Network system and network configuration method

Abstract

Disclosed in embodiments of the present application are a network system and a network configuration method. Unlike conventional wiring modes in which a same port group of a switch must be connected in pairs to a same optical interface group of an optical switch, in a novel wiring mode provided in the embodiments of the present application, a sending port and a receiving port in a same port group of a switch are respectively connected to an input optical interface of one of two different optical switches and an output optical interface of the other optical switch, and each of the two optical switches is responsible for one unidirectional link of the same port group of the switch, which is different from the conventional wiring modes in which one optical switch is responsible for a bidirectional link of the same port group of the switch. In the present application, the wiring mode between the switch and the optical switches is more flexible, is no longer limited to the requirement that the same port group of the switch must be connected in pairs to the same optical interface group of the optical switch, and helps prevent certain ports of the switch from also being in an idle state, thereby improving the utilization rate of ports and network resources.

Description

A network system and network configuration method

[0001] This application claims priority to Russian patent application filed on November 15, 2024, with application number RU2024134227 and entitled "A network system and a method for configuring a network", the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of communications, and more particularly to a network system and a network configuration method. Background Technology

[0003] With the rapid development of large-scale artificial intelligence (AI) models, the computing power demand for AI training is growing exponentially, leading to a growing need for larger-scale computing clusters with higher throughput and lower latency. Optical cross-connect (OXC) devices are increasingly being used in high-performance computing clusters due to their ultra-low forwarding latency and ultra-high throughput. In existing common large-scale hybrid optical-electrical AI clusters, OXC devices are often deployed between cluster basic units (PODs), responsible for topology matching and reconstruction between AI tasks. This provides more flexible and adaptable network resources for different AI tasks, thereby improving task efficiency and performance.

[0004] In existing large-scale optoelectronic hybrid AI cluster architectures, Points of Destination (PODs) are interconnected via OXC devices. This interconnection has the following characteristics: on the top-level electrical switch within a POD responsible for external POD connections, the transmitting and receiving sides of each port are connected to the input and output ports of a pair of ports on the OXC device. Therefore, the OXC device needs to establish a bidirectional connection when establishing a link to enable communication between the two connected electrical switch ports. This architecture and interconnection method are widely used in large-scale clusters. However, in practical applications, port fragmentation inevitably occurs in the network. That is, during AI task scheduling, even with sufficient computing resources, some ports on the electrical switches remain idle. Because they cannot establish links with other PODs' electrical switches via the OXC device, port and network resources are wasted, affecting the deployment and performance of the entire cluster's AI tasks. Summary of the Invention

[0005] This application provides a network system and network configuration method, which makes the cabling between switches and optical switches more flexible and helps to improve the utilization of ports and network resources.

[0006] In a first aspect, embodiments of this application provide a network system, including a first switch, a first optical switch, and a second optical switch. The first switch includes a first port group, which includes a first transmitting port and a first receiving port. The first optical switch includes a first optical interface group, which includes a first input optical interface and a first output optical interface. The second optical switch includes a second optical interface group, which includes a second input optical interface and a second output optical interface. The first transmitting port is used to connect to the first input optical interface, and the first receiving port is used to connect to the second output optical interface.

[0007] This embodiment provides a novel cabling method for network systems. Unlike traditional cabling methods that require the same port group of a switch to be paired with the same optical interface group of an optical switch, in the novel cabling method provided in this application embodiment, the transmitting port and receiving port of the same port group of a switch are respectively used to connect the input optical interface and output optical interface of two different optical switches. Essentially, each of these two optical switches is responsible for a unidirectional link of the same port group of the switch, unlike the traditional cabling method where one optical switch is responsible for a bidirectional link of the same port group of the switch. It can be seen that using the novel cabling method provided in this application embodiment, the cabling between switches and optical switches is more flexible, no longer limited to the requirement that the same port group of a switch must be paired with the same optical interface group of an optical switch. In practical application scenarios, this is more conducive to avoiding some ports of the switch being idle, thereby improving the utilization rate of port and network resources.

[0008] In some possible implementations, the first switch includes a second port group, which includes a second transmitting port and a second receiving port. The second transmitting port is used to connect to a second input optical interface, and the second receiving port is used to connect to a first output optical interface. Based on the one unidirectional link associated with the first port group of the first switch provided in the previous implementation, this implementation further expands the number of port groups of the first switch and provides one unidirectional link associated with the second port group of the first switch. Both unidirectional links are implemented through the first optical interface group and the second optical interface group, so that the first switch can establish topologies with more other switches.

[0009] In some possible implementations, the network system further includes a second switch, a third optical switch, and a fourth optical switch. The first and second switches belong to a first trunking unit (POD), the first and second optical switches belong to a first optical switch group, and the third and fourth optical switches belong to a second optical switch group. The second switch includes a third port group, which includes a third transmit port and a third receive port. The third optical switch includes a third optical interface group, which includes a third input optical interface and a third output optical interface. The fourth optical switch includes a fourth optical interface group, which includes a fourth input optical interface and a fourth output optical interface. The third transmit port is used to connect to the third input optical interface, and the third receive port is used to connect to the fourth output optical interface. In this implementation, the number of switches in a POD can be multiple. For example, the first POD includes a first switch and a second switch. The two switches within the POD can establish topologies through two different optical switch groups, thereby supporting the establishment of multiple links between PODs and facilitating adaptation to more application scenarios based on actual traffic requirements.

[0010] In some possible implementations, the second switch includes a fourth port group, which includes a fourth transmit port and a fourth receive port. The fourth transmit port is used to connect to a fourth input optical interface, and the fourth receive port is used to connect to a third output optical interface. Based on the one unidirectional link associated with the third port group of the second switch provided in the previous implementation, this implementation further expands the number of port groups of the second switch and provides one unidirectional link associated with the fourth port group of the second switch. Both unidirectional links are implemented through the third optical interface group and the fourth optical interface group, so that the second switch can establish topologies with more other switches.

[0011] In some possible implementations, the network system further includes a third switch, which belongs to the second POD. The third switch includes a fifth port group, which includes a fifth transmitting port and a fifth receiving port. The first optical switch includes a fifth optical interface group, which includes a fifth input optical interface and a fifth output optical interface. The second optical switch includes a sixth optical interface group, which includes a sixth input optical interface and a sixth output optical interface. The fifth transmitting port is used to connect to the sixth input optical interface, the fifth receiving port is used to connect to the fifth output optical interface, the first input optical interface is used to connect to the fifth output optical interface, and the sixth input optical interface is used to connect to the second output optical interface. This provides a specific method for establishing a topology between PODs. Because the cabling method between switches and optical switches is more flexible, the method for establishing a topology between PODs is also more flexible, which is beneficial to improving the utilization of port and network resources.

[0012] In some possible implementations, the network system also includes a controller. The controller is used to: determine a first topology connecting the first switch to the third switch via a first optical switch based on the network system's first traffic requirements; determine a second topology connecting the first switch to the third switch via a second optical switch based on the first topology; and configure the first switch, the third switch, the first optical switch, and the second optical switch based on the first and second topologies. A topology reconfiguration method adapted to new cabling methods is provided here. This method uses the physical topology composed of one optical switch from each optical switch group to solve for links equivalent to half of the original link requirements. Then, specific rules are used to establish the physical topology for the other optical switch in each optical switch group, thereby completing the network topology connections. Essentially, the controller only needs to calculate the required half of the topology based on the traffic requirements, and then obtain the other half of the topology through mirror mapping based on the already calculated half, which helps reduce the computational cost of topology construction for the controller.

[0013] In some possible implementations, the controller is specifically used to: acquire the topology reconstruction method for transforming the network system from the original topology to the target topology, wherein the target topology includes a first topology and a second topology, and the original topology is determined by the controller based on the second traffic demand of the network system, which is the traffic demand of the network system prior to the first traffic demand; and configure the first switch, the third switch, the first optical switch, and the second optical switch according to the topology reconstruction method. It should be understood that since the traffic demand between PODs changes in real-time in actual scenarios, the topology established between PODs must also change accordingly as the traffic demand changes. Using the topology reconstruction method provided by this implementation, only the links that differ between the target topology and the original topology need to be adjusted, which can minimize link changes and reduce the impact on running services.

