Topological grouping method and device for multi-track InfiniBand network

Abstract

The invention provides a topology grouping method and device for a multi-track InfiniBand network. The method comprises the following steps: acquiring link information and port information of each device of the multi-track InfiniBand network; grouping the servers based on the link information and the port information of the servers to obtain a plurality of server groups; constructing a corresponding super computing unit based on each server group and the Leaf switch connected with the server group; based on the link information and the port information of the Leaf switch of each super computing unit, constructing a Spine switch group; and constructing a Core switch group based on the link information and the port information of the Spine switch of each Spine switch group. According to the method provided by the invention, through topological grouping of the multi-track InfiniBand network, efficient utilization of cluster resources and stable output of computing performance can be guaranteed.

Description

Technical Field

[0001] This application relates to the field of computer network technology, and in particular to a topology grouping method and apparatus for a multi-track InfiniBand network. Background Technology

[0002] Existing methods for grouping complex network topologies have the following technical problems when dealing with InfiniBand (IB) networks: Fixed layout limits display effectiveness: Traditional topology visualization typically uses a fixed-space layout to display servers, switches, and other devices in layers to represent a hierarchical structure. However, in multi-track IB networks of intelligent computing centers and large-scale computing clusters, the large number of nodes and complex connections result in crisscrossing lines in the topology diagram, leading to a chaotic and difficult-to-identify display that affects the clarity of the visualization and the effective presentation of the topology structure.

[0003] Lack of flexible hierarchical grouping mechanism: Due to their high bandwidth and low latency requirements, multi-track IB networks employ complex full-mesh or parallel multi-track topologies, resulting in extremely dense connections between nodes and switches. Existing hierarchical grouping methods struggle to effectively handle the diversity and density of this structure, failing to reflect the actual logical hierarchy of the network in demonstrations. This leads to information redundancy and hinders network administrators from quickly understanding and locating critical devices.

[0004] Insufficient real-time monitoring and dynamic adaptability: In intelligent computing center environments, real-time device status and traffic monitoring are crucial. However, existing solutions are mostly static displays, lacking dynamic updates and responsiveness. This results in administrators being unable to obtain critical status information in a timely manner during device fault diagnosis, alarm handling, and performance optimization, limiting the visualization effect and fault diagnosis efficiency of cluster operation.

[0005] Performance bottlenecks are difficult to identify: The complexity of multi-track IB networks makes performance bottlenecks and network faults more hidden. Traditional static displays cannot directly provide real-time data such as traffic hotspots and device loads, which increases the difficulty of diagnosis and makes it difficult to detect potential problems in a timely manner, thereby affecting the reliability and performance of the intelligent computing center cluster. Summary of the Invention

[0006] In view of this, this application provides a topology grouping method and apparatus for multi-track InfiniBand networks to solve the above-mentioned technical problems.

[0007] In a first aspect, embodiments of this application provide a topology grouping method for a multi-track InfiniBand network, including: Obtain the link information and port information of each device in the multi-track InfiniBand network during the target time period. Each device includes at least: a server, a Leaf switch, a Spine switch, and a Core switch. Based on the link information and port information of the servers in the multi-track InfiniBand network for the target time period, the servers are grouped to obtain multiple server groups; Based on each server group and its connected Leaf switches, a corresponding supercomputing unit is constructed. The supercomputing unit contains multiple servers and multiple Leaf switches. In this case, a server in a supercomputing unit is only connected to a Leaf switch within the supercomputing unit. Based on the link and port information of the Leaf switches in each supercomputing unit, Spine switch groups are constructed; wherein, the Leaf switches in a supercomputing unit are all connected to the Spine switches in a Spine switch group. Core switch groups are constructed based on the link and port information of the Spine switches in each Spine switch group; wherein, the Spine switches in a Spine switch group are all connected to the Core switches in a Core switch group.

[0008] In one possible implementation, servers are grouped based on the link and port information of the servers in the multi-track InfiniBand network for the target time period, resulting in multiple server groups, including: Call the CollectDeviceLinkPortInfo function to obtain information about all servers and their connected Leaf switches; The servers are grouped based on information about all servers and their connected Leaf switches, resulting in multiple server groups.

