Server cabinet

Abstract

The invention discloses a server cabinet, and relates to the technical field of servers, the server cabinet comprises a plurality of computing nodes and a plurality of switching nodes, the plurality of computing nodes are arranged along a first direction, the plurality of switching nodes are arranged along a second direction, and each computing node is provided with a plurality of first connectors in a third direction. Each computing node is provided with a plurality of first connectors in the first direction, each switching node is provided with a plurality of second connectors in the third direction, and the first connectors are correspondingly connected with the second connectors, so that each computing node and each switching node are orthogonally connected when being blind-inserted into the server cabinet, and therefore, the computing nodes and the switching nodes can be orthogonally connected to form full interconnection; and the use of cables for connecting the computing nodes and the switching nodes is reduced, so that the cost required by full interconnection is reduced. Meanwhile, due to the design of orthogonal connection, the complexity of wiring is reduced, and the fault risk caused by wiring errors or poor connection is reduced, so that the reliability of the system is improved.

Description

Technical Field

[0001] This application relates to the technical field of servers, and particularly to a server cabinet. Background Art

[0002] A whole cabinet server places all business nodes in a server cabinet uniformly. The cabinet as a whole provides power supply, heat dissipation, and related business connection functions. There are business computing nodes and switching nodes for business interconnection in the cabinet, which together form a computing cluster.

[0003] In the related art, the computing nodes and the switching nodes are communicatively connected through cables to form a fully interconnected architecture. Since the number of computing nodes and switching nodes is relatively large, to achieve full interconnection, the number of cables surges, resulting in a sharp increase in connection costs. Moreover, the intersecting cables are prone to failures, affecting the reliability of the server cabinet. Summary of the Invention

[0004] This application provides a server cabinet to at least solve the problems of excessively high costs of fully interconnected cables between computing nodes and switching nodes and relatively low reliability of the server cabinet in the related art.

[0005] This application provides a server cabinet, including a plurality of computing nodes and a plurality of switching nodes. The plurality of computing nodes are arranged along a first direction, and the plurality of switching nodes are arranged along a second direction. Each computing node is provided with a plurality of first connectors in a third direction, and the plurality of first connectors on each computing node are arranged along the second direction. Each switching node is provided with a plurality of second connectors in the third direction, and the plurality of second connectors on each switching node are arranged along the first direction. The first connectors and the second connectors are correspondingly connected so that when each computing node and each switching node are blindly inserted into the server cabinet, they are orthogonally connected, and the first direction, the second direction, and the third direction are perpendicular to each other.

[0006] With the server cabinet provided by the embodiments of the present application, since multiple computing nodes and multiple switching nodes are orthogonally arranged in the cabinet and are plugged into each other through the first connector and the second connector, the computing nodes and the switching nodes can be orthogonally connected to form a full interconnection, reducing the use of cables for connecting the computing nodes and the switching nodes, thereby reducing the cost required for full interconnection. At the same time, the orthogonal connection design reduces the complexity of wiring and the risk of failures caused by wiring errors or poor connections, thereby improving the reliability of the system. Through the optimized connection method, the data transmission path is shorter and the latency is lower, thereby enhancing the performance of the overall system. Since the computing nodes and the switching nodes are arranged in different directions and are connected through the first connector and the second connector in the third direction, the internal space of the cabinet can be utilized more effectively, improving the overall density of the cabinet. The modular design enables the computing nodes and the switching nodes to be independently replaced or upgraded, simplifies the maintenance process, and reduces the downtime. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] To more clearly illustrate the embodiments of the present application, the following will briefly introduce the drawings required for the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0008] Figure 1 FIG. 1 is one of the schematic structural diagrams of a server cabinet provided by the embodiments of the present application;

[0009] Figure 2 is Figure 1 the front view of the shown server cabinet;

[0010] Figure 3 is Figure 1 the rear view of the shown server cabinet;

[0011] Figure 4 FIG. 2 is another schematic structural diagram of a server cabinet provided by the embodiments of the present application;

[0012] Figure 5 FIG. 3 is the schematic structural diagram of a bracket of a server cabinet provided by the embodiments of the present application;

[0013] Figure 6 is Figure 1 the top view of the shown server cabinet;

[0014] Figure 7 FIG. 4 is yet another schematic structural diagram of a server cabinet provided by the embodiments of the present application;

[0015] Figure 8 is Figure 7Schematic structural diagram of the water inlet pipeline of the server cabinet shown

[0016] Among them, the above-mentioned drawings include the following reference numerals:

[0017] 100 - Server cabinet; 10 - Computing node; 11 - First connector; 12 - Third connector; 13 - First processor; 14 - First in-board connector group; 141 - First in-board connector; 20 - Switching node; 21 - Second connector; 22 - Second processor; 23 - Second in-board connector group; 231 - Second in-board connector; 24 - Output port; 30 - Management switch; 31 - Fourth connector; 40 - Power supply module; 41 - Input cable; 42 - Busbar; 421 - First busbar; 422 - Second busbar; 50 - Cooling module; 51 - Water inlet pipeline; 511 - First pipeline; 512 - Second pipeline; 513 - First joint; 514 - Second joint; 52 - Water outlet pipeline; 521 - Third pipeline; 522 - Fourth pipeline; 53 - Total water inlet; 54 - Total water outlet; 60 - Bracket; 61 - First guiding column; 62 - Second guiding column; 101 - In-board wiring; 102 - Cable. Detailed implementation manners

[0018] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the protection scope of the present application.

[0019] It should be noted that the orientation or positional relationship indicated by terms such as "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation to the present application. The terms "installation", "connection", and "coupling" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. The terms "parallel", "perpendicular", and "equal" include the described situations and situations similar to the described situations, and the range of the similar situations is within the acceptable deviation range, where the acceptable deviation range is determined by those of ordinary skill in the art considering the measurements being discussed and the errors associated with the measurements of specific quantities (i.e., the limitations of the measurement system). For example, "parallel" includes absolute parallel and approximate parallel, and the acceptable deviation range of approximate parallel can be, for example, within 5° deviation; "perpendicular" includes absolute perpendicular and approximate perpendicular, and the acceptable deviation range of approximate perpendicular can also be, for example, within 5° deviation. "Equal" includes absolute equality and approximate equality, and the acceptable deviation range of approximate equality can be, for example, that the difference between the two equal ones is less than or equal to 5% of either one. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific situations.

[0020] In order to enable those skilled in the art of the present technology to better understand the solution of the present application, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0021] The server cabinet 100 can be an AI (Artificial Intelligence) integrated cabinet server. The AI integrated cabinet server is a high-performance computing infrastructure designed specifically for artificial intelligence (AI) applications. It integrates the computing nodes 10 and the switching nodes 20 and meets the high computing power requirements of AI training, inference, and large-scale data processing through in-depth optimization of hardware, software, and architecture.

[0022] Figure 1 One of the structural schematic diagrams of a server cabinet provided by an embodiment of the present application. As Figure 1As shown in the figure, the server cabinet 100 provided by the embodiment of the present application includes a plurality of computing nodes 10 and a plurality of switching nodes 20. The computing nodes and the switching nodes are communicatively connected to form a fully interconnected architecture.

[0023] Figure 2 For Figure 1 the front view of the server cabinet shown in the figure. As Figure 2 shown, a plurality of computing nodes 10 are arranged along the first direction x. A plurality of first connectors 11 are provided on each computing node 10 in the third direction z, and the plurality of first connectors 11 on each computing node 10 are arranged along the second direction y.

[0024] Figure 3 For Figure 1 the rear view of the server cabinet shown in the figure. As Figure 3 shown, a plurality of switching nodes 20 are arranged along the second direction y. A plurality of second connectors 21 are provided on each switching node 20 in the third direction z, and the plurality of second connectors 21 on each switching node 20 are arranged along the first direction x.