[0014] In some possible implementations, the network system further includes a third switch and a fourth switch, which belong to the second POD. The third switch includes a fifth port group, which includes a fifth transmitting port and a fifth receiving port. The fourth switch includes a sixth port group, which includes a sixth transmitting port and a sixth receiving port. The first optical switch includes a fifth optical interface group, which includes a fifth input optical interface and a fifth output optical interface. The second optical switch includes a sixth optical interface group, which includes a sixth input optical interface and a sixth output optical interface. The third optical switch includes a seventh optical interface group, which includes a seventh input optical interface and a seventh output optical interface. The second optical switch includes an eighth optical interface group, which includes an eighth input optical interface and an eighth output optical interface. The fifth transmitting port is used to connect to the sixth input optical interface, the fifth receiving port is used to connect to the fifth output optical interface, the first input optical interface is used to connect to the fifth output optical interface, and the sixth input optical interface is used to connect to the second output optical interface. The sixth transmitting port is used to connect to the eighth input optical interface, the sixth receiving port is used to connect to the seventh output optical interface, the third input optical interface is used to connect to the seventh output optical interface, and the eighth input optical interface is used to connect to the fourth output optical interface. In this implementation, there can also be multiple bidirectional links between the two PODs, that is, the traffic between the two PODs can be distributed among multiple sets of electrical switches, thereby achieving load balancing. This is beneficial for adapting to more application scenarios according to actual traffic requirements and also helps to improve the reliability of service transmission.

[0015] In some possible implementations, the network system also includes a controller. The controller is used to: determine a first topology connecting a first switch to a third switch via a first optical switch and a third topology connecting a second switch to a fourth switch via a third optical switch, based on the network system's first traffic requirements; determine a second topology connecting the first switch to the third switch via a second optical switch based on the first topology; and determine a fourth topology connecting the second switch to the fourth switch via a fourth optical switch based on the third topology; and configure the first, second, third, and fourth optical switches, as well as the first, second, third, and fourth optical switches, based on the first, second, third, and fourth topologies. In other words, the controller only needs to calculate half of the required topology based on the traffic requirements, and then obtain the other half of the topology through mirroring, which helps reduce the computational cost of topology construction for the controller.

[0016] In some possible implementations, the controller is specifically used to: first configure the first switch, the third switch, the first optical switch, and the second optical switch according to the first and second topologies, and then configure the second switch, the fourth switch, the third optical switch, and the fourth optical switch according to the third and fourth topologies. It should be understood that since there can be multiple bidirectional links between PODs, meaning that traffic demands between PODs can be evenly distributed across different electrical switches, a step-by-step topology reconfiguration method is provided here. First, the topology related to one optical switch group is adjusted, while the topology related to another optical switch group remains unchanged. After the topology adjustment of one optical switch group is completed, the topology related to the other optical switch group is adjusted. In this way, the entire link adjustment process will not cause a complete interruption of services between PODs, improving the reliability of service transmission during topology reconfiguration.

[0017] In some possible implementations, the controller is specifically used to: acquire the topology reconstruction method for transforming the network system from the original topology to the target topology, wherein the target topology includes a first topology, a second topology, a third topology, and a fourth topology, and the original topology is determined by the controller based on the second traffic demand of the network system, which is the traffic demand of the network system prior to the first traffic demand; and configure the first switch, the second switch, the third switch, the fourth switch, the first optical switch, the second optical switch, the third optical switch, and the fourth optical switch according to the topology reconstruction method. It should be understood that since the traffic demand between PODs changes in real-time in actual scenarios, the topology established between PODs must also change accordingly as the traffic demand changes. Using the topology reconstruction method provided by this implementation, only the links that differ between the target topology and the original topology need to be adjusted, which can minimize link changes and reduce the impact on running services.

[0018] In some possible implementations, the first switch is an electrical switch, and the first optical switch and the second optical switch are optical cross-connect (OXC) devices, which can be used in a network architecture of a hybrid optoelectronic AI cluster with good performance.

[0019] In some possible implementations, the network system further includes an optical module, a first transmitting port for connecting to a first input optical interface via the optical module, and a first receiving port for connecting to a second output optical interface via the optical module.

[0020] Secondly, embodiments of this application provide a network configuration method applied to a network system. The network system includes a first switch, a first optical switch, a second optical switch, and a controller. The first switch includes a first port group, which includes a first transmitting port and a first receiving port. The first optical switch includes a first optical interface group, which includes a first input optical interface and a first output optical interface. The second optical switch includes a second optical interface group, which includes a second input optical interface and a second output optical interface. The first transmitting port is used to connect to the first input optical interface, and the first receiving port is used to connect to the second output optical interface. The method includes: determining, through the controller, a first connection method for the first switch to connect to the first optical switch and the second optical switch, and configuring the first switch, the first optical switch, and the second optical switch according to the first connection method.

[0021] In some possible implementations, the network system further includes a second switch, a third optical switch, and a fourth optical switch. The first and second switches belong to a first POD (Programmable Node), the first and second optical switches belong to a first optical switch group, and the third and fourth optical switches belong to a second optical switch group. The second switch includes a third port group, which includes a third transmitting port and a third receiving port. The third optical switch includes a third optical interface group, which includes a third input optical interface and a third output optical interface. The fourth optical switch includes a fourth optical interface group, which includes a fourth input optical interface and a fourth output optical interface. The third transmitting port is used to connect to the third input optical interface, and the third receiving port is used to connect to the fourth output optical interface. The method further includes: determining, through a controller, a second connection method for the second switch to connect to the third and fourth optical switches, and configuring the second, third, and fourth optical switches according to the second connection method.

[0022] In some possible implementations, the network system further includes a third switch, which belongs to the second POD. The third switch includes a fifth port group, which includes a fifth transmitting port and a fifth receiving port. The first optical switch includes a fifth optical interface group, which includes a fifth input optical interface and a fifth output optical interface. The second optical switch includes a sixth optical interface group, which includes a sixth input optical interface and a sixth output optical interface. The fifth transmitting port is used to connect to the sixth input optical interface, the fifth receiving port is used to connect to the fifth output optical interface, the first input optical interface is used to connect to the fifth output optical interface, and the sixth input optical interface is used to connect to the second output optical interface. The method further includes: determining a first topology for the connection between the first switch and the third switch via the first optical switch based on the first traffic requirements of the network system using a controller; determining a second topology for the connection between the first switch and the third switch via the second optical switch based on the first topology using a controller; and configuring the first switch, the third switch, the first optical switch, and the second optical switch based on the first and second topologies using a controller.

[0023] In some possible implementations, configuring the first switch, third switch, first optical switch, and second optical switch by the controller according to the first and second topologies includes: obtaining a topology reconstruction method for transforming the network system from the original topology to the target topology by the controller, wherein the target topology includes the first and second topologies, the original topology is determined by the controller according to the second traffic demand of the network system, the second traffic demand being the traffic demand of the network system prior to the first traffic demand; minimizing the number of links adjusted by the topology reconstruction method; and configuring the first switch, third switch, first optical switch, and second optical switch by the controller according to the topology reconstruction method.

[0024] In some possible implementations, the network system further includes a third switch and a fourth switch, which belong to the second POD. The third switch includes a fifth port group, which includes a fifth transmitting port and a fifth receiving port; the fourth switch includes a sixth port group, which includes a sixth transmitting port and a sixth receiving port. The first optical switch includes a fifth optical interface group, which includes a fifth input optical interface and a fifth output optical interface. The second optical switch includes a sixth optical interface group, which includes a sixth input optical interface and a sixth output optical interface. The third optical switch includes a seventh optical interface group, which includes a seventh input optical interface and a seventh output optical interface. The second optical switch includes an eighth optical interface group, which includes an eighth input optical interface and an eighth output optical interface. The fifth transmitting port is used to connect to the sixth input optical interface, the fifth receiving port is used to connect to the fifth output optical interface, the first input optical interface is used to connect to the fifth output optical interface, and the sixth input optical interface is used to connect to the second output optical interface. The sixth transmitting port is used to connect to the eighth input optical interface, the sixth receiving port is used to connect to the seventh output optical interface, the third input optical interface is used to connect to the seventh output optical interface, and the eighth input optical interface is used to connect to the fourth output optical interface. The method further includes: determining, by the controller, a first topology connecting the first switch to the third switch via the first optical switch and a third topology connecting the second switch to the fourth switch via the third optical switch, based on the first traffic demand of the network system; determining, by the controller, a second topology connecting the first switch to the third switch via the second optical switch based on the first topology, and a fourth topology connecting the second switch to the fourth switch via the fourth optical switch based on the third topology; and configuring the first switch, the second switch, the third switch, the fourth switch, the first optical switch, the second optical switch, the third optical switch, and the fourth optical switch via the controller based on the first topology, the second topology, the third topology, and the fourth topology.

[0025] In some possible implementations, configuring the first switch, second switch, third switch, fourth switch, first optical switch, second optical switch, third optical switch, and fourth optical switch by the controller according to the first topology, second topology, third topology, and fourth topology includes: configuring the first switch, third switch, first optical switch, and second optical switch by the controller according to the first topology and second topology, and then configuring the second switch, fourth switch, third optical switch, and fourth optical switch according to the third topology and fourth topology.