[0009] In one possible implementation, a corresponding supercomputing unit is constructed based on each server group and its connected Leaf switches; including: Get the port information of each server in a server group; Based on the port information of each server, the link between the server and the Leaf switch is determined; Based on the link between the server and the Leaf switches, determine the set of Leaf switches connected to the server; The set of Leaf switches for the supercomputing unit is obtained by deduplicating the union of the Leaf switch sets for all servers. One Leaf switch is randomly selected from the Leaf switch set of the supercomputing unit as the core switch of the supercomputing unit.

[0010] In one possible implementation, Spine switch packets are constructed based on the link and port information of the Leaf switches in each supercomputing unit; including: Get the ID of each Leaf switch in the Leaf switch set of each supercomputing unit; Based on the identifier of each Leaf switch, determine the link to the Spine switch it is connected to; Based on the links of the Spine switches connected to each Leaf switch, determine the set of Spine switches connected to each Leaf switch; Call the FindLinksBetween function to obtain the unique Spine switch set of the Leaf switch set of the supercomputing unit, so that the connection relationship between each Leaf switch and the Spine switch is correct; The deduplication operation removes duplicate Spine switches from the Spine switch set, generating Spine switch groups.

[0011] In one possible implementation, a Core switch group is constructed based on the link and port information of the Spine switches in each Spine switch group, including: For each Spine switch in a Spine switch group, locate the Core switch links connected to the Spine switches; Call the FindLinksBetween function to obtain the unique set of Core switches for the Spine switch group, so that the connection relationship between each Spine switch and the Core switch is correct; The deduplication operation removes duplicate Core switches from the Core switch set, generating Core switch groups.

[0012] In one possible implementation, the method further includes: Display multiple supercomputing units, Spine switch groups, and Core switch groups.

[0013] Secondly, embodiments of this application provide a topology packet transposition for a multi-track InfiniBand network, including: The acquisition unit is used to acquire the link information and port information of each device in the multi-track InfiniBand network during the target time period. The devices include at least: a server, a Leaf switch, a Spine switch, and a Core switch. The processing unit is used to group servers based on the link information and port information of the servers in the multi-track InfiniBand network for the target time period, thereby obtaining multiple server groups; The first grouping unit is used to construct a corresponding supercomputing unit based on each server group and its connected Leaf switches. The supercomputing unit contains multiple servers and multiple Leaf switches, wherein a server in a supercomputing unit is only connected to a Leaf switch within the supercomputing unit. The second grouping unit is used to construct Spine switch groups based on the link and port information of the Leaf switches in each supercomputing unit; wherein, the Leaf switches in a supercomputing unit are all connected to the Spine switches in a Spine switch group; The third grouping unit is used to construct a Core switch group based on the link information and port information of the Spine switches in each Spine switch group; the Spine switches in the Spine switch group are all connected to the Core switches in a Core switch group.

[0014] Thirdly, embodiments of this application provide an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method of embodiments of this application.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the methods of embodiments of this application.

[0016] Fifthly, embodiments of this application provide a computer program product, including a computer program / instructions, which, when executed by a processor, implement the method of embodiments of this application.

[0017] The method in this application can ensure efficient utilization of cluster resources and stable output of computing performance by topologically grouping the multi-track InfiniBand network. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating the topology grouping method for a multi-track InfiniBand network provided in this application embodiment; Figure 2 A system architecture diagram of a multi-track InfiniBand network provided in the embodiments of this application; Figure 3 A topology diagram of a multi-track InfiniBand network provided in the embodiments of this application; Figure 4 Functional structure diagram of the topology packetization device for a multi-track InfiniBand network provided in the embodiments of this application; Figure 5 This is a structural diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, 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. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0021] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0022] First, a brief introduction to the design concept of the embodiments of this application will be given.

[0023] In modern high-performance computing and large-scale intelligent computing centers, InfiniBand (IB) networks, with their high bandwidth and low latency, have become a key architecture supporting massively parallel computing and data transmission. With the rapid development of artificial intelligence, the application of large models in natural language processing, machine learning, and data analysis has driven a strong demand for computing resources in intelligent computing centers. As the core infrastructure for large model training and inference, the management and optimization of computing power in intelligent computing centers are particularly important.