[0025] The plurality of computing nodes 10 and the plurality of switching nodes 20 form an orthogonal architecture.

[0026] The first connector 11 is correspondingly connected to the second connector 21, so that when each computing node 10 and each switching node 20 are blindly inserted into the server cabinet 100, they are orthogonally connected, and the first direction x, the second direction y, and the third direction z are perpendicular to each other.

[0027] That is, after the computing node 10 and the switching node 20 are pushed into the cabinet, the first connector 11 and the second connector 21 are automatically connected to realize high-speed signal electrical connection.

[0028] In some possible implementation manners, the first direction x is the width direction of the server cabinet 100, the second direction y is the height direction of the server cabinet 100, and the third direction z is the depth direction of the server cabinet 100.

[0029] With the server cabinet 100 provided by the embodiments of the present application, since multiple computing nodes 10 and multiple switching nodes 20 are orthogonally arranged in the cabinet and are plugged into each other through the first connector 11 and the second connector 21, the computing nodes 10 and the switching nodes 20 can be orthogonally connected to form a full interconnection, reducing the use of the cables 102 connecting the computing nodes 10 and the switching nodes 20, thereby reducing the cost required for the full interconnection. At the same time, the design of the orthogonal connection reduces the complexity of the wiring and reduces the risk of failures caused by wiring errors or poor connections, thereby improving the reliability of the system. Through the optimized connection method, the data transmission path is shorter and the delay is lower, thereby improving the performance of the overall system. Since the computing nodes 10 and the switching nodes 20 are arranged in different directions and are connected through the first connector 11 and the second connector 21 in the third direction z, the internal space of the cabinet can be utilized more effectively, and the overall density of the cabinet is increased. The modular design enables the computing nodes 10 and the switching nodes 20 to be independently replaced or upgraded, simplifies the maintenance process, and reduces the downtime.

[0030] In some possible implementation manners, the first connector 11 and the second connector 21 are orthogonal connectors.

[0031] In some possible implementation manners, 16 computing nodes 10 are arranged on the front side of the server cabinet 100, and 12 switching nodes 20 are arranged on the rear side of the server cabinet 100. The computing nodes 10 are arranged vertically, and the switching nodes 20 are arranged horizontally. The computing nodes 10 and the switching nodes 20 are orthogonally plugged into each other at 90 degrees in the cabinet. Each computing node 10 is connected to each switching node 20 to form a full interconnection.

[0032] In the server cabinet 100, the advantages of the vertical plug-in design of the computing nodes 10 are as follows:

[0033] 1. The height design of the computing nodes 10 is flexible. Since the computing nodes 10 are vertically plugged in, the computing nodes 10 are only limited by the height of the cold plate and the height of the panel ports, etc., and are no longer limited by the traditional fixed U height. Therefore, the height of the computing nodes 10 can be adjusted according to actual needs, and the granularity of the height is finer. This flexibility allows more computing nodes 10 to be accommodated within the total height of the cabinet, thereby increasing the computing density.

[0034] 2. The width design of the computing nodes 10 is flexible and is not limited by the traditional fixed 19-inch chassis, allowing the node width to be slightly increased. This design can accommodate more GPUs (Graphics Processing Units), thereby increasing the number of GPUs per unit area.

[0035] 3. Increase the computing capacity and density per unit cross-sectional area of the server cabinet 100. In some possible implementation manners, each computing node 10 is provided with 4 GPUs. By vertically arranging 16 computing nodes 10, each computing node 10 supports 4 GPUs, and the bandwidth of each GPU is 4.8T, so the total bandwidth reaches 19.2T.

[0036] The width of the cabinet is 19 inches, that is, 48.26 cm. 48.26 / 16 = 30 mm. Therefore, the vertical insertion design makes the thickness of each computing node 10 only 30 mm. Compared with the traditional 1U (44.45 mm) design, the density of the computing nodes 10 is greatly improved.

[0037] In the network architecture and bandwidth allocation of the server cabinet 100, the total bandwidth of the computing nodes 10 is equal to the total bandwidth of the switching nodes 20, that is, the bandwidth of a single GPU * the number of GPUs in the computing node 10 * the number of computing nodes 10 = the bandwidth of the chips on the switching node 20 * the number of chips on the switching node 20 * the number of switching nodes 20.

[0038] The bandwidth of each chip on the switching node 20 is 25.6T, and the total bandwidth of the computing nodes 10 is 19.2T * 16 = 307.2T.

[0039] Therefore, the switching nodes 20 are configured as: 307.2T / 25.6T = 12. 12 switching nodes 20 are required to support this bandwidth configuration.

[0040] Each computing node 10 provides 1.6T (19.2T / 12) of bandwidth to each switching node 20, and each GPU provides 0.4T (4.8T / 12) of bandwidth to each switching node 20.

[0041] Therefore, the switching nodes 20 are designed as follows: the height of the computing nodes 10 is 536 mm, which is allocated to 12 switching nodes 20, and the height of each switching node 20 is 44 mm (536 mm / 12).

[0042] In some possible implementation manners, the server cabinet 100 further includes a management switch 30. The management switch 30 and multiple switching nodes 20 are arranged along the second direction y. Each computing node 10 is further provided with a third connector 12 in the third direction z. The third connectors 12 on each computing node 10 and multiple first connectors 11 are arranged along the second direction y. The management switch 30 is provided with multiple fourth connectors 31 in the third direction z. Multiple fourth connectors 31 are arranged along the first direction x. Multiple fourth connectors 31 are connected to the third connectors 12 on multiple computing nodes 10 in a one-to-one correspondence, so that the management switch 30 is orthogonally connected to each computing node 10.

[0043] The introduction of the management switch 30 enables centralized management of multiple computing nodes 10. Through this centralized management, network configuration, monitoring, and troubleshooting can be carried out more efficiently. The management switch 30 can provide real-time performance monitoring and data analysis, helping administrators detect and resolve potential problems in a timely manner and ensuring the stable operation of the system.

[0044] By arranging the management switch 30 and the switching nodes 20 in an array along the second direction y, such that the management switch 30 and the computing nodes 10 also adopt an orthogonal connection design, the connection between the management switch 30 and the computing nodes 10 is more concise, reducing the complexity of redundant cabling, thereby reducing the risk of cabling errors. Due to the simplified cabling, the reliability of the system is improved, reducing failures caused by poor connections or cabling errors. Through the centralized control of the management switch 30, administrators can perform network configuration and adjustment more quickly, improving management efficiency.

[0045] At the same time, the orthogonal connection design allows for the easy addition or replacement of the computing nodes 10 and the management switch 30 without affecting the existing system, enhancing the scalability of the system. The orthogonal arrangement of the management switch 30 and the computing nodes 10 makes the utilization of the internal space of the cabinet more efficient, enabling more devices to be accommodated in a limited space.

[0046] In some possible implementation manners, the management switch 30 and each of the computing nodes 10 are also orthogonally plugged in. The management network interface in the computing node 10 is connected to the management switch 30 through a third connector 12 at the rear.

[0047] The computing node 10 eliminates the traditional method of connecting to the management switch 30 through network cables, which is convenient for management and saves the external interfaces of the computing node 10 at the same time.

[0048] In some possible implementation manners, along the second direction y, the management switch 30 is disposed in the middle of a plurality of switching nodes 20.

[0049] Placing the management switch 30 in the middle of a plurality of switching nodes 20 can shorten the signal transmission path, reduce latency, and improve data transmission efficiency. Placing the management switch 30 in the central position helps to balance the load distribution of network traffic, avoiding overload of the switching nodes 20 on one side, thereby improving the performance and stability of the entire system.