[0026] In some possible implementations, configuring the first switch, second switch, third switch, fourth switch, first optical switch, second optical switch, third optical switch, and fourth optical switch by the controller according to the first topology, second topology, third topology, and fourth topology includes: obtaining the topology reconstruction method of the network system from the original topology to the target topology by the controller, wherein the target topology includes the first topology, second topology, third topology, and fourth topology, and the original topology is determined by the controller according to the second traffic demand of the network system, the second traffic demand being the traffic demand of the network system prior to the first traffic demand; and configuring the first switch, second switch, third switch, fourth switch, first optical switch, second optical switch, third optical switch, and fourth optical switch by the controller according to the topology reconstruction method.

[0027] This application provides a novel cabling method for network systems. Unlike traditional cabling methods that require the same port group of a switch to be paired with the same optical interface group of an optical switch, in this novel cabling method, the transmitting and receiving ports of the same port group of a switch are used to connect the input and output optical interfaces of two different optical switches, respectively. Essentially, each of these two optical switches is responsible for a unidirectional link of the same port group, unlike the traditional cabling method where one optical switch is responsible for a bidirectional link of the same port group. It can be seen that the cabling method provided by this application provides greater flexibility in the cabling between switches and optical switches, no longer limiting the requirement that the same port group of a switch must be paired with the same optical interface group of an optical switch. In practical applications, this is more conducive to preventing some ports of the switch from being idle, thereby improving the utilization rate of ports and network resources. Attached Figure Description

[0028] Figure 1 is a schematic diagram of a leaf-ridge network architecture in an embodiment of this application;

[0029] Figure 2 is a schematic diagram of a photoelectric hybrid AI cluster architecture;

[0030] Figure 3 is a schematic diagram of the architecture of a traditional optoelectronic hybrid network system;

[0031] Figure 4 is a schematic diagram of a port fragmentation phenomenon based on the architecture shown in Figure 3;

[0032] Figure 5 is a schematic diagram of an application scenario of the network system in an embodiment of this application;

[0033] Figure 6 is a schematic diagram of the first implementation of the network system in this application;

[0034] Figure 7 is a schematic diagram of a second implementation of the network system in this application;

[0035] Figure 8 is a schematic diagram of a third implementation of the network system in this application;

[0036] Figure 9 is a schematic diagram of the fourth implementation of the network system in this application;

[0037] Figure 10 is a schematic diagram of the fifth implementation of the network system in this application;

[0038] Figure 11 is a schematic diagram of the sixth implementation of the network system in this application;

[0039] Figure 12 is a schematic diagram of the seventh implementation of the network system in this application;

[0040] Figure 13 is a flowchart illustrating a network configuration method in an embodiment of this application;

[0041] Figure 14 is a comparative diagram of the original topology and the target topology of the network system in an embodiment of this application;

[0042] Figure 15 is a schematic diagram of one implementation of the network system transforming from the original topology to the target topology in an embodiment of this application. Detailed Implementation

[0043] This application provides a network system and network configuration method. Unlike traditional cabling methods that require identical port groups of electrical switches to be paired with identical optical interface groups of optical switches, the novel cabling method provided in this application uses the transmitting and receiving ports of identical port groups of electrical switches to connect to the input and output optical interfaces of two different optical switches, respectively. Essentially, each of these two optical switches is responsible for a unidirectional link of the same port group of the electrical switch. Using this novel cabling method, the cabling between electrical switches and optical switches is more flexible, no longer limited to identical port groups of electrical switches being paired with identical optical interface groups of optical switches. In practical applications, this is more conducive to preventing some ports of electrical switches from being idle, thereby improving the utilization rate of port and network resources.

[0044] It should be noted that the terms "first," "second," etc., in this application specification, claims, and the accompanying drawings are used to distinguish similar objects, not to limit a specific order or sequence. It should be understood that the above terms can be used interchangeably where appropriate so that the embodiments described in this application can be implemented in an order other than that described in this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units 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 devices. Additionally, the terms "system" and "network" are often used interchangeably herein. The term "and / or" in this document merely describes the relationship between related objects, indicating that three relationships can exist; for example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, the character " / " in this document generally indicates that the preceding and following related objects are in an "or" relationship. It should be understood that in the embodiments of this application, "B corresponding to A" means that B is associated with A, and B can be determined based on A. However, it should also be understood that determining B based on A does not mean that B is determined solely based on A; B can also be determined based on A and / or other information.

[0045] To facilitate understanding, the technical terms involved in the embodiments of this application will be explained below.

[0046] (1) Spine-leaf network architecture: Figure 1 is a schematic diagram of a spine-leaf network architecture in an embodiment of this application. As shown in Figure 1, the spine-leaf network architecture is a data center network topology consisting of two layers of switches: the spine layer and the leaf layer. The leaf layer consists of access switches, which aggregate traffic from servers and directly connect to the spine layer switches or the network core. The switches in the leaf layer can also be called top-of-rack (TOR) switches.

[0047] (2) Cluster Basic Unit (POD): A cluster basic unit is essentially a group of computers linked together by a high-speed network to form a whole. Various networking methods can be used within a POD, such as the Spine-leaf architecture shown above. PODs can also be connected to each other via electrical or optical networks.

[0048] (3) Optical cross connect (OXC): Optical cross connect is an optical cross device based on port-level flexible scheduling, which can exchange optical signals between different optical paths.

[0049] Figure 2 is a schematic diagram of a hybrid optoelectronic AI cluster architecture. As shown in Figure 2, in common large-scale optoelectronic hybrid artificial intelligence (AI) clusters, optical switches are often deployed between PODs (Programmable Devices) to perform topology matching and reconstruction between AI tasks. This provides more flexible and adaptable network resources for different AI tasks, thereby improving task efficiency and performance. Since the number of ports on an optical switch is limited, to increase the scale of the AI ​​cluster, multiple optical switches are often needed to be connected in planes or groups. The topmost electrical switch group within each POD needs to be meshed with the optical switches. The number of ports on the optical switches and the number of connections between each optical switch and each POD determine the scale of the cluster.

[0050] Figure 3 is a schematic diagram of the architecture of a traditional optoelectronic hybrid network system. As shown in Figure 3, for the top-level spine switch group within a POD responsible for connecting external PODs, the transmitting and receiving sides of each port are often connected to the input and output ports of a pair of OXC devices. Therefore, when establishing a link, the OXC device also needs to establish a bidirectional connection to enable communication between the connected spine switch ports. Taking Figure 3 as an example, each POD has two spine switches, which can divide the four OXC devices into two groups. The spine1 switch in each POD is connected to the OXC devices in Group 1, and the spine2 switch in each POD is connected to the OXC devices in Group 2. The number of OXC devices in each group is equal to the number of uplink ports on each spine switch. Each spine switch is connected to each OXC device in its corresponding group via a pair of bidirectional optical fibers. That is, the transmitting side of each spine switch port is connected to the input of the OXC device, and the receiving side of each spine switch port is connected to the output of the OXC device. Taking the example that port 1 of the Spine1 switch in Pod2 needs to communicate with port 1 of the Spine1 switch in Pod3, two cross paths as shown in Figure 3 need to be built inside the OXC1 device.

[0051] As shown in Figure 3, the traditional architecture and connection method are widely used in large-scale clusters. However, in actual applications, port fragmentation will inevitably occur in the network. That is, during AI task scheduling, when there are sufficient computing resources, some ports of the switch are also idle. However, since they cannot establish links with other POD switches through OXC devices, port and network resources are wasted, which also affects the deployment and performance of AI tasks in the entire cluster.

[0052] Figure 4 illustrates a port fragmentation phenomenon based on the architecture shown in Figure 3. As shown in Figure 4, the same numbered spine switch ports within different Pods are connected to the same OXC device. Therefore, only switch ports with the same number can establish a path through the OXC device. In this network topology, suppose an AI task arrives that requires Ring communication between three Pods. Therefore, at least one bidirectional link needs to be established between any two Pods in the three Pods for the task to be deployed and executed. However, it is found that the required topology cannot be constructed at this point. Assuming that Pod1 and Pod2 establish a link through port 1 of the spine switch, and Pod2 and Pod3 establish a link through port 2 of the spine switch, the remaining ports of the spine switch in Pod1 are connected to the OXC1 device, and the remaining ports of the spine switch in Pod3 are connected to the OXC2 device. In this case, no further link can be established between Pod1 and Pod3. The same applies to other situations. The reason is that when the number of Pods that need to establish a link is odd, the number of ports that can be allocated in the OXC device is also odd. Since each time a bidirectional link is established, two ports of the OXC device are required, there will inevitably be an odd number of ports left in the OXC device that cannot be used to establish a connection.

