Data communication method and apparatus for network
By employing a heterogeneous network architecture in Scale Up networks and utilizing a combination of direct and switched links, large-scale interconnection of computing chips is achieved, reducing the cost of network chips and data transmission latency, making it suitable for artificial intelligence and high-performance computing scenarios.
Patent Information
- Application Number
- PCT/CN2026/085829
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-12
- Filing Date
- 2026-03-25
- Publication Date
- 2026-11-19
AI Technical Summary
In existing scale-up network architectures, the interconnection scale of computing chips is limited, and network chips are expensive and have high communication latency, making it difficult to achieve low-cost and low-latency data transmission for large-scale networking.
By adopting a heterogeneous network architecture, direct links are established within servers and switching links are established between servers. By utilizing the heterogeneous connection between network chipsets and computing chips, efficient data transmission between computing chips is achieved, reducing port consumption of network chips and data forwarding latency.
It supports interconnection of more computing chips, reduces the cost of network chips and data transmission latency, and is suitable for intensive computing and low-latency scenarios, such as artificial intelligence and high-performance computing.
Smart Images

Figure CN2026085829_19112026_PF_FP_ABST
Abstract
Description
Data communication method and apparatus for networks
[0001] This application claims priority to Chinese Patent Application No. 202510608818.X, filed on May 12, 2025, entitled “Data Communication Method and Apparatus for Networks”, the entire contents of which are incorporated herein by reference. Technical Field
[0002] This application relates to the field of communication technology, and in particular to a data communication method and apparatus for networks. Background Technology
[0003] Scale-up networking is a technique that integrates multiple computing chips (such as graphics processing units, GPUs) into a single node using high-speed connectivity to improve computing performance and resource utilization efficiency. Scale-up networks can enhance performance by increasing the hardware resources within a single node.
[0004] In related technologies, the networking methods of scale-up networks include: a single-layer Clos networking method that uses a single-level network chip for networking, and a two-layer Clos networking method that uses two-level network chips.
[0005] However, the number of interconnected computing chips in a single-layer Cloos networking method is related to the number of ports that the network chip can support, thus limiting the scale of interconnected computing chips. In a two-layer Cloos networking method, communication from one computing chip to another requires two levels of network chips, which are more expensive and result in greater communication latency. Summary of the Invention
[0006] This application provides a data communication method and apparatus for networks, which can achieve low-cost and low-latency data transmission while realizing large-scale networking.
[0007] In a first aspect, this application provides a data communication method for a network, applied to a network controller. The network includes multiple servers and multiple network chipsets. Each server includes multiple computing chips, and a direct link is established between every two computing chips in each server. Each network chip in the network chipsets establishes an exchange link with at least one computing chip in each server. The method includes: determining data to be sent from a source computing chip in a source server to a destination computing chip in a destination server; if a first network chip has established exchange links with both the source and destination computing chips, then determining that data is sent to the destination computing chip via the first network chip; if no network chip has established exchange links with both the source and destination computing chips, then determining that data is sent to the destination computing chip via the first and second computing chips in the source server, or determining that data is sent to the destination computing chip via a third network chip and a second computing chip in the destination server; wherein the second network chip has established exchange links with both the destination and first computing chips simultaneously, and the third network chip has established exchange links with both the source and second computing chips simultaneously.
[0008] The network in this embodiment is a heterogeneous network consisting of direct links inside the server and switching links outside the server. Compared with related technologies, it can support a number of interconnected computing chips that are increased many times over. Moreover, each computing chip port only needs to consume one network chip port, and each data forwarding requires at most one computing chip and one network chip. Compared with related technologies, it reduces the cost of network chips and the latency of data transmission, thereby achieving low cost and low latency while realizing large-scale networking. It is suitable for scenarios that require intensive computing and low latency (such as artificial intelligence (AI) training scenarios and high performance computing (HPC) scenarios).
[0009] In one possible implementation, the identifiers of the computing chips in any two servers are identical, and the identifiers of at least one computing chip in different servers connected to each network chip are identical; the identifier of the first computing chip is the same as that of the destination computing chip, and the identifier of the second computing chip is the same as that of the source computing chip.
[0010] For example, the one-to-one correspondence of the identifiers of computing chips in any two servers means that the number of computing chips in any two servers is the same, and each computing chip in one server has the same identifier as the corresponding computing chip in the other server. For example, each server may have M computing chips, and the identifiers may be from 1 to M.
[0011] The requirement that at least one computing chip in different servers connected to a network chip has the same identifier means that: if a network chip establishes a switching link with one computing chip in each server, then the identifiers of all computing chips connected to the network chip are the same. If a network chip establishes switching links with multiple computing chips in each server, then the number of computing chips connected to the network chip in different servers is the same, and each computing chip connected to the network chip in one server has the same identifier as the corresponding computing chip connected to it in another server.
[0012] In one possible implementation, the identifiers of at least one computing chip connected to each network chip in a network chipset are identical, while the identifiers of the computing chips connected to any two network chipsets are different.
[0013] In one possible implementation, the computing chip is an AI chip.
[0014] Secondly, this application provides a data communication method for a network, applied to a network including multiple servers and multiple network chipsets. Each of the multiple servers includes multiple computing chips, and a direct link is established between every two computing chips in each server. Each network chip in the network chipsets establishes an exchange link with at least one computing chip in each server. The method includes: a source computing chip in a source server of the communication network sending data to a first network chip that has established exchange links with both the source computing chip and the destination computing chip; the first network chip sending data to a destination computing chip in a destination server of the communication network; or, the source computing chip sending data to the first computing chip in the source server; the first computing chip sending data to a second network chip that has established exchange links with both the destination computing chip and the first computing chip; the second network chip sending data to the destination computing chip; or, the source computing chip sending data to a third network chip that has established exchange links with both the source computing chip and the second computing chip in the destination server; the third network chip sending data to the second computing chip; and the second computing chip sending data to the destination computing chip.
[0015] In one possible implementation, the identifiers of the computing chips in any two servers are identical, and the identifiers of at least one computing chip in different servers connected to each network chip are identical; the identifier of the first computing chip is the same as that of the destination computing chip, and the identifier of the second computing chip is the same as that of the source computing chip.
