Processor topology switching system, method and server

Abstract

The application discloses a processor topology switching system, method and server, relates to the technical field of computers, and allows a user to simply input an operation instruction in a management client; the management client can generate a management request according to the operation instruction and send the management request to a management controller; the management controller generates a topology switching configuration file according to the management request to control a topology switching controller, and realizes switching of a logical connection relationship between processors. In the embodiment of the application, the target logical connection relationship in the topology switching configuration file is generated by the management controller, and the user only needs to simply input the operation instruction, so that the operation complexity is low, the cost is low, and the switching error rate can be effectively reduced.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to processor topology switching systems, methods and servers. Background Technology

[0002] With the rapid development of technologies such as artificial intelligence and big data, the performance requirements for servers are gradually increasing. Deploying multiple GPUs (Graphics Processing Units) on a server can improve its performance. The logical connections between multiple GPUs constitute the server's GPU topology, and these topologies can be switched between.

[0003] In related technologies, the GPU topology mode of a server is implemented through hardware-level jumpers or BIOS (Basic Input / Output System) configuration. This method suffers from poor flexibility, complex operation, switching errors, and high cost. Summary of the Invention

[0004] This application provides a processor topology switching system, method, and server to at least address the problems of poor flexibility, complex operation, and high cost in related technologies.

[0005] In a first aspect, this application provides a processor topology switching system, comprising: a management controller and a topology switching controller. The management controller includes: a management request parsing unit and a switching configuration file generation unit, the switching configuration file generation unit being connected to the management request parsing unit and the topology switching controller. The management request parsing unit is configured to: receive a management request, parse the request type of the management request, and, when the request type corresponds to a topology switching request, control the switching configuration file generation unit to generate a topology switching configuration file corresponding to the management request. The management request is generated based on operation instructions input by a user on a management client. The topology switching configuration file includes the target logical connection relationship between processors on the server. The switching configuration file generation unit is configured to: generate the topology switching configuration file corresponding to the management request and send the topology switching configuration file to the topology switching controller. The topology switching controller is configured to: switch the logical connection relationship between processors according to the topology switching configuration file.

[0006] Secondly, this application provides a processor topology switching method applied to a processor topology switching system. The processor topology switching system includes a management controller and a topology switching controller. The management controller includes a management request parsing unit and a switching configuration file generation unit, with the switching configuration file generation unit connected to the management request parsing unit and the topology switching controller. The aforementioned processor topology switching method includes:

[0007] The management request parsing unit receives management requests and parses the request type. The management request is generated based on the operation instructions entered by the user on the management client. When the request type corresponds to a topology switch request, the management request parsing unit controls the switch configuration file generation unit to generate a topology switch configuration file corresponding to the management request. The topology switch configuration file includes the target logical connection relationship between processors on the server. The switch configuration file generation unit generates the topology switch configuration file corresponding to the management request and sends the topology switch configuration file to the topology switch controller. The topology switch controller switches the logical connection relationship between processors according to the topology switch configuration file.

[0008] Thirdly, this application provides a server including the aforementioned processor topology switching system.

[0009] This application allows users to input simple operation commands in the management client. The management client can then generate management requests based on these commands and send them to the management controller. The management controller, in turn, generates a topology switching configuration file based on the management request to control the topology switching controller, thereby switching the logical connection relationships between processors. In this embodiment, the target logical connection relationships in the topology switching configuration file are generated by the management controller. Users only need to input simple operation commands, resulting in low operational complexity, low cost, and effectively reduced switching error rates. Attached Figure Description

[0010] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of a server processor topology provided in an embodiment of this application;

[0012] Figure 2 This is a schematic diagram of another server processor topology provided in an embodiment of this application;

[0013] Figure 3 This is a schematic diagram of a processor topology switching system provided in an embodiment of this application;

[0014] Figure 4 This is a schematic diagram of another processor topology switching system provided in an embodiment of this application;

[0015] Figure 5 A flowchart illustrating the steps of a processor topology switching method provided in this application embodiment. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of this application.

[0017] It should be noted that, in the description of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. The terms "first," "second," etc., in this application are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0018] This application aims to perform topology switching on one or more processors in a server. The server described in this application is typically suitable for application scenarios with high data processing performance requirements, such as deep learning training, inference, and scientific computing. In these application scenarios, the server usually has multiple processors, which may include GPUs and CPUs (Central Processing Units). The logical connection relationship between processors is used to indicate the GPU allocated to the CPU, and the CPU and the GPU allocated to that CPU can cooperate to perform data processing.

[0019] Figure 1 This is a schematic diagram of a server processor topology provided in an embodiment of this application, with reference to... Figure 1 As shown, the server includes two central processing units (CPU1 and CPU2) and four graphics processing units (GPU1 to GPU4). When GPU1 and GPU2 are logically connected to CPU1, and GPU3 and GPU4 are logically connected to CPU2, CPU1, GPU1, and GPU2 work together to process data, and CPU2, GPU3, and GPU4 work together to process data.

[0020] It should be noted that in practical applications, the number of processors and topology of a server are not limited to... Figure 1 As shown.

[0021] The processor topology of the aforementioned servers can be dynamically and flexibly switched according to actual needs. For example, the processor topology of the servers can be... Figure 1 The processor topology shown is adjusted to Figure 2The processor topology shown is as follows: GPU1, GPU2, and GPU3 are logically connected to CPU1, and GPU4 is logically connected to CPU2. At this time, CPU1, GPU1, GPU2, and GPU3 work together to process data, and CPU2 and GPU4 work together to process data.

[0022] Related technologies typically achieve topology switching through hardware-level jumper settings or BIOS configuration. However, this method requires manual configuration by professionals, is complex and costly, and is prone to topology switching errors. Incorrect topology switching can cause GPUs to malfunction or even be damaged, severely impacting server service quality.

[0023] To address the aforementioned technical problems, embodiments of this application allow users to simply input operation commands in the management client. The management client can then generate a management request based on these commands and send it to the management controller. The management controller, in turn, generates a topology switching configuration file based on the management request to control the topology switching controller, thereby enabling the switching of logical connections between processors. In these embodiments, the target logical connection relationships in the topology switching configuration file are generated by the management controller. Users only need to input simple operation commands, resulting in low operational complexity, low cost, and effectively reduced switching error rates.

[0024] Furthermore, the management controller in this application can be a Baseboard Management Controller (BMC), which runs independently of the server's operating system. It can monitor the server's hardware status, control the server's power supply, and access the server's operating system logs through out-of-band management. Therefore, this application achieves switching through the management controller, allowing the server's operating system to continue running. Only a restart of the operating system is required after the switch, avoiding the time consumed by stopping the server's operating system for the switch, and effectively reducing the service interruption duration caused by processor topology switching.

[0025] Furthermore, the management controller and switching controller in this application are commonly used hardware in servers, and therefore will not lead to an increase in hardware costs.

[0026] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] Figure 3 This is a schematic diagram of a processor topology switching system provided in an embodiment of this application, as shown below. Figure 3As shown, the processor topology switching system 100 provided in the embodiments of this application includes a management controller 101 and a topology switching controller 102, which are interconnected so that the management controller 101 can switch the processor topology structure through the topology switching controller 102.