[0053] As described above, traditional large-scale AI cluster optoelectronic hybrid networks suffer from unavoidable network resource fragmentation, affecting the performance and resource utilization of the AI ​​cluster network. This application proposes a novel network system that employs a newly designed cabling method and provides a network resource reconfiguration scheme adapted to the new network system, solving the aforementioned problems while maintaining cluster scale.

[0054] Figure 5 is a schematic diagram of an application scenario of the network system in this application embodiment. As shown in Figure 5, the network system provided in this application embodiment can be applied to large-scale, high-performance distributed AI computing reconfigurable optical network clusters. AI model training is often multi-card training or even larger-scale distributed training. Distributed training allocates different data / models to each accelerator card, completes the AI ​​model calculation on multiple cards within the cluster, and then fuses them within / between nodes.

[0055] The compute node shown in Figure 5 is a server used to process computational tasks. The compute node server is responsible for executing parallel computing tasks, where each node can independently perform computational tasks and share data and results with other nodes.

[0056] The switching network shown in Figure 5 is the cornerstone of the network platform, providing data exchange services for communication between servers. Switches can be classified into electrical switches and optical switches according to their working principle. Electrical switches can be further classified according to network structure, such as spine layer switches and leaf layer switches (TOR switches). Types of optical switches include, but are not limited to, microelectromechanical systems (MEMS), OXC devices, and high-speed optical switches.

[0057] The network controller shown in Figure 5 is a scheduling center that acts as a control surface to schedule network resources for the entire system. The network controller can also be called a controller or scheduler. It stores information such as the system's network topology and resource occupancy status. When an AI task arrives, it allocates available network resources according to the task's requirements and a specific resource scheduling algorithm, and reconstructs a suitable physical topology by scheduling the switching network.

[0058] In some possible scenarios, the electrical switching network in each POD shown in Figure 5 can also be replaced with an optical switching network. That is, the electrical switches in the electrical switching network can be replaced with optical switches. For ease of explanation, the following description will use the optoelectronic hybrid network scenario shown in Figure 5 as an example. If the electrical switches are replaced with optical switches, the cabling method and topology configuration method of the network system provided in the embodiments of this application will still be adopted.

[0059] It should be noted that the electrical switches in an electrical switching network are specifically connected to the optical switches in an optical switching network via optical modules. That is, a port on the electrical switch connects to a port on one side of the optical module, and a port on the other side of the optical module connects to the optical interface of the optical switch. Specifically, the optical module converts electrical signals from the electrical switch into optical signals, which are then transmitted to the optical switch via optical fiber. The optical module also converts optical signals from the optical switch back into electrical signals, which are then transmitted to the electrical switch. As an example, electrical switches can be connected to optical modules in a one-to-one correspondence. As another example, multiple electrical switches can be connected to the same optical module; for example, the 400G ports of two electrical switches can each be connected to the 800G port of one optical module.

[0060] It should be understood that, for ease of illustrating specific wiring methods, the accompanying drawings for the following embodiments no longer show optical modules, but directly demonstrate the connection relationship between electrical switches and optical switches. Based on this, those skilled in the art can understand how electrical switches are connected to optical switches via optical modules. It should also be understood that ports on electrical switches can also be called electrical interfaces, and ports on optical switches can also be called optical interfaces. For ease of description, the terms "ports on electrical switches" and "optical interfaces on optical switches" will be used consistently below. In the embodiments of this application, a pair of transmitting and receiving ports on an electrical switch is referred to as a port group, and a pair of input and output optical interfaces on an optical switch is referred to as an optical interface group. Specifically, the transmitting port on the electrical switch is used to connect to the input optical interface on the optical switch, and the receiving port on the electrical switch is used to connect to the output optical interface on the optical switch. The transmitting port on the electrical switch can also be called an output port, the receiving port on the electrical switch can also be called an input port, the output optical interface on the optical switch can also be called a transmitting optical interface, and the input optical interface on the optical switch can also be called a receiving optical interface.

[0061] The network system provided in the embodiments of this application will be described in detail below.

[0062] Figure 6 is a schematic diagram of the first embodiment of the network system in this application. As shown in Figure 6, the basic structure of the network system consisting of one electrical switch and two optical switches will be described first. For port group 1 of electrical switch 1, port group 1 includes a transmitting port 1-1 and a receiving port 1-2. The transmitting port 1-1 of electrical switch 1 is used to connect to the input optical interface 1-1 in the optical interface group 1 of optical switch 1, and the receiving port 1-2 of electrical switch 1 is used to connect to the output optical interface 1-2 in the optical interface group 1 of optical switch 2. It should be understood that the embodiments of this application do not limit the number of port groups on each electrical switch. For example, electrical switch 1 also includes port group 2, which includes a transmitting port 2-1 and a receiving port 2-2. The transmitting port 2-1 of electrical switch 1 is used to connect to the input optical interface 1-1 in the optical interface group 1 of optical switch 2, and the receiving port 2-2 of electrical switch 1 is used to connect to the output optical interface 1-2 in the optical interface group 1 of optical switch 1.

[0063] As can be seen, unlike traditional cabling methods that require the same port group of an electrical switch to be paired and connected to the same optical interface group of an optical switch, in the novel cabling method provided in this application embodiment, the transmitting port and receiving port of the same port group of an electrical switch are respectively used to connect the input optical interface and output optical interface of two different optical switches. This is equivalent to each of these two optical switches being responsible for a unidirectional link of the same port group of the electrical switch, unlike the traditional cabling method where one optical switch is responsible for a bidirectional link of the same port group of the electrical switch. In this application embodiment, the two optical switches used to connect the transmitting port and receiving port of the same port group of the electrical switch respectively can be referred to as one optical switch group.

[0064] It should be noted that the embodiments of this application do not limit the number of electrical switches in each POD. The following description takes two electrical switches in a POD as an example. If the POD contains more than two electrical switches, those skilled in the art can flexibly expand it based on the novel wiring method provided in the embodiments of this application.

[0065] Figure 7 is a schematic diagram of a second implementation of the network system in this application. As shown in Figure 7, POD1 includes electrical switch 1 and electrical switch 2. The wiring method of electrical switch 1 can be referred to the relevant description of the embodiment shown in Figure 6, and will not be repeated here. For port group 1 of electrical switch 2, the transmitting port 1-1 of electrical switch 2 is used to connect to the input optical interface 1-1 in the optical interface group 1 of optical switch 3, and the receiving port 1-2 of electrical switch 2 is used to connect to the output optical interface 1-2 in the optical interface group 1 of optical switch 4; for port group 2 of electrical switch 2, the transmitting port 2-1 of electrical switch 2 is used to connect to the input optical interface 1-1 in the optical interface group 1 of optical switch 4, and the receiving port 2-2 of electrical switch 2 is used to connect to the output optical interface 1-2 in the optical interface group 1 of optical switch 3. It can be seen that the two optical switches in optical switch group 1 are used to connect to electrical switch 1 in POD1, and the two optical switches in optical switch group 2 are used to connect to electrical switch 2 in POD1. That is, different electrical switches within the same POD are used to connect to different optical switch groups respectively.

[0066] It should be noted that the embodiments of this application do not limit the number of port groups on each power switch. The following description uses an extension of four port groups on the power switch for further explanation. If the number of port groups on the power switch is greater, those skilled in the art can flexibly expand it based on the novel cabling method provided in the embodiments of this application.

[0067] Figure 8 is a schematic diagram of the third implementation of the network system in this application. As shown in Figure 8, the electrical switch 1 includes four port groups: port group 1, port group 2, port group 3, and port group 4. The wiring methods for port groups 1 and 2 on the electrical switch 1 can be referred to the relevant description in the embodiment shown in Figure 6, and will not be repeated here. For port group 3 of the electrical switch 1, the transmitting port 3-1 of the electrical switch 1 is used to connect to the input optical interface 1-1 in the optical interface group 1 of the optical switch 5, and the receiving port 3-2 of the electrical switch 1 is used to connect to the output optical interface 1-2 in the optical interface group 1 of the optical switch 6; for port group 4 of the electrical switch 1, the transmitting port 4-1 of the electrical switch 1 is used to connect to the input optical interface 1-1 in the optical interface group 1 of the optical switch 6, and the receiving port 4-2 of the electrical switch 1 is used to connect to the output optical interface 1-2 in the optical interface group 1 of the optical switch 5. It can be seen that port groups 1 and 2 on the electrical switch 1 are used to connect two optical switches in optical switch group 1, and port groups 3 and 4 on the electrical switch 1 are used to connect two optical switches in optical switch group 3. In this embodiment of the application, all optical switches connected to the same electrical switch can be referred to as one optical switch set. For example, as shown in Figure 8, optical switch set 1 includes optical switch group 1 and optical switch group 3.