[0024] Network topology is crucial for device management, situational awareness, and real-time monitoring in intelligent computing centers. Topology diagrams provide a clear visual representation of the network infrastructure and the connections between devices, offering administrators key insights into network structure and operational status. However, traditional network topology visualizations often employ fixed-space layouts, displaying servers, switches, and other devices in layers to represent a hierarchical structure, or using conventional star, ring, bus, or mesh topologies. In the context of the large scale and complex connections of intelligent computing centers, this approach often results in chaotic visualizations and excessively crossed connections due to the fixed layout and complex topology, reducing the clarity and effectiveness of the topology display. Furthermore, traditional methods lack sufficient flexibility to support the needs of ultra-large-scale networks like intelligent computing centers, increasing the difficulty of network troubleshooting and performance optimization.

[0025] The complex topology and massive equipment array of the multi-track InfiniBand network significantly increase the difficulty of management. In particular, the network topology of the intelligent computing center not only needs to clearly display the network structure, but also needs to display the device status in real time in order to quickly identify and handle faults or performance bottlenecks, and ensure the continuity and accuracy of large model training.

[0026] To address this, this application proposes an intelligent network topology hierarchical grouping method suitable for large-scale multi-track IB network environments. Based on the actual operational characteristics of complex multi-track IB networks, this method intelligently hierarchically and group nodes, optimizing the network structure display and making the topology information more intuitive and clear. Simultaneously, through real-time updated dynamic display views and interactive functions, administrators can obtain device status, performance data, and alarm information in real time, quickly locate faulty devices and critical connections, and optimize network performance promptly. This solution significantly enhances the cluster's device monitoring and troubleshooting capabilities while improving the network management efficiency of intelligent computing centers, providing stable and efficient network support for large model training and high-performance computing tasks.

[0027] After introducing the application scenarios and design concepts of the embodiments of this application, the technical solutions provided by the embodiments of this application will be described below.

[0028] like Figure 1 As shown, this application provides a topology grouping method for a multi-track InfiniBand network, including: Step 101: Obtain the link information and port information of each device in the multi-track InfiniBand network during the target time period. The devices include at least: server, Leaf switch, Spine switch and Core switch. Step 102: Based on the link information and port information of the servers in the multi-track InfiniBand network for the target time period, the servers are grouped to obtain multiple server groups; Step 103: Based on each server group and its connected Leaf switches, construct a corresponding supercomputing unit. The supercomputing unit contains multiple servers and multiple Leaf switches. A server in a supercomputing unit is only connected to a Leaf switch within the supercomputing unit. Step 104: Based on the link and port information of the Leaf switches in each supercomputing unit, construct a Spine switch group; wherein, the Leaf switches in a supercomputing unit are all connected to the Spine switches in a Spine switch group; Step 105: Based on the link information and port information of the Spine switches in each Spine switch group, construct the Core switch group; wherein, the Spine switches in a Spine switch group are all connected to the Core switches in a Core switch group.

[0029] This embodiment, through an adaptive topology grouping method and an efficient data acquisition mechanism, enables precise management of clusters ranging from kilo-card to 10,000-card levels. It achieves a hierarchical distribution of server-Leaf switch-Spine switch-Core switch, allowing large-scale complex topologies to be rendered and displayed in a concise, efficient, and clear manner.

[0030] The method in this embodiment breaks through the limitations of traditional methods in large and complex networks. Especially in large-scale model training and computationally intensive task scenarios, it can ensure the efficient utilization of cluster resources and the stable output of computing performance, providing a reliable technical guarantee for the construction of next-generation intelligent computing centers. It effectively solves the bottleneck problem of existing technologies in the application of large-scale intelligent computing clusters, breaks through the limitations of traditional methods in complex networks, and provides new ideas and technical guarantees for the efficient management of large-scale clusters.

[0031] In some embodiments, servers are grouped based on the link information and port information of the servers in the multi-track InfiniBand network during the target time period, resulting in multiple server groups, including: Call the CollectDeviceLinkPortInfo function to obtain information about all servers and their connected Leaf switches; The servers are grouped based on information about all servers and their connected Leaf switches, resulting in multiple server groups.

