Optoelectronic hybrid interconnect system for large-scale intelligent computing centers
By using a hybrid optoelectronic interconnect architecture for intra-group and inter-group switching systems, the problem of efficient communication and low-latency synchronization under large-scale computing nodes in traditional data center architectures is solved, realizing low-cost and efficient data center network interconnection and adapting to the complex data processing needs of large model training.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-21
- Publication Date
- 2026-03-06
AI Technical Summary
Traditional data center interconnect architectures struggle to meet the demands of efficient communication and low-latency synchronization for large-scale computing nodes, especially in large model training tasks, where existing optoelectronic hybrid architectures are ill-suited for flexible reconfiguration and efficient data processing.
It adopts a hybrid optoelectronic interconnection architecture of intra-group switching system and inter-group switching system. The intra-group switching system consists of multiple electrical switches and servers, providing flexible data synchronization. The inter-group switching system consists of optical switches, which use optical links to achieve high-bandwidth and low-latency data transmission, and adapt to service requirements through the flexible reconfiguration capability of optical switches.
It achieves low-cost, high-efficiency data center network interconnection, supports flexible connection of large-scale computing nodes and low-latency data transmission, has smooth upgrade capabilities, and adapts to complex data processing needs.
Smart Images

Figure CN120200990B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of communication technology, and in particular to an optoelectronic hybrid interconnection system for large-scale intelligent computing centers. Background Technology
[0002] With the continuous development of large-scale model technology, accelerating the development of the artificial intelligence industry, and building a powerful data center (or intelligent computing center) to support large-scale models has become an important research and development direction for the future. The ultra-large-scale data volume and distributed computing method are the key characteristics of large-scale model training tasks, which put forward new requirements for the networking of intelligent computing centers that carry such new services: (1) Oriented to large-scale connection: Large-scale model training services require the participation of a large number of computing nodes, and the networking of intelligent computing centers urgently needs to support the efficient communication and collaboration of tens of thousands of computing nodes; (2) Support for large-capacity transmission: Large-scale model training tasks require training on a large amount of datasets, and the model has many parameters, so the intelligent computing center network needs to improve the capacity and rate of channel transmission; (3) Ensure low-latency synchronization: Frequent parameter synchronization and data interaction between computing nodes make communication latency a bottleneck for services, and the networking of intelligent computing centers needs to ensure low latency of data interaction, thereby ensuring the high efficiency of large-scale cluster training tasks.
[0003] Traditional electrical switching-based data center architectures suffer from bandwidth limitations and latency bottlenecks, making it difficult to meet the explosive traffic growth and high synchronization demands of large model training. Electrical packet switching-based data center architectures include Spine-Leaf networks, Dragonfly networks, and Tours networks. In a Spine-Leaf network, all Leaf layer switches are equally connected to all Spine layer switches, forming a fully interconnected network. While Spine-Leaf networks offer excellent bi-splitting bandwidth performance, their large network diameter and inconsistent communication hop counts between intra-cluster and inter-cluster nodes mean that parameter synchronization during AI training requires waiting for the last node to complete before proceeding to the next iteration, potentially wasting computational resources. Dragonfly networks achieve full connectivity between groups through direct connections, effectively reducing latency, but they struggle to meet the scalability and communication performance requirements of various synchronization algorithms under the business demands of modern intelligent computing centers. The Torus architecture is a multi-dimensional topology with advantages such as simple structure and multiple paths. However, as the network dimension increases, the number of links and latency within the network gradually increase. In addition, the ring structure of the Torus itself can lead to complex deadlock problems. These interconnect architectures are unable to support the explosive growth of traffic within data centers during large-scale model training, and suffer from problems such as bandwidth limitations, high energy consumption, latency bottlenecks, and switching capacity limitations. More efficient data center interconnect solutions are needed to address the bottlenecks of electrically switched networks.
