly connected to the offload card.

In a possible implementation manner, the heterogeneous card supports the first protocol, the unloading card is loaded with the first driver corresponding to the first protocol, the server supports the second protocol, and the cloud instance of the server is loaded with the first driver corresponding to the second protocol The second driver; the communication module is specifically configured to: receive the first operation request from the second virtualization device through the first driver, and the first operation request conforms to the format requirements of the first protocol; convert the first operation request into the second operation request , the second operation request conforms to the format requirements of the second protocol; the second operation request is sent to the second driver through the first virtualization device, so that the cloud instance of the server reads from the memory of the cloud instance of the server based on the second operation request Fetch or write target data to the server's cloud instance's memory.

In a possible implementation manner, the unloading card includes an engine module supporting the second protocol; the recording module is specifically configured to: load the first driver after determining that the heterogeneous card is started; obtain device information of the heterogeneous card through the first driver ; Configure the engine module as the first virtualized device based on the configuration information, and the configuration information is preset information; Receive the enumeration result from the second driver through the first virtualized device, and the enumeration result is used to indicate that the cloud instance of the server has been The first virtualized device is detected; the mapping relationship between the second virtualized device of the heterogeneous card and the first virtualized device of the unloaded card is constructed based on the device information and the enumeration result.

In a possible implementation manner, if the first operation request is a first read request, the communication module is specifically configured to: receive a second sending request from the second driver through the first virtualization device, and the second sending request conforms to the The format requirements of the second protocol, and the second sending request is used to indicate that the cloud instance of the server is about to send the target data; the second sending request is converted into a first sending request, and the first sending request conforms to the format requirements of the first protocol; A driver sends the first sending request to the second virtualization device, so that the heterogeneous card sends the first read request to the offload card through the second virtualization device based on the first sending request.

In a possible implementation manner, the communication module is specifically configured to: receive the target data from the second driver through the first virtualization device; and send the target data to the second virtualization device through the first driver.

In a possible implementation manner, if the first operation request is a first read request and the second operation request is a second read request, the communication module is further configured to: send the second read request to the the second drive, so that the cloud instance of the server reads the target data from the memory of the cloud instance of the server based on the second read request, and transmits it to the first virtualization device; stores the target data in the memory of the offload card; A driver receives the first read request from the second virtualization device; reads target data from the memory of the offload card based on the first read request; sends the target data to the second virtualization device through the first driver.

In a possible implementation manner, if the first operation request is the first write request, the communication module is specifically configured to: receive the first receive request from the second virtualization device through the first driver, and the first receive request conforms to the first receive request. The format requirements of a protocol, and the first receiving request is used to instruct the heterogeneous card to request to obtain the available memory space information of the cloud instance of the server; the first receiving request is converted into a second receiving request, and the second receiving request conforms to the second protocol. Format requirements; send the second receiving request to the second driver through the first virtualization device; receive the available memory space information from the second driver through the first virtualization device; send the available memory space information to the second driver through the first driver The device is virtualized, so that the heterogeneous card sends a first write request to the offload card through the second virtualization device based on the available memory space information.

A sixth aspect of the embodiments of the present application provides an unloading card, one end of the unloading card is connected to a server, the other end of the unloading card is connected to a heterogeneous card, the heterogeneous card supports a first protocol, and both the server and the unloading card are loaded with The first driver corresponding to the first protocol; the unloading card includes: a communication module for receiving an operation request from the heterogeneous card through the first driver of the unloading card, and the operation request conforms to the format requirements of the first protocol; the communication module is also used to pass The through channel sends the operation request to the first driver of the cloud instance of the server, so that the cloud instance of the server reads target data from the memory of the cloud instance of the server or writes the target data to the memory of the cloud instance of the server based on the operation request.

In a possible implementation manner, the communication module is further configured to: receive an initialization request from the first driver of the cloud instance of the server through the pass-through channel; send the initialization request to the heterogeneous card by uninstalling the first driver of the card, so that the Based on the initialization request, the heterogeneous card sends the device information of the heterogeneous card to the first driver of the uninstall card; after receiving the device information of the heterogeneous card through the first driver of the uninstall card, the device information of the heterogeneous card is sent through the pass-through channel The first driver to the cloud instance of the server.

In a possible implementation manner, if the operation request is a read request, the communication module is further configured to: receive a sending request from the first driver of the cloud instance of the server through the pass-through channel, and the sending request conforms to the format requirements of the first protocol , and the sending request is used to indicate that the cloud instance of the server is about to send the target data; the sending request is sent to the heterogeneous card through the first driver of the uninstall card, so that the heterogeneous card sends the read data to the first driver of the uninstall card based on the sending request ask.

In a possible implementation manner, the communication module is further configured to: receive target data from the first driver of the cloud instance of the server through the pass-through channel; and send the target data to the heterogeneous card through the first driver of the offload card.

In a possible implementation manner, the communication module is further configured to: send the read request to the first driver of the cloud instance of the server through the pass-through channel, so that the cloud instance of the server can download the read request from the cloud instance of the server based on the read request of the unloaded card from the cloud instance of the server. Read the target data in the memory, and send the target data to the pass-through channel through the first driver of the cloud instance of the server; store the target data in the memory of the offload card; receive the read data from the heterogeneous card through the first driver of the offload card request; read target data from the memory of the offload card based on a read request of the heterogeneous card; send the target data to the heterogeneous card through the first driver of the offload card.

In a possible implementation manner, if the operation request is a write request, the communication module is further configured to: receive a reception request from the heterogeneous card through the first driver of the unload card, the reception request conforms to the format requirements of the first protocol, and The receiving request is used to instruct the heterogeneous card to request to obtain the available memory space information of the cloud instance of the server; send the receiving request to the first driver of the cloud instance of the server through the pass-through channel; receive the first driver of the cloud instance from the server through the pass-through channel The available memory space information; the available memory space information is sent to the heterogeneous card through the first driver of the unloading card, so that the heterogeneous card sends a write request to the unloading card based on the available memory space information.

A seventh aspect of the embodiments of the present application provides an uninstall card, the uninstall card includes: a processor and a memory; the memory is used for storing computer-executed instructions; the processor is used for executing computer-executed instructions stored in the memory, so that the uninstall card realizes The method as described in the second aspect, any possible implementation manner of the second aspect, the fourth aspect, or any one possible implementation manner of the fourth aspect.

An eighth aspect of the embodiments of the present application provides a computer storage medium, including computer-readable instructions, when the computer-readable instructions are executed, any one of the possible implementation manners of the second aspect and the second aspect is implemented. , the fourth aspect, or the method described in any possible implementation manner of the fourth aspect.

A ninth aspect of the embodiments of the present application provides a computer program product including instructions, which, when run on a computer, enables the computer to execute any of the second aspect, any possible implementation manner of the second aspect, and the fourth aspect or the method described in any possible implementation manner of the fourth aspect.

In the server system provided by the embodiment of the present application, the offload card is connected between the server and the heterogeneous card. Since the first virtualization device of the offload card is directly connected to the cloud instance of the server, the offload card can communicate with the cloud instance of the server based on the first virtualization device. Furthermore, since the second virtual device of the heterogeneous card is directly connected to the offload card, the heterogeneous card can communicate with the offload card based on the second virtual device. Therefore, the offload card can act as a bridge between the heterogeneous card and the cloud instance of the server, and realize the communication between the heterogeneous card and the cloud instance of the server. Since the offload card is directly connected to the expansion of the heterogeneous card, the offload card can be responsible for processing the management and control plane of the second virtualization device of the heterogeneous card, which can save the resources of the server and help to promote the lightweight of the server.

Description of drawings

FIG. 1 is a schematic structural diagram of a server system provided by an embodiment of the present application;

2 is another schematic structural diagram of a server system provided by an embodiment of the present application;

3 is another schematic structural diagram of a server system provided by an embodiment of the present application;

4 is a schematic structural diagram of an unloading card provided by an embodiment of the present application;

FIG. 5 is another schematic structural diagram of an unloading card provided by an embodiment of the present application;

6 is a schematic flowchart of a data processing method provided by an embodiment of the present application;

FIG. 7 is another schematic flowchart of a data processing method provided by an embodiment of the present application;

FIG. 8 is another schematic flowchart of a data processing method provided by an embodiment of the present application;

9 is another schematic flowchart of a data processing method provided by an embodiment of the present application;

10 is another schematic flowchart of a data processing method provided by an embodiment of the present application;

11 is another schematic flowchart of a data processing method provided by an embodiment of the present application;

12 is another schematic flowchart of a data processing method provided by an embodiment of the present application;

13 is another schematic flowchart of a data processing method provided by an embodiment of the present application;

FIG. 14 is another schematic structural diagram of an unloading card provided by an embodiment of the present application;

FIG. 15 is another schematic structural diagram of an unloading card provided by an embodiment of the present application;

FIG. 16 is another schematic structural diagram of an unloading card provided by an embodiment of the present application.

Detailed ways

The terms "first", "second" and the like in the description and claims of the present application and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchanged under appropriate circumstances, and this is only a distinguishing manner adopted when describing objects with the same attributes in the embodiments of the present application. Furthermore, the terms "comprising" and "having" and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, product or device comprising a series of elements is not necessarily limited to those elements, but may include no explicit or other units inherent to these processes, methods, products, or devices.