[0016] In one possible implementation, the identifiers of at least one computing chip connected to each network chip in a network chipset are identical, while the identifiers of the computing chips connected to any two network chipsets are different.
[0017] In one possible implementation, the computing chip is an AI chip.
[0018] Thirdly, this application provides a data communication device for a network, applied to a controller of the network, the network including multiple servers and multiple network chipsets, each of the multiple servers including multiple computing chips, each pair of computing chips in each server establishing a direct link, and each network chip in the network chipsets establishing an exchange link with at least one computing chip in each server; the device includes: a determining module, configured to determine data to be sent from a source computing chip in a source server to a destination computing chip in a destination server; a processing module, configured to determine, if a first network chip has established an exchange link with both the source computing chip and the destination computing chip, to send data to the destination computing chip via the first network chip; the processing module is further configured to determine, if no network chip has established an exchange link with both the source computing chip and the destination computing chip, to send data to the destination computing chip via the first computing chip and the second network chip in the source server, or to determine to send data to the destination computing chip via a third network chip and the second computing chip in the destination server; wherein the second network chip has established an exchange link with both the destination computing chip and the first computing chip, and the third network chip has established an exchange link with both the source computing chip and the second computing chip.
[0019] In one possible implementation, the identifiers of the computing chips in any two servers are identical, and the identifiers of at least one computing chip in different servers connected to each network chip are identical; the identifier of the first computing chip is the same as that of the destination computing chip, and the identifier of the second computing chip is the same as that of the source computing chip.
[0020] In one possible implementation, the identifiers of at least one computing chip connected to each network chip in a network chipset are identical, while the identifiers of the computing chips connected to any two network chipsets are different.
[0021] In one possible implementation, the computing chip is an AI chip.
[0022] Fourthly, this application provides an electronic device comprising: one or more processors; a memory for storing one or more computer programs or instructions; wherein, when the one or more computer programs or instructions are executed by the one or more processors, the one or more processors implement the method as described in any one of the first aspects, or the one or more processors implement the method as described in any one of the second aspects.
[0023] Fifthly, this application provides an electronic device, including a processor for performing the method as described in any one of the first aspects, or for performing the method as described in any one of the second aspects.
[0024] In a sixth aspect, this application provides a communication device, the device comprising: a processing circuit and an interface circuit; wherein the interface circuit is used to couple with a memory external to the communication device and to provide a communication interface for the processing circuit to access the memory; the processing circuit is used to execute program instructions in the memory to implement the method as described in any one of the first aspects, or to implement the method as described in any one of the second aspects.
[0025] In practical implementation, the communication device can be a chip, the input circuit can be an input pin, the output circuit can be an output pin, and the processing circuit can be a transistor, gate circuit, flip-flop, and various logic circuits. The input signal received by the input circuit can be received and input by, for example, but not limited to, a receiver, and the signal output by the output circuit can be output to, for example, but not limited to, a transmitter and transmitted by the transmitter. Furthermore, the input circuit and the output circuit can be the same circuit, which is used as the input circuit and the output circuit at different times. This application does not limit the specific implementation of the processor and various circuits.
[0026] In one implementation, the communication device can be a wireless communication device, i.e., a computer device that supports wireless communication functions. Specifically, the wireless communication device can be a terminal such as a smartphone, or a wireless access network device such as a base station. The network chip can also be called a system-on-a-chip (SoC), or simply a SoC chip. The communication chip may include a baseband processing chip and a radio frequency (RF) processing chip. The baseband processing chip is sometimes also called a modem or baseband chip. The RF processing chip is sometimes called an RF transceiver or RF chip. In physical implementation, some or all of the chips in the communication chip can be integrated within the SoC chip. For example, the baseband processing chip is integrated into the SoC chip, while the RF processing chip is not integrated with the SoC chip. The interface circuit can be the RF processing chip in the wireless communication device, and the processing circuit can be the baseband processing chip in the wireless communication device.
[0027] In another implementation, the communication device can be a component of a wireless communication device, such as an integrated circuit product like a network chip or communication chip. The interface circuit can be an input / output interface, interface circuit, output circuit, input circuit, pin, or related circuit on the chip or chip network. The processor can also be represented as a processing circuit or logic circuit.
[0028] In a seventh aspect, this application provides a computer-readable storage medium storing program code, which, when executed by a processor, implements the method as described in any one of the first aspects, or implements the method as described in any one of the second aspects.
[0029] Eighthly, this application provides a chip comprising: at least one processor. The at least one processor is configured to perform the method as described in any one of the first aspects, or to perform the method as described in any one of the second aspects.
[0030] Optionally, the chip also includes memory. At least one processor is configured to execute code in the memory, which, when executed, causes the chip to implement the method as described in any one of the first aspects, or to implement the method as described in any one of the second aspects.
[0031] Alternatively, the chip described above can also be an integrated circuit.
[0032] Ninthly, this application provides a computer program product containing instructions that, when run on a computer, cause the computer to implement the method as described in any one of the first aspects, or to implement the method as described in any one of the second aspects. Attached Figure Description
[0033] Figure 1 is a schematic diagram of the structure of a communication network provided in an embodiment of this application.
[0034] Figure 2 is a schematic diagram of another communication network provided in an embodiment of this application.
[0035] Figure 3 is a schematic diagram of another communication network provided in an embodiment of this application.
[0036] Figure 4 is a schematic diagram of the structure of a computing chip provided in an embodiment of this application.
[0037] Figure 5 is a schematic diagram of another communication network provided in an embodiment of this application.
[0038] Figure 6 is a flowchart illustrating a data communication method for a network provided in an embodiment of this application.
[0039] Figure 7 is a flowchart illustrating another data communication method for a network provided in an embodiment of this application.
[0040] Figure 8 is a flowchart illustrating another data communication method for a network provided in an embodiment of this application.
[0041] Figure 9 is a flowchart illustrating another data communication method for a network provided in an embodiment of this application.