[0004] Compared to electrical packet switching, optical switching does not parse data packets during forwarding, eliminating the need for transceiver modules to convert between optical and electrical signals. In large-scale networks, this feature effectively reduces the frequent optoelectronic conversion overhead in electrical switching networks, lowering communication latency. However, optical switching equipment and components are relatively expensive, requiring significant investment and presenting greater maintenance challenges. Therefore, a hybrid optoelectronic switching architecture, which replaces some electrical packet switches with optical switches, is considered more suitable for large-scale, high-performance data centers and has already been implemented. This architecture combines the advantages of both electrical packet switching and optical switching, balancing performance and cost. For example, both the Helios and ReSaw architectures employ a two-layer tree topology. Their access switches (electrical switches) are directly connected to servers, handling local communication, while the core switches consist of both electrical packet switches and optical switches. Some uplink ports of the access switches connect to electrical packet switches, and others connect to optical switches. Electrical packet switching is used for short-term bursts of traffic, while optical switching is used for longer-duration traffic with high bandwidth requirements. This type of architecture has multiple switching layers and higher switching latency, but it can achieve data aggregation at a lower cost, making it very suitable for traditional data center operations characterized by "coexistence of large and small flows, with more north-south traffic and less east-west traffic." However, intelligent computing centers require large-scale, arbitrary interconnection between nodes at the same level, placing high demands on east-west communication and exhibiting new characteristics of less north-south traffic and more east-west traffic. Traditional optoelectronic hybrid architectures are not suitable for this. Furthermore, traditional optoelectronic hybrid architectures are difficult to reconfigure flexibly, have weak scalability and smooth upgrade capabilities, and struggle to support large-scale cluster interconnection training operations, thus failing to efficiently handle complex data processing needs. Therefore, how to achieve large-scale GPU card interconnection for large model training needs, and build a flat, non-blocking intelligent computing center network architecture that combines deployment cost advantages with the flexible adaptability of rigid pipelines has become a significant challenge for current intelligent computing centers. Summary of the Invention
[0005] In view of this, embodiments of the present invention provide a novel optoelectronic hybrid interconnect system for large-scale intelligent computing centers, in order to eliminate or improve one or more defects existing in the prior art.
[0006] One aspect of the present invention provides an optoelectronic hybrid interconnection system for large-scale intelligent computing centers, the system comprising: an intra-group switching system and an inter-group switching system, the intra-group switching system comprising a plurality of intra-group switching units, and the inter-group switching system comprising at least one optical switch;
[0007] Each group of switching units includes a first number of electrical switches and at least one server. Each server has at least a first number of computing nodes corresponding to the first number of electrical switches. The first number of computing nodes are interconnected. Each electrical switch is connected to a corresponding computing node among the first number of computing nodes through a downlink port.
[0008] Each optical switch in the inter-group switching system includes multiple pairs of dual-sided ports. These multiple pairs of dual-sided ports are used to provide multiple optical path connections between different pairs of opposite-sided ports through mapping configuration between the opposite-sided ports. The two opposite-sided ports of each optical path connection of the same switch can be connected to the uplink ports of the electrical switches of different intra-group switching units based on the connection relationship between the optical switch ports and the electrical switch ports. The mapping configuration enables the formation of inter-group optical path switching channels between different intra-group switching units.
[0009] In some embodiments of the present invention, a first number of computing nodes of the servers of the group switching unit are interconnected via high-speed interconnect technology.
[0010] In some embodiments of the present invention, the intra-group switching system and the inter-group switching system are interconnected using single-fiber bidirectional optical modules for data transmission.
[0011] In some embodiments of the present invention, the number of downlink ports of each switch in the group switching unit is greater than or equal to the number of servers in the corresponding group switching unit.
[0012] In some embodiments of the present invention, the electrical switches of the group switching unit and the computing nodes inside the server are correspondingly established through a mapping relationship between the electrical switch number and the computing node number.
[0013] In some embodiments of the present invention, the connection relationship between the optical switch port and the electrical switch port includes: the same pair of dual-sided ports in the optical switch are respectively connected to different uplink ports of the electrical switch in the same group of switching units; the mapping configuration includes establishing a one-to-one mapping relationship between different pairs of opposite-sided ports of the dual-sided ports.
[0014] In some embodiments of the present invention, the total number of pairs of ports on both sides of the optical switches in the inter-group switching system is greater than or equal to a second quantity, and at least one optical path connection is established between any two intra-group switching units, wherein the second quantity is N*(N-1) / 2; the number of optical switches in the inter-group switching system satisfies the following formula: Alternatively, the total number of pairs of ports on both sides of the optical switches in the inter-group switching system is less than the second number, and optical path connections are established between the selected intra-group switching units; the number of optical switches in the inter-group switching system satisfies the following formula:
[0015] Where N is the number of intra-group switching units in the intra-group switching system, O is the number of optical switches, and n o Let be the number of pairs of ports on both sides of the o-th optical switch.
[0016] In some embodiments of the present invention, the mapping configuration of the opposite-side ports of the optical switches in the inter-group switching system can switch according to the service transmission requirements of different intra-group switching units, thereby realizing the switching of connection optical paths.
[0017] In this embodiment of the invention, the computing node is used to send and receive data information that needs to be exchanged; the electrical switch of the same group switching unit is used to receive data information sent from the corresponding computing node and forward it to the optical switch connected to the inter-group switching system, or to receive data information sent from the optical switch connected to the inter-group switching system and forward it to the corresponding computing node; the optical switch is used to receive data information from the electrical switch and forward the received data information between different group switching units.