The embodiments of the present application are applied in cloud scenarios. In such scenarios, at least one cloud instance is usually deployed on the server (host), such as VM, bare metal server (BMS), application container engine (docker), and so on. In order to meet various requirements of cloud scenarios, servers are generally connected with heterogeneous cards, such as GPUs, NICs and other devices.

Servers usually support standard protocols, while heterogeneous cards usually support non-standard protocols. In order to communicate with the heterogeneous card, the cloud instance of the server needs to additionally load the dedicated driver of the heterogeneous card (that is, the driver corresponding to the non-standard protocol supported by the heterogeneous card), so as to realize the communication between the cloud instance of the server and the heterogeneous card. . Therefore, if the cloud instance of the server only loads the general driver (that is, the driver corresponding to the standard protocol supported by the server), the communication between the cloud instance of the server and the heterogeneous card cannot be realized, and there are certain ecological problems.

Further, the expansion of the server directly connected to the heterogeneous card will increase the complexity of the management plane. Therefore, the front-end server will occupy a lot of resources, and these resources are used to process the control plane of the virtualized device of heterogeneous cards, which does not conform to the trend of light-weight front-end servers.

Further, the core lightweight technology is being explored and evolved, which is also the future development trend. The continuous optimization and downsizing of the kernel will also bring about compatibility problems with some forms of heterogeneous card drivers, and may not be able to adapt to some hardware devices.

Furthermore, the current technology has security problems. Since the heterogeneous card is directly inserted into the server, all VF devices are exposed to the front-end virtual machine, and each VM will affect each other. There is no particularly good solution for recovery and isolation.

Furthermore, the design is not flexible enough, and it relies too much on heterogeneous card manufacturers. When encountering changes in usage scenarios, it is impossible to meet the diverse needs of cloud customers based on simple design modifications.

In order to solve the above problem, an embodiment of the present application provides a server system. FIG. 1 is a schematic structural diagram of a server system provided by an embodiment of the present application. As shown in FIG. 1 , the server system includes: a server, an offload card and a heterogeneous card, wherein one end of the offload card is connected to the server, and the other end of the offload card is connected to the heterogeneous card. Specifically, the physical connection relationship between the aforementioned server, the offload card, and the heterogeneous card can be implemented in various ways, which will be introduced separately below.

In a possible implementation manner, as shown in FIG. 2 ( FIG. 2 is another schematic structural diagram of the server system provided by the embodiment of the application), the server system includes: a server, an offload card, a heterogeneous card, and an expansion card (for example, , riser card, that is, the function expansion card or riser card of the PCIe interface). The server has a first interface, the expansion card has a second interface, a third interface, a fourth interface and a fifth interface, the uninstall card has a sixth interface and a seventh interface, and the heterogeneous card has an eighth interface. Inside the expansion card, the second interface is connected to the third interface, and the fourth interface is connected to the fifth interface. Outside the expansion card, the first interface of the server is connected to the second interface of the expansion card. The sixth interface of the unloading card is connected to the third interface of the expansion card, the seventh interface of the unloading card is connected to the fourth interface of the expansion card, and the eighth interface of the heterogeneous card is connected to the fifth interface of the expansion card, so as to realize the above physical Connection relationship, that is, the heterogeneous card is connected to the server through the offload card.

In another possible implementation manner, as shown in FIG. 3 ( FIG. 3 is another schematic structural diagram of the server system provided by the embodiment of the application), the server system includes: a server, an offload card, and a heterogeneous card. The server has a first interface, the offload card has a second interface and a third interface, and the heterogeneous card has a fourth interface. The first interface of the server is connected to the second interface of the uninstall card, and the third interface of the uninstall card is connected to the fourth interface of the heterogeneous card, so as to realize the above-mentioned physical connection relationship, that is, the heterogeneous card is connected to the server through the uninstall card.

It should be understood that each of the foregoing interfaces may be PCIe interfaces or other types of interfaces, which are not limited here.

In the above server system, the offload card can be used as a bridge between the server and the heterogeneous card. When the server and the heterogeneous card exchange information, the offload card can centrally manage the information transmitted between the two. The following will introduce the server, uninstall card and heterogeneous card respectively:

At least one cloud instance can be deployed on the server, the cloud instance can be VM-BMS, docker, etc., and the server supports standard virtualization I/O protocols (hereinafter referred to as standard protocols), such as virtio protocol, non-volatile Storage Express (non-volatile memory express, NVMe) protocol and more. Therefore, a server and a cloud instance on the server are usually loaded with a driver corresponding to a standard protocol, that is, a standard driver (also called a generic driver), and then interact with a device supporting the standard protocol through the standard driver.

Heterogeneous cards can be GPUs, NICs and other devices developed by different manufacturers. Since the protocols customized by each manufacturer are different from standard protocols, heterogeneous cards usually support non-standard protocols (ie, non-standard virtualized I/O protocols). Since the unloading card needs to communicate with the heterogeneous card, the unloading card needs to load a driver corresponding to a non-standard protocol supported by the heterogeneous card, that is, a non-standard driver (also called a dedicated driver). Further, the heterogeneous card can support the SRIOV function, that is, the heterogeneous card can create a virtualized device including at least one physical function (physical function, PF) and multiple virtual functions (virtual function, VF). The virtualization device of the heterogeneous card is directly connected to the offload card, that is, the heterogeneous card can communicate with the offload card through its virtualization device.

FIG. 4 is a schematic structural diagram of an unloading card provided by an embodiment of the present application. As shown in FIG. 4 , the offload card is a device supporting a standard protocol, including a hardware system and a processor, and the hardware system and the processor are electrically connected, so the processor can realize information interaction with the hardware system. It is worth noting that unloading the card can be achieved in several forms. For example, the offload card can be an ASIC chip, that is, the ASIC chip includes two parts, a hardware system and a processor. For another example, the unloading card may be a combination of an FPGA chip and an ARM chip, wherein the FPGA chip is used as a hardware system, and the ARM chip is used as a processor, etc., which is not limited here.

Specifically, as shown in FIG. 5 (FIG. 5 is another schematic structural diagram of the unload card provided by the embodiment of the application), after the unload card is started, when the processor of the unload card runs the corresponding code, the processor can be inside the unload card. It is presented as a software system (that is, the steps implemented by the software system can be regarded as being executed by the processor from the perspective of physical hardware). , so that the unloading card works between the server and the heterogeneous card, as a bridge for information transmission between the two, and realizes functions such as data format conversion and transmission of information. The following will briefly introduce the hardware system and the software system in conjunction with Figure 5:

The hardware system includes modules such as a protocol processing module, a protocol conversion module, and a storage module (ie, the memory of the offload card). The protocol processing module includes a plurality of engine modules supporting different standard protocols, for example, at least one engine module (virtio engine) supporting the virtio protocol, at least one engine module (NVMe engine) supporting the NVMe protocol, and so on. In addition, the protocol processing module also includes a pass-through channel (PF bypass) and a PCIe switch (PCIe switch). The engine module and the pass-through channel are mounted on the PCIe switch. It should be noted that the engine module may be configured by the processing module of the software system as a virtualization device directly connected to the cloud instance of the server. Both the virtualization device and the pass-through channel inside the offload card can be used to receive information from or send information to the cloud instance of the server through the PCIe switch. It is worth noting that the virtualization device inside the hardware system works in a non-pass-through scenario. (In this scenario, the offload card needs to convert the data format of the transmitted information), while the pass-through channel works in the pass-through scenario (in this scenario, the offload card does not need to convert the data format of the transmitted information). The protocol conversion module can be used to convert the data format of the information in the non-pass-through scenario, for example, convert the information from the heterogeneous card (the information sent by the heterogeneous card to the cloud instance of the server) into the information that meets the format requirements of the standard protocol , so as to be sent to the server through the engine module, for another example, the information from the virtualized device (that is, the information sent by the cloud instance of the server to the heterogeneous card) is converted into information that conforms to the format requirements of the non-standard protocol, and then sent to the heterogeneous card. card and so on. In addition, the protocol conversion module does not convert the data format of the information in the pass-through scenario. In the non-pass-through and pass-through scenarios, the storage module is used to store the mapping relationship between data and virtualized devices, etc., which will not be introduced here.

The software system includes a processing module and an initialization module. In a non-pass-through scenario, the processing module can be used to configure the engine module, so that the engine module appears as a virtualized device to the outside, so that it can be discovered and used by the server. Further, after the processing module completes the configuration of the engine module to obtain the virtualized device, it can bind the virtualized device inside the offload card to the virtualized device of the heterogeneous card (that is, to construct the virtualized device of the offloaded card and the heterogeneous card. The mapping relationship between the virtualized devices of the card), so that the virtualized device inside the unloading card is used as a dedicated channel for the heterogeneous card. chemical equipment B. Then, the processing module can bind the virtualization device A to the virtualization devices D and E of the heterogeneous card C, and bind the virtualization device B to the virtualization devices G, H and I of the heterogeneous card F, so that the virtualization Device A serves as a dedicated channel for heterogeneous card C, virtualized device B serves as a dedicated channel for heterogeneous card F, and so on. In both non-pass-through and pass-through scenarios, the initialization module is used to load the non-standard driver corresponding to the heterogeneous card, and initialize the heterogeneous card based on the non-standard driver, so that information exchange between the heterogeneous card and the unloading card can be performed. .