[0032] In some embodiments, a corresponding supercomputing unit is constructed based on each server group and its connected Leaf switches; including: Get the port information of each server in a server group; Based on the port information of each server, the link between the server and the Leaf switch is determined; Based on the link between the server and the Leaf switches, determine the set of Leaf switches connected to the server; The set of Leaf switches for the supercomputing unit is obtained by deduplicating the union of the Leaf switch sets for all servers. One Leaf switch is randomly selected from the Leaf switch set of the supercomputing unit as the core switch of the supercomputing unit.

[0033] In some embodiments, Spine switch groups are constructed based on the link and port information of the Leaf switches of each supercomputing unit; including: Get the ID of each Leaf switch in the Leaf switch set of each supercomputing unit; Based on the identifier of each Leaf switch, determine the link to the Spine switch it is connected to; Based on the links of the Spine switches connected to each Leaf switch, determine the set of Spine switches connected to each Leaf switch; Call the FindLinksBetween function to obtain the unique Spine switch set of the Leaf switch set of the supercomputing unit, so that the connection relationship between each Leaf switch and the Spine switch is correct; The deduplication operation removes duplicate Spine switches from the Spine switch set, generating Spine switch groups.

[0034] In this embodiment, different Leaf switches are connected to different sets of Spine switches.

[0035] In some embodiments, a Core switch group is constructed based on the link information and port information of the Spine switches in each Spine switch group, including: For each Spine switch in a Spine switch group, locate the Core switch links connected to the Spine switches; Call the FindLinksBetween function to obtain the unique set of Core switches for the Spine switch group, so that the connection relationship between each Spine switch and the Core switch is correct; The deduplication operation removes duplicate Core switches from the Core switch set, generating Core switch groups.

[0036] In this embodiment, different Spine switches are grouped and connected to different Core switches.

[0037] The Core grouping in this embodiment ensures the stability of the highest-level network connections, providing core support for the overall topology architecture.

[0038] In some embodiments, the method further includes: Display multiple supercomputing units, Spine switch groups, and Core switch groups.

[0039] This embodiment supports real-time display of network topology status, enabling administrators to quickly respond to device changes, traffic fluctuations, and fault alarms, achieving more accurate device monitoring and more agile fault handling.

[0040] The method of this embodiment will be described in detail below through a specific example.

[0041] like Figure 2 As shown, the InfiniBand network is designed with a four-layer architecture, including a control layer, a data layer, an algorithm layer, and a presentation layer. The modules in each layer are interconnected through data flow and control signals, specifically: Control layer: Includes timer module and data acquisition module, responsible for triggering data acquisition at regular intervals to ensure the real-time and effectiveness of network topology data and device performance data.

[0042] Data layer: Includes data interface module and data processing module, responsible for organizing and standardizing the collected raw data to meet the input requirements of the algorithm layer and provide a consistent and standardized data source for subsequent topology grouping.

[0043] Algorithm layer: Composed of network topology grouping algorithm module. This module receives data processed by the data layer, generates the grouped network topology structure through intelligent grouping algorithm, and updates the hierarchical information of network devices as needed.

[0044] Presentation Layer: The topology visualization module receives the grouping results output by the algorithm layer and generates a visualized network topology map in real time. It supports dynamic updates and intuitive structural display, making it easy for users to understand the network status in real time.

[0045] Through the modular design of the above four-layer architecture, a system layout with clear functions and distinct layers is achieved, which can efficiently process and display complex network structures, and ensure real-time updates of network topology and visualization of grouping results.

[0046] The network topology is divided into a server layer and a switch layer, the latter further subdivided into an access layer, an aggregation layer, and a core layer, corresponding to the Leaf layer, Spine layer, and Core layer switches, respectively. Taking the networking design of the A800 large-scale cluster as an example, this network architecture is highly structured, making full use of the high bandwidth and low latency characteristics of the IB network to ensure efficient data transmission and optimal utilization of computing resources.