[0042] Figure 10 is a block diagram of a data communication device for a network provided in an embodiment of the application.
[0043] Figure 11 is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0044] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions 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, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0045] The terms "first," "second," etc., used in the specification, embodiments, claims, and drawings of this application are for distinguishing purposes only and should not be construed as indicating or implying relative importance or order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as including a series of steps or units. A method, system, product, or apparatus is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or apparatuses.
[0046] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.
[0047] This application provides a communication network capable of simultaneously achieving large-scale interconnection, low cost, and low latency. Supported protocols include, but are not limited to: NVLINK, Infinity Fabric, CXL (Compute Express Link), the high-speed serial computer expansion bus standard (Peripheral Component Interconnect Express, PCIe), UB, UALink, ETH-X, and various protocols based on Ethernet (such as ULN of the Ultra Ethernet Consortium (UEC) protocol).
[0048] The communication network includes multiple servers (also referred to as nodes) and multiple network chipsets. Each server includes multiple computing chips, and a direct link (also referred to as an intra-node link) is established between every two computing chips in each server. Each network chipset includes at least one network chip, and each network chip in each network chipset establishes a switching link (also referred to as an external link) with at least one computing chip in each server. As can be seen from the architecture of the communication network provided in this application embodiment, the number of servers is related to the number of ports of the network chips.
[0049] The computing chip may include an AI chip. For example, computing chips include, but are not limited to, GPUs, central processing units (CPUs), tensor processing units (TPUs), or neural processing units (NPUs). The network chip may include a switching chip (also called a switch (SW)).
[0050] Direct links include, but are not limited to: full mesh direct links, 2D-Torus direct links, 3D-Torus direct links, or cube direct links. For example, suppose a computing chip inside a server has 28 112 gigabits per second (G) serial deserializers (SERDES). These 28 112G SERDES can be divided into seven 4-lane networks. This computing chip is connected to seven other computing chips via four lanes each, with a bandwidth of 4 * 112G = 448G between each pair of computing chips.
[0051] Switching links can be connected using optical modules. Optical modules include, but are not limited to, on-board optics (OBO), co-packaged optics (CPO), or pluggable optical modules.
[0052] For example, please refer to Figure 1, which is a schematic diagram of a communication network structure provided in an embodiment of this application. The communication network includes N servers (server 1 to server N) and L network chipsets (network chipset 1 to network chipset L). Each server includes M computing chips (computing chip 1 to computing chip M), and network chipset i includes K i There are network chips, N>1, M>1, 0≤i≤L, K i ≥1. Each server has M interconnected computing chips. Figure 1 illustrates an example where each network chip establishes a switching link with one computing chip in each server. L can be 4, 7, 8, 9, 13, or 15, etc., and M can be 4, 5, 6, 8, 12, 15, or 16, etc. This application embodiment does not limit the values of L and M. Figure 1 exemplarily shows the switching link of one network chip in each network chipset. The switching links of other network chips can be referred to Figure 1, and will not be elaborated upon here.
[0053] In one possible implementation, multiple computing chips in each server are identified, and the identifiers of computing chips in any two servers are identical. At least one computing chip in different servers connected to the network chip is also identical in its identifier.
[0054] For example, the one-to-one correspondence of the identifiers of computing chips in any two servers means that the number of computing chips in any two servers is the same, and each computing chip in one server has the same identifier as the corresponding computing chip in the other server. For example, each server may have M computing chips, and the identifiers may be from 1 to M.
[0055] The requirement that at least one computing chip in different servers connected to a network chip has the same identifier means that if the network chip establishes a switching link with one computing chip in each server, then the identifiers of all the computing chips connected to the network chip are the same. If the network chip establishes switching links with multiple computing chips in each server, then the number of computing chips connected to the network chip in different servers is the same, and each computing chip connected to the network chip in one server has the same identifier as the corresponding computing chip connected to it in another server. For example, if each server has M computing chips, and their identifiers are all from 1 to M, the network chip can connect to computing chip 2 in each server, or it can connect to computing chip 3, computing chip 100, and computing chip 200 in each server, etc.
[0056] In one possible implementation, the identifiers of at least one computing chip connected to each network chip in a network chipset are identical, and the identifiers of computing chips connected to network chips in any two network chipsets are different. In this implementation, the number of network chipsets is the same as the number of computing chips included in the server, and the number of network chips in a network chipset is related to the number of ports of the computing chips. In this embodiment, the establishment of a switching link between each network chip in a network chipset and computing chips with the same identifier in multiple servers can be referred to as a rail. Then, the chips within a server belong to different rails, the network chips in a network chipset belong to the same rail, and the network chips in different network chipsets belong to different rails. The number of rails is the same as the number of computing chips included in the server.
[0057] For example, please refer to Figure 2, which is a schematic diagram of another communication network structure provided in an embodiment of this application. The communication network includes N servers (server 1 to server N), and each server includes 8 computing chips (computing chip 1 to computing chip 8). Figure 2 takes an example where a network chip establishes a switching link with one computing chip in each server, and the identifiers of the computing chips connected to each network chip in the network chip group are the same, while the identifiers of the computing chips connected to any two network chipsets are different. In this case, the network includes 8 network chip groups (network chip group 1 to network chip group 8), and each network chip group includes K network chips, where K > 0.
[0058] As shown in Figure 2, the eight computing chips in each server are interconnected in pairs. Each of the K network chips in the network chipset establishes a switching link with the same-identified computing chips in the N servers, forming eight rails (rails 1 to 8). Rail 1: Each network chip in network chipset 1 establishes a switching link with computing chip 1 in the N servers; Rail 2: Each network chip in network chipset 2 establishes a switching link with computing chip 2 in the N servers; ...; Rail 8: Each network chip in network chipset 8 establishes a switching link with computing chip 8 in the N servers.
[0059] The following is a specific example to further illustrate the communication network provided in this application embodiment. For example, please refer to Figure 3, which is a schematic diagram of another communication network structure provided in this application embodiment. In this example, each network chip has 512 112G channels, divided into 256 2-channel switching ports. Each switching port consists of two 112G channels, resulting in an effective bandwidth of 224G. Since the network chip has 256 switching ports, it can connect to 256 computing chips, that is, it can support 256 servers.