[0050] The centralized position of the management switch 30 makes it easier for maintenance personnel to access and manage the management switch 30 and its connections, simplifying the maintenance and troubleshooting process. At the same time, this layout supports modular design, enabling the convenient upgrade or replacement of the management switch 30 without affecting other components.

[0051] Figure 4This is the second schematic diagram of the structure of a server cabinet provided by an embodiment of the present application. As Figure 4 shown, in some possible implementation manners, the computing node 10 includes a plurality of first processors 13 and a plurality of first in-board connector groups 14. Each first in-board connector group 14 corresponds to one of the first processors 13, and each first in-board connector group 14 includes a plurality of first in-board connectors 141. The first processor 13 is respectively connected to the plurality of first in-board connectors 141 in the corresponding first in-board connector group 14 through in-board traces 101. Each first connector 11 corresponds to one of the first in-board connectors 141 in the plurality of first in-board connector groups 14 respectively. The first connector 11 is respectively connected to the corresponding first in-board connector 141 in each first in-board connector group 14 through a cable 102. That is, a certain first in-board connector 141 in the first in-board connector group 14 corresponding to each first processor 13 is connected to the first in-board connector 141 with the same serial number in the first in-board connector groups 14 corresponding to other first processors 13 through the cable 102 to the same first connector 11.

[0052] This design of the computing node 10 allows each first processor 13 to be independently connected to its corresponding first in-board connector 141 and the external first connector 11, making the system more modular and facilitating upgrade and maintenance.

[0053] By using the first in-board connectors 141 and the cables 102 to connect the first connectors 11, the system can be more easily configured and expanded to adapt to different computing requirements. The use of the in-board traces 101 helps the connection between the first processor 13 and the first in-board connectors 141, optimizes the integrity of signal transmission, reduces signal interference and loss, thereby improving the performance and reliability of the system. The independent connection paths of the first processor 13, the first in-board connectors 141, and the first connectors 11 reduce the possibility of single-point failures and improve the overall reliability of the system.

[0054] Due to the independence of each first processor 13 and the first in-board connectors 141, troubleshooting becomes simpler and more straightforward, and problems can be quickly located and solved. The modular and distributed design of the computing node 10 helps to more evenly distribute heat, optimize the heat dissipation path, and improve the overall heat dissipation efficiency.

[0055] In some possible implementation manners, the first processor 13 is a GPU.

[0056] In some possible implementations, the switching node 20 includes a plurality of second processors 22 and a plurality of second in-board connector groups 23. Each second in-board connector group 23 corresponds to one second processor 22. Each second in-board connector group 23 includes a plurality of second in-board connectors 231. The second processor 22 is respectively connected to the plurality of second in-board connectors 231 in the corresponding second in-board connector group 23 through in-board traces 101. Each second connector 21 corresponds to one of the second in-board connectors 231 in the plurality of second in-board connector groups 23 respectively. The second connector 21 is respectively connected to the corresponding second in-board connector 231 in the second in-board connector group 23 through a cable 102. That is, a certain second in-board connector 231 in the second in-board connector group 23 corresponding to each second processor 22 is connected to the second in-board connector 231 with the same serial number in the second in-board connector groups 23 corresponding to other second processors 22 through the cable 102 and is connected to the same second connector 21.

[0057] This design of the switching node 20 allows each second processor 22 to be independently connected to its corresponding second in-board connector 231 and the external second connector 21, making the system more modular and facilitating upgrading and maintenance.

[0058] By connecting the second in-board connector 231 and the cable 102 to the second connector 21, the system can be more easily configured and expanded to meet different computing requirements. The use of the in-board traces 101 helps connect the second processor 22 to the second in-board connector 231, optimizes the integrity of signal transmission, reduces signal interference and loss, and thus improves the performance and reliability of the system. The independent connection paths of the second processor 22, the second in-board connector 231, and the second connector 21 reduce the possibility of single-point failures and improve the overall reliability of the system.

[0059] Due to the independence of each second processor 22 and the second in-board connector 231, fault troubleshooting becomes simpler and more straightforward, enabling quick location and solution of problems. The modular and distributed design of the switching node 20 helps distribute heat more evenly, optimizes the heat dissipation path, and improves the overall heat dissipation efficiency.

[0060] The connection between the first processor 13 and the second processor 22 is achieved through the above-mentioned interconnection topology network, enabling each first processor 13 on each computing node 10 to be electrically connected and fully interconnected with each second processor 22 on each switching node 20. There are 16 computing nodes 10, 4 first processors 13, 12 switching nodes 20, and 2 second processors 22. The computing node 10 is connected to 12 first connectors 11, and the bandwidth of each first processor 13 is evenly distributed to the 12 first connectors 11 for full interconnection.

[0061] In some possible implementations, the server cabinet 100 further includes a plurality of output ports 24. Along the third direction z, the output ports 24 are disposed on the side of the switching node 20 away from the computing node 10, and the output ports 24 are connected to the second processor 22 through the in-board trace 101.

[0062] The output ports 24 are centrally arranged on the back of the switching node 20, making the external connections more neat and orderly, simplifying the wiring management, and improving the cleanliness inside the cabinet. The output ports 24 are located on the back of the switching node 20, facilitating quick access and operation by maintenance personnel, simplifying the connection and troubleshooting processes, and reducing the maintenance time. By concentrating the output ports 24 and related connections on the back of the switching node 20, the airflow can be better managed, the heat dissipation path can be optimized, and the overall heat dissipation efficiency of the cabinet can be improved. The use of the in-board trace 101 helps to reduce signal interference and loss, ensuring the integrity and reliability of the signal transmission from the second processor 22 to the output ports 24. Through the optimized wiring and connection design, the risk of failures caused by wiring errors or poor connections is reduced, thereby improving the reliability of the system. The output port 24 design on the back makes the system expansion easier, and adding or replacing connections will not interfere with other components inside the cabinet. Concentrating the output ports 24 on the back helps to more effectively utilize the internal space of the cabinet, providing more installation space for other components.

[0063] To achieve a non-blocking network architecture, the bandwidth of the output ports 24 needs to be the same as the input bandwidth of the switching node 20, which is 25.6T. A single switching node 20 includes 2 second processors 22 to meet this requirement.

[0064] Twelve switching nodes 20 are arranged at the rear side of the server cabinet 100. The switching nodes 20 are arranged horizontally, and a management switch 30 is arranged in the middle of the cabinet. There are 32 output ports 24 at the back end of each switching node 20 panel. The output ports 24 are OSFP (Octal Small Form-factor Pluggable, a network port in a packaging form), and the rate of each output port 24 is 800G. 32 * 800G = 25.6T, which is used for output. At this time, the output bandwidth = the bandwidth of the switching node 20 = 25.6T. The input and output bandwidths of each second processor 22 are the same, and the 25.6T bandwidth of the entire second processor 22 is evenly divided.

[0065] Since the output of the second processor 22 directly runs a wire from the second processor 22 to the output ports 24 through the in-board trace 101, there is no problem with the mating tolerance for the output ports 24 externally. Each second processor 22 is externally connected to 16 output ports 24, resulting in a non-blocking external connection.

[0066] In some possible implementations, the first connector 11 is actively connected to the computing node 10 to be floatingly connected to the panel of the computing node 10.

[0067] During installation, the active connection makes the alignment and connection of the computing node 10 easier, reducing the possibility of installation errors. The active connection allows the computing node 10 to move and adjust more flexibly during installation and maintenance. This flexibility makes it easier to make configuration changes in a limited space. The active connection design makes the plugging and unplugging of the computing node 10 more convenient, simplifies the maintenance and replacement process, reduces downtime, and improves the maintainability of the system. The active connection can reduce the physical stress caused by thermal expansion or mechanical vibration, thereby reducing the risk of wear and damage to the first connector 11 and the second connector 21. By reducing physical stress and simplifying the connection process, the active connection helps to improve the overall reliability of the system and reduce the risk of failures caused by poor connections.