[0028] The management controller 101 includes a management request parsing unit 1011 and a switching configuration file generation unit 1012. The switching configuration file generation unit 1012 is connected to the management request parsing unit 1011 and the topology switching controller 102. Thus, the connection between the switching configuration file generation unit 1012 and the topology switching controller 102 enables the connection between the management controller 101 and the topology switching controller 102.

[0029] The aforementioned management request parsing unit 1011 is used to parse management requests and forward them to the business processing unit based on the parsing results. In the processor topology switching scenario, the management request parsing unit 1011 is used to: receive management requests, parse the request type of the management request, and when the request type corresponds to a topology switching request, control the switching configuration file generation unit 1012 to generate a topology switching configuration file corresponding to the management request. The management request is generated based on the operation instructions input by the user on the management client 103. The management client 103 is communicatively connected to the management controller 101. The management request is a request sent by the management client 103 to the management controller 101. The topology switching configuration file includes the target logical connection relationship between processors on the server.

[0030] The request type can be represented by one or more flag bits in the header information of the management request. The number of flag bits is related to the number of request types supported by the management controller 101. When the number of request types is denoted as M and the number of flag bits is denoted as N, the two satisfy the following relationship: For example, when the number of flag bits is 1, M=2, indicating that the management controller 101 supports two types of management requests. Among them, flag bit 1 indicates that the management request is a topology switching request to realize the processor topology switching.

[0031] Reference Figure 4 As shown, the management client 103 is connected to the management request parsing unit 1011. The parsing process for the above-mentioned request type includes: when the management request parsing unit 1011 receives the management request sent by the management client, it extracts the flag bit from the specified bits of the header information of the management request, and determines whether the request type is a topology switching request based on the specific value of the flag bit. When it is a topology switching request, the management request parsing unit 1011 sends the management request to the switching configuration file generation unit 1012.

[0032] After receiving the management request, the switching configuration file generation unit 1012 generates a topology switching configuration file corresponding to the management request and sends the topology switching configuration file to the topology switching controller 102. The topology switching configuration file includes the target logical connection relationship between processors on the server.

[0033] Specifically, the aforementioned topology switching configuration file can be extracted from the topology switching configuration file or dynamically generated based on the topology mode carried in the management request. This application embodiment does not limit the specific generation strategy of the topology switching configuration file.

[0034] Of course, the topology switching configuration file generation unit 1012 can generate the topology switching configuration file based on the priority of various generation strategies. For example, it can prioritize extracting the topology switching configuration file from the management request. If the topology switching configuration file does not exist in the management request, it can then dynamically generate the topology switching configuration file according to the topology pattern in the management request. In this way, the target logical connection relationship set by the user can be used first, so that the topology structure after the switch accurately meets the user's needs. When the user has not set the target logical connection relationship, it can also be dynamically generated, which helps to improve the switching success rate.

[0035] When the topology switching controller 102 receives the topology switching configuration file sent by the switching configuration file generation unit 1012, it is used to: switch the logical connection relationship between processors according to the topology switching configuration file. Specifically, the topology switching controller 102 can replace the configuration file in the target path with the received topology switching configuration file, and run the loading of the topology switching configuration file to make the topology switching configuration file effective, so that during the execution of the computing task, each processor will collaboratively process data according to the target logical connection relationship in the topology switching configuration file.

[0036] In summary, this embodiment allows users to input simple operation commands in the management client 103. The management client 103 then generates a management request based on the user's input and sends it to the management request parsing unit 1011 of the management controller 101 for parsing. When the parsed request type indicates that the management request is a topology switching request, the management request parsing unit 1011 controls the switching configuration file generation unit 1012 to generate a topology switching configuration file based on the management request. Thus, the topology switching controller 102 switches the logical connection relationships between processors based on the topology switching configuration file. It can be seen that the target logical connection relationships in the topology switching configuration file are generated by the switching configuration file generation unit 1012. Users only need to input simple operation commands and do not need to input complex content. Therefore, the operation complexity is low, the cost is low, and the switching error rate can be effectively reduced.

[0037] Of course, this application embodiment also supports user input of topology switching configuration files, which the management controller 101 can add to the management request, thus accurately meeting the user's needs for processor topology.

[0038] In some possible implementations, the data format of the topology switching request is: vendor function number, request type, request data, and response data, wherein the vendor function number is used to indicate the vendor that made the management request, and each vendor has one or more vendor function numbers.

[0039] The request data is used to indicate the relevant data required for the request.

[0040] The response data may include, but is not limited to: successful switchover, switchover failure, and the reason for the switchover failure.

[0041] Among some possible implementations, Figure 4 This is a schematic diagram of another processor topology switching system provided in an embodiment of this application, referring to... Figure 4 As shown, the management controller 101 also includes a storage unit 1013, and a switching configuration file generation unit 1012 is connected to the storage unit 1013. The storage unit 1013 is used to store any information required by the management controller 101 in the process of managing the server. The storage unit 1013 can be integrated into the management controller 101, or it can be located outside the management controller 101 and connected to the switching configuration file generation unit 1012 in the management controller 101.

[0042] The management request parsing unit 1011 is also used to: parse the target topology pattern from the management request and send the target topology pattern to the switching configuration file generation unit 1012.

[0043] The target topology pattern can be generated by the management client 103 based on the user's operation instructions in the management client 103. The user can input the target topology pattern in the management client 103. The management client 103 can provide a UI (User Interface) interactive interface to facilitate the user's input of operation instructions, or it can only provide an instruction input interface.

[0044] In the UI interface, users can control the system through UI controls. Specifically, the management client 103 can display a mode selection box, allowing users to choose a topology mode as the target topology mode from the list of options in the mode selection box. Alternatively, the current topology mode can be displayed next to the mode selection box to prompt the user to select a topology mode other than the current one as the target topology mode. The management client 103 can also differentiate the current topology mode in the list of options in the mode selection box, allowing users to distinguish the current topology mode based on the display method. Thus, users can select other topology modes besides the current one from the list of options as the target topology mode. It can be seen that this embodiment of the application can assist users in selecting a mode by displaying the current topology mode, improving the effectiveness of topology switching and avoiding switching caused by invalid user operations.

[0045] In the command input interface, users can enter machine-recognizable commands, which are usually composed of a series of characters, including but not limited to letters and special characters.

[0046] The management client 103 can generate a management request based on the target topology pattern input by the user. The target topology pattern can be represented by the pattern flag bit located in the first bit of the management request. Therefore, when the management request parsing unit 1011 receives the management request, it can extract the pattern flag bit from the first bit of the management request to determine the target topology pattern and send it to the switching configuration file generation unit 1012.

[0047] Correspondingly, the switching configuration file generation unit 1012 is also used for:

[0048] The topology switching configuration file corresponding to the target topology mode is determined from one or more configuration files stored in the storage unit 1013. Specifically, each configuration file in the storage unit 1013 corresponds to its own topology mode. Therefore, the switching configuration file generation unit 1012 can match the target topology mode with the topology modes of each configuration file in the storage unit 1013, and use the configuration file whose topology mode matches the target topology mode as the topology switching configuration file.