[0060] Each server comprises 8 computing chips, resulting in a total of 256 * 8 = 2048 computing chips in the communication network. These 8 computing chips belong to different rails and connect to different network chipsets, thus the network comprises 8 network chipsets. Each computing chip in a server has 28 internal 112G channels, connected to the other 7 computing chips in the server via 4 of these internal channels (direct links). Each computing chip has 32 external 112G channels, divided into 16 two-channel ports. Therefore, each network chipset comprises 16 network chips, all belonging to the same rail. The 16 ports of each computing chip connect to the 16 network chips within a single network chipset (switching links). The total number of network chips in the network is 16 * 8 = 128.
[0061] In this application embodiment, the architecture of the computing chip includes, but is not limited to, single-master die (wafer), dual-master die, or chiplet packaging architecture. In single-master die and dual-master die architectures, the direct connection links and switching links of the computing chip are directly output from the computing chip, without the need for an external data processing unit (DPU) or remote network interface card (RNIC). In the chiplet packaging architecture, the direct connection links and switching links of the computing chip can be directly output through the input / output (IO) die.
[0062] For example, please refer to Figure 4, which is a schematic diagram of a computing chip structure provided in an embodiment of this application. Figures 4(a), 4(b), and 4(c) respectively illustrate a single-main die, a dual-main die, and a chiplet packaging architecture. As shown in Figure 4(a), the computing chip includes high bandwidth memory (HBM) and one die, and the direct connection links and switching links of the computing chip are directly output from the computing chip. As shown in Figure 4(b), the computing chip includes HBM and two dies, and the direct connection links and switching links of the computing chip are directly output from the computing chip. As shown in Figure 4(c), the computing chip includes HBM and an IO die, and the direct connection links and switching links of the computing chip are directly output from the IO die.
[0063] For example, the communication network provided in this application embodiment can be a scale-up network, such as an AI scale-up network.
[0064] In one possible implementation, the communication network may further include a controller, which can be used to determine the data transmission path within the communication network. The controller may include: a CPU, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a neural processing unit (NPU), or a digital signal processor (DSP), etc. There can be at least one controller. When there are multiple controllers, communication connections are established between them.
[0065] For example, please refer to Figure 5, which is a schematic diagram of another communication network structure provided in an embodiment of this application. Figure 5 uses the architecture shown in Figure 2 as an example, with the controller being a CPU located inside a server. Each server includes two controllers, and all controllers establish connections with each other. Each CPU includes an operating system, on which control software can run. The CPU can use the control software to send its respective routing information to each computing chip and network chip in the communication network, and data is transmitted on each data link according to the routing information. The routing information can also be synchronized between the controllers.
[0066] For example, the routing information of a computing chip may include multiple routing entries, each of which may include, but is not limited to, a destination address, a next-hop address (e.g., output port information), etc. The computing chip or network chip can query its respective routing table based on the destination address of the data to determine the next hop and send the data to the next hop. If the computing chip determines that the destination address of the data is itself, it does not need to send the data.
[0067] In the communication network provided in this application embodiment, computing chips within the same server communicate via direct links. Computing chips across servers communicate via switched links, or via a combination of switched links and direct links.
[0068] The following describes a data communication method for a network provided in an embodiment of this application. For example, please refer to FIG6, which is a flowchart illustrating a data communication method for a network provided in an embodiment of this application. This method can be applied to a network controller. The network structure can be referred to the foregoing description. As shown in FIG6, the method may include the following processes:
[0069] 101. Determine the data sent from the source computing chip in the source server to the destination computing chip in the destination server.
[0070] The source server is any one of multiple servers in the network, and the source computing chip is any one of those computing chips. The destination computing chip is a different server from the source server, and it is any one of those computing chips. The source server and the destination server are different.
[0071] 102. If the first network chip has established an exchange link with both the source computing chip and the destination computing chip, then it is determined that data will be sent to the destination computing chip through the first network chip.
[0072] The first network chip belongs to multiple network chipsets. In process 102, data interaction between the source computing chip and the destination computing chip only requires link exchange. For example, the identifiers of at least one computing chip connected to each network chip in a network chipset are identical, and the identifiers of computing chips connected to network chips in any two network chipsets are different. Referring to the aforementioned definition of a guide rail, the source computing chip and the destination computing chip belong to the same guide rail, and the source computing chip and the destination computing chip are computing chips with the same identifier in the source server and the destination server, respectively.
[0073] As shown in Figure 2, Figure 3 or Figure 5, the source computing chip can be computing chip 1 in server 100 (source server), the destination computing chip can be computing chip 1 in server 200 (destination server), and the first network chip can be the network chip in network chipset 1.
[0074] Referring to the description in Figure 5 above, the controller can indicate that data is sent to the destination computing chip via the routing information sent to each computing chip and each network chip. This process 102 can be regarded as part of the process of determining routing information, and it can be executed before or simultaneously with process 101. This embodiment of the application does not limit this.
[0075] 103. If no network chip has established an exchange link with both the source computing chip and the destination computing chip, then it is determined that data is sent to the destination computing chip through the first computing chip and the second network chip in the source server, or data is sent to the destination computing chip through the third network chip and the second computing chip in the destination server.
[0076] The second and third network chips belong to multiple network chip groups. The first computing chip is a non-source computing chip in the source server, and the second computing chip is a non-destination computing chip in the destination server. The first computing chip has a direct link with the source computing chip, and the second network chip has exchanged links with both the first and destination computing chips. The third network chip has exchanged links with both the source and second computing chips, and the second computing chip has a direct link with the destination computing chip.
[0077] For example, in a network chipset, each network chip is connected to at least one computing chip with the same identifier, and the computing chips connected to any two network chipsets have different identifiers. Referring to the aforementioned definition of the guide rails, the source computing chip and the destination computing chip belong to different guide rails, and the source computing chip and the destination computing chip are computing chips with different identifiers in the source server and the destination server, respectively. The first computing chip and the destination computing chip belong to the same guide rail, and the first computing chip and the destination computing chip are computing chips with the same identifier in the source server and the destination server, respectively. The source computing chip and the second computing chip belong to the same guide rail, and the source computing chip and the second computing chip are computing chips with the same identifier in the source server and the destination server, respectively.