In the embodiment of the present application, the cloud computing system can work in a non-pass-through scenario and a pass-through scenario, and the two scenarios will be introduced separately below. It should be noted that, depending on which mode the cloud computing system works in, scenarios can be set in advance for the cloud computing system according to actual needs. For the convenience of description, the non-standard protocol supported by the heterogeneous card is referred to as the first protocol, the standard protocol supported by the server is referred to as the second protocol, the driver corresponding to the non-standard protocol is referred to as the first driver, and the standard protocol is referred to as the first driver. The corresponding driver is referred to as the second driver, the virtualization device of the unload card is referred to as the first virtualization device, and the virtualization device of the heterogeneous card is referred to as the second virtualization device.

First, the non-pass-through scenarios are introduced. In a non-pass-through scenario, the offload card can record the mapping relationship between the first virtualization device of the offload card and the second virtualized device of the heterogeneous card (that is, the initialization process), and then realize the cloud instance of the server and the heterogeneous card according to the mapping relationship. The communication between the second virtualization devices (ie the data transmission process), wherein the first virtualization device of the offload card is directly connected to the cloud instance of the server, and the second virtual device of the heterogeneous card is directly connected to the offload card. The initialization process of the non-pass-through scenario is described below. FIG. 6 is a schematic flowchart of a data processing method provided by an embodiment of the present application. As shown in FIG. 6 , the method includes:

601. After the server is started, load a second driver.

In this embodiment, since the server is connected to one side of the uninstall card, the uninstall card and the server can be started synchronously or asynchronously. For example, if the unloading card does not have its own power supply, the server can only control the unloading card to start after the server is powered on. For another example, if the unloading card has its own power supply, the server and the unloading card can be started synchronously through their own power supplies, etc., which is not limited here.

It is worth noting that, after the server is started, both the server and the cloud instance of the server can load the second driver corresponding to the second protocol supported by the server. Furthermore, after the unloading card is started, the processor inside the unloading card starts to run, thereby starting the software system to support the function of software-hardware coordination, that is, the hardware system and the software system can cooperate with each other.

602. After it is determined that the heterogeneous card is activated, the unloading card loads the first driver.

Similarly, since the heterogeneous card is connected to the other side of the unloading card, the unloading card and the heterogeneous card can be started synchronously or asynchronously. For details, please refer to the relevant description of the startup process of the server and the unloading card in step 301. Repeat. After the unloading card determines that the heterogeneous card is started, that is, after the unloading card determines that the heterogeneous card is powered on, the unloading card loads the first driver corresponding to the first protocol supported by the heterogeneous card through the initialization module (the first driver corresponding to the heterogeneous card). ).

603. The uninstall card acquires the device information of the heterogeneous card through the first driver.

After loading the first driver, the unloading card may send an initialization request to the heterogeneous card through the first driver, so that the heterogeneous card performs single-root I/O virtualization (SRIOV) initialization based on the initialization request . During the SRIOV initialization process of the heterogeneous card, a second virtualization device (including at least one PF and multiple VFs) that is directly connected to the offload card can be created. communication. After the heterogeneous card completes the SRIOV initialization, the device information of the heterogeneous card can be returned to the unloading card through the second virtualization device. The device information of the heterogeneous card includes the device identification (identity, ID) and device type of each PF and VF, etc. information.

604. The uninstall card configures the engine module as the first virtualization device based on the configuration information, where the configuration information is preset information.

In the hardware system of the offload card, an engine module supporting the second protocol is included, for example, the aforementioned virtio engine, NVMe engine, and so on. After loading the first driver, the unloading card obtains configuration information for configuring the first virtualization device through the processing module. The configuration information is preset information, including information related to the device type of the first virtualized device (command channel), information related to the number of the first virtualized device (the number of queues), and the like. After obtaining the configuration information, the processing module may configure the engine module as the first virtualization device based on the configuration information.

For example, suppose the server supports the virtio protocol. After loading the first driver, the unloading card obtains the configuration information corresponding to the virtio protocol through the processing module, and the content contained in the configuration information (that is, which first virtualization device under the virtio protocol should be configured and how many first virtual devices are to be configured) Virtualization device) has been set in advance, such as the configuration information contains the command channel corresponding to virtio-gpu and the number of queues corresponding to virtio-gpu. After obtaining the configuration information, the processing module can determine the virtio engine in the virtio engine, NVMe engine and other modules, and configure the virtio engine as the Nth Nth such as virtio-gpu1, virtio-gpu2, ..., virtio-gpuN based on the configuration information For a virtualized device, the value of N is related to the number of queues.

It should be noted that the first virtualization device of the offload card is directly connected to the cloud instance of the server, that is, the offload card can communicate with the cloud instance of the server through the first virtualization device.

It is worth noting that in the process of the offload card sending information to the cloud instance of the server, the information sent by the offload card through the first virtualization device is forwarded to the second server of the cloud instance of the server via the second driver of the server (host). drive to reach the cloud instance of the server. This process can be described as the first virtualization device of the uninstall card sending information to the second driver of the cloud instance of the server, which will not be described in detail later.

Similarly, when the cloud instance of the server sends information to the offload card, the information sent by the second driver of the cloud instance of the server is forwarded to the first virtualization device of the offload card via the second driver of the server, thereby reaching the offload card. . This process can be described as the second driver of the cloud instance of the server sending information to the first virtualization device of the offload card, which will not be described in detail later.

It should be understood that the foregoing examples only take the second protocol supported by the cloud instance of the server as the virtio protocol for illustrative illustration, and do not limit the type of the second protocol supported by the cloud instance of the server in this application.

It should also be understood that in the foregoing example, under the virtio protocol, only the first virtualized device is virtio-gpu as an example for schematic illustration, and the first virtualized device may also be other types of devices such as virtio-blk and virtio-net .

It should also be understood that step 603 and step 604 may be performed synchronously or asynchronously. When the two are executed asynchronously, step 603 can be executed before step 604, and step 603 can also be executed after step 604, which is not limited here.

605. The uninstall card receives an enumeration result from the second driver through the first virtualization device, where the enumeration result is used to indicate that the cloud instance of the server has detected the first virtualization device.

After the processing module of the offload card completes the configuration operation to obtain the first virtualized device, the first virtualized device can be discovered by the cloud instance of the server. After the cloud instance of the server discovers (detects) the first virtualized device, the second driver of the cloud instance can deliver an enumeration result to the first virtualized device of the offload card, where the enumeration result is used to indicate the cloud of the server The instance has detected the first virtualized device.

As in the above example, after the processing module completes the configuration of the virtio engine, the virtio engine can present the cloud instances of the server as virtio-gpu1, virtio-gpu2, ..., virtio-gpuN and other devices. After the cloud instance of the server (through the second driver of the cloud instance) detects virtio-gpu1, virtio-gpu2, ..., virtio-gpuN, it can send the enumeration result to the uninstall card to notify the uninstall card that the cloud instance of the server has virtio-gpu1, virtio-gpu2, …, virtio-gpuN detected.

606. The offload card builds a mapping relationship between the second virtualized device of the heterogeneous card and the first virtualized device of the offloaded card based on the device information of the heterogeneous card and the enumeration result.

In step 603, after receiving the device information of the heterogeneous card through the first driver, the uninstall card can send the device information to the processing module through the first driver.

After receiving the enumeration result and the device information of the heterogeneous card, the processing module of the offload card can construct the second virtualized device of the heterogeneous card and the first virtualized device of the offload card based on the device information of the heterogeneous card and the enumeration result. The mapping relationship between the devices is stored, and the mapping relationship is stored in the storage module.

For example, let the device information of the heterogeneous card be used to indicate that the heterogeneous card contains two second virtualization devices (VF1 and VF2 respectively), and the enumeration result is used to indicate that the cloud instance of the server has detected three first virtualization devices devices (virtio-gpu1, virtio-gpu2 and virtio-gpu3 respectively). The processing module can establish the following mapping relationship based on the device information and enumeration results of the heterogeneous cards: virtio-gpu1 corresponds to VF1, virtio-gpu2 corresponds to VF1, and virtio-gpu3 corresponds to VF2.

Similarly, if another heterogeneous card is connected to the unloading card, the same initialization process as step 601 to step 606 can also be performed for the heterogeneous card. Then, the unloading card is loaded with first drivers corresponding to multiple heterogeneous cards, and the unloading card can record the mapping relationship between multiple first virtualization devices and multiple second virtualization devices of different heterogeneous cards. For example, if the heterogeneous card 1 and the heterogeneous card 2 are connected to the unloading card successively, after the initialization is completed, the following mapping relationship can be obtained: virtio-gpu1 corresponds to the VF1 of the heterogeneous card 1, and virtio-gpu2 corresponds to the heterogeneous card 1 virtio-gpu3 corresponds to VF2 of heterogeneous card 1, virtio-gpu4 corresponds to VF3 of heterogeneous card 2, virtio-gpu5 corresponds to VF4 of heterogeneous card 2, and so on. If the cloud instance of the server sends certain information to virtio-gpu1, the offload card can forward the information to VF1 based on the correspondence between virtio-gpu1 and VF1, so the information can be successfully sent to heterogeneous card 1.