[0068] At the same time, the active connection supports modular design, enabling the computing node 10 to be upgraded or replaced independently of other components, enhancing the scalability of the system. For technicians, the active connection provides a better user experience, making the operation and management of the system more intuitive and convenient.

[0069] In some possible implementations, the second connector 21 is actively connected to the switching node 20 to be floatingly connected to the panel of the switching node 20.

[0070] During installation, the active connection makes the alignment and connection of the switching node 20 easier, reducing the possibility of installation errors. The active connection allows the switching node 20 to move and adjust more flexibly during installation and maintenance. This flexibility makes it easier to make configuration changes in a limited space. The active connection design makes the plugging and unplugging of the switching node 20 more convenient, simplifies the maintenance and replacement process, reduces downtime, and improves the maintainability of the system. The active connection can reduce the physical stress caused by thermal expansion or mechanical vibration, thereby reducing the risk of wear and damage to the first connector 11 and the second connector 21. By reducing physical stress and simplifying the connection process, the active connection helps to improve the overall reliability of the system and reduce the risk of failures caused by poor connections.

[0071] At the same time, the active connection supports modular design, enabling the switching node 20 to be upgraded or replaced independently of other components, enhancing the scalability of the system. For technicians, the active connection provides a better user experience, making the operation and management of the system more intuitive and convenient.

[0072] In some possible implementation manners, the first connector 11 and the second connector 21 adopt a floating connection and are fixed on the computing node 10 and the switching node 20, which can avoid the problem that the direct rigid fixation of the first connector 11 and the second connector 21 to the circuit boards of the computing node 10 and the switching node 20 causes tolerances in orthogonal docking. In this way, when the computing node 10 and the switching node 20 are orthogonally inserted, there is a certain floating amount for both the first connectors 11 and the second connectors 21 on both sides, avoiding the loosening of other computing nodes 10 when one computing node 10 is unplugged or plugged in because multiple computing nodes 10 are all connected to the switching node 20 at the same time.

[0073] As Figure 5 shown, in some possible implementation manners, the server cabinet 100 further includes a bracket 60. Along the third direction z, the bracket 60 is located between the computing node 10 and the switching node 20. A plurality of first guiding columns 61 are provided on one side of the bracket 60 facing the computing node 10, and corresponding first guiding holes are provided on one side of the panel of the computing node 10 facing the bracket 60. The first guiding columns 61 are docked with the first guiding holes, so that a plurality of computing nodes 10 are blindly inserted into the server cabinet 100. A plurality of second guiding columns 62 are provided on one side of the bracket 60 facing the switching node 20, and corresponding second guiding holes are provided on one side of the panel of the switching node 20 facing the bracket 60. The second guiding columns 62 are docked with the second guiding holes, so that a plurality of switching nodes 20 are blindly inserted into the server cabinet 100, thereby enabling the first connector 11 and the second connector 21 to be correspondingly connected.

[0074] The bracket 60 is used to limit the positions of the computing node 10 and the switching node 20 during blind insertion, align the insertion positions of the computing node 10 and the switching node 20, and ensure that the connectors can be correctly plugged in.

[0075] In some possible implementation manners, the outer frame of the bracket 60 is square. On one side of the bracket 60 facing the computing node 10, a plurality of first guiding columns 61 are provided at the upper and lower side frames, and corresponding first guiding holes are provided at corresponding positions on the rear panel of the computing node 10. The two are docked, so that a plurality of computing nodes 10 can be docked with the plurality of first guiding columns 61 according to the designed positions for alignment. On one side of the bracket 60 facing the switching node 20, a plurality of second guiding columns 62 are provided at the left and right side frames, and corresponding second guiding holes are provided at corresponding positions on the rear panel of the switching node 20. The two are docked, so that a plurality of switching nodes 20 can be docked with the plurality of second guiding columns 62 according to the designed positions for alignment. Therefore, it is ensured that the computing node 10 and the switching node 20 can be accurately aligned when blindly inserting and docking high-speed signal connectors, and when performing power supply blind insertion connection and liquid cooling pipe blind insertion connection.

[0076] In some possible implementations, the server cabinet 100 further includes at least two power supply modules 40, and the at least two power supply modules 40 are arranged along the third direction z. When the computing nodes 10 are blindly inserted into the server cabinet 100, at least one power supply module 40 is electrically connected to the multiple computing nodes 10. When the switching nodes 20 are blindly inserted into the server cabinet 100, at least one power supply module 40 is electrically connected to the multiple switching nodes 20.

[0077] That is, after the computing nodes 10 and the switching nodes 20 are pushed into the cabinet, the power supply module 40 is automatically connected to the computing nodes 10 and the switching nodes 20 to achieve automatic power supply.

[0078] This design not only improves the wiring efficiency and maintenance convenience of the system, but also provides better support for the efficient operation of the data center, ensuring continuous and stable power supply.

[0079] Allocating the power supply load to different power supply modules 40 can effectively share the power load, reduce the pressure on a single module, and extend the service life of the power supply module 40. Different power supply modules 40 can be configured according to the different requirements of the computing nodes 10 and the switching nodes 20, providing a flexible power management solution to adapt to different workloads and conditions. By separately managing the power supply of the computing nodes 10 and the switching nodes 20, the energy consumption can be more finely monitored and optimized, improving the energy efficiency of the entire system.

[0080] At the same time, this design supports modularization, enabling the power supply module 40 to be upgraded or replaced independently of other components, enhancing the scalability of the system. By using multiple power supply modules 40, the system can achieve redundant power supply design. If one power supply module 40 fails, another module can continue to supply power, ensuring the continuous operation of the system and improving the overall reliability. The design of multiple power supply modules 40 allows for operation on a single module without affecting the operation of the entire system during maintenance or replacement, reducing the downtime. By separating the power supply paths of the computing nodes 10 and the switching nodes 20, the impact of power failures on the entire system can also be reduced, improving the security of the system.

[0081] In some possible implementations, the power supply module 40 is a powershelf (intelligent power management system). Since the density of the server cabinet 100 is relatively high and the numbers of both the computing nodes 10 and the switching nodes 20 are relatively high, the power of the entire cabinet will be very high, and it is difficult for a single power supply module 40 to support it. At least two power supply modules 40 are required. One power supply module 40 supplies power to the multiple computing nodes 10, and the other power supply module 40 supplies power to the multiple switching nodes 20.

[0082] Place two power supply modules 40 horizontally opposite to each other in the same plane, which can effectively utilize the depth space generated by the orthogonality between the computing nodes 10 and the switching nodes 20 in the orthogonal cabinet, thereby saving the height U number of the server cabinet 100. Compared with the way of stacking multiple intelligent power management systems vertically in the traditional cabinet, the overall height of the cabinet can be reduced.

[0083] In some possible implementation manners, along the second direction y, the power supply module 40 is arranged on one side of the plurality of computing nodes 10 and the plurality of switching nodes 20.

[0084] Concentrating the power supply modules 40 on one side makes the power wiring more concise and orderly, reduces the complexity of cable crossing and winding, and improves the tidiness inside the cabinet. The power supply module 40 is located on one side of the computing node 10 and the switching node 20, which is convenient for maintenance personnel to access and operate quickly, and simplifies the maintenance and fault troubleshooting process of the power supply system. By concentrating the power supply modules 40 on one side, the internal space of the cabinet can be utilized more effectively, providing more installation space for other components.