[0047] The specific grouping steps are as follows: Step 1 Initialization: Collect the network graph G, device set D, link set L, and port set P;

[0048] Step 2: Build the server and Leaf switch group (SU) For each server : 2.1 Extract server port information:

[0049] 2.2 Locate relevant links based on port information:

[0050] 2.3 Determine the set of Leaf switches connected to the server:

[0051] 2.4 Randomly select one Leaf switch as the core switch for SU packets:

[0052] 2.5 Construct and save SU groups:

[0053] Step 3: Construct Spine Switch Packets (SGs) Retrieve all Leaf switch IDs in the SU group:

[0054] 3.2 Locate and deduplicate links connected to the Spine switch based on Leaf switch information:

[0055] 3.3 Generate a Spine switch set:

[0056] 3.4 Constructing Spine Groups:

[0057] Step 4: Build Core Switch Packets (CGs) For the set of Spine switches in each SG group: 4.1 Locate the Core switch link connected to the Spine switch:

[0058] 4.2 Construct the Core switch set after deduplication:

[0059] 4.3 Build and save the Core group:

[0060] Step 5 Output: Return the set of all groups SU, SG, CG.

[0061] In this design, servers are organized into supercomputing units (SUs), each containing multiple servers and Leaf switches. Each server connects to multiple Leaf switches via multiple network interface cards (NICs), creating redundant connections to improve network reliability and fault recovery capabilities. The Leaf switches within each SU are logically grouped to support efficient interconnection between servers and the IB network. The Leaf switches further connect to Spine switches, which are organized into multiple SG groups to optimize traffic distribution within the IB network. Finally, global traffic is processed through core switches, which are divided into multiple CG groups, providing a highly integrated network backbone for the entire large model cluster. Figure 3The diagram illustrates the connection methods in the topology diagram. SU groups contain multiple compute nodes and Leaf switches, with each compute node's network interface card (NIC) connected to a Leaf switch. Spine switches are connected to Leaf switches in such a way that the first Leaf switch in each SU group connects to SG-1, the second to SG-2, and so on; a Leaf switch in an SU group will connect to all SG groups. Core switches are connected to Spine switches in such a way that the first Spine server in each SG connects to CG-1, the second to CG-2, and so on; a Spine switch in an SG group will connect to all SG groups.

[0062] Access Layer (Edge Layer): As the first layer of the network, the access layer consists of Leaf switches, whose ports are directly connected to the servers. The access layer is primarily responsible for aggregating and initially processing the data traffic generated by the servers. This layer is designed for efficient data transmission, ensuring that data traffic can be quickly and stably delivered to the next layer of the network.

[0063] Aggregate Layer: Composed of Spine switches, the aggregation layer is responsible for receiving and further processing data traffic from the access layer. This layer not only performs data routing and filtering operations but also forwards the processed data to the core layer. The aggregation layer plays a crucial role in the network architecture, ensuring smooth data transmission between different network nodes and effectively alleviating network congestion under conditions of high data traffic.

[0064] Core Layer: The core layer is the core part of the entire network architecture, composed of core switches, and is responsible for handling global traffic across the entire network. The core layer is tightly connected to the aggregation layer, providing high-bandwidth, low-latency backbone connections to ensure the entire intelligent computing center can work collaboratively in an efficient and stable environment.

[0065] By employing the aforementioned layered logic, the challenge of managing complex network topologies in intelligent computing clusters is resolved, particularly in multi-track InfiniBand network environments, ensuring efficient and stable data transmission between different network layers. This design not only improves overall network performance but also significantly enhances network stability, reduces potential failure risks arising from network architecture complexity, and greatly supports the smooth execution of large-scale computing tasks within intelligent computing clusters.