[0078] As shown in Figures 2, 3, or 5, the source computing chip can be computing chip 3 in server 55 (source server), the first computing chip can be computing chip 5 in server 55, the destination computing chip can be computing chip 5 in server 280 (destination server), the second network chip is the network chip in network chipset 5. Alternatively, the second computing chip can be computing chip 3 in server 280, and the third network chip is the network chip in network chipset 3.
[0079] Referring to the description in Figure 5 above, the controller can indicate, through routing information sent to each computing chip and each network chip, whether the data is sent to the destination computing chip via the first computing chip and the second network chip, or via the third network chip and the second computing chip. This process 103 can be considered as part of the process of determining routing information, and it can be executed before or simultaneously with process 101. This embodiment of the application does not limit this.
[0080] In this embodiment, data transmission from the source computing chip in the source server to the destination computing chip in the destination server is divided into two scenarios: Scenario 1. The source computing chip and the destination computing chip have established a switching link with the same network chip (refer to the aforementioned process 102); Scenario 2. No network chip has established a switching link with both the source computing chip and the destination computing chip (refer to the aforementioned process 103). The following describes the data transmission process of Scenario 1 and Scenario 2 when this method is applied to a network.
[0081] For example, please refer to Figure 7, which is a flowchart illustrating another data communication method for a network provided in an embodiment of this application. This method can be applied to a network, and the network structure can be referred to the foregoing description. Figure 7 illustrates the data transmission process in scenario 1. In this method, the data interaction between the source computing chip and the destination computing chip only requires link exchange. The method may include the following processes:
[0082] 201. The source computing chip and the first network chip send data. The first network chip establishes an exchange link with both the source computing chip and the destination computing chip.
[0083] The first network chip belongs to multiple network chipsets, and it is any network chip in any network chipset in the network.
[0084] For example, the source computing chip can determine that the next hop is the first network chip by querying its routing table and send data to the first network chip. The process of querying the routing table can be performed by a hardware module (such as an I / O die or a communication module) in the source computing chip without the need for software intervention.
[0085] The source computing chip sends data to the first network chip via a switching link. As shown in Figures 2, 3, or 5, assuming the source computing chip is computing chip 1 in server 100 (source server) and the destination computing chip is computing chip 1 in server 200 (destination server), then the first network chip is the network chip in network chipset 1. Computing chip 1 in server 100 sends data to the network chip in network chipset 1 via a switching link.
[0086] 202. The first network chip sends data to the destination computing chip.
[0087] The first network chip can determine the next hop as the destination computing chip by consulting its routing table and then send data to the destination computing chip. The destination computing chip determines that the destination address of the data is itself and will not continue sending data. The process of consulting the routing table can be performed by the hardware module (such as the I / O die or communication module) in the first network chip, without the need for software intervention.
[0088] The first network chip sends data to the destination computing chip via a switching link. As shown in Figures 2, 3, or 5, assuming the source computing chip is computing chip 1 in server 100 (source server) and the destination computing chip is computing chip 1 in server 200 (destination server), then the first network chip is the network chip in network chipset 1. The network chip in network chipset 1 sends data to computing chip 1 in server 200 via a switching link.
[0089] For example, please refer to Figure 8, which is a flowchart illustrating another data communication method for a network provided in an embodiment of this application. This method can be applied to a network, and the network structure can be referred to the foregoing description. Figure 8 illustrates a data transmission process in scenario 2. In this method, starting with the source computing chip, the data first undergoes heterogeneous identifier exchange (also known as track switching operation within the source server) via a direct connection link inside the source server, and then undergoes homogeneous identifier exchange via a switching link outside the server. This method may include the following processes:
[0090] 301. The source computing chip sends data to the first computing chip in the source server.
[0091] The first computing chip is a non-source computing chip in the source server, and the first computing chip has a direct connection link with the source computing chip.
[0092] For example, the source computing chip can determine that the next hop is the first computing chip by querying its routing table and send data to the first computing chip. The process of querying the routing table can be performed by a hardware module (such as an I / O die or a communication module) in the source computing chip without the need for software intervention.
[0093] The source computing chip sends data to the first computing chip via a direct link. As shown in Figures 2, 3, or 5, assuming the source computing chip is computing chip 3 in server 55 (source server) and the destination computing chip is computing chip 5 in server 280 (destination server), then the first computing chip is computing chip 5 in server 55, and the second network chip is the network chip in network chipset 5. Computing chip 3 in server 55 sends data to computing chip 5 in server 55 via a direct link.
[0094] 302. The first computing chip sends data to the second network chip, and the second network chip establishes an exchange link with the destination computing chip and the first computing chip at the same time.
[0095] The second network chip is any network chip in a plurality of network chipsets that has established a switching link with the first computing chip and the target computing chip.
[0096] For example, the first computing chip can determine that the next hop is the second network chip by querying its routing table and send data to the second network chip. The process of querying the routing table can be performed by a hardware module (e.g., an I / O die or a communication module) in the first computing chip without the need for software intervention.
[0097] The first computing chip sends data to the second network chip via a switching link. As shown in Figures 2, 3, or 5, assuming the source computing chip is computing chip 3 in server 55 (source server) and the destination computing chip is computing chip 5 in server 280 (destination server), then the first computing chip is computing chip 5 in server 55, and the second network chip is the network chip in network chipset 5. Computing chip 5 in server 55 sends data to the network chip in network chipset 5 via a switching link.
[0098] 303. The second network chip sends data to the target computing chip.
[0099] For example, the second network chip can determine the next hop as the destination computing chip by querying its routing table and send data to the destination computing chip. The destination computing chip determines that the destination address of the data is itself and will not continue sending data. The process of querying the routing table can be performed by a hardware module (e.g., an I / O die or a communication module) in the second network chip, without the need for software intervention.