[0049] If a topology switching configuration file is not found in one or more configuration files in storage unit 1013, a topology switching configuration file is dynamically generated based on the target topology mode. At this time, all processors can also be acquired to determine the target logical connection relationship between all processors based on the target topology mode, and a topology switching configuration file can be generated based on this target logical connection relationship. For example, when the target topology mode is Balanced, all graphics processors can be evenly distributed among the central processing units (CPUs) to ensure that each CPU has the same number of graphics processors. This topology mode can balance the load on each CPU, making it suitable for distributed training and multi-tasking scenarios.

[0050] As can be seen, the embodiments of this application can generate topology switching configuration files based on two strategies using the target topology mode. Users only need to input the topology mode to achieve switching, which can reduce the complexity of user input operation commands. In addition, the two topology switching configuration file generation strategies help improve the switching success rate.

[0051] In another example of this application embodiment, the management request parsing unit 1011 is further configured to: parse the topology switching configuration file from the management request, wherein the topology switching configuration file is generated by the management client 103 based on the operation instructions input by the user. Correspondingly, the switching configuration file generation unit 1012 is further configured to: send the topology switching configuration file to the topology switching controller 102 when it receives the topology switching configuration file sent by the management request parsing unit 1011; and determine the topology switching configuration file according to the target topology mode when it does not receive the topology switching configuration file sent by the management request parsing unit 1011. This application embodiment uses the topology switching configuration file carried in the management request or dynamically generates the topology switching configuration file according to priority, which not only can meet the user's switching needs as accurately as possible, but also avoids topology switching failure due to the failure to generate the topology switching configuration file.

[0052] The aforementioned topology switching configuration file can be entered by the user in the management client 103, or it can be generated by the management client 103 based on the user's input operation instructions. The process by which the aforementioned switching configuration file generation unit 1012 determines the topology switching configuration file according to the target topology mode includes the following: determining the topology switching configuration file corresponding to the target topology mode from one or more configuration files stored in the storage unit 1013; and dynamically generating the topology switching configuration file according to the target topology mode when there is no topology switching configuration file in one or more configuration files in the storage unit 1013.

[0053] In one possible implementation, the switching configuration file generation unit 1012 is further configured to generate a topology switching configuration file by: first, calculating the number of graphics processors corresponding to each central processing unit (CPU) based on the target topology pattern; then, establishing a target logical connection relationship between the CPU and the graphics processors based on the number of graphics processors corresponding to each CPU and the CPU's requirement for graphics processors; and finally, generating a topology switching configuration file based on the target logical connection relationship. Thus, this application, while allocating graphics processors according to the target topology pattern, strives to meet the CPU's requirements for graphics processors, thereby contributing to improved CPU processing performance.

[0054] The number of graphics processors (GPUs) corresponding to each central processing unit (CPU) is related to the target topology mode. For example, in a balanced topology mode, the number of GPUs corresponding to each CPU is the same. Therefore, the number of GPUs corresponding to each CPU is the rounded result of the ratio of the total number of GPUs to the total number of CPUs on the server. The rounding result can be an up-rounded result, a down-rounded result, or a rounded-to-the-round value. In a cascaded topology mode, all GPUs can be assigned to one CPU.

[0055] After obtaining the number of graphics processors (GPUs) for each central processing unit (CPU), GPUs can be allocated based on the CPU's requirements for GPUs. These requirements include CPU memory requirements, floating-point performance requirements, and parallel processing requirements. For example, memory requirements can be greater than or equal to a preset memory threshold; floating-point performance requirements can be greater than or equal to a preset computational threshold; and parallel processing requirements can be greater than or equal to a preset data volume threshold.

[0056] The allocation of graphics processors based on the CPU's demand information for graphics processors is specifically manifested as follows: for each CPU, graphics processors that meet its demand information are selected as candidate graphics processors for that CPU, and a certain number of graphics processors are selected from the candidate graphics processors as the graphics processors for that CPU, so as to establish a target logical connection relationship between the two.

[0057] In one possible implementation, the aforementioned central processing units (CPUs) can be prioritized to allocate corresponding graphics processors (GPUs) to each CPU in descending order of priority. CPUs with higher priority are given priority in GPU allocation to those that meet their needs. The CPU priority is positively correlated with the CPU's workload and task priority; the greater the workload and the higher the task priority, the higher the CPU's priority. Conversely, the lower the workload and the lower the task priority, the lower the CPU's priority. This approach prioritizes the needs of CPUs with higher workloads and more critical tasks, ensuring a higher success rate for their task execution.

[0058] In one embodiment of this application, reference is made to Figure 4 As shown, the management controller 101 further includes a feasibility verification unit 1014, which is connected to the storage unit 1013 and the switching configuration file generation unit 1012. The feasibility verification unit 1014 is used to: obtain preset topology switching information from the storage unit 1013, the preset topology switching information including the switchability relationship between one or more topology modes; perform feasibility verification on the management request according to the preset topology switching information to obtain a feasibility verification result; and when the feasibility verification result is that the feasibility verification is passed, control the switching configuration file generation unit 1012 to determine the topology switching configuration file according to the management request. In this way, the switching between topology modes can be restricted by the preset topology switching information to pre-determine whether it is feasible, avoid topology switching failure, poor topology switching result, and abnormal topology structure after switching, which may lead to abnormal server operation.

[0059] In one example, the preset topology switching information includes multiple directed topology pattern pairs. Each directed topology pattern pair includes a first topology pattern to a second topology pattern, indicating that switching from the first topology pattern to the second topology pattern is supported. When there is no directed topology pattern pair from the second topology pattern to the first topology pattern in the preset topology switching information, it means that switching from the second topology pattern to the first topology pattern is not supported.

[0060] When performing feasibility verification based on the aforementioned preset topology switching information, the feasibility verification unit 1014 needs to obtain the current topology mode. The current topology mode can be carried in the management request or obtained by the feasibility verification unit 1014 from the storage unit 1013. Thus, the current topology mode and the target topology mode can be constructed as a pair to be verified to determine whether the pair exists in the preset topology switching information. If it exists, the feasibility verification result is "feasibility verification passed," and a topology switching configuration file can be generated according to the management request. If it does not exist, the feasibility verification result is "feasibility verification failed," and a topology switching configuration file does not need to be generated according to the management request; the topology switching fails.

[0061] When the current topology is stored in storage unit 1013, the amount of data required for management requests can be reduced. At this time, refer to... Figure 4 As shown, the feasibility verification unit 1014 is also connected to the management request parsing unit 1011. The feasibility verification unit 1014 is further used for:

[0062] The system obtains the current topology mode and the target topology mode sent by the management request parsing unit 1011 from the storage unit 1013, and determines whether the current topology mode can be switched to the target topology mode according to the preset topology switching information. If the current topology mode can be switched to the target topology mode, the system determines that the management request verification is successful. If the current topology mode cannot be switched to the target topology mode, the system determines that the management request verification fails.