[0068] It should be noted that the number of optical interface groups required on each optical switch in the optical switch group depends on the number of electrical switches that each optical switch needs to connect to, and is not limited here. The following is a further description using the example of each optical switch needing to connect to 3 electrical switches. If each optical switch needs to connect to more than 3 electrical switches, those skilled in the art can flexibly expand based on the novel cabling method provided in the embodiments of this application.

[0069] Figure 9 is a schematic diagram of the fourth implementation of the network system in this application. As shown in Figure 9, each of the two optical switches in optical switch group 1 includes three port groups. POD1 includes electrical switch 1 and electrical switch 2, POD2 includes electrical switch 3 and electrical switch 4, and POD3 includes electrical switch 5 and electrical switch 6. The wiring method for port groups 1 and 2 on electrical switch 1 can be referred to the relevant description in the embodiment shown in Figure 6, and will not be repeated here. For port group 1 of electrical switch 3, the transmitting port 1-1 of electrical switch 3 is used to connect to the input optical interface 2-1 in the optical interface group 2 of optical switch 1, and the receiving port 1-2 of electrical switch 3 is used to connect to the output optical interface 2-2 in the optical interface group 2 of optical switch 2; for port group 2 of electrical switch 3, the transmitting port 2-1 of electrical switch 3 is used to connect to the input optical interface 2-1 in the optical interface group 2 of optical switch 2, and the receiving port 2-2 of electrical switch 3 is used to connect to the output optical interface 2-2 in the optical interface group 2 of optical switch 1. For port group 1 of electrical switch 5, the transmitting port 1-1 of electrical switch 5 is used to connect to the input optical interface 3-1 in the optical interface group 3 of optical switch 1, and the receiving port 1-2 of electrical switch 5 is used to connect to the output optical interface 3-2 in the optical interface group 3 of optical switch 2. For port group 2 of electrical switch 5, the transmitting port 2-1 of electrical switch 5 is used to connect to the input optical interface 3-1 in the optical interface group 3 of optical switch 2, and the receiving port 2-2 of electrical switch 5 is used to connect to the output optical interface 3-2 in the optical interface group 3 of optical switch 1. Following a similar wiring method, the wiring methods for optical switch 3 and optical switch 4 in optical switch group 2 to connect to electrical switch 2 in POD1, electrical switch 4 in POD2, and electrical switch 6 in POD3 can also be obtained, which will not be elaborated here.

[0070] Based on the above descriptions of the embodiments, a general cabling method is provided below for the network system in this application embodiment. For example, the network system includes P Pods, K optical switches, each Pod containing T electrical switches, and each electrical switch having M port groups. The K optical switches can be divided into... There are several optical switch groups, each containing two optical switches. The two optical switches in each group are numbered 1 and 2. Furthermore, all... The optical switch group is divided into T sets of optical switches, and each set of optical switches contains Each optical switch group is numbered from 1 to 1 within the optical switch set. Here we use O i,j,k Let $k$ be the optical switch in the $j$-th optical switch group within the $i$-th optical switch set. k∈{1,2}, K=T×M. Using S… p,iS represents the i-th power switch in the p-th Pod. p,i,m,k S represents p,i The k-th port of the m-th port group, where... k∈{1,2}

[0071] For the electrical switch S p,i For all ports m, Implement the following connection method: If k=1, then connect the transmitting port of port group m to port O. i,m,1 Connect the input optical interface of the p-th optical interface group to the receive port of the m-port group and connect it to O. i,m,2 Connect the output optical interface of the p-th optical interface group; if k=2, then connect the receiving port of the m-port group to O. i,m,1 Connect the output optical interface of the p-th optical interface group to the transmit port of the m-port group and connect it to O. i,m,2 The input optical interface of the p-th optical interface group is connected.

[0072] It should be noted that the new wiring method provided in this application embodiment allows for more flexible wiring between electrical switches and optical switches. It is no longer limited to the same port group of the electrical switch being connected in pairs to the same optical interface group of the optical switch. In practical application scenarios, this is more conducive to avoiding some ports of the electrical switch being idle, thereby improving the utilization rate of ports and network resources.

[0073] The following description is based on specific embodiments. It should be understood that the topology connection between PODs provided in the embodiments shown below, including Figure 10, is only a possible example. In actual application scenarios, the number of PODs, the number of electrical switches in the PODs, and the ports of the electrical switches and the optical interfaces of the optical switches used for the links established between the PODs all depend on the actual needs. This application does not limit the specific requirements.

[0074] Figure 10 is a schematic diagram of the fifth implementation of the network system in this application. As shown in Figure 10, taking a network system including 3 PODs, and each POD including 1 power switch as an example, according to the AI ​​task requirements, a Ring topology needs to be built between the 3 pods, that is, a topology needs to be established between POD1 and POD2, between POD2 and POD3, and between POD3 and POD1. For ease of explanation, the topology established between PODs is represented by "→" below.

[0075] The topology established between electrical switch 1 in POD1 and electrical switch 3 in POD2 is shown by the gray arrow in Figure 10. Specifically, the transmission port 1-1 of electrical switch 1 → the input port 1-1 of optical switch 1 → the output port 2-2 of optical switch 1 → the receiving port 2-2 of electrical switch 3, and the transmission port 2-1 of electrical switch 3 → the input port 2-1 of optical switch 2 → the output port 1-2 of optical switch 2 → the receiving port 1-2 of electrical switch 1.

[0076] The topology established between electrical switch 3 in POD2 and electrical switch 5 in POD3 is shown by the black arrow in Figure 10. Specifically, the transmitting port 1-1 of electrical switch 3 → the input port 2-1 of optical switch 1 → the output port 3-2 of optical switch 1 → the receiving port 2-2 of electrical switch 5, and the transmitting port 2-1 of electrical switch 5 → the input port 3-1 of optical switch 2 → the output port 2-2 of optical switch 2 → the receiving port 1-2 of electrical switch 3.

[0077] The topology established between electrical switch 1 in POD1 and electrical switch 5 in POD3 is shown by the dashed arrow in Figure 10. Specifically, the transmission port 2-1 of electrical switch 1 → the input port 1-1 of optical switch 2 → the output port 3-2 of optical switch 2 → the receiving port 1-2 of electrical switch 5, and the transmission port 1-1 of electrical switch 5 → the input port 3-1 of optical switch 1 → the output port 1-2 of optical switch 1 → the receiving port 2-2 of electrical switch 1.

[0078] As can be seen from the embodiment shown in Figure 10, each optical switch only needs to be responsible for one unidirectional link between two PODs. At this time, the task can be scheduled and deployed normally, and there will be no waste of port resources due to network link failure, task stagnation waiting for more resources to be released, thereby causing a decrease in the overall utilization and performance of the cluster.

[0079] It should be noted that, since the new cabling method provided in this application embodiment changes the traditional port-to-port bidirectional link method, this application embodiment also proposes a topology reconfiguration method adapted to the new cabling method, which helps reduce the computational cost of the controller in constructing the topology. The core idea of ​​this topology reconfiguration method is as follows: First, based on specific rules, the traffic demand matrix is ​​preprocessed to make its link requirements legal. Then, the physical topology composed of one optical switch in each optical switch group is used to solve for links equivalent to half of the original link requirements. Finally, specific rules are used to establish the physical topology for the other optical switch in each optical switch group, thereby completing the network topology connections. Specific embodiments are described below.

[0080] Figure 11 is a schematic diagram of the sixth implementation of the network system in this application. As shown in Figure 11, taking the establishment of a topology between electrical switch 1 in POD1 and electrical switch 3 in POD2 as an example, the controller can first calculate the topology of one unidirectional link between POD1 and POD2 as shown by the solid arrow in Figure 11 based on the traffic demand between POD1 and POD2, namely, the transmitting port 1-1 of electrical switch 1 → the input port 1-1 of optical switch 1 → the output port 2-2 of optical switch 1 → the receiving port 1-2 of electrical switch 3. Next, the controller performs mirror mapping based on the topology shown by the solid arrow in Figure 11 to obtain the topology of another unidirectional link between POD1 and POD2 as shown by the dashed arrow in Figure 11, namely, the transmitting port 1-1 of electrical switch 3 → the input port 2-1 of optical switch 2 → the output port 1-2 of optical switch 2 → the receiving port 1-2 of electrical switch 1, thereby determining the topology of a bidirectional link between POD1 and POD2. Then, the controller can control electrical switch 1, electrical switch 3, optical switch 1, and optical switch 2 to configure them according to the determined topology of a single bidirectional link, thereby establishing the physical topology between POD1 and POD2. It should be understood that the embodiments of this application do not limit the actual application to calculating the topology of the unidirectional link related to which optical switch in the optical switch group first. For example, as shown in Figure 11, the topology of the unidirectional link related to optical switch 1 can be calculated first, and then extended to obtain the topology of the unidirectional link related to optical switch 2. Alternatively, the topology of the unidirectional link related to optical switch 2 can be calculated first, and then extended to obtain the topology of the unidirectional link related to optical switch 2.