[0100] The second network chip sends data to the destination computing chip via a switching link. As shown in Figures 2, 3, or 5, assuming the source computing chip is computing chip 3 in server 55 (source server) and the destination computing chip is computing chip 5 in server 280 (destination server), then the first computing chip is computing chip 5 in server 55, and the second network chip is the network chip in network chipset 5. The network chip in network chipset 5 sends data to computing chip 5 in server 280 via a switching link.
[0101] For example, please refer to Figure 9, which is a flowchart illustrating another data communication method for a network provided in an embodiment of this application. This method can be applied to a network, and the network structure can be referred to the foregoing description. Figure 8 illustrates another data transmission process under scenario 2. In this method, starting with the source computing chip, the data first undergoes same-identity exchange via an external switching link to the server, and then undergoes different-identity exchange via a direct connection link inside the server (also known as a track-switching operation within the source server). This method may include the following processes:
[0102] 401. The source computing chip sends data to the third network chip, and the third network chip establishes an exchange link with the source computing chip and the second computing chip in the destination server at the same time.
[0103] The third network chip is any network chip in a multi-chipset that has established a switching link with both the source computing chip and the second computing chip. The second computing chip is a non-destination computing chip in the destination server, and it has a direct connection link with the destination computing chip.
[0104] For example, the source computing chip can determine that the next hop is a third network chip by querying its routing table and send data to the third network chip. The process of querying the routing table can be performed by a hardware module (such as an I / O die or a communication module) in the source computing chip without the need for software intervention.
[0105] The source computing chip sends data to the third network chip via a switching link. As shown in Figure 2 or Figure 3, assuming the source computing chip is computing chip 3 in server 55 (source server) and the destination computing chip is computing chip 5 in server 280 (destination server), then the second computing chip is computing chip 3 in server 280, and the third network chip is the network chip in network chipset 3. Computing chip 3 in server 55 sends data to the network chip in network chipset 3 via a switching link.
[0106] 402. The third network chip sends data to the second computing chip.
[0107] For example, the third network chip can determine that the next hop is the second computing chip by querying its routing table and send data to the second computing chip. The process of querying the routing table can be performed by a hardware module (such as an I / O die or a communication module) in the third network chip without the need for software intervention.
[0108] The third network chip sends data to the second computing chip via a switching link. As shown in Figure 2 or Figure 3, assuming the source computing chip is computing chip 3 in server 55 (source server) and the destination computing chip is computing chip 5 in server 280 (destination server), then the second computing chip is computing chip 3 in server 280, and the third network chip is the network chip in network chipset 3. The network chip in network chipset 3 sends data to computing chip 3 in server 280 via a switching link.
[0109] 403. The second computing chip sends data to the target computing chip.
[0110] For example, the second computing chip can determine the next hop as the destination computing chip by querying its routing table and send data to the destination computing chip. The destination computing chip determines that the destination address of the data is itself and will not continue sending data. The process of querying the routing table can be performed by a hardware module (e.g., an I / O die or a communication module) in the second computing chip, without the need for software intervention.
[0111] The second computing chip sends data to the destination computing chip via a direct link. As shown in Figure 2 or Figure 3, assuming the source computing chip is computing chip 3 in server 55 (source server) and the destination computing chip is computing chip 5 in server 280 (destination server), then the second computing chip is computing chip 3 in server 280, and the third network chip is the network chip in network chipset 3. Computing chip 3 in server 280 sends data to computing chip 5 in server 280 via a direct link.
[0112] In summary, the data communication method for a network provided in this application includes a network comprising multiple servers and multiple network chipsets. Each server includes multiple computing chips, and a direct link is established between every two computing chips in each server. Each network chip in each network chipset establishes an exchange link with at least one computing chip in each server. For data that needs to be transmitted from a source computing chip in a source server to a destination computing chip in a destination server, if a first network chip has established exchange links with both the source and destination computing chips, the data will be transmitted to the destination computing chip through the first network chip. If no network chip has established exchange links with both the source and destination computing chips, the data will be transmitted to the destination computing chip through the first computing chip in the source server and a second network chip that has established exchange links with both the first and destination computing chips. Alternatively, the data will be transmitted to the destination computing chip through a third network chip that has established exchange links with both the source and destination computing chips and the second computing chip in the destination server. This network is heterogeneous, consisting of direct links within servers and switching links outside servers. Compared to related technologies, it supports a significantly increased number of interconnected computing chips. When transmitting data, each computing chip's port only requires one network chip's port, and forwarding each data transmission requires at most one computing chip and one network chip. Compared to related technologies, this reduces the cost of network chips and the latency of data transmission. Thus, it achieves large-scale networking while enabling low-cost and low-latency data transmission, making it suitable for scenarios requiring intensive computing and low latency (such as AI training scenarios and HPC scenarios).
[0113] The order of the methods provided in the embodiments of this application can be adjusted appropriately, and the process can also be added or removed as appropriate. Any variations that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application, and the embodiments of this application do not limit this.
[0114] The foregoing primarily describes the data communication method for a network provided in the embodiments of this application from the perspective of the device. It is understood that, in order to achieve the above functions, the device includes corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, based on the algorithm steps of the examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0115] This application embodiment can divide the device into functional modules according to the above method example. For example, each function can be divided into its own functional module, or two or more functions can be integrated into one device. The integrated modules can be implemented in hardware or software functional modules. It should be noted that the module division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.
[0116] Figure 10 is a block diagram of a data communication device for a network according to an embodiment of the application. When each functional module is divided according to its corresponding function, the data communication device 500 for a network may include a determining module 501 and a processing module 502. Exemplarily, the data communication device for a network may be a controller, or other combined devices, components, etc., having the functions described above for a data communication device for a network. Exemplarily, the data communication device 500 for a network may be a processor (or processing circuit), such as a baseband processor, which may include one or more central processing units (CPUs). The data communication device 500 for a network may also be a processor (or processing circuit) of a chip system, which may include one or more central processing units.