Further, the server system of the embodiment of the present application also has a reloading function for the first virtualization device. Take the server supporting the virtio protocol as an example. Assuming that the virtio protocol supported by the server is the virtio1.0 version, after the aforementioned steps 601 to 606 (wherein, in the configuration process of the virtio engine, the configuration information corresponding to virtio1.0 is used), so that the virtio engine presents It is virtio-gpu under virtio1.0. If the version of the virtio protocol supported by the server is changed, for example, from virtio1.0 to virtio2.0, the processing module can obtain the configuration information corresponding to virtio2.0, and based on the configuration information, the virtio engine that has been configured will be reconfigured. The configuration makes the virtio engine appear as virtio-gpu under virtio2.0. Similarly, if you need to change the device type presented by the virtio engine, for example, switch virtio-gpu to virtio-blk, the processing module can obtain new configuration information, and reconfigure the configured virtio engine based on the configuration information, so that The virtio engine is presented as virtio-blk. It should be noted that even if the first virtualized device is reloaded, the mapping relationship between the reloaded first virtualized device and the second virtualized device of the heterogeneous card will not be changed, that is, the first virtualized device before reloading is retained. A mapping relationship between a virtualized device and a second virtualized device of a heterogeneous card.

After the above initialization process, the data transmission process can be performed. For the server side, what the cloud instance of the server can see is the first virtualized device presented by the uninstall card, so the cloud instance of the server can send information to the first virtualized device according to its own needs. After receiving the information through the first virtualization device, the uninstall card can send the information to the first virtualization device of the heterogeneous card based on the mapping relationship between the first virtualization device of the uninstall card and the second virtualization device of the heterogeneous card. Second, the virtualized device, which is equivalent to the cloud instance of the server is actually sending information to the heterogeneous card, so as to realize the communication between the cloud instance of the server and the heterogeneous card.

In a non-pass-through scenario, the data transmission process includes a data transmission process in a transmit (transmit, TX) direction and a data transmission process in a receive (receive, RX) direction. The data transmission process in the TX direction includes two modes, and the first mode is introduced first. FIG. 7 is another schematic flowchart of a data processing method provided by an embodiment of the present application. As shown in FIG. 7 , the method includes:

701. The uninstall card receives a second sending request from the second driver through the first virtualization device, the second sending request conforms to the format requirements of the second protocol, and the second sending request is used to indicate that the cloud instance of the server is about to send the target data .

After initialization, what the cloud instance of the server can see is the first virtualized device presented by the offload card. When the cloud instance of the server needs to send target data to the first virtualization device, a second sending request can be generated, the second sending request conforms to the format requirements of the second protocol, and the second sending request is used to indicate that the cloud instance of the server is about to send target data. After obtaining the second sending request, the cloud instance of the server sends the second sending request to the first virtualization device through the second driver.

For example, when the cloud instance of the server wants to send target data to virtio-gpu1, a second sending request may be generated, and the second sending request conforms to the format requirements of the second protocol (ie, the virtio protocol). Then, the cloud instance of the server sends the second sending request to virtio-gpu1 through the second driver.

702. The uninstall card converts the second sending request into a first sending request, where the first sending request conforms to the format requirements of the first protocol.

After receiving the second sending request through the first virtualization device, the unloading card sends the second sending request to the protocol conversion module through the first virtualization device, so that the protocol conversion module obtains the first virtualization device and the heterogeneous configuration from the storage module. The mapping relationship between the second virtualization devices of the card, and based on the mapping relationship, the second sending request is converted into a first sending request, and the first sending request conforms to the format requirements of the first protocol.

For example, if the protocol conversion module receives the second sending request from virtio-gpu1, the mapping relationship between virtio-gpu1 and the VF1 of the heterogeneous card 1 can be obtained from the storage module. After determining that the second sending request is a request sent to the heterogeneous card 1, the data format of the second sending request can be converted to obtain the first sending request, and the first sending request conforms to the first protocol supported by the heterogeneous card 1. format requirement. For another example, if the protocol conversion module receives the second sending request from virtio-gpu5, the mapping relationship between virtio-gpu5 and the VF4 of the heterogeneous card 2 can be obtained from the storage module. After it is determined that the second sending request is a request sent to the heterogeneous card 2, the data format of the second sending request can be converted to obtain the first sending request, and the first sending request conforms to the first protocol supported by the heterogeneous card 2. format requirements, etc.

It is worth noting that when the protocol conversion module converts the data format, it needs to obtain the mapping table of the data format from the storage module, so as to complete the conversion of the data format based on the mapping table. Specifically, the mapping table records the correspondence between the data format of the first protocol and the data format of the second protocol, so the protocol conversion module can convert the second sending request (the data format conforming to the second protocol) based on the mapping table request) into the first sending request (complies with the data format requirements of the first protocol).

Further, the aforementioned conversion of the data format includes conversion of the address of the target data and the like.

703. The offload card sends the first sending request to the second virtualization device through the first driver, so that the heterogeneous card sends the first read request to the offload card through the second virtualization device based on the first sending request.

After obtaining the first sending request, the offload card sends the first sending request to the first driver through the protocol conversion module, so that the first driver sends the first sending request to the second virtualization device of the heterogeneous card.

For example, if the protocol conversion module receives the second sending request from virtio-gpu1, the mapping relationship between virtio-gpu1 and the VF1 of the heterogeneous card 1 can be obtained from the storage module. After determining that the second sending request is a request sent to the heterogeneous card 1, the data format of the second sending request can be converted to obtain the first sending request, and the first sending request conforms to the first protocol supported by the heterogeneous card 1. format requirement. Then, the protocol conversion module sends the first sending request to the first driver corresponding to the heterogeneous card 1, so that the first driver sends the first sending request to the VF1.

704. The uninstall card receives the first read request from the second virtualization device through the first driver, where the first read request conforms to the format requirement of the first protocol.

After the heterogeneous card receives the first sending request, it can be determined that the cloud instance of the server is about to send the target data to the heterogeneous card. Therefore, the heterogeneous card generates the first read request, and the first read request conforms to the format requirements of the first protocol, and is used to indicate that the heterogeneous card needs to read the target data. After obtaining the first read request, the heterogeneous card sends the first read request to the offload card through the second virtualization device.

705. The uninstall card converts the first read request into a second read request, where the second read request conforms to the format requirements of the second protocol.

After receiving the first read request from the second virtualization device, the first driver of the offload card can send the first read request to the protocol conversion module, so that the protocol conversion module obtains the first virtualized data of the offload card from the storage module based on the The mapping relationship between the device and the second virtualized device of the heterogeneous card, and based on the mapping relationship, the first read request is converted into a second read request, and the second read request conforms to the format requirements of the second protocol.

For example, in the first driver installed on the uninstall card, if the first driver corresponding to the heterogeneous card 1 receives the first read request from the VF1 of the heterogeneous card 1, it can send the first read request to the protocol conversion module, And notify the protocol conversion module that the first read request comes from the VF1 of the heterogeneous card 1 (for example, the first read request may carry the identifier of the VF1, etc.). The protocol conversion module can obtain the mapping relationship between virtio-gpu1 and the VF1 of the heterogeneous card 1 from the storage module. After determining that the first read request is a request sent to the cloud instance through virtio-gpu1, it can perform data processing on the first read request. The format is converted to obtain a second read request, and the second read request conforms to the format requirements of the second protocol (ie, the virtio protocol) supported by the cloud instance of the server. Then, the protocol conversion module sends the second read request to virtio-gpu1.

706. The uninstall card sends the second read request to the second driver through the first virtualization device.

After obtaining the second read request, the offload card sends the second read request to the first virtualization device through the protocol conversion module, so that the first virtualization device sends the second read request to the second driver of the cloud instance of the server.

707. The cloud instance of the server reads the target data from the memory of the cloud instance of the server based on the second read request.

After the cloud instance of the server obtains the second read request, the target data can be read from the memory of the cloud instance of the server based on the second read request.

708. The uninstall card receives the target data from the second driver through the first virtualization device.

After the cloud instance of the server obtains the target data, the second driver can return the target data to the first virtualization device.

709. The uninstall card sends the target data to the second virtualization device through the first driver.

After receiving the target data through the first virtualization device, the offload card sends the target data to the protocol conversion module through the first virtualization device, so that the protocol conversion module obtains the second data between the first virtualization device and the heterogeneous card from the storage module. The mapping relationship between virtualized devices, but the data format conversion of the target data is not performed. Based on the mapping relationship, the target data is directly sent to the first driver corresponding to the heterogeneous card, and then the target data is sent to the heterogeneous card through the first driver. The second virtualized device that constitutes the card, that is, the heterogeneous card successfully reads the target data. So far, the data transmission process in the TX direction is completed.

For example, if the protocol conversion module receives the target data from virtio-gpu1, the mapping relationship between virtio-gpu1 and the VF1 of the heterogeneous card 1 can be obtained from the storage module. After it is determined that the target data is the data sent to the heterogeneous card 1, the data format conversion of the target data is not performed. Then, the protocol conversion module sends the target data to the first driver corresponding to the heterogeneous card 1, so that the first driver sends the target data to the VF1.

The above is the description of the first mode of the data transmission process in the TX direction, and the second mode will be introduced below. FIG. 8 is another schematic flowchart of a data processing method provided by an embodiment of the present application. As shown in FIG. 8 , the method includes:

801. The uninstall card receives a second sending request from the second driver through the first virtualization device, the second sending request conforms to the format requirements of the second protocol, and the second sending request is used to indicate that the cloud instance of the server is about to send the target data .

802. The offload card converts the second sending request into a first sending request, where the first sending request conforms to the format requirements of the first protocol.

803. The offload card sends the first sending request to the second virtualization device through the first driver, so that the heterogeneous card sends the first read request to the offload card through the second virtualization device based on the first sending request.

For the description of steps 701 to 703 , reference may be made to the relevant descriptions of the foregoing steps 801 to 803 , which will not be repeated here.

