Switching network module for high-radix networks

Abstract

A switched network module for a high radix network has a plurality of multi-fiber connector interfaces, where some of the multi-fiber connector interfaces can be connected to network devices in a network having an internal mesh implemented in two levels using multi-fiber cables. The first hierarchy is configured for rearrangement, and the second hierarchy is configured for recombining individual optical fibers of different groups of optical fibers. The optical paths of each transmitter and receiver are matched so as to provide an appropriate optical connection from the transmitting fiber to the receiving fiber, and any complex network topology can be achieved by at least a reduced 1 / N point-to-point interconnection, where N = the number of channels per multi-fiber connector interface.

Description

Technical Field

[0001] Disclosed is an apparatus and method for improving the scalability of a data center network using a mesh network topology, switches with various cardinalities, hierarchies, and oversubscription ratios. The disclosed apparatus and method reduce the number of manual network connections while simplifying wiring installation, thereby improving the flexibility and reliability of a data center at a reduced cost. Background Art

[0002] The importance of using optical fiber to transmit communication signals has grown rapidly due to its high bandwidth, low attenuation and other significant advantages, including radiation resistance, small size and light weight. Data center architectures using optical fiber are evolving to meet global traffic demands and the increasing number of users and applications. The rise of cloud data centers, especially hyperscale clouds, has significantly changed enterprise information technology (IT) business structures, network systems and topologies. In addition, the requirements of cloud data centers are affecting technology roadmaps and standardization.

[0003] The widespread adoption of server virtualization and advances in data processing and storage technologies have led to the growth of east-west traffic within data centers. The traditional three-layer switch architecture, including the core layer, aggregation layer, and access layer (CAA), cannot provide the low and balanced latency channels required for east-west traffic. In addition, since the CAA architecture utilizes the spanning tree protocol to disable redundant paths and build a loop-free topology, the CAA architecture does not fully utilize network capacity.

[0004] The Folded Clos network (FCN) or Spine-and-Leaf architecture is a topology that is better suited to overcome the limitations of three-tier CAA networks. The Clos network is a multi-stage circuit-switched network proposed by Charles Clos in 1953. Initially, the network was designed to increase the capacity of crossbar switches. Due to the development and adoption of very large scale integrated circuit (VLSI) technology, the Clos network became less applicable. Complex optical interconnect topologies were initially used in high-performance computing (HPC) and later in cloud data centers, making this architecture relevant again. The Folded Clos network topology utilizes two types of switching nodes: spine and leaf. Each spine is connected to each leaf. The network can be scaled horizontally to enable communication between a large number of servers while minimizing latency and non-uniformity by simply adding more spine and leaf switches.

[0005] The FCN depends on the switch radix k, which is the ratio of leaf switch server downlinks to spine switch uplinks, and the number of tiers or layers of the network, m. The choice of (k,m) has a significant impact on the number of switches, the reliability and latency of the network, and the deployment cost of the data center network.Figure 1 The relationship between the number of servers for different levels of oversubscription is shown, assuming that all switches have similar cardinality and the total oversubscription ratio is 1:1.

[0006] Figure 2 An example of two FCNs with a similar number of hosts using different radixes and levels is shown. The higher radix (32 in this example) connects 32 edge switches to 16 spine switches in a two-tier network, as shown in part (a) of the figure. A two-tier FCN provides the lowest latency, but at the expense of requiring a denser network (512 interconnects). By using a three-tier network, the interconnect layout is simplified (256 interconnects). However, this requires more switches and introduces more latency into the network. Over the past few years, the need for flatter networks to handle the growing amount of traffic between machines has favored an increase in the radix of switch application-specific integrated circuits (ASICs). Currently, ASICs can handle 256 radix switches at 100Gb / s per port. These switches support 64x400GbE, 128x200GbE, or 256x100GbE, enabling flat networks with up to three tiers.

[0007] Based on the industry telecommunications infrastructure standard TIA-942-A, the locations of leaf switches and spine switches can be tens or hundreds of meters apart. Typically, spine switches are located in the main distribution area (MDA), while leaf switches are located in the equipment distribution area (EDA) or horizontal distribution area (HDA).