[0085] The centrally arranged power supply modules 40 contribute to better management of the air flow inside the cabinet, optimize the heat dissipation path, improve the overall heat dissipation efficiency, and ensure that the equipment operates at the best temperature. Centralized management of the power supply modules 40 helps better monitor and manage power distribution, improve the security of the system, and reduce the impact of power failures on the entire system.

[0086] At the same time, by optimizing the wiring and reducing the cable length, the risk of failures caused by wiring errors or poor connections is reduced, thereby improving the reliability of the system. This layout also supports modular design, enabling the power supply modules 40 to be upgraded or replaced independently of other components, enhancing the scalability of the system.

[0087] In some possible implementation manners, two power supply modules 40 are designed on the top side of the server cabinet 100 to supply power to the entire cabinet. One power supply module 40 supplies power to all the computing nodes 10, and the other power supply module 40 supplies power to all the switching nodes 20.

[0088] The directions in which the two power supply modules 40 are inserted into the PSU (Power Supply Unit) are both outward, which is convenient for plugging and unplugging.

[0089] In some possible implementation manners, the server cabinet 100 further includes at least two input cables 41. Along the third direction z, the input cables 41 are arranged in the middle of at least two power supply modules 40 and are connected to the power supply modules 40 in a one-to-one correspondence.

[0090] The input cable 41 is centrally arranged in the middle of the power supply module 40, making the power input wiring more orderly, reducing the complexity of cable crossing and winding, and improving the cleanliness inside the cabinet. By reducing the cable length and optimizing the wiring path, the risk of failures caused by wiring errors or poor connections is reduced, thereby improving the reliability of the system. By optimizing the layout of the input cable 41, the internal space of the cabinet can be utilized more effectively, providing more installation space for other components.

[0091] The design of the input cable 41 at the middle position makes installation and maintenance more convenient. Technicians can more easily access and manage the power cables, simplifying the operation process. The input cable 41 is connected to the power supply module 40 in a one-to-one correspondence, providing a flexible power management solution that can be adjusted and configured according to different power requirements. Centralized management of the input cable 41 helps to better monitor and manage the power input, improve the security of the system, and reduce the impact of power failures on the entire system. This design supports modularity, enabling the input cable 41 and the power supply module 40 to be upgraded or replaced independently of other components, enhancing the scalability of the system.

[0092] In some possible implementation manners, along the second direction y, the input cable 41 is arranged on a side of the power supply module 40 away from the computing node 10 and the switching node 20. The input cable 41 exits in the middle above the cabinet, thus avoiding the traditional way of exiting from the back of the cabinet and interacting with the cables on the back sides of other cabinets. A hole can be opened at the middle position on the top of the cabinet to provide space for the input cable 41 to pass through.

[0093] The input cable 41 is arranged on the surface of the two power supply modules 40 facing the inside of the cabinet as the input of AC (Alternating Current).

[0094] Figure 7 This is the third structural schematic diagram of a server cabinet provided by an embodiment of the present application. As Figure 7 shown, in some possible implementation manners, the server cabinet 100 further includes at least two busbars 42, and the power supply module 40 is electrically connected to the computing node 10 or the switching node 20 through the corresponding busbar 42.

[0095] The busbar 42 can effectively centralize and distribute power, reduce the power loss in traditional cable connections, and improve the power transmission efficiency. Using the busbar 42 instead of cable connections can significantly reduce the wiring complexity inside the cabinet, making the inside of the cabinet cleaner and more orderly. By reducing the use of cables, the busbar 42 can release more cabinet space, providing more installation space for other components. The design of the busbar 42 makes the management of power connections simpler, facilitating system maintenance and expansion, and reducing the downtime.

[0096] The bus bar 42 provides a more stable power connection, reducing the risk of power failures caused by loose or damaged cable connections, thereby improving the reliability of the system. The bus bar 42 provides a safer way of power connection, reducing the risk of cable overheating or short - circuiting, and enhancing the safety of the entire system. The bus bar 42 can carry higher currents, meeting the power supply requirements of high - power devices, ensuring that the computing nodes 10 and the switching nodes 20 can obtain a stable power supply.

[0097] Moreover, the use of the bus bar 42 supports modular design, enabling the power supply module 40 to be upgraded or replaced independently of other components, enhancing the scalability of the system.

[0098] In some possible implementation manners, the bus bar 42 includes a first bus bar 421 and a second bus bar 422. One side of the first bus bar 421 extends along the second direction y to connect with the power supply module 40, and the other side extends along the first direction x to connect with multiple computing nodes 10 respectively, providing power to the computing nodes 10. The second bus bar 422 extends along the second direction y and is connected to multiple switching nodes 20 respectively, providing power to each switching node 20.

[0099] One side of the first bus bar 421 extends along the second direction y, and the other side extends along the first direction x to connect with the computing nodes 10, that is, the L - shaped first bus bar 421. After being led out from the power supply module 40, it is arranged parallel to the arrangement direction of the multiple computing nodes 10 and connected to the computing nodes 10. The second bus bar 422 extends along the second direction y to connect with the switching nodes 20, that is, the straight - shaped second bus bar 422, and is arranged parallel to the arrangement direction of the multiple switching nodes 20 and connected to the switching nodes 20. This layout makes the power wiring more reasonable and efficient, reducing the situation of cable crossing and chaos. By using dedicated bus bars, the first bus bar 421 and the second bus bar 422, to connect the computing nodes 10 and the switching nodes 20 respectively, the risk of power interruption caused by poor connection or cable failure is reduced, and the overall reliability of the system is improved. The separate first bus bar 421 and second bus bar 422 make fault troubleshooting easier. Technicians can quickly identify and solve power problems of specific nodes, reducing the maintenance time.

[0100] The design of the first bus bar 421 and the second bus bar 422 allows independent management of the power requirements of the computing nodes 10 and the switching nodes 20, providing greater flexibility to adapt to different system configurations and expansion requirements.

[0101] In some possible implementation manners, one end of the first bus bar 421 is connected to the top power supply module 40, extends downward along the second direction y for a certain length, and then extends along the first direction x, thus avoiding the power supply module 40.

[0102] In some possible implementations, the power supply module 40 and the corresponding bus bar 42 are electrically connected through a power receiving clamp, and the computing node 10 or the switching node 20 and the corresponding bus bar 42 are electrically connected through a power receiving clamp.

[0103] The power receiving clamp provides a firm electrical connection, reducing the risk of power interruption due to loose connections or poor contacts, and improving the overall reliability of the system. The design of the power receiving clamp makes the installation and disassembly of the bus bar 42 more convenient and fast. Technicians can more easily install, maintain, and replace the system, reducing the downtime. Using the power receiving clamp connection allows for more flexible configuration and re-wiring, supporting quick adjustments to adapt to different system requirements and configuration changes.

[0104] The power receiving clamp can absorb a certain amount of mechanical stress, reducing the physical stress caused by vibration or thermal expansion, thereby reducing the risk of wear and damage to the connectors and related components. The power receiving clamp provides a safer electrical connection method, reducing the risk of electric arcs and short circuits, and improving the safety of the entire system.

[0105] The use of the power receiving clamp supports modular design, enabling the power supply module 40, the computing node 10, and the switching node 20 to be upgraded or replaced independently of other components, enhancing the scalability of the system. By reducing the complexity of traditional cable connections, the power receiving clamp helps to optimize the use of the internal space of the cabinet, providing more installation space for other components.

[0106] In some possible implementations, along the second direction y, the bus bar 42 corresponding to the computing node 10 is located on the side of the switching node 20 close to the power supply module 40, that is, the first bus bar 421 is located between the switching node 20 and the power supply module 40. Along the first direction x, the bus bar 42 corresponding to the switching node 20 is located on the side of the computing node 10, that is, the second bus bar 422 is located on one side in the width direction of the server cabinet 100.