804. The uninstall card sends the second read request to the second driver through the first virtualization device.

After the offload card sends the first sending request to the second virtualization device of the heterogeneous card. The unloading card can generate a second read request, and the second read request conforms to the format requirements of the second protocol and is used to instruct the unloading card to read target data. Then, the offload card sends the second read request to the second driver of the cloud instance of the server through the first virtualization device.

805. The cloud instance of the server reads the target data from the memory of the cloud instance of the server based on the second read request.

After the cloud instance of the server obtains the second read request, the target data is read from the memory of the cloud instance of the server based on the second read request.

806. The cloud instance of the server sends the target data to the first virtualization device through the second driver.

After obtaining the target data, the cloud instance of the server returns the target data to the first virtualization device through the second driver.

807. The uninstall card stores the target data in the memory of the uninstall card.

After receiving the target data through the first virtualization device, the uninstall card sends the target data to the operation module (included in the hardware system, not shown in FIG. 5 ), so that the operation module stores the target data in the storage module of the uninstall card. middle.

808. The offload card receives the first read request from the second virtualization device through the first driver.

After receiving the second sending request, the heterogeneous card can determine that the cloud instance of the server is about to send the target data, so a first read request can be generated, and the first read request conforms to the format requirements of the first protocol. Then, the heterogeneous card sends the first read request to the first driver of the offload card through the second virtualization device.

809. The offload card reads the target data from the memory of the offload card based on the first read request.

Since the target data has been pre-stored in the storage module of the uninstall card, after receiving the first read request, the first driver of the uninstall card can send the first read request to the operation module, so that the operation module based on the first read request Get the target data in the storage module.

810. The uninstall card sends the target data to the second virtualization device through the first driver.

After obtaining the target data, the uninstall card sends the target data to the first driver through the operation module, so that the first driver returns the target data to the second virtualization device of the heterogeneous card, that is, the heterogeneous card successfully reads the target data. So far, the data transmission process in the TX direction is completed.

The above is the description of the data transmission process in the TX direction, and the following describes the data transmission process in the RX direction. FIG. 9 is another schematic flowchart of a data processing method provided by an embodiment of the present application. As shown in FIG. 9 , the method includes:

901. The uninstall card receives a first reception request from the second virtualization device through the first driver, the first reception request conforms to the format requirements of the first protocol, and the first reception request is used to instruct the heterogeneous card to request to obtain the cloud instance of the server available memory space information.

After initialization, the heterogeneous card can actively request the available memory space information from the cloud instance of the server, so as to determine which part of the memory in the cloud instance of the server can be used for writing target data. Specifically, the heterogeneous card may first generate a first receiving request, the first receiving request conforms to the format requirements of the first protocol, and is used to instruct the heterogeneous card to request to obtain the available memory space information of the cloud instance of the server. Then, the heterogeneous card sends the first reception request to the offload card through the second virtualization device.

902. The uninstall card converts the first receiving request into a second receiving request, where the second receiving request conforms to the format requirements of the second protocol.

903. The uninstall card sends the second reception request to the second driver through the first virtualization device.

For the description of step 902 and step 903, reference may be made to the relevant description part of the above-mentioned step 705 and step 706, which will not be repeated here.

904. The uninstall card receives the available memory space information from the second driver through the first virtualization device.

After the cloud instance of the server obtains the second receiving request, it can obtain its own available memory space information based on the second receiving request. Then, the cloud instance of the server returns the available memory space information to the first virtualization device through the second driver.

905. The uninstall card sends the available memory space information to the second virtualization device through the first driver.

After receiving the available memory space information through the first virtualization device, the unloading card can send the available memory space information to the protocol conversion module through the first virtualization device, so that the protocol conversion module obtains the first virtualization device and the protocol conversion module from the storage module. The mapping relationship between the second virtualization devices of the card is constructed, but the data format conversion of the available memory space information is not performed. Based on the mapping relationship, the available memory space information is directly sent to the first driver corresponding to the heterogeneous card, and then through the The first driver sends the available memory space information to the second virtualization device of the heterogeneous card, that is, the heterogeneous card successfully obtains the available memory space information of the cloud instance of the server.

After obtaining the available memory space information of the cloud instance of the server, the heterogeneous card can determine which memory location of the cloud instance to write the target data to based on the available memory space information, thereby generating the first write request, which conforms to the first protocol format requirements, and is used to indicate that the heterogeneous card needs to write the target data. Then, the heterogeneous card sends the first write request to the offload card through the second virtualization device.

906. The uninstall card receives, through the first driver, the first write request from the second virtualization device, where the first write request conforms to the format requirement of the first protocol.

907. The offload card converts the first write request into a second write request, where the second write request conforms to the format requirements of the second protocol.

908. The uninstall card sends the second write request to the second driver through the first virtualization device.

909. The cloud instance of the server writes the target data to the memory of the cloud instance of the server based on the second write request.

For the description of steps 906 to 909, reference may be made to the relevant descriptions of the above-mentioned steps 704 and 707, which will not be repeated here. So far, the cloud instance of the server has successfully written the target data, and the data transmission process in the RX direction is completed.

In the server system provided by the embodiment of the present application, the offload card is connected between the server and the heterogeneous card. When the server system works in a non-passthrough scenario, the cloud instance of the server is loaded with the second driver corresponding to the second protocol supported by the server, and the unloading card is loaded with the first driver corresponding to the first protocol supported by the heterogeneous card. Specifically, after receiving the first operation request from the second virtualization device of the heterogeneous card through the first driver, the uninstall card can convert the first operation request into a second operation request, and the first operation request conforms to the format of the first protocol requirements, the second operation request conforms to the format requirements of the second protocol. Then, the uninstall card sends the first operation request to the second driver of the cloud instance of the server through the first virtualization device, so that the cloud instance of the server completes the corresponding operation based on the first operation request. Therefore, in the server system provided by the embodiments of the present application, even if the cloud instance of the server only loads the second driver, the communication between the cloud instance of the server and the heterogeneous card can be realized.

Further, since the offload card directly connects to the expansion of the heterogeneous card, the offload card can be responsible for processing the management and control plane of the second virtualization device of the heterogeneous card, does not occupy any additional resources of the front-end server, and can save the resources of the server. Conducive to promoting the lightweight of the server.

Furthermore, security problems can also be solved very well. For example, if an abnormal situation occurs when a virtual machine escapes safely, the management plane of the uninstall card isolates and shields the abnormal queue channel, so that it will not affect other virtual machines, and will cause exceptions. Error information is saved in local memory, and a new workflow can be started immediately after failure recovery.

Further, the unloading card that supports software and hardware collaboration can exist in the form of FPGA+CPU, making full use of the dynamic reconfigurability of FPGA, and designing the engine module to be more flexible, not only supporting all current virtualized I/Os For protocol processing, by modifying the design and reloading the image files of some modules, it is possible to support diverse scenarios in the future without changing the hardware architecture.

The above is a specific description of the non-pass-through scenario, and the pass-through scenario will be introduced below. In the pass-through scenario, the initialization process and the data transmission process are also included. The initialization process of the pass-through scenario is described below. FIG. 10 is another schematic flowchart of a data processing method provided by an embodiment of the present application. As shown in FIG. 10 , the method includes:

1000. After determining that the heterogeneous card is activated, the server and the unloading card both load the first driver.

When both the server and the unloading card determine that the heterogeneous card is started, both the server and the cloud instance of the server can load the first driver corresponding to the first protocol supported by the heterogeneous card. Similarly, the unloading card can also load the heterogeneous card through the initialization module. The first driver corresponding to the first protocol supported by the card.

1001. The uninstall card receives an initialization request of the first driver of the cloud instance from the server through the pass-through channel.

When the offload card works in the pass-through scenario, one end of the pass-through channel of the off-load card (that is, the PF bypass in Figure 5) is directly connected to the cloud instance of the server, and the other end is directly connected to the heterogeneous card. The information from the cloud instance of the server is directly transparently transmitted to the heterogeneous card, or the information from the heterogeneous card is directly transparently transmitted to the cloud instance of the server.

Specifically, after the cloud instance of the server loads the first driver, an initialization request may be sent to the pass-through channel of the unload card.

1002. The uninstall card sends an initialization request to the heterogeneous card through the first driver of the uninstall card.

After receiving the initialization request through the pass-through channel, the unload card can send the initialization request to the first driver of the unload card through the pass-through channel, so that the first driver of the unload card sends the initialization request to the heterogeneous card.

1003. Based on the initialization request, the heterogeneous card sends the device information of the heterogeneous card to the first driver that uninstalls the card.

After receiving the initialization request, the heterogeneous card can start to initialize. After the initialization is completed, the device information of the heterogeneous card may be returned to the first driver that uninstalls the card. The device information of the heterogeneous card includes information such as ID, device type and the like of the heterogeneous card.

1004. After receiving the device information of the heterogeneous card through the first driver of the uninstall card, the uninstall card sends the device information of the heterogeneous card to the first driver of the cloud instance of the server through the pass-through channel.

After the unloading card receives the device information of the heterogeneous card through the first driver of the unloading card, it can send the device information of the heterogeneous card to the pass-through channel through the first driver of the unloading card, so that the pass-through channel returns the device information of the heterogeneous card to the The first driver of the cloud instance of the server, that is, the cloud instance of the server obtains the device information of the heterogeneous card. So far, the initialization process in the pass-through scenario is completed.