[0008] The architecture has been proven to deliver high bandwidth and low latency (only two hops to reach the destination), providing low oversubscription connectivity. However, for a large number of switches, the spine-and-leaf architecture requires a complex mesh with a large number of fibers and connectors, which increases the cost and complexity of the installation.

[0009] Future data centers will require more flexible and adaptable networks than the traditional grids implemented today to accommodate highly distributed computing, machine learning (ML) training loads, advanced virtualization, and data replication.

[0010] Deploying a new data center or expanding a data center network with hundreds or thousands of servers is not an easy task. Spine-to-leaf switches require a large number of interconnects, such as Figure 3 100. In this example, the switching fabric 100 may have 572 paths. Each line in illustration 120 may represent a group of eight or twelve optical fibers terminated in a multi-fiber MPO connector. The optical fibers may be ribbonized in the form of a conventional flat ribbon or a rollable ribbon. Illustration 110 shows an enlarged view of a small area of ​​the switching fabric 100.

[0011] Interconnected switches of 100 or more can be prone to errors, which in many cases can be exacerbated by tight deployment deadlines or lack of installer training. While spine-and-leaf topologies are resilient to misaligned connections, large interconnect errors can have a noticeable impact through performance degradation, leading to the loss of some server links. Managing large-scale network configurations often requires dedicated personnel to check interconnects, which can cause delays and increase deployment costs.

[0012] As shown in the prior art, the use of transpose boxes can help reduce installation errors. However, the prior art cannot easily adapt to different network topologies, switch bases, or oversubscription levels.

[0013] A new grid method and apparatus is disclosed herein that utilizes a modular, flexible and better organized interconnect mapping that can be quickly and reliably deployed in a data center.

[0014] In US8621111, US2012 / 0250679 A1 and US2014 / 0025843 A1, a method for providing scalability in a data transmission network using a transpose box is disclosed. The box can connect a first layer and a second layer of a network. The box facilitates the deployment of the network. However, a dedicated box is required for the selected network. As described in the application, the network topology determines the type of transpose box to be used. A change in topology may require the replacement of the transpose box. Based on the description, if the number of spine or leaf switches, oversubscription, or other parameters of the network change, a different box needs to be used.

[0015] Once the topology is chosen, the application provides a method for expansion. This requires using cables to connect the ports of one box to the ports of another box. This increases the loss of the network and does not effectively accommodate the expansion of the network.

[0016] This approach disclosed in US2014 / 0025843 A1 may be well suited for large data centers that have already chosen the type of network architecture to implement and can prepare and maintain an inventory of different types of transpose boxes as needed. Wider deployment of mesh networks in data centers requires a more flexible or modular approach.

[0017] In W2019099771A1, an interconnect box is disclosed. This application shows an exemplary wiring for connecting various spine and leaf switches using a 1RU module that can be mounted on a rack. The ports of these modules are internally connected using internal multi-fiber cables containing a specific grid. However, the module seems to be tuned to a specific topology, such as providing a grid between four spine and leaf switch ports. The application does not describe how the device can be used for topologies with a variable number of leaf switches or spine switches or with a variable number of ports.

[0018] In US20150295655A, an optical interconnect assembly is described that uses multiple leaf-side multiplexers and demultiplexers on each side of the network, one on the spine side and the other near the leaf side. Each multiplexer and demultiplexer is configured to work together in a desired topology. However, the application does not demonstrate the flexibility and scalability of this approach.

[0019] US11269152 describes a method to circumvent the limitations of optical shuffle boxes, which, according to the application, are not easily adaptable to reconfiguration or expansion of a switch network. The application describes an apparatus and method for patching a network link using multiple patch panels. At least two chassis are required to connect switches from one layer of the network to another. Each rack is capable of accommodating multiple modules, such as boxes arranged in a vertical configuration. The connection from the first-level switch to one side of the module is made using a branch cable. One side of the branch cable is terminated with an MPO (24 fibers) and the other side is terminated with an LC or other duplex connector. One side of the module has one or two MPO ports, and an additional six duplex LC connectors or newer ultra-small form factor (VSFF) connectors.