[0063] The current topology mode is written into the storage unit 1013 by the switching configuration file generation unit 1012 after the last successful topology switch. The switching configuration file generation unit 1012 is connected to the aforementioned storage unit 1013.

[0064] In other embodiments, reference is made to... Figure 4As shown, the management controller 101 further includes a switching mode verification unit 1015, which is connected to the management request parsing unit 1011 and the switching configuration file generation unit 1012. The switching mode verification unit 1015 is used to: receive the topology switching configuration file sent by the management request parsing unit 1011 and parse the topology switching configuration file to obtain the actual topology mode; match the actual topology mode with the target topology mode to obtain a matching result; when the matching result indicates that the actual topology mode and the target topology mode match, control the switching configuration file generation unit 1012 to send the topology switching configuration file to the topology switching controller 102; when the matching result indicates that the actual topology mode and the target topology mode do not match, the switching configuration file generation unit 1012 will not send the topology switching configuration file to the topology switching controller 102. In this embodiment, when both the target topology mode and the topology switching configuration file exist in the management request, it determines whether they match to avoid topology switching when the management request is incorrect, thus helping to improve the accuracy of topology switching.

[0065] In some optional implementations, when the switching mode verification unit 1015 determines that the actual topology mode and the target topology mode do not match, it can also send a request error indication message to the management client 103. Upon receiving the request error indication message, the management client 103 displays the request error indication message to prompt the user to modify it. This can improve the success rate of topology switching.

[0066] In some embodiments of this application, reference is made to Figure 4 As shown, the management controller 101 further includes a system restart unit 1016, which is connected to the topology switching controller 102. The system restart unit 1016 is used to: upon receiving a successful switch message from the topology switching controller 102, determine that the topology switch was successful and restart the server's operating system; if no successful switch message is received from the topology switching controller 102 within a preset time period, determine that the topology switch failed, and do not restart the server's operating system. In this way, the system can be accurately restarted based on the successful switch message to ensure the target logical connection relationship in the aforementioned topology switching configuration file takes effect.

[0067] Specifically, while sending the topology switching configuration file to the topology switching controller 102, the switching configuration file generation unit 1012 also sends a timer start message to the system restart unit 1016. The system restart unit 1016 starts the timer, and the timeout duration is the aforementioned preset duration. If the system restart unit 1016 receives a switching success message before the timer expires, it restarts the server's operating system; if it does not receive a switching success message before the timer expires, it does not restart the server's operating system.

[0068] It is understandable that the aforementioned switching configuration file generation unit 1012 can also estimate the preset duration based on the data volume of the topology switching configuration file. The preset duration is positively correlated with the number of topology switching configuration files, and the switching configuration file generation unit 1012 sends the preset duration and the timer start message together to the system restart unit 1016. In this way, accurate estimation of the timeout duration is achieved to set an appropriate timeout duration, which helps to save time.

[0069] To further accurately implement operating system reboot, refer to... Figure 4 As shown, the system restart unit 1016 is also connected to the feasibility verification unit 1014 and the switching mode verification unit 1015. The system restart unit 1016 is further configured to: determine the processor's topology switching result based on at least one of the following: the feasibility verification result, the switching result message sent by the topology switching controller 102, or the matching result; and restart the operating system when the topology switching result indicates successful switching. Successful switching includes: the feasibility verification result being "feasibility verification passed," the matching result indicating that the actual topology mode and the target topology mode match, and the switching result message being a "successful switching message." When the feasibility verification result is "failed" or the matching result is "mismatched," the topology switching result can be determined as a "switching failure" without waiting for the switching result message, allowing for a quicker determination of the switching result and subsequent processing, thus saving time.

[0070] Among some possible implementations, refer to Figure 4 As shown, the processor topology switching system 100 further includes a switching result feedback unit 1017, connected to the feasibility verification unit 1014, the switching mode verification unit 1015, and the management client 103. The switching result feedback unit 1017 is used for:

[0071] When the topology switching result indicates a switching failure, a switching failure message is generated and sent to the management client 103 for display. A switching failure includes at least one of the following: a feasibility verification result indicating a failure, a matching result indicating a mismatch between the actual and target topology modes, or a switching failure message. Thus, switching failure can be determined immediately upon failure of feasibility verification or a mismatch between the actual and target topology modes, without waiting for a switching result message. This allows for rapid feedback of the switching failure message to alert the user.

[0072] Of course, the failover message can also include the reason for the failover, such as: failure of feasibility verification, mismatch between the actual topology mode and the target topology mode, or failure of the topology switch controller 102 to switch over. This allows users to adjust their operation commands accordingly based on the reason for the failover, thereby improving the success rate of the next topology switchover.

[0073] Furthermore, in some possible implementations, refer to Figure 4 As shown, the handover result feedback unit 1017 is further configured to: determine the cause of handover failure and generate handover adjustment instruction information based on the cause of handover failure; add the handover failure instruction information and the handover adjustment instruction information to the handover failure message, wherein the handover failure instruction information is used to indicate handover failure. Thus, this application intelligently generates handover adjustment instruction information to instruct the user to adjust operation commands according to the handover adjustment instruction information, eliminating the need for the user to analyze the cause of handover failure to determine the adjustment strategy for operation commands, thereby further reducing user operation complexity and lowering the user operation threshold.

[0074] It should be noted that the above-mentioned handover failure indication information can be represented by a flag bit, for example, 0 to indicate handover failure. Of course, in the handover success message, the handover success indication information is represented by a flag bit of 1. The handover adjustment indication information is used to adjust the target topology mode, or one or more target logical connection relationships in the topology handover configuration file, etc.

[0075] Corresponding to the handover result feedback unit 1017 sending handover adjustment instruction information to the management client 103, the management client 103 is also configured to: display the handover adjustment instruction information and update the management request according to the handover adjustment instruction information, so that upon receiving a user's confirmation operation on the management request, it sends the management request to the management request parsing unit 1011, so that the management request parsing unit 1011 parses it, the configuration unit generates a topology handover configuration file, and the topology handover controller 102 performs topology handover according to the topology handover configuration file. In this way, the management client 103 can automatically update the management request, further reducing the user's operation of updating the management request, simplifying the complexity of user operations, and improving user operation efficiency.

[0076] Of course, in practical applications, there may be multiple switching and adjustment instructions, which may generate multiple updated management requests. The management client 103 can display multiple updated management requests at the same time so that the user can select one as the management request to be executed.

[0077] In one embodiment of this application, reference is made to Figure 4 As shown, the handover result feedback unit 1017 is also connected to the storage unit 1013. The handover result feedback unit 1017 is used to: determine the handover failure level according to the reason for the handover failure when the topology handover result indicates a handover failure, and record the reason for the handover failure in the log file in the storage unit 1013 according to the reason for the handover failure level. In this way, the reason for the handover failure can be recorded in the log file to facilitate subsequent analysis and backtracking of the reason for the handover failure, avoid the recurrence of the same reason for the handover failure, and help improve the handover success rate.