In the cut-through scenario, the data transmission process also includes a data transmission process in the TX direction and a data transmission process in the RX direction. The data transmission process in the TX direction includes two modes, and the first mode is introduced first. FIG. 11 is another schematic flowchart of a data processing method provided by an embodiment of the present application. As shown in FIG. 11 , the method includes:

1101. The offload card receives a sending request from the first driver of the cloud instance of the server through the pass-through channel, the sending request conforms to the format requirements of the first protocol, and the sending request is used to indicate that the cloud instance of the server is about to send the target data.

After initialization, since the heterogeneous card is directly connected to the cloud instance of the server through the pass-through channel inside the unloading card, on the server side, the cloud instance of the server can directly see the heterogeneous card. When the cloud instance of the server needs to send the target data to the heterogeneous card, a sending request can be generated, and the sending request conforms to the format of the first protocol and is used to indicate that the cloud instance of the server is about to send the target data. Then, the cloud instance of the server sends the sending request to the pass-through channel of the offload card through its own first driver.

1102. The uninstall card sends a sending request to the heterogeneous card through the first driver of the uninstall card.

After receiving the sending request through the pass-through channel, the unloading card can send the sending request to the first driver of the unloading card through the pass-through channel, so that the first driver of the unloading card sends the sending request to the heterogeneous card.

1103. Based on the sending request, the heterogeneous card sends a read request to the first driver for unloading the card, where the read request conforms to the format requirements of the first protocol.

After the heterogeneous card receives the sending request, it can determine based on the sending request that the cloud instance of the server is about to send the target data, so the heterogeneous card generates a read request, which conforms to the format requirements of the first protocol and is used to indicate that the heterogeneous card needs to send the target data. Read target data. Then, the heterogeneous card sends a read request to the first driver of the offload card.

1104. The uninstall card sends the read request to the first driver of the cloud instance of the server through the pass-through channel.

After the uninstall card receives the read request of the heterogeneous card through the first driver of the uninstall card, it can send the read request to the pass-through channel through the first driver of the unload card, so that the pass-through channel sends the read request to the first driver of the cloud instance of the server .

1105. The cloud instance of the server reads the target data from the memory of the cloud instance of the server based on the read request.

After the cloud instance of the server receives the read request, it reads the target data from its own memory based on the read request.

1106. The offload card receives the target data of the first driver of the cloud instance from the server through the pass-through channel.

After the cloud instance of the server obtains the target data, it can send the target data to the direct channel of the offload card through its own first driver.

1107. The uninstall card sends the target data to the heterogeneous card through the first driver of the uninstall card.

After the unloading card receives the target data through the pass-through channel, it can send the target data to the first driver of the unloading card through the pass-through channel, so that the first driver of the unloading card sends the target data to the heterogeneous card, that is, the heterogeneous card is successfully read. Get the target data. So far, the data transmission process in the TX direction is completed.

The above is the description of the first mode of the data transmission process in the TX direction, and the second mode will be introduced below. FIG. 12 is another schematic flowchart of a data processing method provided by an embodiment of the present application. As shown in FIG. 12 , the method includes:

1201. The uninstall card receives a sending request from the first driver of the cloud instance of the server through the pass-through channel, the sending request conforms to the format requirements of the first protocol, and the sending request is used to indicate that the cloud instance of the server is about to send the target data.

1202. The uninstall card sends a sending request to the heterogeneous card through the first driver of the uninstall card.

For the description of steps 1201 to 1202, reference may be made to the relevant descriptions of the above-mentioned steps 1101 and 1102, which will not be repeated here.

1203. The offload card is further configured to send the read request to the first driver of the cloud instance of the server through the pass-through channel.

After the unloading card sends the sending request to the heterogeneous card, a read request can be generated first, and the read request conforms to the format requirements of the first protocol and is used to instruct the unloading card to read target data. Then, the offload card sends the read request to the first driver of the cloud instance of the server through the pass-through channel.

1204. The cloud instance of the server reads the target data from the memory of the cloud instance of the server based on the read request from the offload card.

1205. The offload card receives the target data of the first driver of the cloud instance from the server through the pass-through channel.

For the description of steps 1204 to 1205, reference may be made to the relevant descriptions of the above-mentioned steps 1105 and 1106, which will not be repeated here.

1206. The uninstall card stores the target data in the memory of the uninstall card.

After the unloading card receives the target data through the pass-through channel, it can send the target data to the operation module of the unloading card through the pass-through channel, so that the operation module stores the target data in the memory of the unloading card.

1207. The uninstall card receives a read request from the heterogeneous card through the first driver of the uninstall card.

1208. The offload card reads the target data from the memory of the offload card based on the read request from the heterogeneous card.

After the unloading card receives the read request from the heterogeneous card through the first driver of the unloading card, since the target data is stored in the memory of the unloading card, the read request can be sent to the operation module through the first driver, so that the operation module is based on the read request. Request to read target data from the memory of the offload card.

1209. The uninstall card sends the target data to the heterogeneous card through the first driver of the uninstall card.

After the operation module of the uninstall card obtains the target data, it sends the target data to the first driver of the uninstall card, so that the first driver sends the target data to the heterogeneous card, that is, the heterogeneous card successfully reads the target data. So far, the data transmission process in the TX direction is completed.

The above is the description of the data transmission process in the TX direction, and the following describes the data transmission process in the RX direction. FIG. 13 is another schematic flowchart of a data processing method provided by an embodiment of the present application. As shown in FIG. 13 , the method includes:

1301. The uninstall card receives a reception request from the heterogeneous card through the first driver of the uninstall card, the reception request conforms to the format requirements of the first protocol, and the reception request is used to instruct the heterogeneous card to request to obtain the available memory space information of the cloud instance of the server .

After initialization, the heterogeneous card can actively request available memory space information from the cloud instance of the server, so as to determine which part of the memory in the cloud instance of the server can be used for writing target data. Specifically, the heterogeneous card may first generate a receiving request, the receiving request conforms to the format requirements of the first protocol, and is used to instruct the heterogeneous card to request to obtain the available memory space information of the cloud instance of the server. Then, the heterogeneous card sends the first reception request to the offload card.

1302. The uninstall card sends the receiving request to the first driver of the cloud instance of the server through the pass-through channel.

After the unloading card receives the receiving request of the heterogeneous card through the first driver of the unloading card, it can send the receiving request to the pass-through channel through the first driver of the unloading card, so that the pass-through channel sends the receiving request to the first server of the cloud instance of the server. drive.

1303. The uninstall card receives the information about the available memory space of the first driver of the cloud instance from the server through the pass-through channel.

After the cloud instance of the server receives the receiving request, it can obtain its own available memory space information based on the receiving request. Then, the cloud instance of the server sends the available memory space information to the pass-through channel of the offload card through its own first driver.

1304. The uninstall card sends the available memory space information to the heterogeneous card through the first driver of the uninstall card.

After receiving the available memory space information through the pass-through channel, the unloading card can send the available memory space information to the first driver of the unloading card through the pass-through channel, so that the first driver of the unloading card sends the available memory space information to the heterogeneous card.

1305. The heterogeneous card sends a write request to the unloading card based on the available memory space information, and the write request meets the format requirements of the first protocol.

After obtaining the available memory space information, the heterogeneous card can determine which memory location of the cloud instance to write the target data to based on the available memory space information, thereby generating a write request. The write request conforms to the format requirements of the first protocol and is used to indicate the The composition card needs to write the target data. Then, the heterogeneous card sends a write request to the offload card.

1306. The uninstall card sends the write request to the first driver of the cloud instance of the server through the pass-through channel.

1307. Based on the write request, the cloud instance of the server writes the target data to the memory of the cloud instance of the server.

For the description of steps 1306 to 1307, reference may be made to the relevant descriptions of the above-mentioned steps 1104 and 1105, which will not be repeated here. So far, the cloud instance of the server has successfully written the target data, and the data transmission process in the RX direction is completed.

In the server system provided by the embodiment of the present application, the offload card is connected between the server and the heterogeneous card. When the server system works in the pass-through scenario, both the cloud instance of the server and the offload card are loaded with the first driver corresponding to the first protocol supported by the heterogeneous card. Specifically, after the uninstall card receives the operation request from the heterogeneous card through the first driver, the uninstall card sends the operation request to the first driver of the cloud instance of the server through the pass-through channel, so that the cloud instance of the server completes the corresponding operation request based on the operation request. operate. Therefore, the offload card can be used as a bridge between the cloud instance of the server and the heterogeneous card, providing a direct channel between the two, so as to realize the communication between the cloud instance of the server and the heterogeneous card.

FIG. 14 is another schematic structural diagram of the unloading card provided by the embodiment of the application. As shown in FIG. 14 , one end of the unloading card is connected to the server, and the other end of the unloading card is connected to the heterogeneous card. The unloading card includes: a recording module 1401, For recording the mapping relationship between the first virtualization device of the uninstall card and the second virtualization device of the heterogeneous card, the communication module 1402 is used to realize the relationship between the cloud instance of the server and the second virtualization device of the heterogeneous card according to the mapping relationship. Communication between the two, wherein the first virtual device of the offload card is directly connected to the cloud instance of the server, and the second virtual device of the heterogeneous card is directly connected to the offload card.