[0081] It should be noted that the embodiment shown in Figure 11 above illustrates a scenario with one bidirectional link between two PODs. In practical applications, there can also be multiple bidirectional links between two PODs, meaning that traffic between the two PODs can be distributed among multiple sets of electrical switches to achieve load balancing. In other words, each of the two PODs requires multiple electrical switches to establish a topology. As an example, the controller determines that eight bidirectional links can be established between the two PODs based on the traffic demand. Assuming each POD includes two electrical switches, four bidirectional links can be established between one set of electrical switches in the two PODs, and the other four bidirectional links can be established between the other set of electrical switches in the two PODs. It should be understood that in practical applications, the specific number of bidirectional links between the two PODs, the number of electrical switches in each POD requiring bidirectional links, and the specific number of bidirectional links for each electrical switch in the POD all depend on actual needs and are not limited here. For ease of explanation, an embodiment with two bidirectional links between two PODs for load balancing is provided below.

[0082] Figure 12 is a schematic diagram of the seventh implementation of the network system in this application. As shown in Figure 12, taking the establishment of a topology between electrical switch 1 in POD1 and electrical switch 3 in POD2, and the establishment of a topology between electrical switch 2 in POD1 and electrical switch 4 in POD2 as an example, the controller can first calculate the topology of two unidirectional links between POD1 and POD2 as shown by the solid arrows in Figure 12 based on the traffic demand between POD1 and POD2, namely, the transmitting port 1-1 of electrical switch 1 → the input port 1-1 of optical switch 1 → the output port 2-2 of optical switch 1 → the receiving port 1-2 of electrical switch 3, and the transmitting port 1-1 of electrical switch 2 → the input port 1-1 of optical switch 3 → the output port 2-2 of optical switch 3 → the receiving port 1-2 of electrical switch 4. Next, the controller performs a mirror mapping based on the topology shown by the solid arrows in Figure 12 to obtain the topology of the other two unidirectional links between POD1 and POD2, as shown by the dashed arrows in Figure 12. Specifically, the topology is: transmit port 1-1 of electrical switch 3 → input port 2-1 of optical switch 2 → output port 1-2 of optical switch 2 → receive port 1-2 of electrical switch 1; and transmit port 1-1 of electrical switch 4 → input port 2-1 of optical switch 4 → output port 1-2 of optical switch 4 → receive port 1-2 of electrical switch 2. This determines the topology of the two bidirectional links between POD1 and POD2. Furthermore, the controller can control electrical switches 1, 2, 3, and 4, as well as optical switches 1, 2, 3, and 4, to configure them according to the determined topology of the two bidirectional links, thus establishing the physical topology between POD1 and POD2.

[0083] It should be understood that, for scenarios similar to Figure 12, where each POD includes multiple electrical switches and each electrical switch includes multiple port groups, the topology between POD1 and POD2 shown in Figure 12 only provides one possible implementation. In practical applications, a topology different from that shown in Figure 12 can also be established between POD1 and POD2. On one hand, this application embodiment does not limit which two electrical switches establish a bidirectional link between two PODs. For example, a bidirectional link could be established between electrical switch 1 in POD1 and electrical switch 4 in POD2 via optical switch group 1, and between electrical switch 2 in POD1 and electrical switch 3 in POD2 via optical switch group 2. Other possible embodiments are not listed here. On the other hand, this application embodiment also does not limit which two port groups of the two electrical switches each bidirectional link is established between. For example, a bidirectional link could be established between port group 1 of electrical switch 1 and port group 2 of electrical switch 3, and between port group 1 of electrical switch 2 and port group 2 of electrical switch 4. Other possible embodiments are not listed here. On the other hand, the embodiments of this application do not limit the number of bidirectional links established between two electrical switches. The specific number depends on traffic demand and the number of available port groups for each electrical switch. For example, two bidirectional links can be established between electrical switch 1 in POD1 and electrical switch 3 in POD2, and two bidirectional links can be established between electrical switch 2 in POD1 and electrical switch 4 in POD2. Correspondingly, if extended to a scenario different from Figure 12, where each POD includes multiple electrical switches and each electrical switch includes multiple port groups, the topology between PODs can be flexibly configured according to the description herein.

[0084] It should be noted that, since the traffic demand between PODs changes in real-time in actual scenarios, the topology established between PODs must also change accordingly as the traffic demand changes. In this embodiment, the topology determined based on the current traffic demand is called the original topology, which represents the current connection state between PODs in the network system; the topology determined based on the new traffic demand is called the target topology, which represents the new target connection state between PODs in the network system. There are various implementation methods for the network system to change from the original topology to the target topology, which will be described below.

[0085] In a first possible implementation, after determining the target topology based on the new traffic demand, the controller can first control the relevant devices in the network system to disconnect all links already established in the original topology, and then control the relevant devices in the network system to configure them according to the links in the target topology.

[0086] In a second possible implementation, after determining the target topology based on the new traffic demands, the controller first calculates a topology reconstruction method to transform the original topology into the target topology. This topology reconstruction method aims to minimize the number of links adjusted when transforming from the original to the target topology; the links that differ between the target and original topologies are the links that need adjustment. Subsequently, the controller controls the relevant devices in the network system to configure themselves according to the topology reconstruction method.

[0087] It can be seen that both of the above possible implementation methods can realize the transformation of the network system from the original topology to the target topology. Compared with the first possible implementation method, the second possible implementation method can meet the traffic requirements while minimizing link changes and reducing the impact on running services.

[0088] The complete process of network configuration by the controller is described below with reference to a specific embodiment. Figure 13 is a schematic flowchart of a network configuration method according to an embodiment of this application. As shown in Figure 13, the network configuration method includes the following steps.

[0089] 1. Obtain the network system's traffic demand matrix.

[0090] Specifically, the controller can obtain the network system's traffic demand matrix in various ways. For example, for black-box traffic, the traffic demand matrix can be obtained by performing real-time or periodic traffic detection on the network system. As another example, for white-box traffic, where more information is available, the specific traffic volume between PODs can be calculated using resource scheduling results and model partitioning strategies within the current cluster.

[0091] It's important to note that in real-world scenarios, the input traffic demand may exceed or fall short of the actual bandwidth available on the links, resulting in wasted bandwidth. Therefore, the controller typically needs to preprocess the traffic demand matrix to ensure it meets the constraints of the bandwidth available in the actual topology. For example, if the actual bandwidth available between PODs is 200G, and the traffic demand between PODs is a total of 250G (one 100G bidirectional link and one 150G bidirectional link), then the traffic demand of these two bidirectional links can be proportionally reduced to ensure that the traffic demand between PODs does not exceed 200G. Conversely, if the actual bandwidth available between PODs is 200G, and the traffic demand is a total of 150G (one 100G bidirectional link and one 50G bidirectional link), then the traffic demand of these two bidirectional links can be proportionally increased to ensure that the traffic demand between PODs is close to or equal to 200G.

[0092] Assume the preprocessed traffic demand matrix is ​​C ij, where element c i,j This represents the number of connections required to connect the i-th Pod to the j-th Pod, where element c j,i This represents the number of connections required to connect the j-th Pod to the i-th Pod. The traffic demand matrix then satisfies the following condition:

[0093] 2. Determine the link requirements for half of the traffic demand matrix.

[0094] Since the traffic of each electrical switch is handled by one optical switch group, when considering half of the link demand of each optical switch group, it is only necessary to satisfy half of the process demand matrix. Here, matrix A is used. ij Let C represent the traffic demand matrix. ij Half of the link requirements, using Let C represent the traffic demand matrix. ij The other half of the link requirement should specifically meet the following conditions:

[0095] 3. Calculate the topology associated with one of the optical switches in each optical switch group.