[0020] Similarly, the second-tier switches are connected to modules in other racks. The wiring required to connect the switches is performed using a plurality of jumper assemblies configured to connect to a plurality of optical modules. The jumpers are specially designed to fix their relative positions as they must maintain the correct (linear) sequence. US11269152 describes a patching method and it can make the network more scalable in terms of network base. However, network deployment remains challenging and prone to interconnection errors. Utility Model Content

[0021] A device having a plurality of multi-fiber connector interfaces, wherein some of these multi-fiber connector interfaces are capable of being connected to network devices in a network using multi-fiber cables, the network having an internal grid implemented in two levels. The first level is configured for rearrangement and the second level is configured for recombination of individual fibers of different fiber groups. The optical paths of each transmitter and receiver are matched to provide appropriate optical connections from the transmitting fiber to the receiving fiber, and complex arbitrary network topologies can be implemented by reducing the number of point-to-point interconnections by at least 1 / N, where N = the number of channels per multi-fiber connector interface. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 The number of servers is shown as a function of the switch radix and the number of switch layers of the network.

[0023] Figure 2 An example of two FCNs with similar number of hosts using different cardinalities and levels is shown.

[0024] Figure 3 The interconnection of an example grid containing 576 interconnects, each having 12 or 8 optical fibers, is shown.

[0025] FIG. 4( a ) shows a front view of the disclosed module 400 .

[0026] FIG. 4( b ) shows a rear view of the module 400 .

[0027] Figure 5 A top view of module 400 is shown.

[0028] Figure 6 The interconnection of the modules 400 is shown.

[0029] Figure 7 An interconnection scheme from input ports to output ports of region 305 is illustrated.

[0030] Figure 8 is a top view of the submodule 500 showing the interconnection arrangement.

[0031] Fig. 9 Eight possible configurations are shown that can be implemented in the submodules.

[0032] Fig.10 The diagram illustrates a simple method of using module 400 to implement a network with 16 leaf switches and up to 16 spine switches.

[0033] Fig.11 (a) shows an example of the interconnection between the ridge port and the module 400.

[0034] Fig.11 (b) shows Fig.11 (a) Example interconnection table.

[0035] Fig.12 (a) shows an example of the interconnection between the ports of the module 400 and the spine chassis ports (sixteen spines, each spine having two line cards).

[0036] Fig.12 (b) shows an example of interconnection between ports of module 400 and spine chassis ports (eight spines with four line cards per spine).

[0037] Fig.12 (c) shows an example of interconnection between ports of module 400 and spine chassis ports (four spines with eight line cards each).

[0038] Fig.12(d) shows an example of interconnection between ports of module 400 and spine chassis ports (two spines, with each spine having sixteen line cards).

[0039] Fig.13 The figure shows a method for implementing a two-level FCN.

[0040] Fig.14 The figure shows a method for implementing a two-level FCN.

[0041] Fig.15 The figure shows a method for implementing a two-level FCN.

[0042] Fig.16 The figure shows a method for implementing a two-level FCN.

[0043] Fig.17 The figure shows a method for implementing a two-level FCN. DETAILED DESCRIPTION

[0044] This document discloses a modular apparatus and general method for deploying an optical network having multiple tiers and cardinalities. The modules and methods can be used with standalone, stacked, or rack network switches as long as the modular connections utilize 16 MPO connectors. In particular, switches having Ethernet-specified SR or DR transceivers in ports, such as 40GBASE-SR8, 800GBASE-SR8, or 800GBASE-DR8, can use these modules without changing the connections. Other types of transceivers (such as 400GBASE-FR4 / LR4) can also be used by combining four transceiver ports with a hardbox or breakout box. The module is a switching network module having multiple multi-fiber connector interfaces, some of which are capable of connecting to network devices in a network using multi-fiber cables, the network having an internal mesh implemented in two tiers. The first tier is configured for rearrangement and the second tier is configured for recombination of individual fibers of different fiber groups. The optical paths of each transmitter and receiver are matched to provide appropriate optical connections from the transmitting fiber to the receiving fiber, and complex arbitrary network topologies can be achieved by reducing point-to-point interconnects by at least 1 / N, where N = the number of channels per multi-fiber connector interface.

[0045] FIG4(a) shows a front view of the disclosed module 400, which is a key element to facilitate the deployment, remodeling and expansion of optical networks. In this embodiment, the module has 64 MPO connector ports, which can be divided into a front portion and a rear portion as shown. Alternatively, the 64 ports can be located on one side of the device (not shown here).