[0078] The handover failure level describes the degree of impact of the cause of the handover failure. The greater the impact, the higher the handover failure level; the smaller the impact, the lower the handover failure level. One or more handover failure causes correspond to one handover failure level.

[0079] In one example, a log file can be set up for each failover level to store the failover records corresponding to that level. These records include not only the reason for the failover but also the time of failure and the management request. This allows for periodic counting of failover records in the log file for each failover level, and statistical analysis of the number of failover records for each level. If the number of failover records is high at higher failover levels, further analysis for that level should be initiated to reduce its frequency. If the number of failover records is low at higher failover levels, further analysis may not be necessary.

[0080] Among some possible implementations, refer to Figure 4 As shown, the management controller 101 further includes a resource management control unit 1018, which is connected to the topology switching controller 102 and the switching profile generation unit 1012. The resource management control unit 1018 is used to: upon receiving a switching success message from the topology switching controller 102, allocate target resources required for the operation of the server's central processing unit (CPU) according to the topology switching profile sent by the switching profile generation unit 1012. In this way, the target resources required for the operation of each CPU are associated with the corresponding graphics processor, ensuring that the CPU's target resources are sufficient without being wasted.

[0081] In some possible implementations, the resource management control unit 1018 is also used to: extract the number of graphics processors corresponding to each central processing unit from the topology switching configuration file; adjust resource parameters according to the number of graphics processors corresponding to the central processing unit, the resource parameters including power supply parameters and / or heat dissipation parameters, the power supply parameters and heat dissipation parameters being positively correlated with the number of graphics processors; and adjust the target resources of the central processing unit according to the resource parameters, the target resources including power supply resources and / or heat dissipation resources.

[0082] The aforementioned power supply parameters indicate the ability to supply power to the central processing unit (CPU), including power output. The heat dissipation parameters indicate the cooling capacity provided to the CPU, including the number of heatsinks and fans. Both the power supply and heat dissipation parameters are positively correlated with the number of graphics processing units (GPUs) in the CPU. This ensures normal power supply and cooling for the CPU while conserving resources.

[0083] In one embodiment of this application, reference is made to Figure 4 As shown, the management controller 101 also includes a server performance detection unit 1019. The server performance detection unit 1019 is connected to the management client 103 and the topology switching controller 102. The server performance detection unit 1019 is used to: upon receiving a successful switching message from the topology switching controller 102, detect server performance, obtain performance parameters, and send the performance parameters to the management client 103. In this way, server performance can be detected in real time after a successful topology switch, allowing users to decide whether to continue using the current processor topology based on server performance. When server performance is poor, users can continue to switch processor topologies; when server performance is good, they can continue to use the current processor topology. This ensures server performance as much as possible and avoids service quality degradation caused by topology switching.

[0084] Of course, when the client receives performance parameters for multiple topology modes, it can analyze these parameters to determine the topology mode with the highest performance, which will then be the recommended topology mode. This recommended topology mode can then be displayed the next time the user switches processor topologies, allowing the user to select it directly.

[0085] Alternatively, the topology modes can be sorted in descending order according to their performance parameters, so that users can choose the topology mode with the highest performance parameters first.

[0086] Among some possible implementations, refer to Figure 4As shown, the management controller 101 further includes a link configuration unit 1010, which is connected to the management request parsing unit 1011. The management request parsing unit 1011 is also used to send a management request to the link configuration unit 1010 when the request type corresponds to a link configuration request. The link configuration request can also be generated by operation instructions entered by the user in the management client 103. The user can enter the request type and request content corresponding to the link configuration request in the management client 103 to generate the link configuration request. For example, when the management client 103 provides a UI interface, the user can select a link configuration request from the list of options in the request type selection box in the UI interface. At this time, the management client 103 displays the configuration interface for the link configuration request, so that the user can enter the request content of the link configuration request in the configuration interface.

[0087] The link configuration unit 1010 is used to configure the link between the handover configuration file generation unit 1012 and the topology handover controller 102. The link configuration request carries the link's validity period and transmission performance parameters. This application can also ensure the success rate of subsequent topology handover configuration file transmission by allowing the user to configure the link.

[0088] The aforementioned effective time is used to indicate that the switching configuration file generation unit 1012 needs to transmit the topology switching configuration file to the topology switching controller 102 within the effective time.

[0089] The transmission performance parameter is usually the transmission speed, which is used to instruct the switching configuration file generation unit 1012 to transmit the data to the topology switching controller 102 at the specified transmission speed.

[0090] In one possible implementation, refer to Figure 4 As shown, the management client 103 also includes a link configuration estimation unit 1031; the link configuration estimation unit 1031 is used to: estimate the effective time and transmission performance parameters based on the number of processors on the server, and display them, and generate a link configuration request based on the effective time and transmission performance parameters when receiving confirmation from the user regarding the effective time and transmission performance parameters.

[0091] Specifically, the management client 103 can also estimate the data volume of the topology switching configuration file based on the number of processors on the server, and the data volume of the topology switching configuration file is positively correlated with the number of processors. Based on this data volume, the effective time and transmission performance parameters are estimated and displayed. When the user confirms the input of the effective time and transmission performance parameters, a link configuration request can be generated based on the effective time and transmission performance parameters and sent to the management request parsing unit 1011.

[0092] The aforementioned effective time is positively correlated with the amount of data in the aforementioned topology switching configuration file, while the transmission performance parameters are negatively correlated with the amount of data in the aforementioned topology switching configuration file.

[0093] In practical applications, the valid time input by the user can be received first, indicating the user's expectation that the topology switching configuration file will be transmitted between the configuration file generation unit 1012 and the topology switching controller 102 within that valid time. The ratio of the data volume of the topology switching configuration file to the valid time is used as the transmission performance parameter, that is, the amount of data transmitted per unit time. This method can ensure, as far as possible, that the topology switching configuration file is transmitted within a certain time.

[0094] In another approach, user-inputted transmission performance parameters can be received first to represent the user's desired transmission performance, and the ratio of the topology switching configuration file's data size to the transmission performance can be used as the effective time. This method can ensure the transmission speed of the topology switching configuration file as much as possible.

[0095] In addition, users can adjust the effective time and transmission performance parameters based on the estimated effective time and transmission performance parameters of the management client 103, and generate a link configuration request with the adjusted effective time and transmission performance parameters.

[0096] Figure 5 This application provides a flowchart of a processor topology switching method, applied to the aforementioned processor topology switching system 100. The processor topology switching system 100 includes a management controller 101 and a topology switching controller 102. The management controller 101 includes a management request parsing unit 1011 and a switching configuration file generation unit 1012. The switching configuration file generation unit 1012 is connected to both the management request parsing unit 1011 and the topology switching controller 102. The connection between the switching configuration file generation unit 1012 and the topology switching controller 102 can be implemented using a UART (Universal Asynchronous Receiver / Transmitter). Specifically, the switching configuration file generation unit 1012 includes a transceiver, and the topology switching controller 102 includes a transceiver; communication between the two transceivers is achieved through these two transceivers.