[0107] The power receiving clamp of the computing node 10 is arranged at the upper part of the high-density switching area to minimize the path of the first bus bar 421, reducing the cost and power supply loss of the first bus bar 421.

[0108] The power receiving clamp of the switching node 20 is arranged at the left part of the high-density switching area, separated from the high-density area.

[0109] This layout makes the power wiring more direct and efficient, reducing the length and complexity of the cables, thereby reducing power loss and signal interference. By reducing the cable length and optimizing the wiring path, the risk of failures due to wiring errors or poor connections is reduced, improving the overall reliability of the system. The reduction in the use of cables can free up more cabinet space, providing more installation space for other components.

[0110] The reasonable layout of the bus bar 42 makes installation and maintenance more convenient. Technicians can more easily access and manage the power connections, simplifying the operation process. By optimizing the power wiring, the air flow inside the cabinet can be better managed, heat accumulation can be reduced, the overall heat dissipation efficiency can be improved, and the equipment can be ensured to operate at the optimal temperature. By optimizing the wiring and reducing cable crossovers, the impact of power failures on the entire system is reduced, and the safety of the system is improved.

[0111] In some possible implementations, along the third direction z, the bus bar 42 is located between the computing node 10 and the switching node 20.

[0112] The bus bar 42 is located between the computing node 10 and the switching node 20, which can shorten the power transmission path, make the power wiring more direct and efficient, and reduce the length and complexity of the cables. By reducing the cable length and optimizing the wiring path, the risk of failures caused by wiring errors or poor connections is reduced, thereby improving the overall reliability of the system. This layout makes installation and maintenance more convenient. Technicians can more easily access and manage the power connections, simplifying the operation process.

[0113] The bus bar 42 is located between the computing node 10 and the switching node 20, reducing the use of cables, freeing up more cabinet space, and providing more installation space for other components. By optimizing the power wiring, the air flow inside the cabinet can be better managed, heat accumulation can be reduced, the overall heat dissipation efficiency can be improved, and the equipment can be ensured to operate at the optimal temperature. This design supports modularity, enabling the computing node 10 and the switching node 20 to be upgraded or replaced independently of other components, enhancing the scalability of the system. This layout allows for flexible configuration and rewiring, supporting quick adjustments to adapt to different system requirements and configuration changes.

[0114] Since both the computing node 10 and the switching node 20 have high power consumption, a heat sink needs to be set in the server cabinet 100 for heat dissipation. When designing an efficient cooling system to support the high-power computing node 10 and switching node 20, adopting a cold plate liquid cooling system is an effective solution.

[0115] Figure 6 For Figure 1 The top view of the shown server cabinet. As Figure 6 shown, in some possible implementations, the server cabinet 100 further includes a cooling module 50. The cooling module 50 includes a water inlet pipeline 51 and a water outlet pipeline 52. When the computing node 10 and the switching node 20 are blindly inserted into the server cabinet 100, the water inlet pipeline 51 is respectively communicated with the computing node 10 and the switching node 20. When the computing node 10 and the switching node 20 are blindly inserted into the server cabinet 100, the water outlet pipeline 52 is respectively communicated with the computing node 10 and the switching node 20.

[0116] After pushing the computing node 10 and the switching node 20 into the cabinet, the water inlet pipeline 51 and the water outlet pipeline 52 are automatically connected to the computing node 10 and the switching node 20, realizing automatic liquid supply cooling.

[0117] By directly introducing the coolant into the computing node 10 and the switching node 20, the cooling module 50 can effectively take away the heat generated by the device, improve the heat dissipation efficiency, ensure that the device operates at the optimal temperature, thereby improving the performance and reliability of the computing node 10 and the switching node 20, and extending their service life. By integrating the cooling module 50, the demand for external cooling equipment can be reduced, the utilization of the internal space of the cabinet can be optimized, and more installation space can be provided for other components. This cooling design can be expanded and adjusted as needed to adapt to different device configurations and heat load requirements, providing greater flexibility.

[0118] Liquid cooling systems are generally more energy-efficient than traditional air cooling systems because liquids have higher heat conduction efficiency, can remove heat more quickly, and reduce the dependence on air conditioning systems. Liquid cooling systems generally produce less noise than air cooling systems because the need for fans is reduced, thus providing a quieter working environment for the data center. Liquid cooling systems can support higher power densities, enabling more computing nodes 10 and switching nodes 20 to be deployed in a limited space to meet the needs of high-performance computing.

[0119] In some possible implementation manners, the water inlet pipeline 51 and the water outlet pipeline 52 are made of corrosion-resistant and highly thermally conductive materials to improve the durability and thermal conduction efficiency of the system.

[0120] In some possible implementation manners, the water inlet pipeline 51 includes a first pipeline 511 and a second pipeline 512, the water outlet pipeline 52 includes a third pipeline 521 and a fourth pipeline 522. Along the second direction y, the server cabinet 100 is provided with a total water inlet 53 and a total water outlet 54. The first pipeline 511 extends along the second direction y, one end of the first pipeline 511 is connected to the total water inlet 53, the other end of the first pipeline 511 is connected to the second pipeline 512, and multiple switching nodes 20 are respectively communicated with the first pipeline 511. The second pipeline 512 extends along the first direction x, and multiple computing nodes 10 are respectively communicated with the second pipeline 512. The third pipeline 521 extends along the first direction x, and multiple computing nodes 10 are respectively communicated with the third pipeline 521. One end of the third pipeline 521 is connected to the fourth pipeline 522, the fourth pipeline 522 extends along the second direction y, and the end of the fourth pipeline 522 away from the third pipeline 521 is connected to the total water outlet 54, and multiple switching nodes 20 are respectively communicated with the fourth pipeline 522. Along the first direction x, the input cable 41 is arranged between the total water inlet 53 and the total water outlet 54.

[0121] By setting the total water inlet 53 and the total water outlet 54 at the top of the cabinet, the vertical space can be better utilized, avoiding occupying the horizontal space inside the cabinet, thus providing more installation space for other components. The design of connecting the total water inlet 53 and the total water outlet 54 at the top makes the pipeline layout more concise and direct, reducing the bending and crossing of pipelines, and lowering the installation complexity and material costs. Setting the total water inlet 53 and the total water outlet 54 at the top also facilitates the connection, inspection and maintenance of pipelines by technicians, reducing the operation difficulty and maintenance time. This design supports modularization and flexible configuration, enabling the cabinet to be more easily connected to different types of CDUs to adapt to different cooling requirements. By setting the inlet and outlet of the coolant at the top, the contact between the coolant and the hot air inside the cabinet can also be reduced, lowering the thermal interference and improving the cooling effect.

[0122] The extended design of the first pipeline 511, the second pipeline 512, the third pipeline 521 and the fourth pipeline 522 reduces the length and bending of the pipelines, which can reduce the fluid resistance, improve the flow efficiency of the coolant, and thus improve the overall cooling efficiency. Since the number and complexity of pipeline connections are reduced, the leakage risk is lowered and the reliability of the system is improved.

[0123] By designing independent inlet and outlet paths for the computing node 10 and the switching node 20 respectively, the flow rate and temperature of the coolant can be more precisely controlled, improving the overall cooling efficiency. The independent cooling paths reduce the heat interference between different nodes, ensuring that each node can obtain the required cooling effect, thereby improving the reliability of the system. The independent pipeline design allows for flexible adjustment of the cooling configuration to adapt to different equipment configurations and heat load requirements, providing greater flexibility and scalability. The independent cooling path design reduces the heat cross-interference between the computing node 10 and the switching node 20, ensuring that each node can operate at the optimal temperature.