[0018] The optoelectronic hybrid interconnection system of this invention for large-scale intelligent computing centers offers several advantages in terms of flexibility. Intra-group switching systems utilize electrical switches to provide flexible switching methods between computing nodes within the same group. Simultaneously, inter-group optical switches provide high-capacity, low-latency data transmission channels between computing nodes across groups. Optical paths can be flexibly established using a service-oriented dynamic resource scheduling method, providing adaptable connectivity and switching capabilities for various services. Regarding construction costs, the deployed optical switches have the ability to be smoothly upgraded. When the number of optical path connections provided by the actual deployed optical switches exceeds the basic optical path connection requirement, this invention can provide a larger capacity transmission channel between intra-group switching units. When the number of optical path connections provided by the actual deployed optical switches is less than the basic optical path connection requirement, the flexible and switchable capabilities of the optical switches can be used to construct adaptable optical paths for services. Compared to other data center network interconnection systems, the system of this invention is not only lower in construction cost but also more flexible.
[0019] Additional advantages, objects, and features of the invention will be set forth in part in the description which follows, and will also become apparent in part to those skilled in the art upon studying the description, or may be learned by practice of the invention. The objects and other advantages of the invention can be realized and obtained by means of the structures specifically pointed out in the description and drawings.
[0020] Those skilled in the art will understand that the objectives and advantages achievable with the present invention are not limited to those specifically described above, and that the above and other objectives achievable with the present invention will become clearer from the following detailed description. Attached Figure Description
[0021] The accompanying drawings, which are provided to further illustrate the invention and form part of this application, are not intended to limit the scope of the invention.
[0022] Figure 1 This is a schematic diagram of the structure of an intra-group switching unit in an optoelectronic hybrid interconnect system according to an embodiment of the present invention.
[0023] Figure 2 This is a schematic diagram of the interconnection between computing nodes of a server in one embodiment of the present invention.
[0024] Figure 3 This is a schematic diagram of the internal optical path connection of the optical switch in an inter-group switching system according to an embodiment of the present invention.
[0025] Figure 4 This is a schematic diagram illustrating the interconnection between the intra-group switching unit of the intra-group switching system and the optical switch of the inter-group switching system in an optoelectronic hybrid interconnection system according to an embodiment of the present invention.
[0026] Figure 5 This is a schematic diagram illustrating the interconnection between the intra-group switching unit of the intra-group switching system and the optical switch of the inter-group switching system in another embodiment of the optoelectronic hybrid interconnection system of the present invention.
[0027] Figure 6 This is a schematic diagram illustrating the interconnection between the intra-group switching unit of the intra-group switching system and the optical switch of the inter-group switching system in another embodiment of the optoelectronic hybrid interconnection system of the present invention.
[0028] Figure 7 This is a schematic diagram illustrating the interconnection between the intra-group switching unit of the intra-group switching system and the optical switch of the inter-group switching system in another embodiment of the optoelectronic hybrid interconnection system of the present invention. Detailed Implementation
[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.
[0030] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.
[0031] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.
[0032] It should also be noted that, unless otherwise specified, the term "connection" in this article can refer not only to a direct connection, but also to an indirect connection involving an intermediary.
[0033] In the following description, embodiments of the invention will be illustrated with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar parts, or the same or similar steps.
[0034] To address the limitations of existing intelligent computing centers in meeting the demands of large-scale model training with numerous connections, high bandwidth, and low latency under time-varying conditions, this invention proposes a novel optoelectronic hybrid interconnect system for large-scale intelligent computing centers. This system achieves a low-cost topology supporting large-scale connections while leveraging the rigid pipeline characteristics of optical interconnects. The proposed novel optoelectronic hybrid interconnect system comprises an intra-group switching system and an inter-group switching system. The intra-group switching system includes multiple intra-group switching units, each comprising a server and multiple electrical switches (also known as electrical packet switches). Each server has multiple interconnected computing nodes. The inter-group switching system includes one or more optical switches. The electrical switches within the intra-group switching units provide flexible data synchronization for the computing nodes. The optical switches between groups utilize the high bandwidth and low latency characteristics of optical links to provide a rigid channel for data exchange between computing nodes within the group, while also leveraging the flexible reconfiguration capabilities of optical switches to flexibly adapt to services. Preferably, each intra-group switching unit has multiple interconnected servers. The interconnection of computing nodes within a single server improves the data processing and communication efficiency within that server; the interconnection between servers ensures efficient data exchange between servers within the same group.