In a possible implementation manner, the heterogeneous card supports the first protocol, the unloading card is loaded with the first driver corresponding to the first protocol, the server supports the second protocol, and the cloud instance of the server is loaded with the first driver corresponding to the second protocol The second driver; the communication module 1402 is specifically configured to: receive the first operation request from the second virtualization device through the first driver, the first operation request conforms to the format requirements of the first protocol; convert the first operation request into the second operation request, the second operation request conforms to the format requirements of the second protocol; the second operation request is sent to the second driver through the first virtualization device, so that the cloud instance of the server based on the second operation request, from the memory of the cloud instance of the server Read target data or write target data to the server's cloud instance's memory.

In a possible implementation manner, the unloading card includes an engine module supporting the second protocol; the recording module 1401 is specifically configured to: load the first driver after determining that the heterogeneous card is started; obtain the device of the heterogeneous card through the first driver information; configure the engine module as the first virtualization device based on the configuration information, and the configuration information is preset information; receive the enumeration result from the second driver through the first virtualization device, and the enumeration result is used to indicate the cloud instance of the server The first virtualized device has been detected; the mapping relationship between the second virtualized device of the heterogeneous card and the first virtualized device of the unloaded card is constructed based on the device information and the enumeration result.

In a possible implementation manner, if the first operation request is the first read request, the communication module 1402 is specifically configured to: receive, through the first virtualization device, a second sending request from the second driver, and the second sending request The format requirements of the second protocol are met, and the second sending request is used to indicate that the cloud instance of the server is about to send the target data; the second sending request is converted into a first sending request, and the first sending request meets the format requirements of the first protocol; The first driver sends the first sending request to the second virtualization device, so that the heterogeneous card sends the first read request to the offload card through the second virtualization device based on the first sending request.

In a possible implementation manner, the communication module 1402 is specifically configured to: receive the target data from the second driver through the first virtualization device; and send the target data to the second virtualization device through the first driver.

In a possible implementation manner, if the first operation request is a first read request and the second operation request is a second read request, the communication module 1402 is further configured to: send the second read request through the first virtualization device to the second driver, so that the cloud instance of the server reads the target data from the memory of the cloud instance of the server based on the second read request, and transmits it to the first virtualization device; stores the target data in the memory of the offload card; The first driver receives the first read request from the second virtualization device; reads the target data from the memory of the offload card based on the first read request; sends the target data to the second virtualization device through the first driver.

In a possible implementation manner, if the first operation request is the first write request, the communication module 1402 is specifically configured to: receive the first receive request from the second virtualization device through the first driver, and the first receive request conforms to the The format requirements of the first protocol, and the first receiving request is used to instruct the heterogeneous card to request to obtain the available memory space information of the cloud instance of the server; the first receiving request is converted into a second receiving request, and the second receiving request conforms to the second protocol. send the second receiving request to the second driver through the first virtualization device; receive the available memory space information from the second driver through the first virtualization device; send the available memory space information to the second driver through the first virtual device The second virtualization device enables the heterogeneous card to send a first write request to the offload card through the second virtualization device based on the available memory space information.

FIG. 15 is another schematic structural diagram of the uninstall card provided by the embodiment of the application. As shown in FIG. 15 , one end of the uninstall card is connected to the server, the other end of the uninstall card is connected to the heterogeneous card, and the heterogeneous card supports the first protocol. Both the server and the unloading card are loaded with a first driver corresponding to the first protocol; the unloading card includes: a communication module 1501 for receiving an operation request from the heterogeneous card through the first driver of the unloading card, the operation request conforming to the first protocol Format requirements; the communication module 1501 is further configured to send the operation request to the first driver of the cloud instance of the server through the pass-through channel, so that the cloud instance of the server reads the target data from the memory of the cloud instance of the server or sends the target data to the server based on the operation request. The memory of the cloud instance writes the target data.

In a possible implementation manner, the communication module 1501 is further configured to: receive an initialization request from the first driver of the cloud instance of the server through the direct channel; send the initialization request to the heterogeneous card by uninstalling the first driver of the card to Make the heterogeneous card send the device information of the heterogeneous card to the first driver of the unload card based on the initialization request; after receiving the device information of the heterogeneous card through the first driver of the unload card, send the device information of the heterogeneous card through the pass-through channel The first driver of the cloud instance sent to the server.

In a possible implementation manner, if the operation request is a read request, the communication module 1501 is further configured to: receive a sending request from the first driver of the cloud instance of the server through the pass-through channel, and send the request in a format that conforms to the first protocol request, and the sending request is used to indicate that the cloud instance of the server is about to send the target data; the sending request is sent to the heterogeneous card through the first driver of the offload card, so that the heterogeneous card sends the request to the first driver of the offload card based on the sending request read request.

In a possible implementation manner, the communication module 1501 is further configured to: receive target data from the first driver of the cloud instance of the server through the pass-through channel; and send the target data to the heterogeneous card by uninstalling the first driver of the card.

In a possible implementation manner, the communication module 1501 is further configured to: send the read request to the first driver of the cloud instance of the server through the pass-through channel, so that the cloud instance of the server can download the read request from the cloud instance of the server based on the read request of the offload card. Read the target data in the memory of the server, and send the target data to the pass-through channel through the first driver of the cloud instance of the server; store the target data in the memory of the offload card; receive the data from the heterogeneous card through the first driver of the offload card read request; read target data from the memory of the unloaded card based on the read request of the heterogeneous card; send the target data to the heterogeneous card through the first driver of the unloaded card.

In a possible implementation manner, if the operation request is a write request, the communication module 1501 is further configured to: receive a reception request from the heterogeneous card through the first driver of the uninstall card, and the reception request conforms to the format requirements of the first protocol, And the receiving request is used to instruct the heterogeneous card to request to obtain the available memory space information of the cloud instance of the server; send the receiving request to the first driver of the cloud instance of the server through the pass-through channel; receive the first driver of the cloud instance from the server through the pass-through channel. The information of the available memory space of the driver; the information of the available memory space is sent to the heterogeneous card through the first driver of the unloading card, so that the heterogeneous card sends a write request to the unloading card based on the information of the available memory space.

It should be noted that the information exchange, execution process and other contents among the modules/units of the above-mentioned apparatus are based on the same concept as the method embodiments of the present application, and the technical effects brought by them are the same as those of the method embodiments of the present application, and the specific contents can be Reference is made to the descriptions in the method embodiments shown in the foregoing embodiments of the present application, which will not be repeated here.

FIG. 16 is a schematic structural diagram of an unloading card provided by an embodiment of the present application. As shown in FIG. 16 , one embodiment of the offload card in this embodiment of the present application may include one or more central processing units 1601 , a memory 1602 , an input/output interface 1603 , a wired or wireless network interface 1604 , and a power supply 1605 .

The memory 1602 may be ephemeral storage or persistent storage. Furthermore, the central processing unit 1601 may be configured to communicate with the memory 1602 to execute a series of instruction operations in the memory 1602 on the offload card.

In this embodiment, the central processing unit 1601 may perform the operations performed by the unloading card in the foregoing embodiments shown in FIG. 6 to FIG. 13 , and details are not repeated here.

In this embodiment, the division of specific functional modules in the central processing unit 1601 may be similar to the division of modules such as the recording module and the communication module described above in FIG. 14 , which will not be repeated here.

In this embodiment, the division of specific functional modules in the central processing unit 1601 may also be similar to the division of modules such as the communication module described in FIG. 6 , which is not repeated here.

The embodiments of the present application also relate to a computer storage medium, including computer-readable instructions, when the computer-readable instructions are executed, the methods described in FIG. 6 to FIG. 13 are implemented.

Embodiments of the present application also relate to providing a computer program product containing instructions, which, when executed on a computer, cause the computer to execute the methods described in FIGS. 6 to 13 .

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

In the several embodiments provided in this application, it should be understood that the disclosed system, apparatus and method may be implemented in other manners. For example, the apparatus 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 may be combined or Can be integrated into another system, or some features can be ignored, or not implemented. On the other hand, the shown or discussed mutual coupling or direct coupling or communication connection may be through some interfaces, indirect coupling or communication connection of devices or units, and may be in electrical, mechanical or other forms.

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

In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit. The above-mentioned integrated units may be implemented in the form of hardware, or may be implemented in the form of software functional units.

The integrated unit, if implemented in the form of a software functional unit and sold or used as an independent product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the present application can be embodied in the form of software products in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, and the computer software products are stored in a storage medium , including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, Read-Only Memory (ROM, Read-Only Memory), Random Access Memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program codes .