[0124] The layout of the first pipeline 511, the second pipeline 512, the third pipeline 521 and the fourth pipeline 522 makes the maintenance and management of the cooling system simpler. Technicians can more easily identify and solve the cooling problems of specific nodes, reducing the maintenance time. By optimizing the pipeline layout, the internal space of the cabinet can be more effectively utilized, providing more installation space for other components.

[0125] In some possible implementation manners, at the total water inlet 53 and the total water outlet 54 at the top of the cabinet, high-efficiency sealing and insulating materials are used for sealing to prevent water leakage and heat loss.

[0126] In some possible implementation manners, the first pipeline 511 and the second pipeline 512 are in an L-shaped layout, and the third pipeline 521 and the fourth pipeline 522 are in an L-shaped layout, which can connect the horizontally placed switching node 20 and the vertically placed computing node 10 at the same time.

[0127] Figure 8 is Figure 7 a schematic structural diagram of the water inlet pipeline of the server cabinet shown. As Figure 8 shown, in some possible implementation manners, the first pipeline 511 is connected to the switching node 20 through the first joint 513, the second pipeline 512 is connected to the computing node 10 through the second joint 514, and the diameter of the first joint 513 is smaller than that of the second joint 514. The third pipeline 521 is connected to the computing node 10 through the third joint, and the fourth pipeline 522 is connected to the switching node 20 through the fourth joint, and the diameter of the fourth joint is smaller than that of the third joint.

[0128] By using the first joint 511 and the fourth joint with a smaller diameter to connect the switching node 20, the flow rate of the coolant at the switching node 20 can be restricted, while the second joint 512 and the third joint with a larger diameter allow more coolant to flow into the computing node 10. This design helps to optimize the distribution of the coolant according to the heat dissipation requirements of different nodes. The computing node 10 usually generates more heat than the switching node 20, so it requires higher cooling capacity. The larger diameters of the second joint 514 and the third joint allow more coolant to pass through, thereby improving the cooling efficiency of the computing node 10. Ensuring that the computing node 10 obtains sufficient cooling capacity helps to maintain its high-performance operation and meet the requirements of high-performance computing. By reducing the excessive use of the coolant and optimizing the flow distribution, the maintenance cost of the cooling system is reduced.

[0129] By precisely controlling the flow rate of the coolant, unnecessary coolant consumption can be reduced, thereby reducing the energy consumption of the entire cooling system. The optimized coolant distribution reduces the risk of overcooling or overheating, ensuring that each node can operate at an appropriate temperature, thereby improving the reliability of the system. The joint design with different diameters can be customized according to the specific requirements of the nodes, simplifying the design and installation process of the cooling system. This design allows for flexible adjustment of the cooling configuration to adapt to different equipment configurations and heat load requirements, providing greater flexibility.

[0130] In some possible implementation manners, the first joint 513 and the second joint 514 use quick connectors with adjustable diameters to adjust the flow rate and water pressure according to actual needs, ensuring a balanced distribution of the coolant between the computing node 10 and the switching node 20.

[0131] In some possible implementation manners, the first joint 513 and the second joint 514 can also select joints with a quick-disconnect function, which is convenient for maintenance and component replacement and reduces the downtime.

[0132] In some possible implementations, flow regulating valves or flow meters can also be added to the first pipeline 511, the second pipeline 512, the third pipeline 521, and the fourth pipeline 522 to monitor and adjust the flow rate of each node, ensuring the stability of the system; a pressure regulating device can also be added to prevent pressure instability caused by flow conflicts.

[0133] In some possible implementations, an intelligent monitoring system can also be set up in the server cabinet 100, and various sensors can be connected to monitor the temperature, flow rate, pressure, etc. of the coolant in real time, providing data support for optimizing the cooling efficiency and troubleshooting. A pump group, a heat exchanger, etc. can also be set up in the water inlet pipeline 51. By optimizing the flow rate and temperature of the coolant and combining dynamic adjustment strategies, energy consumption can be reduced and the overall energy efficiency of the system can be improved.

[0134] In some possible implementations, along the third direction z, the water inlet pipeline 51 and the water outlet pipeline 52 are located between the computing node 10 and the switching node 20. Along the first direction x, the first pipeline 511 and the fourth pipeline 522 are respectively located on both sides of the computing node 10. Along the second direction y, the second pipeline 512 and the third pipeline 521 are arranged on the side of the plurality of computing nodes 10 and the plurality of switching nodes 20 away from the power supply module 40.

[0135] By arranging the water inlet pipeline 51 and the water outlet pipeline 52 between the computing node 10 and the switching node 20, the flow of the coolant can be managed more directly, ensuring that heat can be quickly removed and improving the cooling efficiency. This layout helps to achieve a more uniform temperature distribution between the computing node 10 and the switching node 20, reduce the formation of hot spots, and ensure that the equipment operates at the optimal temperature.

[0136] Concentrating the water inlet pipeline 51 and the water outlet pipeline 52 between the computing node 10 and the switching node 20, and distributing the first pipeline 511 and the fourth pipeline 522 on both sides of the computing node 10 simplifies the wiring and installation process of the cooling system, making the system cleaner and easier to manage. This cooling design can support a higher power density, enabling more computing nodes 10 and switching nodes 20 to be deployed in a limited space to meet the needs of high-performance computing. By optimizing the flow path of the coolant, the risk of equipment failure caused by overheating is reduced, thereby improving the overall reliability of the system. This design allows for flexible adjustment of the cooling configuration to adapt to different equipment configurations and heat load requirements, providing greater flexibility and scalability. By optimizing the pipeline layout, the internal space of the cabinet can be more effectively utilized, providing more installation space for other components.

[0137] The symmetrical design of the first pipeline 511 and the fourth pipeline 522 ensures the symmetrical layout of the water inlet and outlet pipelines, so as to evenly distribute the coolant and reduce problems such as pressure loss and uneven flow rate.

[0138] In some possible implementations, along the second direction y, the first bus bar 421 and the second pipeline 512 and the third pipeline 521 on the computing node 10 are respectively disposed on the upper and lower sides.

[0139] By disposing the first bus bar 421 and the cooling pipeline on the upper and lower sides respectively, the force during plugging and unplugging can be effectively balanced, preventing physical stress and potential damage caused by uneven unilateral force, and helping to extend the service life of the connector and related components. The balanced force distribution reduces the system instability caused by the plugging and unplugging operations, ensuring that the computing node 10 remains stable during installation and maintenance. Moreover, this layout makes it easier and more intuitive for technicians to perform the plugging and unplugging operations, reducing the possibility of operation errors and improving the maintenance efficiency. By reducing physical stress and optimizing the component layout, the risk of failure caused by mechanical stress can be reduced, improving the reliability of the system.

[0140] At the same time, disposing the cooling pipeline and the first bus bar 421 on the upper and lower sides respectively helps to better organize the internal space of the cabinet, avoid mutual interference between components, and improve the overall space utilization rate. This layout can also optimize the flow path of the coolant, ensuring that the coolant can more effectively cover the key components of the computing node 10 and improving the heat dissipation efficiency. This design also supports modularization, enabling the computing node 10 to be upgraded or replaced independently of other components, enhancing the scalability of the system.

[0141] The above has introduced in detail a server cabinet provided by the present application. Specific examples are used in this article to elaborate on the principle and implementation manner of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application. It should be noted that for those of ordinary skill in the art in this technical field, without departing from the principle of the present application, several improvements and modifications can still be made to the present application, and these improvements and modifications also fall within the protection scope of the claims of the present application.