[0066] Employing a unique four-layer architecture—core, aggregation, access, and device layers—this design presents the connections within complex topologies more clearly and systematically. This layered design, combined with the characteristics of multi-track IB networks, ensures an intuitive and scalable network view even with high-density nodes and multi-track connections, effectively enhancing the hierarchy and readability of the topology display. The introduction of Canvas drawing technology enables high-performance visualization rendering, supporting smooth operation and dynamic interaction of complex topologies in large-scale clusters. Compared to traditional methods, it significantly reduces page load time and operation latency, ensuring that administrators can quickly and easily obtain topology information even in large clusters, improving system response efficiency and user experience. Based on modular design and standard interfaces, it supports the access of various network structures and new devices, exhibiting excellent scalability and compatibility. Facing the ever-expanding needs of intelligent computing centers, this solution can easily adapt to cluster topologies of different sizes and architectures, solving the problem of insufficient scalability in complex scenarios using traditional methods.

[0067] Based on the same inventive concept, this application provides a topology packetization device for a multi-track InfiniBand network, see reference. Figure 4 As shown, the topology packetization device for the multi-track InfiniBand network provided in this application embodiment includes at least: The acquisition unit 201 is used to acquire the link information and port information of each device in the multi-track InfiniBand network during the target time period. The devices include at least: a server, a Leaf switch, a Spine switch, and a Core switch. Processing unit 202 is used to group servers based on the link information and port information of the servers in the multi-track InfiniBand network of the target time period, and obtain multiple server groups; The first grouping unit 203 is used to construct a corresponding supercomputing unit based on each server group and its connected Leaf switches. The supercomputing unit contains multiple servers and multiple Leaf switches, wherein a server in a supercomputing unit is only connected to a Leaf switch within the supercomputing unit. The second grouping unit 204 is used to construct a Spine switch group based on the link information and port information of the Leaf switches in each supercomputing unit; wherein, the Leaf switches in a supercomputing unit are all connected to the Spine switches in a Spine switch group; The third grouping unit 205 is used to construct a Core switch group based on the link information and port information of the Spine switches in each Spine switch group; the Spine switches in the Spine switch group are all connected to the Core switches in a Core switch group.

[0068] It should be noted that the principle of the topology grouping device 200 for multi-track InfiniBand network provided in this application embodiment to solve the technical problem is similar to the method provided in this application embodiment. Therefore, the implementation of the topology grouping device 200 for multi-track InfiniBand network provided in this application embodiment can refer to the implementation of the method provided in this application embodiment, and the repeated parts will not be described again.

[0069] Based on the same inventive concept, embodiments of this application also provide an electronic device, such as... Figure 5 As shown, it includes a memory and a processor. The memory stores an executable program, and the processor executes the executable program to implement the steps of the topology grouping method for the multi-track InfiniBand network provided in the above embodiments.

[0070] The aforementioned processor can be a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The aforementioned PLD can be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0071] Since the electronic device described in this application embodiment is an electronic device equipped with a memory that implements the topology grouping method of the multi-track InfiniBand network disclosed in this application embodiment, those skilled in the art can understand the structure and variations of the electronic device described in this application embodiment based on the topology grouping method of the multi-track InfiniBand network described in this application embodiment, and therefore will not be described again here.

[0072] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the topology grouping method for the multi-track InfiniBand network provided in the above embodiments.

[0073] The storage medium in this embodiment may be included in an electronic device; or it may exist independently and not be assembled into an electronic device. The storage medium carries one or more computer programs, which, when executed, implement the steps of the topology grouping method for a multi-track InfiniBand network provided in the above embodiment.

[0074] It should be understood that the various solutions in this embodiment have the same technical effects as those in the above method embodiments, and will not be repeated here.

[0075] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. Optionally, specific examples in this embodiment can refer to the examples described in any embodiment of this application, which will not be repeated here. Obviously, those skilled in the art should understand that the various modules or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, this application is not limited to any particular hardware and software combination.

[0076] This application also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the topology grouping method for the multi-track InfiniBand network provided in the above embodiments.