[0117] For example, the network includes multiple servers and multiple network chipsets, each of the multiple servers includes multiple computing chips, a direct link is established between every two computing chips in each server, and each network chip in the network chipset establishes a switching link with at least one computing chip in each server.
[0118] The determination module 501 is used to determine the data sent from the source computing chip in the source server to the destination computing chip in the destination server.
[0119] The processing module 502 is used to determine that data is sent to the destination computing chip through the first network chip if the first network chip has established an exchange link with both the source computing chip and the destination computing chip.
[0120] The processing module 502 is also used to determine, if no network chip has established an exchange link with both the source computing chip and the destination computing chip, to send data to the destination computing chip through the first computing chip and the second network chip in the source server, or to determine to send data to the destination computing chip through the third network chip and the second computing chip in the destination server.
[0121] In this configuration, the second network chip establishes a switching link with both the destination computing chip and the first computing chip, and the third network chip establishes a switching link with both the source computing chip and the second computing chip.
[0122] In combination with the above scheme, the identifiers of the computing chips in any two servers are identical, and the identifiers of at least one computing chip in different servers connected to each network chip are identical; the identifiers of the first computing chip and the destination computing chip are identical, and the identifiers of the second computing chip and the source computing chip are identical.
[0123] In combination with the above scheme, the identifiers of at least one computing chip connected to each network chip in the network chipset are identical, and the identifiers of the computing chips connected to any two network chipsets are different.
[0124] Based on the above scheme, the computing chip is an AI chip.
[0125] Optionally, the data communication device for a network provided in Figure 10 may further include a storage module 503, primarily used for storing software programs. The processing module 502 can call the software programs stored in the storage module 503 to implement any of the methods in the embodiments of this application. The storage module 503 includes media with storage functions such as hard disks, RAM, and ROM.
[0126] Figure 11 is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. The electronic device 600 can be a controller or a functional module within a controller, a computing chip or a functional module within a computing chip, or a network chip or a functional module within a network chip. As shown in Figure 11, the electronic device 600 includes a processor 601, a transceiver 602, and a communication line 603.
[0127] The processor 601 is used to execute any step in the aforementioned method embodiments, and when performing processes such as sending data, it can selectively call the transceiver 602 and the communication line 603 to complete the corresponding operation.
[0128] Furthermore, the electronic device 600 may also include a memory 604. The processor 601, memory 604, and transceiver 602 can communicate with each other via a communication line 603.
[0129] Transceiver 602 is used to communicate with other devices or other communication networks, such as Ethernet, radio access network (RAN), wireless local area network (WLAN), etc. Transceiver 602 can be a module, circuit, transceiver, or any device capable of enabling communication.
[0130] The transceiver 602 is mainly used for sending and receiving data, and may include a transmitter and a receiver to send and receive data, respectively. For example, data can be obtained through the transceiver 602. Operations other than sending and receiving data are implemented by the processor, such as determining the destination server and destination computing chip of the data.
[0131] The transceiver 602 includes the transceiver module 501 shown in Figure 9.
[0132] The communication line 603 may include a path for transmitting information between various components of the electronic device 600 (e.g., processor 601, transceiver 602, and memory 604).
[0133] In one design, the processor can be viewed as a logic circuit, and the transceiver as an interface circuit.
[0134] Memory 604 is used to store instructions, which may be computer programs. When the instructions stored in memory 604 are executed by processor 601, processor 601 performs various steps of the data communication method for a network according to embodiments of this application. Specifically, processor 601 may execute methods 101 to 102, or methods 201 to 202, or methods 301 to 303, or methods 401 to 403 as described above.
[0135] The memory 604 can be a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM).
[0136] It should be noted that the memory 604 can exist independently of the processor 601, or it can be integrated with the processor 601. The memory 604 can be used to store instructions, program code, or some data, etc. The memory 604 can be located inside or outside the electronic device 600, without restriction.
[0137] The processor 601 includes the processing module 502 shown in Figure 10.
[0138] The processor 601 can be a general-purpose CPU, microprocessor, application-specific integrated circuit (ASIC), GPU, or one or more integrated circuits to execute relevant programs to implement the data communication method for networks according to the embodiments of this application. It should be noted that the processor here is not limited to a general-purpose CPU, but also includes processors (or accelerators) with multiple parallel processing units, such as GPUs and NPUs, which are capable of performing a large number of parallel tasks.
[0139] The processor 601 can also be an integrated circuit chip with signal processing capabilities.
[0140] The processor 601 described above can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory 604. The processor 601 reads the information in memory 604 and, in conjunction with its hardware, executes the data communication method for the network in the embodiments of this application.
[0141] As an optional implementation, the electronic device 600 also includes an output device 605 and an input device 606. For example, the input device 606 is a device such as a keyboard, mouse, microphone, or joystick, and the output device 605 is a device such as a display screen or speaker.
[0142] It should be noted that the electronic device 600 can be a chip system or a device with a similar structure to that shown in Figure 11. The chip system may include chips or other discrete components. Actions, terms, etc., involved in the various embodiments of this application can be referenced interchangeably without limitation. The message names or parameter names in the messages used for interaction between devices in the embodiments of this application are merely examples; other names may be used in specific implementations without limitation. Furthermore, the composition structure shown in Figure 11 does not constitute a limitation on the electronic device 600. In addition to the components shown in Figure 11, the electronic device 600 may include more or fewer components than shown in Figure 11, or combine certain components, or have different component arrangements.
[0143] The processor and transceiver described in this application can be implemented on integrated circuits (ICs), analog ICs, radio frequency integrated circuits, mixed-signal ICs, application-specific integrated circuits (ASICs), printed circuit boards (PCBs), electronic devices, etc. The processor and transceiver can also be manufactured using various IC process technologies, such as complementary metal oxide semiconductors (CMOS), n-metal-oxide-semiconductor (NMOS), p-type metal oxide semiconductors (PMOS), bipolar junction transistors (BJTs), bipolar CMOS (BiCMOS), silicon germanium (SiGe), gallium arsenide (GaAs), etc.
[0144] According to the method provided in the embodiments of this application, this application also provides a computer program product, which includes computer program code. When the computer program code is run on a computer, it causes the computer to execute any of the methods described in the embodiments of this application.