Claims

1. A server cabinet, characterized in that, It includes multiple computing nodes and multiple switching nodes. The multiple computing nodes are arranged along a first direction, and the multiple switching nodes are arranged along a second direction. Each of the computing nodes is provided with multiple first connectors in a third direction, and the multiple first connectors on each computing node are arranged along the second direction. Each of the switching nodes is provided with multiple second connectors in the third direction, and the multiple second connectors on each switching node are arranged along the first direction. The first connectors are correspondingly connected to the second connectors so that when each computing node and each switching node are blindly inserted into the server cabinet, they are orthogonally connected. The first direction, the second direction, and the third direction are perpendicular to each other.

2. The server cabinet according to claim 1, wherein, The server cabinet further includes a management switch. The management switch and the multiple switching nodes are arranged along the second direction. Each of the computing nodes is further provided with a third connector in the third direction, and the third connectors on each computing node and the multiple first connectors are arranged along the second direction. The management switch is provided with multiple fourth connectors in the third direction, and the multiple fourth connectors are arranged along the first direction. The multiple fourth connectors are correspondingly connected to the third connectors on the multiple computing nodes one by one so that the management switch is orthogonally connected to each computing node.

3. The server cabinet according to claim 2, characterized in that, Along the second direction, the management switch is arranged in the middle of the multiple switching nodes.

4. The server cabinet according to claim 2, characterized in that, The computing node includes multiple first processors and multiple in-board connector groups for the first processors. Each in-board connector group for the first processors corresponds to one of the first processors. Each in-board connector group for the first processors includes multiple in-board connectors for the first processors. The first processor is respectively connected to the multiple in-board connectors for the first processors in the corresponding in-board connector group for the first processors through in-board traces. Each of the first connectors respectively corresponds to one of the in-board connectors for the first processors in the multiple in-board connector groups for the first processors. A certain in-board connector for the first processors in the in-board connector group for the first processors corresponding to each first processor and the in-board connectors for the first processors with the same serial number in the in-board connector groups for the first processors corresponding to other first processors are connected to the same first connector through a cable.

5. The server cabinet according to claim 4, characterized in that, The switching node includes multiple second processors and multiple in-board connector groups for the second processors. Each in-board connector group for the second processors corresponds to one of the second processors. Each in-board connector group for the second processors includes multiple in-board connectors for the second processors. The second processor is respectively connected to the multiple in-board connectors for the second processors in the corresponding in-board connector group for the second processors through in-board traces. Each of the second connectors respectively corresponds to one of the in-board connectors for the second processors in the multiple in-board connector groups for the second processors. A certain in-board connector for the second processors in the in-board connector group for the second processors corresponding to each second processor and the in-board connectors for the second processors with the same serial number in the in-board connector groups for the second processors corresponding to other second processors are connected to the same second connector through a cable.

6. The server cabinet according to claim 5, characterized in that, The server cabinet further includes a plurality of output ports. Along the third direction, the output ports are arranged on a side of the switching node away from the computing node, and the output ports are connected to the second processor via intra-board wiring.

7. The server cabinet according to claim 1, wherein, The first connector is movably connected to the computing node so as to be floatingly connected to a panel of the computing node, and the second connector is movably connected to the switching node so as to be floatingly connected to a panel of the switching node.

8. The server cabinet according to claim 1, characterized in that, The server cabinet also includes a bracket, and along the third direction, the bracket is located between the computing node and the switching node, a plurality of first guide columns are provided on the side of the bracket facing the computing node, a corresponding first guide hole is provided on the side of the panel of the computing node facing the bracket, and the first guide column is docked with the first guide hole to enable the plurality of computing nodes to be blindly inserted into the server cabinet, a plurality of second guide columns are provided on the side of the bracket facing the switching node, a corresponding second guide hole is provided on the side of the panel of the switching node facing the bracket, and the second guide column is docked with the second guide hole to enable the plurality of switching nodes to be blindly inserted into the server cabinet, thereby enabling the first connector to be correspondingly connected with the second connector.

9. The server cabinet according to claim 8, characterized in that, The server cabinet also includes at least two power supply modules, and at least two of the power supply modules are arranged along the third direction. When the computing node is blindly plugged into the server cabinet, at least one of the power supply modules is electrically connected to the computing node, and when the switching node is blindly plugged into the server cabinet, at least one of the power supply modules is electrically connected to the switching node.

10. The server cabinet according to claim 9, characterized in that, Along the second direction, the power supply module is arranged on one side of the plurality of computing nodes and the plurality of switching nodes.

11. The server cabinet according to claim 9, wherein, The server cabinet also includes at least two input cables. Along the third direction, the input cables are arranged in the middle of at least two of the power supply modules and are connected to the power supply modules one-to-one. Along the second direction, the input cables are arranged on a side of the power supply module away from the computing node and the switching node.

12. The server cabinet according to claim 9, wherein, The server cabinet further includes at least two bus bars, and the power supply module is electrically connected to the computing node or the switching node via the corresponding bus bars.

13. The server cabinet according to claim 12, wherein, The bus includes a first bus and a second bus. One side of the first bus extends along the second direction to be connected to the power supply module, and the other side of the first bus extends along the first direction to be respectively connected to multiple computing nodes. The second bus extends along the second direction and is respectively connected to multiple switching nodes.

14. The server cabinet according to claim 12, characterized in that, The power supply module is electrically connected to the corresponding busbar via a power clamp, and the computing node or the switching node is electrically connected to the corresponding busbar via a power clamp.

15. The server cabinet according to claim 12, characterized in that, Along the second direction, the bus correspondingly connected to the computing node is located on a side of the switching node close to the power supply module, and along the first direction, the bus correspondingly connected to the switching node is located on one side of the computing node.

16. The server cabinet according to claim 12, wherein There are two power supply modules. Along the third direction, the bus bar is located between the computing node and the switching node, and the bus bar is located between the two power supply modules.

17. The server cabinet according to claim 11, characterized in that, The server cabinet further includes a cooling module. The cooling module includes a water inlet pipeline and a water outlet pipeline. When the computing node and the switching node are blindly inserted into the server cabinet, the water inlet pipeline is respectively communicated with the computing node and the switching node, and the water outlet pipeline is respectively communicated with the computing node and the switching node.

18. The server cabinet according to claim 17, characterized in that, The water inlet pipeline includes a first pipeline and a second pipeline. The water outlet pipeline includes a third pipeline and a fourth pipeline. Along the second direction, the server cabinet is provided with a total water inlet and a total water outlet. The first pipeline extends along the second direction. One end of the first pipeline is connected to the total water inlet, and the other end of the first pipeline is connected to the second pipeline. A plurality of switching nodes are respectively communicated with the first pipeline. The second pipeline extends along the first direction. A plurality of computing nodes are respectively communicated with the second pipeline. The third pipeline extends along the first direction. A plurality of computing nodes are respectively communicated with the third pipeline. One end of the third pipeline is connected to the fourth pipeline. The fourth pipeline extends along the second direction. The end of the fourth pipeline far from the third pipeline is connected to the total water outlet. A plurality of switching nodes are respectively communicated with the fourth pipeline. Along the first direction, the input cable is arranged between the total water inlet and the total water outlet.

19. The server cabinet according to claim 18, characterized in that, The first pipeline is communicated with the switching node through a first joint. The second pipeline is communicated with the computing node through a second joint. The diameter of the first joint is smaller than that of the second joint. The third pipeline is communicated with the computing node through a third joint. The fourth pipeline is communicated with the switching node through a fourth joint. The diameter of the fourth joint is smaller than that of the third joint.

20. The server cabinet according to claim 18, wherein, Along the third direction, the water inlet pipeline and the water outlet pipeline are located between the computing node and the switching node. Along the first direction, the first pipeline and the fourth pipeline are respectively located on both sides of the computing node. Along the second direction, the second pipeline and the third pipeline are arranged on one side of the plurality of computing nodes and the plurality of switching nodes away from the power supply module.

Images