[0035] The optoelectronic hybrid interconnection system for large-scale intelligent computing centers provided by this invention includes an intra-group switching system and an inter-group switching system. The intra-group switching system includes multiple intra-group switching units (or multiple groups), and the inter-group switching system includes at least one optical switch.
[0036] Each group switching unit includes a first number of electrical switches and at least one server. Each server has at least a first number of computing nodes corresponding to the first number of electrical switches, wherein the first number of computing nodes are interconnected. Any electrical switch in each group switching unit is connected to one computing node on all the servers in the group. For example, a correspondence can be pre-set so that each electrical switch in the group switching unit is connected to a corresponding computing node among the first number of computing nodes in the group through a downlink port.
[0037] The number of intra-group switching units can be represented by N, where N is a natural number. For example, N = 5, meaning the intra-group switching system includes 5 intra-group switching units, or the intra-group switching system is divided into 5 groups, numbered G1, G2, ..., G... i ,…,G N Gi This represents the i-th group. Each group contains exchange units numbered E1, E2, ..., E... m ,…,E M M power switches and numbered S1, S2, ..., S l ,…,S L L servers, of which E i Let S be the i-th power switch. i Let be the i-th server. Each switch in the group can contain P switching ports, and each server includes servers numbered C1, C2, ..., C... m ,…,C M M computing nodes, where C i Let C be the i-th computation node. i The computing nodes can connect to the group numbered E via the network port. i It is connected to the downlink port of the power switch.
[0038] like Figure 1 As shown, M=8, L=32, that is, each intra-group switching unit of the intra-group switching system contains 8 electrical switches (E1~E8) and 32 servers (S1~S2). 32 The number of ports P of the power switch is, for example, 64, and each server includes 8 compute nodes (C1 to C8). Figure 1 In the example, C1 = 1, ..., C8 = 8), the compute nodes can be GPUs within the server, and compute nodes within the same service are interconnected, such as... Figure 2 As shown, the connection can be direct or indirect. For example, compute nodes within the same service can communicate with each other using high-speed interconnect technology (such as NVLink). In this embodiment of the invention, any electrical switch in one group switching unit is connected to one compute node on all servers within the group.
[0039] The inter-group switching system may include one or more optical switches, each optical switch including multiple pairs of dual-sided ports (e.g., n pairs). These multiple pairs of dual-sided ports are pre-configured with mapping to provide multiple optical path connections between opposite ports of the dual-sided ports. Each optical path connection of the same optical switch allows two opposite ports to connect to the uplink ports of electrical switches in different intra-group switching units. This mapping configuration enables the formation of inter-group optical path switching channels between different intra-group switching units. In this embodiment of the invention, the intra-group switching system and the inter-group switching system are interconnected using single-fiber bidirectional optical modules for data transmission. In this embodiment of the invention, the mapping configuration includes establishing a one-to-one mapping relationship between different pairs of opposite ports of the dual-sided ports.
[0040] Intra-group switching units and inter-group optical switches can be used for:
[0041] Computing nodes are used to send and receive data that requires interaction;
[0042] The electrical switch within the same group of switching units is used to receive data information sent from the corresponding computing node and forward it to the optical switch connected to the inter-group switching system, or to receive data information sent from the optical switch connected to the inter-group switching system and forward it to the corresponding computing node.
[0043] Optical switches are used to receive data from electrical switches and forward the received data between switching units in different groups.
[0044] The aforementioned novel optoelectronic hybrid interconnection system for large-scale intelligent computing centers comprises an intra-group switching unit where each electrical switch has P ports, of which L are downlink ports for connecting to computing nodes and (PL) are uplink ports for interconnecting with inter-group optical switches. In some embodiments of the present invention, the number L of downlink ports in each electrical switch within the intra-group switching unit is greater than or equal to the number of servers in the corresponding intra-group switching unit. In addition to the downlink ports connecting to computing nodes in the servers and the uplink ports connecting to optical switches, the electrical switches may also have redundant or idle uplink / downlink ports. The specific number is not limited by the present invention, as long as it meets the connection requirements with computing nodes and optical switches.
[0045] In this embodiment of the invention, the electrical switches of the group's switching unit and the computing nodes inside the server are correspondingly established through a mapping relationship between the electrical switch number and the computing node number. Any electrical switch of one group's switching unit can be connected to one computing node on all the servers within the group in the following manner: Connect the group's servers S1, S2, S3, ..., S... L Map the computing nodes within the group to the group's power switch, and map computing node C... i Connect to the corresponding power switch E via the network port. i Connecting. As an example, a feasible mapping method is to connect compute nodes with the same number on different servers to the same electrical switch via network ports. That is, all compute nodes numbered 1 in the group of 32 servers are connected to the electrical packet switch numbered 1 through the first network port of each server; all compute nodes numbered C2 (e.g., C2=2) are connected to the electrical packet switch numbered E2 (e.g., E2=2) through the second network port of each server; and so on up to all compute nodes numbered C... i (e.g. C) i The computing nodes (i) are connected to server E through the i network ports of each server. i (e.g. E) i =i) are connected to the electrical switch. This mapping method is only an example and the present invention is not limited thereto.