[0046] For ease of illustration, we assume that ports 420 to 451, each representing an MPO connector, are located on the front side of the module, i.e., facing the leaf switch, as shown in part (a) of the figure. On the other side of the module, ports 440 to 491 (opposite to ports 420-451), each representing an MPO connector, are connected to the spine switch. The MPO size allows the module width W to range from 12 inches to 19 inches and the height H to range from 0.4 to 0.64 inches. The MPO connector can be placed vertically as shown to save more space. As described later in this application, machine-readable labels 410 and 412 can help deploy or inspect network interconnects. Moreover, if necessary, the lateral rails 405 on both sides of the module will enable the module to be inserted into the rack structure. Alternatively, using brackets 406, the module can be attached directly to the rack. By using the specified height range of this embodiment, up to four modules can be stacked in less than 1.6RU depending on density requirements.

[0047] Figure 5 A top view of the module is shown, showing the attached machine readable tags 410 and 412. A laser scanner or camera can read the tags. The read code can be linked to a database with an interconnection diagram of all modules in the network. This information can be displayed on a portable device, tablet, phone or augmented reality lens to facilitate deployment. For more specific information on this, see RS16563 and 25512.

[0048] Figure 6 305 and 310. The interconnection scheme of the module according to the present invention is shown. In order to simplify the module structure, the grid is divided into two areas, 305 and 310. Area 305 reorders the fiber groups, for example, 380 can be a standard or reelable ribbon, or just a cable unit of 16 fibers. Figure 7 As shown, the interconnection method for region 305 (from ports 420 to 451 to ports 320 to 351 ) needs to increase the mesh diversity and provide the best combination of individual optical fibers in region 310 . Figure 6 It shows how a single optical fiber (eg, 370) is paired with a group of optical fibers (eg, 380).

[0049] Region 310 contains four submodules 500 that create an interconnected grid of eight groups of optical fibers. Figure 8 An interconnect diagram of one submodule 500 is shown, indicating how each optical fiber from the port group 510 to 545 connected to the zone 305 is distributed to ports 550 to 585 .

[0050] For an MPO transmitting eight parallel channels, the mesh of submodules 500 has 64 duplex fiber interconnects. Millions of interconnects can be implemented to achieve full duplex bandwidth communication between ports, including eight such as Fig. 9Any of these configurations may be used to implement submodule 500 as long as they are consistently employed in module 400. Table I lists the interconnect configurations for the eight configurations.

[0051] The fiber groups from regions 305 to 310 are connected at interface 307. Interface 307 can be implemented using mechanical splices, multi-fiber connectors, or quality fusion splices. The latter is more suitable for achieving lower losses. Similarly, the fiber group from region 310, 32 groups of fibers from four 500 modules, are connected to ports 460 to 490 using fusion or mechanical splices or multi-fiber connectors (represented by 302).

[0052] The resulting interconnection diagram of module 400 is shown in Table II, including the meshing of fiber groups (region 305) and the meshing of individual fibers (region 310). As described in the table, each output port meshes with exactly eight input ports. For example, the first eight ports 460 to 467 mesh with ports 420, 424, 428, 432, 436, 440, 444, and 448. For example, the last eight ports 484 to 491 mesh with ports 423, 427, 431, 435, 439, 443, 447, and 451.

[0053]

[0054] Table I shows the grid configuration of the submodule 500 interconnection table for eight configurations.

[0055]

[0056] Table II Grid configuration of a module 400.

[0057] Therefore, according to the utility model, the device mixes Ethernet physical medium dependent (PMD) channels with other transceiver PMD channels to facilitate the interconnection of spine and leaf switches and distribute business flows between multiple redundant paths. The two-step grid combined in each module 400 by combining parts 305 and 310 increases the degree of fiber connection within each module. The device simplifies network deployment because an important part of the network complexity is transferred from the structured wiring switching network to one or more modules 400. Using modules 400 and following simple rules to connect a group of uplinks or downlinks horizontally or vertically makes the installation cleaner and the cable management is greatly improved, as shown in the following description of this application.