Claims

A server system, characterized in that the server system comprises: a server, an uninstall card and a heterogeneous card, the uninstall card is connected to the server, and the heterogeneous card is connected to the uninstall card;

The offload card is used to record the mapping relationship between the first virtualization device of the offload card and the second virtual device of the heterogeneous card, and realize the cloud instance of the server and the heterogeneous card according to the mapping relationship The communication between the second virtualization devices of the offload card, wherein the first virtualization device of the offload card is directly connected to the cloud instance of the server, and the second virtual device of the heterogeneous card is directly connected to the offload card.
The server system according to claim 1, wherein the server includes a first interface, the uninstall card includes a second interface and a third interface, and the heterogeneous card includes a fourth interface;

The first interface is connected to the second interface, and the third interface is connected to the fourth interface.
The server system according to claim 1, wherein the server system further comprises an expansion card, the server includes a first interface, and the expansion card includes a second interface, a third interface, a fourth interface and a fifth interface an interface, the unloading card includes a sixth interface and a seventh interface, and the heterogeneous card includes an eighth interface;

The first interface is connected to the second interface, the second interface is connected to the third interface, the fourth interface is connected to the fifth interface, and the sixth interface is connected to the third interface connected, the seventh interface is connected with the fourth interface, and the eighth interface is connected with the fifth interface.
The server system according to any one of claims 1 to 3, wherein the heterogeneous card supports a first protocol, the unloading card is loaded with a first driver corresponding to the first protocol, and the server A second protocol is supported, and the cloud instance of the server is loaded with a second driver corresponding to the second protocol;

The uninstall card is specifically configured to receive, through the first driver, a first operation request from the second virtualization device, where the first operation request conforms to a format requirement of the first protocol;

The uninstall card is specifically configured to convert the first operation request into a second operation request, and the second operation request conforms to the format requirements of the second protocol;

The offload card is specifically configured to send the second operation request to the second driver through the first virtualization device, so that the cloud instance of the server, based on the second operation request, downloads the server from the server. The memory of the cloud instance of the server reads the target data or writes the target data to the memory of the cloud instance of the server.
The cloud computing system according to claim 4, wherein the offload card comprises an engine module supporting the second protocol;

The unloading card is specifically configured to load the first driver after determining that the heterogeneous card is activated;

The uninstall card is specifically used to obtain the device information of the heterogeneous card through the first driver;

The offload card is specifically configured to configure the engine module as the first virtualization device based on configuration information, where the configuration information is preset information;

The uninstall card is specifically configured to receive an enumeration result from the second driver through the first virtualization device, where the enumeration result is used to indicate that the cloud instance of the server has detected the first virtualization equipment;

The offload card is specifically configured to construct a mapping relationship between the second virtualization device of the heterogeneous card and the first virtualization device of the offload card based on the device information and the enumeration result.
The cloud computing system according to any one of claims 1 to 5, wherein if the first operation request is a first read request, the uninstall card is specifically used to pass the first virtualization device, receiving a second sending request from the second driver, the second sending request conforming to the format requirements of the second protocol, and the second sending request is used to indicate that the cloud instance of the server is about to send the target data;

The unloading card is specifically configured to convert the second sending request into a first sending request, where the first sending request conforms to the format requirements of the first protocol;

The offload card is specifically configured to send the first sending request to the second virtualization device through the first driver, so that the heterogeneous card sends the request through the second virtual device based on the first sending request. The virtualization device sends the first read request to the offload card.
The cloud computing system according to any one of claims 1 to 6, wherein the uninstall card is specifically configured to receive target data from the second driver through the first virtualization device;

The offload card is specifically configured to send the target data to the second virtualization device through the first driver.
The cloud computing system according to claim 6, wherein, if the first operation request is a first read request and the second operation request is a second read request, the uninstall card is further used to pass the The first virtualization device sends the second read request to the second driver, so that the cloud instance of the server reads the data from the memory of the cloud instance of the server based on the second read request target data, and transmit it to the first virtualization device;

The unloading card is also used to store the target data in the memory of the unloading card;

The offload card is further configured to receive the first read request from the second virtualization device through the first driver;

The offload card is further configured to read the target data from the memory of the offload card based on the first read request;

The offload card is further configured to send the target data to the second virtualization device through the first driver.
The cloud computing system according to any one of claims 1 to 5, wherein if the first operation request is a first write request, the uninstall card is specifically configured to receive data from the the first receiving request of the second virtualization device, the first receiving request complies with the format requirements of the first protocol, and the first receiving request is used to instruct the heterogeneous card to request to obtain the cloud of the server Information about the available memory space of the instance;

The unloading card is specifically configured to convert the first receiving request into a second receiving request, where the second receiving request conforms to the format requirements of the second protocol;

The uninstall card is specifically configured to send the second receiving request to the second driver through the first virtualization device;

The uninstall card is specifically configured to receive the available memory space information from the second driver through the first virtualization device;

The offload card is specifically configured to send the available memory space information to the second virtualization device through the first driver, so that the heterogeneous card uses the second virtualization device based on the available memory space information The device sends the first write request to the offload card.
A data processing method, wherein the method is applied to the server system, the server system includes: a server, an offload card and a heterogeneous card, the offload card is connected to the server, the heterogeneous card the card is connected to the unloading card;

The uninstall card records the mapping relationship between the first virtualization device of the uninstall card and the second virtualization device of the heterogeneous card;

The offload card implements communication between the cloud instance of the server and the second virtualization device of the heterogeneous card according to the mapping relationship, wherein the first virtualization device of the offload card is directly connected to the cloud of the server For example, the second virtual device of the heterogeneous card is directly connected to the offload card.
The method according to claim 10, wherein the heterogeneous card supports a first protocol, the unloading card is loaded with a first driver corresponding to the first protocol, the server supports a second protocol, and The cloud instance of the server is loaded with a second driver corresponding to the second protocol, and the offload card implements communication between the cloud instance of the server and the second virtualization device of the heterogeneous card according to the mapping relationship include:

receiving, by the uninstall card, a first operation request from the second virtualization device through the first driver, where the first operation request conforms to a format requirement of the first protocol;

The uninstall card converts the first operation request into a second operation request, and the second operation request conforms to the format requirements of the second protocol;

The offload card sends the second operation request to the second driver through the first virtualization device, so that based on the second operation request, the cloud instance of the server downloads the request from the cloud instance of the server. The memory of the server reads the target data or writes the target data to the memory of the cloud instance of the server.
The method according to claim 11, wherein the offload card includes an engine module supporting the second protocol, and the offload card records the difference between the first virtualization device of the offload card and the heterogeneous card. The mapping relationship of the second virtualized device includes:

After determining that the heterogeneous card is activated, the unloading card loads the first driver;

The uninstall card obtains the device information of the heterogeneous card through the first driver;

The offload card configures the engine module as the first virtualization device based on configuration information, where the configuration information is preset information;

receiving, by the uninstall card, an enumeration result from the second driver through the first virtualization device, where the enumeration result is used to indicate that the cloud instance of the server has detected the first virtualization device;

The offload card constructs a mapping relationship between the second virtualization device of the heterogeneous card and the first virtualization device of the offload card based on the device information and the enumeration result.
The method according to any one of claims 10 to 12, characterized in that, if the first operation request is a first read request, the uninstall card receives data from the second virtualization through the first driver. Before the first operation request of the device, the method further includes:

The offload card receives a second sending request from the second driver through the first virtualization device, the second sending request conforms to the format requirements of the second protocol, and the second sending request uses to indicate that the cloud instance of the server is about to send the target data;

The offload card converts the second sending request into a first sending request, where the first sending request conforms to the format requirements of the first protocol;

The offload card sends the first sending request to the second virtualization device through the first driver, so that the heterogeneous card sends the request through the second virtualization device based on the first sending request The first read request is sent to the offload card.
The method according to any one of claims 10 to 13, wherein the method further comprises:

receiving, by the uninstall card, target data from the second driver through the first virtualization device;

The offload card sends the target data to the second virtualization device through the first driver.
The method according to claim 13, wherein if the first operation request is a first read request and the second operation request is a second read request, the uninstall card uses the first driver to After the first sending request is sent to the second virtualization device, the method further includes:

The offload card sends the second read request to the second driver through the first virtualization device, so that the cloud instance of the server reads the data from the cloud instance of the server based on the second read request. reading the target data in the memory and transmitting it to the first virtualization device;

The unloading card stores the target data in the memory of the unloading card;

The offload card receives the first read request from the second virtualization device through the first driver;

The offload card reads the target data from the memory of the offload card based on the first read request;

The offload card sends the target data to the second virtualization device through the first driver.
The method according to any one of claims 10 to 12, characterized in that, if the first operation request is a first write request, the offload card receives data from the second virtualization through the first driver. Before the first operation request of the device, the method further includes:

The offload card receives a first receive request from the second virtualization device through the first driver, the first receive request conforms to the format requirements of the first protocol, and the first receive request is used for Instructing the heterogeneous card to request to obtain the available memory space information of the cloud instance of the server;

The offload card converts the first receiving request into a second receiving request, where the second receiving request conforms to the format requirements of the second protocol;

sending, by the uninstall card, the second receiving request to the second driver through the first virtualization device;

receiving, by the uninstall card, the available memory space information from the second driver through the first virtualization device;

The offload card sends the available memory space information to the second virtualization device through the first driver, so that the heterogeneous card sends information to the second virtualization device through the second virtualization device based on the available memory space information. The offload card sends the first write request.
An unloading card, characterized in that one end of the unloading card is connected to a server, the other end of the unloading card is connected to a heterogeneous card, and the unloading card comprises:

a recording module, configured to record the mapping relationship between the first virtualization device of the uninstall card and the second virtualization device of the heterogeneous card;

A communication module, configured to implement communication between the cloud instance of the server and the second virtualization device of the heterogeneous card according to the mapping relationship, wherein the first virtualization device of the offload card is directly connected to the server's Cloud instance, the second virtual device of the heterogeneous card is directly connected to the offload card.
An unloading card, characterized in that the unloading card comprises: a processor and a memory;

the memory is used to store computer-executable instructions;

The processor is configured to execute computer-executable instructions stored in the memory to cause the offload card to implement the method of any one of claims 10 to 16.
A computer storage medium, characterized by comprising computer-readable instructions, which, when executed, implement the method according to any one of claims 10 to 16.
A computer program product comprising instructions which, when run on a computer, cause the computer to perform the method of any one of claims 10 to 16.