[0046] In this embodiment of the invention, the inter-group exchange system may include systems numbered A1, A2, A3, ..., A o ,…,A O O optical switches, A o Let be the o-th optical switch. Each optical switch may include multiple pairs of dual-sided switching ports, for example, n. o For two-sided ports, i.e., each side has n o There are 2n ports in total. o One dual-sided port, n o This represents the number of ports on one side of the o-th optical switch. Different optical switches can have the same or different number of ports on both sides. For example, if an inter-group optical switch has n pairs of ports on both sides, the ports on the left side can be represented as P. L1 ,P L2 ,…,P Li ,…,P Ln The port on the right can be represented as P. R1 ,P R2 ,…,P Ri ,…,P Rn P Li P represents the i-th port on the left. Ri P represents the i-th port on the right. Li -P Ri It consists of a pair of dual-sided ports.
[0047] In this embodiment of the invention, the multiple pairs of dual-sided ports are used to form multiple sets of dual-sided ports through mapping configuration between different pairs of opposite-sided ports. These multiple sets of dual-sided ports are used to provide multiple optical path connections between different pairs of opposite-sided ports. That is, each set of dual-sided ports is formed by opposite-sided ports in different pairs of ports, and an optical path connection is formed between these sets of opposite-sided ports. Each set of dual-sided ports of the same optical switch (i.e., the two opposite-sided ports of each optical path connection) can be connected to the uplink port of the electrical switch of the switching unit in different groups based on the connection relationship between the optical switch port and the electrical switch port. This mapping configuration enables the formation of inter-group optical path switching channels between different groups of switching units. As an example, inside the optical switch, if P... Li -P Ri If P represents a pair of two-sided ports, then Li -P RjA pair of dual-sided ports (i≠j) can be used to represent a group of ports consisting of port i on the left and port j on the right. These dual-sided ports can form optical path connections, but ports on the same side cannot be connected. At any given time, an optical switch with n port groups can provide n optical path connections. To facilitate flexible mapping configuration of multiple pairs of dual-sided ports, in this embodiment of the invention, preferably, the connection relationship between the optical switch ports and the electrical switch ports can be predetermined or determined in real time. This connection relationship can include: the same pair of dual-sided ports of the same optical switch can be connected to different uplink ports of the electrical switch within the same group of switching units. More specifically, the same pair of dual-sided ports of the same optical switch can be connected to different uplink ports of the same electrical switch within the same group of switching units, or to uplink ports of different electrical switches within the same group of switching units. In some embodiments of the invention, in addition to the switching ports connected to the electrical switches, the optical switch can also be provided with redundant or idle switching ports. The specific number is not limited by the invention, as long as it can meet the connection requirements with the electrical switches.
[0048] Figure 3 The image shows an example of the internal optical path connection of an optical switch in an embodiment of the present invention. Figure 3 This demonstrates an optical switch with n=5 ports, comprising 2n dual-sided ports, 5 ports on the left and 5 on the right. The left and right ports are matched into 5 pairs of dual-sided ports by numbering: P L1 -P R1 P L2 -P R2 P L3 -P R3 P L4 -P R4 P L5 -P R5 These five pairs of dual-sided ports, through mapping configurations of different pairs of opposite-sided ports, form five sets of dual-sided ports for providing optical path connections. More specifically, the port mapping configuration is as follows: these five sets of dual-sided ports are mapped and configured as P L1 -P R2 P L2 -P R3 P L3 -P R4 P L4 -P R5 P L5 -P R1The optical switch is configured with five optical paths: port 1 on the left and port 2 on the right, forming a single-fiber bidirectional optical module for data transmission with the same wavelength as the electrical packet switch within the same group. The optical paths of the optical switch are flexible and reconfigurable. To facilitate flexible mapping configuration of multiple pairs of dual-sided ports, the following optical paths are provided: port 2 on the left and port 3 on the right; port 3 on the left and port 4 on the right; port 4 on the left and port 5 on the right; and port 5 on the left and port 1 on the right. A total of five optical paths are provided. Figure 3 In the context of this optical switch, the same pair of dual-sided ports, such as P... L1 and P R1 It can be connected to different uplink ports of the same switch in the same group or to uplink ports of different switches.