[0058] Stacking of several modules 400 can implement networks with different configurations and cardinality with different numbers of spine and leaf switches. For example, Fig.10A stack of four modules 400 is shown. Part (a) of the figure shows one side of the module connected to a leaf switch. For simplicity, we label it the front side. Part (b) shows the opposite side of the same module 400, the back side, which is connected to a spine switch.

[0059] Fig.10 The figure in assumes that 32 leaf switches need to be connected to the switching network shown in part (c) of the figure, each leaf switch having 4 MPO uplinks. In this illustrative example, the uplinks of the leaf switches are connected horizontally in groups of four until the last port of each module 400 is used. For example, port 710 is connected to the first fourth uplink of the first leaf switch, and 712 is connected to the first fourth uplink of the second leaf switch. The last four ports 714 of the first module 400 in the stack are connected to the first fourth uplink of the eighth leaf switch. Following the previous description and part (d) of the figure, we can consider module ports 710, 712 and 714 to be connected to four uplinks from leaf switches L1, L2 and L8. The first port of the second module 400, i.e., port 718, is connected to the uplink of the ninth leaf switch (L9), and the last port of the bottom module 400 in the stack, i.e., port 720, is connected to four uplinks from the thirty-second leaf switch (L32).

[0060] The ridge ports are allocated on the back side of the stacking module 400, such as Fig.10 For example, if independent spine switches are used, 730, 732, 734, and 736 correspond to the ports of the first, second, third, and thirty-second spine switches, respectively. Fig.10 In the figure, they are labeled S1, S2, S3 and S32. A more detailed description of the connection from the module to the spine is as follows Fig.11 Part (a) of the figure shows the connection of ports S1, S2 and S32 to independent spine switches. Part (b) of the same figure shows the interconnection table of all ports.

[0061] Alternatively, a rack switch can be used to implement the spine. Although more expensive than a standalone system, a rack switch can provide several advantages, such as scalability, reliability, and performance. The port connections for a spine using a rack switch can follow Fig.12Various configurations described in . For example, using sixteen spine switches, each with two line cards, all S1 and S2 ports can be connected to the first spine, S3 and S4 are connected to the second main spine, and S31 and S32 ports are connected to the sixteenth spine, as shown in part (a) of the figure. Using eight spine switches, each with four line cards, all S1, S2, S3, and S4 ports can be connected to the first spine, S5, S6, S7, S8 are connected to the second spine, and S29, S30, S31, S32 are connected to the last spine, as shown in part (b) of the figure. Using four spine switches, each with eight line cards, ports S1 to S8 can be connected to the first spine, S9 to S16 can be connected to the second spine, and S25 to S32 are connected to the last spine, as shown in part (c) of the figure. Otherwise, by using two spines with 16 line cards, each of S1 to S16 will be connected to the first spine, and the remaining will be connected to the second spine, as shown in part (d) of the figure.

[0062] Using module 400 and the described method, each spine switch having all thirty-two leaf switches is interconnected with thirty-two spine switches, such as Fig.10 As shown in parts (c) and (d) of . Network deployment using stacks of modules 400 enables a significantly wider range of configuration alternatives while providing the number of spine-leaf connections required with various bandwidth and oversubscription requirements (from 1:1 to 1:12). In addition, using stacks of modules 400 increases resilience to interconnect errors during network deployment. Fig.12 In all tables shown, the arrangement of the columns has no effect on network performance. For example, swapping the S1 with S2, S3, or S32 ports on the back side of a stack of module 400 preserves the interconnect mapping to the leaf ports connected to the front side of the module stack. In general, module 400 reduces the complexity of scaling out, scaling up, or even de-scaling networks, as shown in the following examples.

[0063] The interconnects within module 400 can transmit signals of any wavelength in the range of 830nm-1650nm. In addition, the signals assigned to each wavelength can propagate in one direction, such as from a transmitter to a receiver, or in a bidirectional propagation manner, such as using a bidirectional transceiver.