[0097] Based on the processor topology switching system 100 described above, the processor topology switching method of this application includes, but is not limited to, the following steps:

[0098] S201, the management request parsing unit 1011 receives a management request and parses the request type of the management request. The management request is generated based on the operation instructions entered by the user on the management client 103.

[0099] Specifically, the user inputs an operation command on the management client 103, the management client 103 generates a management request based on the operation command, and sends the management request to the management request parsing unit 1011.

[0100] In this embodiment of the application, a management request is any management request that requires the management controller 101 to manage the server, including but not limited to topology switching requests and link configuration requests.

[0101] A topology switch request may include a target topology pattern and / or a topology switch configuration file. The topology switch configuration file may be a file imported by the user through the UI of the management client 103.

[0102] S202, when the request type corresponds to a topology switching request, the management request parsing unit 1011 controls the switching configuration file generation unit 1012 to generate a topology switching configuration file corresponding to the management request. The topology switching configuration file includes the target logical connection relationship between processors on the server.

[0103] When the control switching configuration file generation unit 1012 generates the topology switching configuration file corresponding to the management request, the management request parsing unit 1011 needs to send different information to the topology switching configuration file. Specifically, the management request parsing unit 1011 can send the target topology pattern parsed from the management request to the switching configuration file generation unit 1012, so that the switching configuration file generation unit 1012 can generate the topology switching configuration file according to the target topology pattern. The management request parsing unit 1011 can also send the topology switching configuration file parsed from the management request to the switching configuration file generation unit 1012. Of course, the management request parsing unit 1011 can also send both the topology switching configuration file and the target topology pattern to the switching configuration file generation unit 1012 simultaneously.

[0104] S203, The configuration file generation unit 1012 generates the topology switching configuration file corresponding to the management request and sends the topology switching configuration file to the topology switching controller 102.

[0105] When the switching configuration file generation unit 1012 receives the target topology mode, it can select the topology switching configuration file corresponding to the target topology mode from the configuration files corresponding to one or more topology modes respectively; or, the switching configuration file generation unit 1012 sends the received topology switching configuration file to the topology switching controller 102.

[0106] S204, Topology switching controller 102 switches the logical connection relationship between processors according to the topology switching configuration file.

[0107] In some possible implementations, the management controller 101 further includes a storage unit 1013, and the switching configuration file generation unit 1012 is connected to the storage unit 1013; the processor topology switching method further includes a management request parsing unit 1011 parsing the target topology mode from the management request and sending the target topology mode to the switching configuration file generation unit 1012.

[0108] Accordingly, the switching configuration file generation unit 1012 determines the topology switching configuration file based on the management request, including:

[0109] From one or more configuration files stored in storage unit 1013, determine the topology switching configuration file corresponding to the target topology mode; if the topology switching configuration file does not exist in the configuration files, dynamically generate the topology switching configuration file according to the target topology mode.

[0110] Among some possible implementations, the processor topology switching method also includes:

[0111] The management request parsing unit 1011 parses the topology switching configuration file from the management request. The topology switching configuration file is generated by the management client 103 based on the operation instructions input by the user.

[0112] Accordingly, the switching configuration file generation unit 1012 determines the topology switching configuration file based on the management request, including:

[0113] When the topology switching configuration file is received from the management request parsing unit 1011, the switching configuration file generation unit 1012 sends the topology switching configuration file to the topology switching controller 102; when the topology switching configuration file is not received from the management request parsing unit 1011, the switching configuration file generation unit 1012 determines the topology switching configuration file according to the target topology mode. That is to say, the management request does not contain a topology switching configuration file.

[0114] In some possible implementations, the switching configuration file generation unit 1012 determines the topology switching configuration file based on the management request, including:

[0115] Calculate the number of GPUs corresponding to each CPU based on the target topology pattern; establish the target logical connection relationship between the CPU and GPUs based on the number of GPUs corresponding to each CPU and the CPU's GPU requirements; and generate a topology switching configuration file based on the target logical connection relationship.

[0116] For example, first create a blank configuration file, then add the target logical connection relationships to this blank configuration file to obtain the topology switching configuration file described above. The target logical connection relationships can exist in the topology switching configuration file in graphical and / or text form. The way these target logical connection relationships exist can be set by the user in the management client 103.

[0117] The requirement information described here refers to the CPU's requirements for the graphics processor (GPU) allocated to it. Assigning a GPU that does not meet these requirements to the CPU may affect the collaborative processing performance of the CPU and GPU, thereby reducing server performance. Therefore, this embodiment of the application requires assigning a GPU that meets the requirements to the CPU to maximize server performance.

[0118] In some possible implementations, the management controller 101 further includes: a feasibility verification unit 1014, which is connected to the storage unit 1013 and the switching configuration file generation unit 1012; the processor topology switching method further includes:

[0119] The feasibility verification unit 1014 obtains preset topology switching information from the storage unit 1013. The preset topology switching information includes the switchable relationship between one or more topology modes. The feasibility verification unit 1014 performs feasibility verification on the management request based on the preset topology switching information to obtain the feasibility verification result. When the feasibility verification result is that the feasibility verification is passed, the feasibility verification unit 1014 controls the switching configuration file generation unit 1012 to determine the topology switching configuration file based on the management request.

[0120] In some possible implementations, the feasibility verification unit 1014 is also connected to the management request parsing unit 1011; the processor topology switching method further includes:

[0121] The feasibility verification unit 1014 obtains the current topology mode and the target topology mode sent by the management request parsing unit 1011 from the storage unit 1013, and determines whether the current topology mode can be switched to the target topology mode according to the preset topology switching information; if the current topology mode can be switched to the target topology mode, the feasibility verification unit 1014 determines that the management request verification is successful; if the current topology mode cannot be switched to the target topology mode, the feasibility verification unit 1014 determines that the management request verification fails.

[0122] In some possible implementations, the management controller 101 further includes a switching mode verification unit 1015, which is connected to the management request parsing unit 1011 and the switching configuration file generation unit 1012. The processor topology switching method further includes: the switching mode verification unit 1015 receiving the topology switching configuration file sent by the management request parsing unit 1011 and parsing the topology switching configuration file to obtain the actual topology mode; the switching mode verification unit 1015 matching the actual topology mode with the target topology mode to obtain a matching result; when the matching result indicates that the actual topology mode and the target topology mode match, the switching mode verification unit 1015 controls the switching configuration file generation unit 1012 to send the topology switching configuration file to the topology switching controller 102.

[0123] In some possible implementations, the management controller 101 further includes a system restart unit 1016, which is connected to the topology switching controller 102. The processor topology switching method further includes: when the system restart unit 1016 receives a switching success message sent by the topology switching controller 102, it indicates that the topology switching controller 102 has successfully performed the switching action, but the server's operating system still needs to be restarted so that the target logical connection relationship after the switch takes effect.

[0124] In some possible implementations, the system restart unit 1016 is also connected to the feasibility verification unit 1014 and the switching mode verification unit 1015. When the system restart unit 1016 receives a switching success message sent by the topology switching controller 102, it restarts the server's operating system, including:

[0125] The system restart unit 1016 determines the processor's topology switching result based on at least one of the feasibility verification result, the switching result message sent by the topology switching controller 102, or the matching result. When the topology switching result indicates that the switching is successful, the system restart unit 1016 restarts the operating system. Successful switching includes: the feasibility verification result being that the feasibility verification is passed, the matching result indicating that the actual topology mode and the target topology mode match, and the switching result message being a switching success message.