[0049] In some embodiments of the present invention, the total number of pairs of ports on both sides of the optical switch in the inter-group switching system can be configured to be equal to N*(N-1) / 2, where N is the number of intra-group switching units in the intra-group switching system. In this case, an optical path connection can be established between any two intra-group switching units. At this time, the number of optical switches in the inter-group switching system satisfies the following formula:
[0050]
[0051] Where O represents the number of optical switches, and n o Let be the number of pairs of ports on both sides of the o-th optical switch.
[0052] Figure 4 This invention provides an example of a novel optoelectronic hybrid interconnect system. For example... Figure 4 As shown, the number of switching units in the group is N=5, and the number of ports on each optical switch is equal, n=n. o =5, o=1,2,…,O, the number of optical switches satisfies:
[0053]
[0054] If all optical switches have the same number of ports on one side, the number of optical switches can be calculated as follows:
[0055] In this scenario, an optical path connection is established between any two intra-group switching units via an optical switch, providing an inter-group optical switching channel for any two intra-group switching units. Specifically, Figure 4In the two optical switches, the first pair of dual-side ports of each optical switch are connected to four different uplink ports of the electrical switch in switching unit G1 within the group. Similarly, the second, third, fourth, and fifth pairs of dual-side ports of each optical switch are connected to four different uplink ports of the electrical switches in switching units G2, G3, G4, and G5 within the group. A one-to-one mapping relationship (mapping configuration) is established between the different pairs of opposite-side ports of the two optical switches. For example, if one optical switch forms five pairs of dual-side ports, P... L1 -P R2 P L2 -P R3 P L3 -P R4 P L4 -P R5 P L5 -P R1 Another optical switch forms 5 sets of dual-sided ports, namely P L1 -P R3 P L2 -P R4 P L3 -P R5 P L4 -P R1 P L5 -P R2 An optical path connection is established between any two intra-group switching units via an optical switch, such as between intra-group switching units G1 and G2 via an optical switch. Figure 4 A set of ports P of the optical switch on the left side of the middle L1 -P R2 An optical path connection was established, and the switching units G1 and G3 within the group communicated via... Figure 4 A set of ports P of the optical switch on the right side of the middle L1 -P R3 Optical path connections have been established. As can be clearly seen in the diagram, optical path connections have also been established between each other in other groups through optical switches. These will not be listed one by one here.
[0056] In other embodiments of the present invention, the total number of pairs of ports on both sides of the optical switches in the inter-group switching system can be configured to be greater than N*(N-1) / 2. In this case, one or more optical path connections can be established between any two intra-group switching units, which can provide a larger inter-group optical circuit switching bandwidth. In this case, the number of optical switches in the inter-group switching system satisfies the following formula:
[0057]
[0058] When the number of ports on each side of each optical switch is equal (n) o =n), the number of optical switches can be calculated as:
[0059] Figure 5 An example of a novel optoelectronic hybrid interconnect system provided in this application. For example... Figure 5 As shown, the number of switching units in the group is N=3, and the number of ports on each optical switch is equal, n=n. o =3, the number of optical switches O=2, that is, it satisfies Right now
[0060] Figure 5 In the two optical switches, the first pair of dual-side ports of each optical switch are connected to four different uplink ports of the electrical switch in the group switching unit G1, respectively. Similarly, the second and third pairs of dual-side ports of each optical switch are connected to four different uplink ports of the electrical switches in the group switching units G2 and G3, respectively. Through the one-to-one mapping relationship (mapping configuration) established between the different pairs of opposite-side ports of the dual-side ports of the two optical switches, the three pairs of dual-side ports formed by each of the two optical switches are P L1 -P R2 P L2 -P R3 P L3 -P R1 and P L1 -P R3 P L2 -P R1 P L3 -P R2 Each optical switch's mapped grouped dual-sided ports establish one optical path connection between any two switching units within the group. Therefore, two optical switches' mapped grouped dual-sided ports establish two optical path connections between any two switching units within the group, providing more data transmission channel options for cross-group computing nodes. These transmission channels can be used simultaneously or partially, and can be adapted for different services by switching optical paths within the switch. In another embodiment of the invention, different numbers of optical path connections can be established between different switching units within the group based on service requirements, thus establishing more or fewer optical path connections between specific switching units within the group. For example, only one optical path connection can be established between switching units G1 and G2 within the group, while three optical path connections can be established between switching units G2 and G3 within the group. This is merely an example, and the invention is not limited thereto.