[0064] An important metric to characterize the degree of reduction in module complexity is to estimate the aggregate data rate per module using Da = fxNfxNcxD, where Nf is the number of fibers used per connector, e.g., Nf = 16, Nc is the number of adapters in module 400, e.g., Nc = 32, D is the data rate per fiber in one direction, and f takes into account bidirectional communication if bidirectional transceivers are used. For example, using the typical case shown in this document, Nf = 16, Nc = 32, f = 1, Da = 512D. For current transceivers operating at DD = 100Gbps / lambda, Da = 51.2Tbps. Assuming next-generation transceivers operating at D = 200Gbps / lambda, Da = 102.4Tbps. Using VSFF connectors such as SN-MT or MMC, 3 NIMs can be accommodated in 1RU, resulting in a mesh data rate density of 307.2Tbps per RU for mesh connections between switches. A rack filled with 400 modules (assuming 50Ru) can provide up to 15Pbps of mesh connectivity.

[0065] Network deployment example for a network using module 400

[0066] Figures 13 to 15 The examples in show implementations of two-tier and three-tier FCNs using various cardinalities, oversubscriptions, and sizes of modules 400. Tables III to VI show a detailed description of the number of modules required for each network and the estimated rack space required for the modules.

[0067] Starting from a two-layer FCN, Fig.13 A switching fabric 850 with 32 spine switches and 64 leaf switches implemented using eight modules 400 is shown. The connection scheme of the stack is shown from both sides of the module 400, one side is labeled as the front side, 855, and the other side is labeled as the back side, 857. The 855 side is connected to the 64 leaf switches via 4 MPO uplinks, each connector having 16 fibers. The switches are labeled Li, where i is the index of the switch. For this example, i ranges from 1 to 64. As shown on the 855 front side, the leaf switches are connected horizontally following the method shown previously. For example, the fourth L1 uplink is connected adjacently in the first four ports of the first module 400. All L64 uplinks are connected to the last four ports of the eighth module 400. Starting from the back side 857 side of the same module stack, the 32 spine switches are connected vertically as shown. Based on the size of the disclosed module 400, the switching fabric can be implemented in less than 4RU.

[0068] Table II shows that the two-tier networks can be scaled out to a large number of leaf switches. The table also shows the number of modules 400 required to implement these networks and the rack space used for these modules. Fig.14 An example of this table is shown. A switching fabric 870 with up to 32 spine and 512 leaf switches is implemented using 64 modules 400. The connection scheme of the stack is shown from both sides of the module 400, one side is labeled as the front side, 875, and the other side is labeled as the back side, 877. The 875 side is connected to the 512 leaf switches via four MPO uplinks. For example, the fourth L1 uplink is connected adjacently in the first four ports of the first module 400. All L512 uplinks are connected to the last four ports of the sixty-fourth module 400. Starting from the back side 877 side of the same module stack, the 32 spine switches are connected vertically as shown. As previously described in Fig.12 As described in , when using a chassis switch with 16 line cards, only two spine switches are required.

[0069]

[0070] Table III, Parameters for a two-tier FCN with 16 spine switches. For the number of switches and servers, the number of modules 400 required and the size of the switches and servers in rack units are shown.

[0071] As shown in Table III, using a two-layer network, the network can scale to support thousands of leaf switches that can interconnect tens or thousands of servers. Scaling beyond this number requires the use of a three-layer FCN. Figure 2 (b) shows the topology of a three-layer network with 16 spine switches and 32 leaf switches. In a three-layer network, spines do not need to be connected to all leaf switches, but to groups of these leaf switches, called PODs.

[0072] Module 400 can also be used to implement a three-layer FCN, such as Fig.15 In this figure, a three-layer network with 1024 spine switches (or 64 racks with 64 line cards) and 2048 leaf switches is shown. The network has 64 PODs, and each POD of the network has 32 switching meshes and leaf switches. The mesh of each POD can be fully implemented with four stacked modules 400, as previously shown (see Fig.10 ). Therefore, 64x4=256 modules are needed to implement all PODs. The connection between the spine and the 64 PODs requires another set of 256 modules 400, because each spine switch only needs to connect to one leaf switch in the POD.

[0073] Fig.16 The front side of the module shows the interconnection method for leaf and spine switches, and Fig.17The back side of the module is shown in FIG. Following the above method, the uplink level of the leaf switch in each POD is filled with modules. For example, in Fig.16 In FIG. 4 , the four MPO uplinks from the first leaf switch of POD 1, namely L1p1, occupy the first MPO port of the first module 400. The four MPO uplinks from the leaf switch 32 of POD 64, namely L32p64, occupy the last four MPO ports of the last module 400 in the stack.