[0077] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions targeted in the blocks may occur in a different order than those targeted in the drawings. For example, two consecutively represented blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0078] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. Multitasking and parallel processing may be advantageous in certain environments. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this application. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

Claims

1. A topology grouping method for a multi-track InfiniBand network, characterized in that, include: Obtain the link information and port information of each device in the multi-track InfiniBand network during the target time period. Each device includes at least: a server, a Leaf switch, a Spine switch, and a Core switch. Based on the link information and port information of the servers in the multi-track InfiniBand network for the target time period, the servers are grouped to obtain multiple server groups; Based on each server group and its connected Leaf switches, a corresponding supercomputing unit is constructed. The supercomputing unit contains multiple servers and multiple Leaf switches. In this case, a server in a supercomputing unit is only connected to a Leaf switch within the supercomputing unit. Based on the link and port information of the Leaf switches in each supercomputing unit, Spine switch groups are constructed; wherein, the Leaf switches in a supercomputing unit are all connected to the Spine switches in a Spine switch group. Core switch groups are constructed based on the link and port information of the Spine switches in each Spine switch group; wherein, the Spine switches in a Spine switch group are all connected to the Core switches in a Core switch group.

2. The method according to claim 1, characterized in that, Based on the link and port information of the servers in the multi-track InfiniBand network for the target time period, the servers are grouped to obtain multiple server groups, including: Call the CollectDeviceLinkPortInfo function to obtain information about all servers and their connected Leaf switches; The servers are grouped based on information about all servers and their connected Leaf switches, resulting in multiple server groups.

3. The method according to claim 2, characterized in that, Based on each server group and its connected Leaf switches, a corresponding supercomputing unit is constructed, including: Get the port information of each server in a server group; Based on the port information of each server, the link between the server and the Leaf switch is determined; Based on the link between the server and the Leaf switches, determine the set of Leaf switches connected to the server; The set of Leaf switches for the supercomputing unit is obtained by deduplicating the union of the Leaf switch sets for all servers. One Leaf switch is randomly selected from the Leaf switch set of the supercomputing unit as the core switch of the supercomputing unit.

4. The method according to claim 1, characterized in that, Based on the link and port information of the Leaf switches in each supercomputing unit, Spine switch groups are constructed, including: Get the ID of each Leaf switch in the Leaf switch set of each supercomputing unit; Based on the identifier of each Leaf switch, determine the link to the Spine switch it is connected to; Based on the links of the Spine switches connected to each Leaf switch, determine the set of Spine switches connected to each Leaf switch; Call the FindLinksBetween function to obtain the unique Spine switch set of the Leaf switch set of the supercomputing unit, so that the connection relationship between each Leaf switch and the Spine switch is correct; The deduplication operation removes duplicate Spine switches from the Spine switch set, generating Spine switch groups.

5. The method according to claim 1, characterized in that, Based on the link and port information of the Spine switches in each Spine switch group, a Core switch group is constructed, including: For each Spine switch in a Spine switch group, locate the Core switch links connected to the Spine switches; Call the FindLinksBetween function to obtain the unique set of Core switches for the Spine switch group, so that the connection relationship between each Spine switch and the Core switch is correct; The deduplication operation removes duplicate Core switches from the Core switch set, generating Core switch groups.

6. The method according to claim 1, characterized in that, The method further includes: Display multiple supercomputing units, Spine switch groups, and Core switch groups.

7. A topology packet transpose for a multi-track InfiniBand network, characterized in that, include: The acquisition unit is used to acquire the link information and port information of each device in the multi-track InfiniBand network during the target time period. The devices include at least: a server, a Leaf switch, a Spine switch, and a Core switch. The processing unit is used to group servers based on the link information and port information of the servers in the multi-track InfiniBand network for the target time period, thereby obtaining multiple server groups; The first grouping unit is used to construct a corresponding supercomputing unit based on each server group and its connected Leaf switches. The supercomputing unit contains multiple servers and multiple Leaf switches, wherein a server in a supercomputing unit is only connected to a Leaf switch within the supercomputing unit. The second grouping unit is used to construct Spine switch groups based on the link and port information of the Leaf switches in each supercomputing unit; wherein, the Leaf switches in a supercomputing unit are all connected to the Spine switches in a Spine switch group; The third grouping unit is used to construct a Core switch group based on the link information and port information of the Spine switches in each Spine switch group; the Spine switches in the Spine switch group are all connected to the Core switches in a Core switch group.

8. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method as claimed in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the method as described in any one of claims 1-6.

10. A computer program product, characterized in that, Includes a computer program / instruction that, when executed by a processor, implements the method as described in any one of claims 1-6.

Images