[0096] Specifically, taking the example of first calculating the topology related to the odd-numbered optical switches in each optical switch group, let the binary variables... This indicates the current state of the t-th power switch in the i-th Pod. Whether the sending port in the port group is connected to the j-th Pod through the first optical switch of the k-th optical switch group, or the t-th electrical switch of the j-th Pod. The receiving port in each port group, i.e., whether a link exists. A value of 1 indicates existence, and a value of 0 indicates non-existence; in this case, k is odd. Let... This indicates the target connection state of the network system and the current connection state of the network system. Correspondingly, A and u are known quantities, and x is an unknown quantity. The following conditions should be met to adapt to traffic requirements:

[0097] In some possible scenarios, consider distributing traffic between Pods evenly across power switches for load balancing. Then, The following constraints also need to be satisfied:

[0098] The goal of this topology solution model is to minimize the number of link adjustments, reducing the impact on existing tasks. In other words, by designing an objective function to calculate how to transform from the current connection state to the target connection state, we can determine the specific topology reconfiguration scheme based on the traffic demand matrix, the current connection state, and the target connection state, ensuring that the number of link adjustments is minimized. Therefore, the objective function is:

[0099] The reconstructed physical topology can be obtained using the above modeling scheme. There are several ways to solve this modeling scheme. As an example, a direct solution approach can be used, such as employing open-source or commercial mathematical solvers. Another example is to transform the modeling into a minimum cost flow problem. For instance, if the original integer programming problem only contains two odd-numbered optical switches, then the problem is equivalent to a minimum cost flow (MCF) problem. Since there are multiple optical switches in reality, it can be recursively decomposed into individual optical switches and then solved using the MCF problem. This method significantly improves the solution speed compared to a direct solution.

[0100] 4. Map to obtain the topology related to another optical switch in each optical switch group.

[0101] Based on the topology associated with one optical switch in each optical switch group obtained in step 3, the topology associated with the other optical switch in each optical switch group can be obtained through mirroring. Specifically, for all i,j,k,t, if This indicates that links need to be constructed using optical switches with odd-numbered addresses within each optical switch group. Therefore, it is necessary to directly construct links through the even-numbered optical switches in each optical switch group. Recorded as

[0102] It should be noted that after the controller determines the topology reconstruction method from the original topology to the target topology, if the entire topology reconstruction is executed at once, it is possible that all links between certain nodes of a certain task will be interrupted for a period of time, forcing the task to be interrupted. Therefore, this application embodiment also provides a step-by-step topology reconstruction method, which can minimize the interruption of running services. It should be understood that this step-by-step topology reconstruction method is applicable to the scenario where load balancing can be achieved between PODs as described in the embodiment shown in Figure 12.

[0103] Specifically, this step-by-step topology reconfiguration method can be divided into T steps, executed sequentially, where step t involves reconfiguring the t-th optical switch group. In step t, the original topology and the target topology are compared, retaining identical link connections and identifying the different links that need to be reconfigured. This topology reconfiguration method ensures that the task will not be interrupted due to the reconfiguration of optical switches when the original traffic matrix meets the following condition: when the inter-POD link requirement exceeds one link. i≠j,C ij >1.

[0104] It should be understood that this step-by-step topology reconfiguration method can guarantee no interruption because the traffic demand between PODs is evenly distributed across different electrical switches. When adjusting the topology related to the t-th optical switch group, only the links of the electrical switch numbered t among all PODs are changed. When adjusting the next optical switch group, the links of the previous optical switch group have already been changed. Therefore, as long as the number of bidirectional links between PODs is greater than 1, it can be guaranteed that the links between PODs will not be interrupted due to the reconfiguration of optical switches. A specific embodiment will be used for further explanation below.

[0105] Figure 14 is a schematic diagram comparing the original topology and the target topology of the network system in an embodiment of this application. As shown in Figure 14, all solid lines represent the original topology between POD1 and POD2, and all dashed lines represent the target topology between POD1 and POD2. Specifically, electrical switch 1 in POD1 establishes a link with electrical switch 3 in POD2 through optical switch group 1, and electrical switch 2 in POD1 establishes a link with electrical switch 4 in POD2 through optical switch group 2. It should be understood that if all solid lines in Figure 14 are transformed into dashed lines at once, all solid lines must be disconnected first, and then the dashed lines must be re-established. During this time, POD1 and POD2 are in a disconnected state, and services cannot communicate and will be forced to stop. However, the step-by-step topology reconstruction method proposed in this embodiment can transform the original topology to the target topology in two steps.

[0106] Figure 15 is a schematic diagram of one implementation method for transforming the network system from the original topology to the target topology in this application embodiment. First, the links related to optical switch group 1 are adjusted. Specifically, all solid-line links related to optical switch group 1 in Figure 14 are interrupted, and then all dashed-line links related to optical switch group 1 in Figure 14 are established, resulting in the target topology related to optical switch group 1 as shown in scenario (a) of Figure 15. During this process, the original topology related to optical switch group 2 is retained. Therefore, the operation in the first step will not cause a complete interruption of services between POD1 and POD2. Second, the links related to optical switch group 2 are adjusted. Specifically, all solid-line links related to optical switch group 2 in Figure 14 are interrupted, and then all dashed-line links related to optical switch group 2 in Figure 14 are established, resulting in the target topology related to optical switch group 2 as shown in scenario (b) of Figure 15. During this process, since the target topology related to optical switch group 1 has already been established, the operation in the second step will also not cause a complete interruption of services between POD1 and POD2. It can be seen that by adopting this step-by-step topology reconstruction method, the service between POD1 and POD2 will not be completely interrupted during the entire process of link adjustment, thus improving the reliability of service transmission during topology reconstruction.

[0107] It should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A network system, characterized in that, The network system includes a first switch, a first optical switch, and a second optical switch; The first switch includes a first port group, which includes a first transmitting port and a first receiving port; the first optical switch includes a first optical interface group, which includes a first input optical interface and a first output optical interface; the second optical switch includes a second optical interface group, which includes a second input optical interface and a second output optical interface; the first transmitting port is used to connect to the first input optical interface, and the first receiving port is used to connect to the second output optical interface.

2. The network system according to claim 1, characterized in that, The first switch includes a second port group, which includes a second transmitting port and a second receiving port; the second transmitting port is used to connect to the second input optical interface, and the second receiving port is used to connect to the first output optical interface.

3. The network system according to claim 1 or 2, characterized in that, The network system also includes a second switch, a third optical switch, and a fourth optical switch; the first switch and the second switch belong to the first cluster basic unit (POD), the first optical switch and the second optical switch belong to the first optical switch group, and the third optical switch and the fourth optical switch belong to the second optical switch group; The second switch includes a third port group, which includes a third transmitting port and a third receiving port; the third optical switch includes a third optical interface group, which includes a third input optical interface and a third output optical interface; the fourth optical switch includes a fourth optical interface group, which includes a fourth input optical interface and a fourth output optical interface; the third transmitting port is used to connect to the third input optical interface, and the third receiving port is used to connect to the fourth output optical interface.

4. The network system according to claim 3, characterized in that, The second switch includes a fourth port group, which includes a fourth transmitting port and a fourth receiving port; the fourth transmitting port is used to connect to the fourth input optical interface, and the fourth receiving port is used to connect to the third output optical interface.

5. The network system according to any one of claims 1 to 4, characterized in that, The network system also includes a third switch, which belongs to the second POD. The third switch includes a fifth port group, which includes a fifth transmitting port and a fifth receiving port. The first optical switch includes a fifth optical interface group, which includes a fifth input optical interface and a fifth output optical interface. The second optical switch includes a sixth optical interface group, which includes a sixth input optical interface and a sixth output optical interface. The fifth transmitting port is used to connect to the sixth input optical interface, the fifth receiving port is used to connect to the fifth output optical interface, the first input optical interface is used to connect to the fifth output optical interface, and the sixth input optical interface is used to connect to the second output optical interface.

6. The network system according to claim 5, characterized in that, The network system further includes a controller, the controller being configured to: A first topology is determined based on the first traffic requirement of the network system, which connects the first switch to the third switch via the first optical switch. Based on the first topology, a second topology is determined in which the first switch is connected to the third switch via the second optical switch; Configure the first switch, the third switch, the first optical switch, and the second optical switch according to the first topology and the second topology.

7. The network system according to claim 6, characterized in that, The controller is specifically used for: The topology reconstruction method of the network system from the original topology to the target topology is obtained, wherein the target topology includes the first topology and the second topology, and the original topology is determined by the controller according to the second traffic requirement of the network system, wherein the second traffic requirement is the traffic requirement of the network system prior to the first traffic requirement; Configure the first switch, the third switch, the first optical switch, and the second optical switch according to the topology reconstruction method.