[0074] From opposite sides of the stack, the columns of the module stack connect to the line card MPO ports of the spine switch. Fig.17 As shown, the line card MPO port S1c1 of spine switch 1 is connected to the eight MPO ports in the first column of the stack. The MPO port of the second line card of the same switch, S1c2, is connected to the eight MPO ports in the second column of the stack. The MPO port of the 32nd line card of the last spine switch, S32c16, is connected to the eight ports in the last column of the stack.

[0075] Such a switch with 1024 spines (or 64 racks with 16 line cards) and 2048 leaves requires 512 modules 400, with the equivalent rack space of the modules 400 being equal to or less than 220RU.

[0076] Table IV shows a method for scaling the network to a larger number of leaves (and servers) in terms of the number of modules 400 and rack space required.

[0077] As shown in this section, the module 400 and the disclosed interconnection method for two-tier and three-tier FCNs simplifies the deployment of optical networks of different sizes and configurations. Because the cable groups representing uplinks / downlinks for the same switch are closely connected, and also because of the high degree of meshing in the network, the risk of interconnection errors during deployment is greatly reduced. For example, in Fig.16 All Lipj connections are interchangeable in , where i is the leaf uplink index ranging from 1 to 4 and j is the POD index ranging from 1 to 64. During network deployment, unexpected changes from L1p1 to L1p2, L1p1 to L1p3, L1p1 to L1p4, or any combination within the group will have no impact on network operation. The topology will still connect all leaf switches from the POD to the spine switch using the same number of paths and the same allocated bandwidth. Similarly, in Fig.17 In the example, all the columns of the ridges are interchangeable, which can be seen from Fig.10 The level of redundancy provided by the stacking of modules 400 greatly reduces the risk of switch fabric failure or performance degradation caused by errors in the interconnect.

[0078]

[0079] Table IV, Parameters for a Layer 3 FCN with 1:1 oversubscription for 1024 spine switches. For the number of switches and servers, the number of modules 400 required and the size of the switches and servers in rack units are shown.

[0080] Although the utility model has been described as having a preferred design, the utility model can be further modified within the spirit and scope of the present disclosure. Therefore, the present application is intended to cover any variation, use or modification of the present utility model using its general principles. Further, the present application is intended to cover deviations from the present disclosure within known or customary practices in the relevant field of the present utility model, and it falls within the limitations of the attached claims.

Claims

1. A switching network module for a high-radix network, having an interface with multiple multi-fiber connectors, characterized in that: Some of the multi-fiber connector interfaces are capable of connecting to network devices in a network using a multi-fiber cable, the fiber optic cable comprising an internal grid implemented in two levels, wherein the first level is configured for rearrangement and the second level is configured for recombination of individual optical fibers of different fiber groups, further wherein the optical path of each transmitter and receiver is matched to provide an appropriate optical connection from the transmitting fiber to the receiving fiber, and wherein complex arbitrary network topologies can be implemented with at least 1 / N reduction in point-to-point interconnections, wherein for an MPO with 8 fibers, N=4.

2. The switching network module according to claim 1, characterized in that: The switch fabric modules are further configured to be stacked to provide two-level or three-level CLOS network topologies of various switch bases.

3. The switching network module according to claim 1, characterized in that: The switch fabric module is further configured to implement networks with different levels of oversubscription ranging from 1:1 to 1:

12.

4. The switching network module according to claim 1, characterized in that: The switch fabric module is further configured to scale the optical network from eight switches to thousands of leaf switches.

5. The switching network module according to claim 1, characterized in that: The switch network module is further configured to provide redundant paths, thereby reducing the risk of switch network failure due to interconnection errors.

6. The switching network module according to claim 1, characterized in that: The switch fabric module is further configured to have a small form factor, enabling three modules to be stacked in one RU.

7. The switching network module according to claim 1, characterized in that: The switch module further includes a machine-readable tag that, when read by a tag reader such as a laser scanner or camera, is capable of providing an interconnection map of the network to a portable device.

8. The switching network module according to claim 1, characterized in that: The switching network module is further configured to effectively distribute the service load of the switch.

Images