[0061] In other embodiments of the present invention, the total number of pairs of ports on both sides of the optical switches in the inter-group switching system can be configured to be less than N*(N-1) / 2. In this case, optical path connections cannot be established simultaneously between any two intra-group switching units, but optical path connections can be established between selected intra-group switching units, and the flexible switching capability of the optical switches can be used to construct adapted optical paths for services. In this case, the number of optical switches in the inter-group switching system satisfies the following formula: o = 1, 2, ..., O.
[0062] Figure 6 An example of a novel optoelectronic hybrid interconnect system provided in this application. For example... Figure 6 As shown, the number of switching units in the group is N=5, and the number of ports on each optical switch is equal, n=n. o =5, the number of optical switches satisfies:
[0063]
[0064] At this point, the number of optical switches can be calculated as follows: That is, O = 1;
[0065] Where, n o =n=5,o=1.
[0066] Figure 6 The first to fifth pairs of dual-side ports of the optical switch are respectively connected to two different uplink ports of the electrical switches in the group's switching units G1 to G5, through... Figure 6 The left-side mapping configuration establishes a one-to-one mapping relationship between different pairs of opposite-side ports on both sides of the optical switch. This optical switch forms 5 groups of dual-side ports P. L1 -P R2 P L2 -P R3 P L3 -P R1 P L4 -P R5 and P L5 -P R1 Optical path connections were established between two intra-group switching units G1 and G2, G2 and G3, G3 and G4, G4 and G5, and G5 and G1, respectively. This means optical path connections were established between specific intra-group switching units, not between any two intra-group switching units. In this case, the flexible switching capability of Optical Switching System (OCS) can be utilized to construct adapted optical paths for services between other intra-group switching units through internal optical path switching within the switch, enabling service transmission between other intra-group switching units. For example, by... Figure 6 Change the mapping configuration on the left side to Figure 6The mapping configuration shown on the right is used for optical path switching. The modified mapping configuration forms 5 sets of dual-side ports P L1 -P R3 P L2 -P R4 P L3 -P R1 P L4 -P R5 and P L5 -P R2 Optical paths can be established between selected intra-group switching units G1 and G3, G2 and G4, G3 and G1, G4 and G5, and G5 and G2, respectively, to meet the service transmission requirements between these intra-group switching units.
[0067] In an inter-group switching system with multiple optical switches, the number of switching ports on different optical switches can be the same or different. Figure 4-6 This is an example of different optical switches in the system having the same number of switching ports. Figure 7 This is an example of different optical switches in the system having different numbers of switching ports.
[0068] like Figure 7 As shown, the number of switching units in the group is N=5, and the system contains 3 optical switches (O=3). The number of ports on one side of the optical switches is n1=n2=3, n3=4. The number of optical switches satisfies the following conditions. Appendix Figure 7 The case where the number of ports on the optical switch is different is given. Figure 7 It can be seen that there is an optical path connection between any two different intra-group switching units, which can provide an inter-group optical path switching channel for the corresponding intra-group switching units.
[0069] In embodiments of the present invention, regardless of whether according to In a hybrid optoelectronic interconnect system, optical switches are deployed according to... In a hybrid optoelectronic interconnect system, optical switches can be deployed, and the mapping configuration of the opposite ports of the optical switches can be flexibly switched to meet the transmission needs of different switching units within the same group.
[0070] Addressing the characteristics of intelligent computing centers—low north-south traffic and high east-west traffic—optical switches, as aggregation layer switches, offer more suitable transmission channels for data transmission in intelligent computing services due to their high bandwidth and low latency. Because intelligent computing services exhibit certain periodicity, burstiness, and discontinuity, when the actual number of optical path connections provided by the optical switch is less than the basic connection requirements (e.g., only one optical path connection exists between any group's switching units), the optical paths can be reconfigured in time according to service requirements through the mapping configuration of the opposite-side ports of the optical switch, thereby finding suitable optical path connections for the service. Furthermore, when the number of possible optical path connections provided by the optical switch exceeds the basic connection requirements between groups, the reconfigurability of the optical switch can be utilized to provide more connections than needed between two group's switching units, thus providing greater transmission bandwidth.
[0071] The optoelectronic hybrid interconnection system for large-scale intelligent computing centers described above is designed based on the data interaction requirements of large-scale model training operations. An intra-group switching unit includes interconnections within computing nodes and interconnections between servers within the group. Intra-server computing node interconnections improve the data processing and communication efficiency within a single server; inter-server interconnections ensure efficient data exchange between servers within the same group. The inter-group switching system is responsible for data flow and coordination between different groups, forming an overall efficient, scalable, and stable network architecture that supports large-scale cluster computing and complex data processing needs.