[0145] This application also provides a computer-readable storage medium. All or part of the processes in the above method embodiments can be executed by a computer or a device with data communication capabilities, using computer programs or instructions to control related hardware. The computer program or set of instructions can be stored in the computer-readable storage medium. When executed, the computer program or set of instructions can include the processes described in the above method embodiments. The computer-readable storage medium can be an internal storage unit of the controller, computing chip, or network chip in any of the foregoing embodiments, such as a hard disk or memory of the controller, computing chip, or network chip. The computer-readable storage medium can also be an external storage device of the controller, computing chip, or network chip, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. Further, the computer-readable storage medium can include both internal storage units of the controller, computing chip, or network chip and external storage devices. The computer-readable storage medium is used to store the computer program or instructions and other programs and data required by the controller, computing chip, or network chip. The aforementioned computer-readable storage medium can also be used to temporarily store data that has been output or will be output.
[0146] This application also provides a chip, which includes a processor and a data interface. The processor reads instructions stored in a memory through the data interface and executes the data communication method for network use in this application.
[0147] Optionally, as one implementation, the chip may further include a memory storing instructions, and the processor is used to execute the instructions stored in the memory. When the instructions are executed, the processor is used to execute the data communication method for the network in the embodiments of this application.
[0148] It should be understood that the processor in the embodiments of this application can be a CPU, GPU, NPU, or other processor (or accelerator) with multiple parallel processing units, which can execute a large number of parallel tasks. The processor can also be other general-purpose processors, DSPs, ASICs, FPGAs, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.
[0149] It should also be understood that the memory in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache.
[0150] By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM).
[0151] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0152] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.
[0153] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.
[0154] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0155] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.
[0156] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.
[0157] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A data communication method for a network, characterized by, A controller applied to the network, the network including multiple servers and multiple network chipsets, each of the multiple servers including multiple computing chips, a direct link being established between every two computing chips in each server, and a switching link being established between each network chip in the network chipsets and at least one computing chip in each server; the method includes: Determine the data sent from the source computing chip in the source server to the destination computing chip in the destination server; If the first network chip has established an exchange link with both the source computing chip and the destination computing chip, then it is determined that the data is sent to the destination computing chip through the first network chip. If no network chip has established an exchange link with both the source computing chip and the destination computing chip, then it is determined that the data is sent to the destination computing chip through the first computing chip and the second network chip in the source server, or it is determined that the data is sent to the destination computing chip through the third network chip and the second computing chip in the destination server. In this configuration, the second network chip establishes a switching link with both the destination computing chip and the first computing chip, and the third network chip establishes a switching link with both the source computing chip and the second computing chip.
2. The method of claim 1, wherein, The identifiers of the computing chips in any two servers are identical, and the identifiers of at least one computing chip in different servers connected to each network chip are identical. The first computing chip has the same identifier as the target computing chip, and the second computing chip has the same identifier as the source computing chip.
3. The method according to claim 2, characterized in that, The identifiers of the at least one computing chip connected to each network chip in the network chipset are identical, and the identifiers of the computing chips connected to any two network chipsets are different.
4. The method according to any one of claims 1 to 3, characterized in that, The computing chip is an artificial intelligence chip.
5. A data communication method for a network, characterized in that, The method is applied to the network, which includes multiple servers and multiple network chipsets. Each of the multiple servers includes multiple computing chips. A direct link is established between every two computing chips in each server. Each network chip in the network chipset is connected to at least one computing chip in each server via a switching link. The source computing chip in the source server of the communication network sends data to a first network chip that has established a switching link with both the source computing chip and the destination computing chip, and the first network chip sends the data to the destination computing chip in the destination server of the communication network. Alternatively, the source computing chip sends the data to a first computing chip in the source server, the first computing chip sends the data to a second network chip that has established a switching link with both the destination computing chip and the first computing chip, and the second network chip sends the data to the destination computing chip. Alternatively, the source computing chip sends the data to a third network chip that has established a switching link with both the source computing chip and the second computing chip in the destination server; the third network chip sends the data to the second computing chip; and the second computing chip sends the data to the destination computing chip.
6. The method according to claim 5, characterized in that, The identifiers of the computing chips in any two servers are identical, and the identifiers of at least one computing chip in different servers connected to the network chip are identical. The first computing chip has the same identifier as the target computing chip, and the second computing chip has the same identifier as the source computing chip.
7. The method according to claim 6, characterized in that, The identifiers of the at least one computing chip connected to each network chip in the network chipset are identical, and the identifiers of the computing chips connected to any two network chipsets are different.
8. The method according to any one of claims 5 to 7, characterized in that, The computing chip is an artificial intelligence chip.
9. A data communication device for a network, characterized in that, A controller applied to the network, the network including multiple servers and multiple network chipsets, each of the multiple servers including multiple computing chips, a direct link being established between every two computing chips in each server, and a switching link being established between each network chip in the network chipsets and at least one computing chip in each server; the device includes: The determination module is used to determine the data sent from the source computing chip in the source server to the destination computing chip in the destination server. The processing module is configured to determine, if the first network chip has established an exchange link with both the source computing chip and the destination computing chip, to send the data to the destination computing chip through the first network chip; The processing module is further configured to, if no network chip has established an exchange link with both the source computing chip and the destination computing chip, determine to send the data to the destination computing chip through the first computing chip and the second network chip in the source server, or determine to send the data to the destination computing chip through the third network chip and the second computing chip in the destination server. In this configuration, the second network chip establishes a switching link with both the destination computing chip and the first computing chip, and the third network chip establishes a switching link with both the source computing chip and the second computing chip.
10. A communication device, characterized in that, The device includes: One or more processors; Memory, used to store one or more computer programs or instructions; When the one or more computer programs or instructions are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 4, or the one or more processors implement the method as described in any one of claims 5 to 8.
11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program code that, when executed by a computer or processor, causes the computer or processor to perform the method as described in any one of claims 1 to 4, or causes the computer or processor to perform the method as described in any one of claims 5 to 8.