[0126] In some possible implementations, the aforementioned topology switching system further includes a switching result feedback unit 1017, which is connected to the feasibility verification unit 1014, the switching mode verification unit 1015, and the management client 103. The processor topology switching method further includes: when the topology switching result indicates a switching failure, the switching result feedback unit 1017 generates a switching failure message to be sent to the management client 103 for display. The switching failure includes at least one of the following: the feasibility verification result is a feasibility verification failure, the matching result indicates that the actual topology mode and the target topology mode do not match, or the switching result message is a switching failure message.

[0127] Among some possible implementations, the processor topology switching method also includes:

[0128] The handover result feedback unit 1017 determines the reason for the handover failure and generates handover adjustment instruction information based on the reason for the handover failure. The handover result feedback unit 1017 adds the handover failure instruction information and the handover adjustment instruction information to the handover failure message. The handover failure instruction information is used to indicate the handover failure.

[0129] In some possible implementations, the switching result feedback unit 1017 is also connected to the storage unit 1013, and the processor topology switching method further includes:

[0130] When the topology switching result indicates a switching failure, the switching result feedback unit 1017 determines the switching failure level based on the switching failure reason and records the switching failure reason in the log file in the storage unit 1013 according to the switching failure level.

[0131] Among some possible implementations, the processor topology switching method also includes:

[0132] The management client 103 displays the switch adjustment instruction information and updates the management request according to the switch adjustment instruction information, so that when it receives the user's confirmation operation on the management request, it sends the management request to the management request parsing unit 1011.

[0133] In some possible implementations, the management controller 101 further includes a resource management control unit 1018, which is connected to the topology switching controller 102 and the switching profile generation unit 1012. The processor topology switching method further includes: when the resource management control unit 1018 receives a switching success message sent by the topology switching controller 102, it allocates the target resources required for the central processing unit to run the server according to the topology switching profile sent by the switching profile generation unit 1012.

[0134] Of course, if the resource management control unit 1018 does not receive a successful switch message from the topology switch controller 102, it indicates a switch failure, including feasibility verification failure, mismatch between the current topology mode and the target topology mode, and switch failure of the topology switch controller 102. In this case, the resource management control unit 1018 does not need to allocate the target resources required for the server's central processing unit to run.

[0135] In some possible implementations, the processor topology switching method described above also includes:

[0136] The resource management control unit 1018 extracts the number of graphics processors corresponding to each central processing unit from the topology switching configuration file; the resource management control unit 1018 adjusts the resource parameters according to the number of graphics processors corresponding to the central processing unit, the resource parameters including power supply parameters and / or heat dissipation parameters, the power supply parameters and heat dissipation parameters being positively correlated with the number of graphics processors; the resource management control unit 1018 adjusts the target resources of the central processing unit according to the resource parameters, the target resources including power supply resources and / or heat dissipation resources.

[0137] The aforementioned power supply parameters can be calculated using a first preset formula, which may be a formula established based on historical experience. Alternatively, the power supply parameters corresponding to the number of graphics processors can be found in a preset mapping table.

[0138] The aforementioned heat dissipation parameters can be calculated using a second preset formula, which may be a formula established based on historical experience. Alternatively, the heat dissipation parameters corresponding to the number of graphics processors can be found in a preset mapping table.

[0139] In some possible implementations, the management controller 101 further includes: a server performance detection unit 1019, which is connected to the management client 103 and the topology switching controller 102; the processor topology switching method further includes:

[0140] When the server performance detection unit 1019 receives a successful switching message from the topology switching controller 102, it detects the server performance, obtains performance parameters, and sends the performance parameters to the management client 103.

[0141] In some possible implementations, the management controller 101 further includes a link configuration unit 1010, which is connected to the management request parsing unit 1011. The processor topology switching method further includes the management request parsing unit 1011 sending a management request to the link configuration unit 1010 when the request type corresponds to a link configuration request. The link configuration unit 1010 configures the link between the switching configuration file generation unit 1012 and the topology switching controller 102, and the link configuration request carries the link's validity period and transmission performance parameters.

[0142] In some possible implementations, the management client 103 further includes: a link configuration estimation unit 1031, and the processor topology switching method further includes:

[0143] The link configuration estimation unit 1031 estimates the effective time and transmission performance parameters based on the number of processors on the server, displays them, and generates a link configuration request based on the effective time and transmission performance parameters when it receives confirmation from the user regarding the effective time and transmission performance parameters.

[0144] It is understood that the above-described processor topology switching method corresponds to the aforementioned processor topology switching system 100, allowing users to simply input operation commands in the management client 103. The management client 103 can generate a management request based on the operation commands and send it to the management controller 101. The management controller 101 generates a topology switching configuration file based on the management request to control the topology switching controller 102, thereby realizing the switching of logical connection relationships between processors. In this embodiment, the target logical connection relationship in the topology switching configuration file is generated by the management controller 101. Users only need to input simple operation commands, resulting in low operation complexity, low cost, and effective reduction of switching error rate.

[0145] This application also provides a server, including the aforementioned processor topology switching system 100.

[0146] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0147] The processor topology switching system, processor topology switching method, and server provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this application. It should be noted that those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. A processor topology switching system, characterized in that, include: The system includes a management client, a management controller, and a topology switching controller. The management controller is a Baseboard Management Controller (BMC), which runs independently of the server's operating system and implements processor topology switching through out-of-band management. The management controller comprises a management request parsing unit, a storage unit, a switching configuration file generation unit, a feasibility verification unit, a switching mode verification unit, and a system restart unit. The switching configuration file generation unit is connected to the management request parsing unit, the storage unit, and the topology switching controller. The feasibility verification unit is connected to the storage unit and the switching configuration file generation unit. The switching mode verification unit is connected to the management request parsing unit and the switching configuration file generation unit. The system restart unit is connected to the topology switching controller. The feasibility verification unit is also connected to the management request parsing unit. The management request parsing unit is used to: receive a management request from the management client, extract a flag bit from a specified bit in the header information of the management request, parse the request type of the management request according to the flag bit, and when the request type corresponds to a topology switching request, control the switching configuration file generation unit to generate a topology switching configuration file corresponding to the management request. The management request is generated based on the operation instructions input by the user on the management client, and the topology switching configuration file includes the target logical connection relationship between processors on the server. The feasibility verification unit is configured to: obtain preset topology switching information from the storage unit, the preset topology switching information including a switchable relationship between one or more topology modes; obtain the current topology mode from the storage unit, and receive the target topology mode sent by the management request parsing unit, and determine whether the current topology mode can be switched to the target topology mode according to the preset topology switching information, so as to perform feasibility verification on the management request and obtain a feasibility verification result; when the feasibility verification result is that the feasibility verification is passed, control the switching configuration file generation unit to determine the topology switching configuration file according to the management request; The switching configuration file generation unit is used to: determine the topology switching configuration file corresponding to the target topology mode from one or more configuration files stored in the storage unit according to the management request; or, if the topology switching configuration file does not exist in the configuration files, calculate the number of graphics processors corresponding to each central processing unit (CPU) according to the target topology mode, wherein, when the target topology mode is a balanced mode, the number of graphics processors corresponding to each CPU is the integer result of the ratio of the total number of graphics processors of the server to the total number of CPUs; when the target topology mode is a cascaded mode, allocate all graphics processors to one of the CPUs; for each CPU, select graphics processors that meet the requirements information as candidate graphics processors for that CPU, and select a number of graphics processors from the candidate graphics processors as the graphics processors for that CPU, so as to establish a target logical connection relationship between the CPU and the graphics processors; the requirements information includes the CPU's memory requirements, floating-point operation performance requirements, and parallel operation requirements for the graphics processors; dynamically generate the topology switching configuration file according to the target logical connection relationship, and send the topology switching configuration file to the topology switching controller; The topology switching controller is used to: switch the logical connection relationship between processors according to the topology switching configuration file; The switching mode verification unit is used to: receive the topology switching configuration file sent by the management request parsing unit, and parse the topology switching configuration file to obtain the actual topology mode; The actual topology pattern and the target topology pattern are matched to obtain a matching result; When the matching result indicates that the actual topology mode and the target topology mode match, the switching configuration file generation unit is controlled to send the topology switching configuration file to the topology switching controller; The system restart unit is used to restart the operating system of the server when it receives a successful switch message from the topology switch controller.