[0072] Furthermore, this invention addresses the needs for flexibility and construction costs. In terms of flexibility, the intra-group switching system's electrical switches provide flexible switching methods between computing nodes within the same group. Simultaneously, the inter-group optical switches provide high-capacity, low-latency data transmission channels between computing nodes across groups. Optical paths can be flexibly established using a service-oriented dynamic resource scheduling method, providing adaptable connectivity and switching capabilities for services. Regarding construction costs, the deployed optical switches have the capability for smooth upgrades. When the number of optical path connections provided by the actual deployed optical switches exceeds the basic optical path connection requirement, this invention can provide a larger capacity transmission channel between intra-group switching units; when the number of optical path connections provided by the actual deployed optical switches is less than the basic optical path connection requirement, the flexible and switchable capabilities of the optical switches can be used to build adaptable optical paths for services. Compared to other data center network interconnection systems, this invention has lower construction costs and greater advantages.
[0073] Those skilled in the art will understand that the exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether 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 invention. When implemented in hardware, it can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the desired tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave.
[0074] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.
[0075] In this invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or in place of features of other embodiments.
[0076] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations of the embodiments of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. An optoelectronic hybrid interconnect system for a large-scale cognitive center, characterized by, The system comprises: an intra-group switching system comprising a plurality of intra-group switching units, and an inter-group switching system comprising at least one optical switch; Each intra-group switching unit comprises a first number of electrical switches and at least one server, each server having at least a first number of computing nodes corresponding to the first number of electrical switches, wherein the first number of computing nodes are in communication with each other, and each electrical switch is connected to a corresponding computing node of the first number of computing nodes through a downlink port; Each optical switch in the inter-group switching system comprises a plurality of pairs of bilateral ports, the plurality of pairs of bilateral ports being configured to provide a plurality of optical path connections between different pairs of opposite ports through mapping between the different pairs of opposite ports, and two opposite ports of each optical path connection of the same switch are capable of connecting uplink ports of electrical switches of different intra-group switching units based on a connection relationship between optical switch ports and electrical switch ports, and the mapping configuration enables the formation of inter-group optical path switching channels between different intra-group switching units.
2. The system of claim 1, wherein, The first number of computing nodes of the servers of the intra-group switching units are in communication with each other through high-speed interconnection technology.
3. The system of claim 1, wherein, The intra-group switching system and the inter-group switching system are interconnected for data transmission using a single-fiber bidirectional optical module.
4. The system of claim 1, wherein, The number of downlink ports of each electrical switch in the intra-group switching unit is greater than or equal to the number of servers in the corresponding intra-group switching unit.
5. The system of claim 1, wherein, The electrical switches and the computing nodes inside the servers of the intra-group switching units are corresponded through a mapping relationship between electrical switch numbers and computing node numbers.
6. The system of claim 1, wherein, The connection relationship between the optical switch ports and the electrical switch ports comprises: the same pair of bilateral ports in the optical switch are connected to different uplink ports of the electrical switches in the same intra-group switching unit, respectively. The mapping configuration comprises establishing a one-to-one mapping relationship between different pairs of opposite ports of the bilateral ports.
7. The system of claim 1, wherein, The total number of pairs of bilateral ports of the optical switches of the inter-group switching system is greater than or equal to a second number, and at least one optical path connection is established between any two intra-group switching units, wherein the second number is N*(N-1) / 2, and N is the number of intra-group switching units in the intra-group switching system. The number of optical switches in the inter-group switching system satisfies the following formula: ; wherein, O is the number of optical switches, is the number of optical switches, o is the number of optical switches.
8. The system of claim 1, wherein, The total number of pairs of bilateral ports of the optical switches of the inter-group switching system is less than the second number, and optical path connections are established between selected intra-group switching units, wherein the second number is N*(N-1) / 2, and N is the number of intra-group switching units in the intra-group switching system. The number of optical switches in the inter-group switching system satisfies the following formula: ; wherein, O is the number of optical switches, is the number of optical switches, o is the number of pairs of bilateral ports in the optical switch.
9. The system of claim 7 or 8, wherein, The mapping configuration of the opposite ports of the optical switches of the inter-group switching system switches for different intra-group switching unit service transmission requirements, thereby realizing the switching of connected optical paths.
10. The system of claim 1, wherein The computing nodes are configured to send and receive data information requiring interaction. The electrical switch of the same group interchanging unit is used to receive data information sent from the corresponding computing node and forward to the connected optical switch of the inter-group switching system, or used to receive data information sent from the connected optical switch of the inter-group switching system and forward to the corresponding computing node; The optical switch is used to receive data information from the electrical switch and forward the received data information between different group interchanging units.
Citation Information
Patent Citations
High performance computing interconnection network system and communication method
CN106789750A
Optical network system
US20210176543A1