8. The network system according to claim 3 or 4, characterized in that, The network system further includes a third switch and a fourth switch, which belong to the second POD; the third switch includes a fifth port group, which includes a fifth transmitting port and a fifth receiving port; the fourth switch includes a sixth port group, which includes a sixth transmitting port and a sixth receiving port; the first optical switch includes a fifth optical interface group, which includes a fifth input optical interface and a fifth output optical interface; the second optical switch includes a sixth optical interface group, which includes a sixth input optical interface and a sixth output optical interface; the third optical switch includes a seventh optical interface group, which includes a seventh input optical interface and a seventh output optical interface; the second optical switch includes an eighth optical interface group, which includes an eighth input optical interface and an eighth output optical interface; The fifth transmitting port is used to connect to the sixth input optical interface, the fifth receiving port is used to connect to the fifth output optical interface, the first input optical interface is used to connect to the fifth output optical interface, and the sixth input optical interface is used to connect to the second output optical interface; the sixth transmitting port is used to connect to the eighth input optical interface, the sixth receiving port is used to connect to the seventh output optical interface, the third input optical interface is used to connect to the seventh output optical interface, and the eighth input optical interface is used to connect to the fourth output optical interface.

9. The network system according to claim 8, characterized in that, The network system further includes a controller, the controller being configured to: Based on the first traffic demand of the network system, a first topology is determined whereby the first switch connects to the third switch via the first optical switch, and a third topology is determined whereby the second switch connects to the fourth switch via the third optical switch. Based on the first topology, a second topology is determined to connect the first switch to the third switch via the second optical switch, and based on the third topology, a fourth topology is determined to connect the second switch to the fourth switch via the fourth optical switch. Configure the first switch, the second switch, the third switch, the fourth switch, the first optical switch, the second optical switch, the third optical switch, and the fourth optical switch according to the first topology, the second topology, the third topology, and the fourth topology.

10. The network system according to claim 9, characterized in that, The controller is specifically used for: First, configure the first switch, the third switch, the first optical switch, and the second optical switch according to the first topology and the second topology. Then, configure the second switch, the fourth switch, the third optical switch, and the fourth optical switch according to the third topology and the fourth topology.

11. The network system according to claim 9 or 10, characterized in that, The controller is specifically used for: The topology reconstruction method of the network system from the original topology to the target topology is obtained, wherein the target topology includes the first topology, the second topology, the third topology and the fourth topology, and the original topology is determined by the controller according to the second traffic demand of the network system, wherein the second traffic demand is the traffic demand of the network system prior to the first traffic demand; Configure the first switch, the second switch, the third switch, the fourth switch, the first optical switch, the second optical switch, the third optical switch, and the fourth optical switch according to the topology reconstruction method.

12. The network system according to any one of claims 1 to 11, characterized in that, The first switch is an electrical switch, and the first optical switch and the second optical switch are optical cross-connect (OXC) devices.

13. The network system according to any one of claims 1 to 12, characterized in that, The network system further includes an optical module, wherein the first transmitting port is used to connect to the first input optical interface through the optical module, and the first receiving port is used to connect to the second output optical interface through the optical module.

14. A network configuration method, characterized in that, The network configuration method is applied to a network system, which includes a first switch, a first optical switch, a second optical switch, and a controller. The first switch includes a first port group, which includes a first transmitting port and a first receiving port; the first optical switch includes a first optical interface group, which includes a first input optical interface and a first output optical interface; the second optical switch includes a second optical interface group, which includes a second input optical interface and a second output optical interface. The first transmitting port is used to connect to the first input optical interface, and the first receiving port is used to connect to the second output optical interface; The method includes: determining, through the controller, a first connection method for the first switch to connect the first optical switch and the second optical switch, respectively, and configuring the first switch, the first optical switch and the second optical switch according to the first connection method.

15. The method according to claim 14, characterized in that, The network system also includes a second switch, a third optical switch, and a fourth optical switch; the first switch and the second switch belong to the first cluster basic unit (POD), the first optical switch and the second optical switch belong to the first optical switch group, and the third optical switch and the fourth optical switch belong to the second optical switch group; The second switch includes a third port group, which includes a third transmitting port and a third receiving port; the third optical switch includes a third optical interface group, which includes a third input optical interface and a third output optical interface; the fourth optical switch includes a fourth optical interface group, which includes a fourth input optical interface and a fourth output optical interface. The third transmitting port is used to connect to the third input optical interface, and the third receiving port is used to connect to the fourth output optical interface; The method further includes: determining, through the controller, a second connection method for the second switch to connect to the third optical switch and the fourth optical switch, respectively, and configuring the second switch, the third optical switch and the fourth optical switch according to the second connection method.

16. The method according to claim 14 or 15, characterized in that, The network system also includes a third switch, which belongs to the second POD. The third switch includes a fifth port group, which includes a fifth transmitting port and a fifth receiving port. The first optical switch includes a fifth optical interface group, which includes a fifth input optical interface and a fifth output optical interface. The second optical switch includes a sixth optical interface group, which includes a sixth input optical interface and a sixth output optical interface. The fifth transmitting port is used to connect to the sixth input optical interface, the fifth receiving port is used to connect to the fifth output optical interface, the first input optical interface is used to connect to the fifth output optical interface, and the sixth input optical interface is used to connect to the second output optical interface. The method further includes: The controller determines a first topology connecting the first switch and the third switch via the first optical switch based on the first traffic demand of the network system. The controller determines, based on the first topology, a second topology connecting the first switch to the third switch via the second optical switch; The controller configures the first switch, the third switch, the first optical switch, and the second optical switch according to the first topology and the second topology.

17. The method according to claim 16, characterized in that, The configuration of the first switch, the third switch, the first optical switch, and the second optical switch by the controller according to the first topology and the second topology includes: The controller obtains the topology reconstruction method for transforming the network system from the original topology to the target topology, wherein the target topology includes the first topology and the second topology, and the original topology is determined by the controller based on the second traffic requirement of the network system, which is the traffic requirement of the network system prior to the first traffic requirement; the number of links adjusted through the topology reconstruction method is minimized; The controller configures the first switch, the third switch, the first optical switch, and the second optical switch according to the topology reconstruction method.

18. The method according to claim 15, characterized in that, The network system further includes a third switch and a fourth switch, which belong to the second POD; the third switch includes a fifth port group, which includes a fifth transmitting port and a fifth receiving port; the fourth switch includes a sixth port group, which includes a sixth transmitting port and a sixth receiving port; the first optical switch includes a fifth optical interface group, which includes a fifth input optical interface and a fifth output optical interface; the second optical switch includes a sixth optical interface group, which includes a sixth input optical interface and a sixth output optical interface; the third optical switch includes a seventh optical interface group, which includes a seventh input optical interface and a seventh output optical interface; the second optical switch includes an eighth optical interface group, which includes an eighth input optical interface and an eighth output optical interface; The fifth transmitting port is used to connect to the sixth input optical interface, the fifth receiving port is used to connect to the fifth output optical interface, the first input optical interface is used to connect to the fifth output optical interface, and the sixth input optical interface is used to connect to the second output optical interface; the sixth transmitting port is used to connect to the eighth input optical interface, the sixth receiving port is used to connect to the seventh output optical interface, the third input optical interface is used to connect to the seventh output optical interface, and the eighth input optical interface is used to connect to the fourth output optical interface; The method further includes: The controller determines, based on the first traffic demand of the network system, a first topology connecting the first switch to the third switch via the first optical switch and a third topology connecting the second switch to the fourth switch via the third optical switch; The controller determines a second topology based on the first topology, in which the first switch connects to the third switch via the second optical switch, and determines a fourth topology based on the third topology, in which the second switch connects to the fourth switch via the fourth optical switch. The controller configures the first switch, the second switch, the third switch, the fourth switch, the first optical switch, the second optical switch, the third optical switch, and the fourth optical switch according to the first topology, the second topology, the third topology, and the fourth topology.

19. The method according to claim 18, characterized in that, The configuration of the first switch, the second switch, the third switch, the fourth switch, the first optical switch, the second optical switch, the third optical switch, and the fourth optical switch by the controller according to the first topology, the second topology, the third topology, and the fourth topology includes: The controller first configures the first switch, the third switch, the first optical switch, and the second optical switch according to the first topology and the second topology, and then configures the second switch, the fourth switch, the third optical switch, and the fourth optical switch according to the third topology and the fourth topology.

20. The method according to claim 18 or 19, characterized in that, The configuration of the first switch, the second switch, the third switch, the fourth switch, the first optical switch, the second optical switch, the third optical switch, and the fourth optical switch by the controller according to the first topology, the second topology, the third topology, and the fourth topology includes: The controller obtains the topology reconstruction method of the network system from the original topology to the target topology, wherein the target topology includes the first topology, the second topology, the third topology and the fourth topology, and the original topology is determined by the controller according to the second traffic requirement of the network system, which is the traffic requirement of the network system before the first traffic requirement; The controller configures the first switch, the second switch, the third switch, the fourth switch, the first optical switch, the second optical switch, the third optical switch, and the fourth optical switch according to the topology reconstruction method.

Images