2. The system according to claim 1, characterized in that, The management request parsing unit is further configured to: parse the target topology pattern from the management request and send the target topology pattern to the switching configuration file generation unit.

3. The system according to claim 2, characterized in that, The management request parsing unit is also used for: The topology switching configuration file is parsed from the management request. The topology switching configuration file is generated by the management client based on the operation instructions input by the user. The switching configuration file generation unit is also used for: Upon receiving the topology switching configuration file sent by the management request parsing unit, the topology switching configuration file is sent to the topology switching controller; If the topology switching configuration file is not received from the management request parsing unit, the topology switching configuration file is determined according to the target topology mode.

4. The system according to claim 1, characterized in that, The feasibility verification unit is also used for: When the current topology mode can be switched to the target topology mode, it is determined that the management request verification is successful; If the current topology mode cannot be switched to the target topology mode, the management request verification is determined to have failed.

5. The system according to claim 1, characterized in that, The system restart unit is also connected to the feasibility verification unit and the switching mode verification unit, and the system restart unit is further used for: The topology switching result of the processor is determined based on at least one of the feasibility verification results, the switching result message sent by the topology switching controller, or the matching results. When the topology switching result indicates a successful switch, the operating system is restarted. A successful switch includes: the feasibility verification result being that the feasibility verification is passed, the matching result indicating that the actual topology mode and the target topology mode match, and the switching result message being a successful switch message.

6. The system according to claim 5, characterized in that, It also includes a switching result feedback unit, connected to the feasibility verification unit, the switching mode verification unit, and the management client. The switching result feedback unit is used for: When the topology switching result indicates a switching failure, a switching failure message is generated and sent to the management client for display. The switching failure includes at least one of the following: the feasibility verification result is a feasibility verification failure, the matching result indicates that the actual topology mode and the target topology mode do not match, or the switching result message is a switching failure message.

7. The system according to claim 6, characterized in that, The switching result feedback unit is also used for: Determine the reason for the handover failure, and generate handover adjustment instruction information based on the reason for the handover failure; Add the handover failure indication information and the handover adjustment indication information to the handover failure message. The handover failure indication information is used to indicate that the handover failed.

8. The system according to claim 7, characterized in that, The switching result feedback unit is also connected to the storage unit, and the switching result feedback unit is used for: When the topology switching result indicates a switching failure, the switching failure level is determined based on the reason for the switching failure, and the reason for the switching failure is recorded in the log file in the storage unit according to the switching failure level.

9. The system according to claim 8, characterized in that, The management client is also used for: The switching and adjustment instruction information is displayed, and the management request is updated according to the switching and adjustment instruction information, so that when the user confirms the management request, the management request is sent to the management request parsing unit.

10. The system according to any one of claims 1 to 3, characterized in that, The management controller further includes a resource management control unit, which is connected to the topology switching controller and the switching configuration file generation unit; The resource management control unit is used to: upon receiving a successful switching message from the topology switching controller, allocate target resources required for the operation of the central processing unit of the server according to the topology switching configuration file sent by the switching configuration file generation unit, wherein the target resources include power supply resources and / or heat dissipation resources.

11. The system according to claim 10, characterized in that, The resource management and control unit is also used for: Extract the number of graphics processors corresponding to each central processing unit from the topology switching configuration file; The resource parameters are adjusted according to the number of graphics processors corresponding to the central processing unit. The resource parameters include power supply parameters and / or heat dissipation parameters, and the power supply parameters and the heat dissipation parameters are positively correlated with the number of graphics processors. The target resources of the central processing unit are adjusted according to the resource parameters.

12. The system according to any one of claims 1 to 3, characterized in that, The management controller further includes a server performance detection unit, which is connected to the management client and the topology switching controller, and is used for: Upon receiving a successful switchover message from the topology switchover controller, the server performance is detected to obtain performance parameters, which are then sent to the management client.

13. The system according to any one of claims 1 to 3, characterized in that, The management controller further includes a link configuration unit, which is connected to the management request parsing unit. The management request parsing unit is further configured to: when the request type corresponds to a link configuration request, send the management request to the link configuration unit; The link configuration unit is used to configure the link between the switching configuration file generation unit and the topology switching controller. The link configuration request carries the effective time and transmission performance parameters of the link.

14. The system according to claim 13, characterized in that, The management client includes: a link configuration estimation unit; The link configuration estimation unit is used to: estimate the validity time and the transmission performance parameters based on the number of processors on the server, display them, and generate the link configuration request based on the validity time and the transmission performance parameters when receiving confirmation from the user regarding the validity time and the transmission performance parameters.

15. A processor topology switching method, characterized in that, A processor topology switching system according to any one of claims 1-14, comprising: a management controller and a topology switching controller, the management controller comprising: a management request parsing unit and a switching configuration file generation unit, the switching configuration file generation unit being connected to the management request parsing unit and the topology switching controller; the method comprising: The management request parsing unit receives a management request and parses the request type of the management request, which is generated based on the operation instructions entered by the user on the management client. When the management request parsing unit receives a topology switch request corresponding to the request type, it controls the switch configuration file generation unit to generate a topology switch configuration file corresponding to the management request. The topology switch configuration file includes the target logical connection relationship between processors on the server. The switching configuration file generation unit generates the topology switching configuration file corresponding to the management request and sends the topology switching configuration file to the topology switching controller; The topology switching controller switches the logical connection relationships between processors according to the topology switching configuration file.

16. A server, characterized in that, Includes the processor topology switching system according to any one of claims 1 to 14.

Images