IP Library Granted Patent US 12,587,768
Granted Patent B2
US 12,587,768 · App. 18/346,441 · Granted Mar 24, 2026

Network device for network fabrics with harmonic connections

Inventors: Giacomo Bernardi (Bologna, IT); Ratul Mahajan (Seattle, WA); Saurabh Kumar (Los Altos, CA)
Assignee: Amazon Technologies, Inc.
H04Q11/0005H04Q2011/005
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,587,768
App. No.
18/346,441
Granted
Mar 24, 2026
Kind
B2
Abstract

A network device operating as a network node in a network fabric has a set of connection ports to provide a bandwidth capacity for the network device. The connection ports include fabric ports to connect with respective strands using multipoint optical connections. Each strand can connect network nodes of the network fabric according to a harmonic specifying a node distance between adjacent connection points on the strand. The connection ports also include server ports to connect with servers and/or external networks. The network device also includes processing logic to distribute traffic for a destination node to network nodes along corresponding strands of the fabric ports.

Claims (35)

1 . A network device operating as a network node in a network fabric, the network device comprising:

a plurality of connection ports providing a bandwidth capacity for the network device, the plurality of connection ports including:

a set of fabric ports operable to connect with respective strands, each strand connecting network nodes of the network fabric according to a harmonic specifying a node distance between adjacent connection points on the strand, and each strand implemented using reconfigurable multipoint optical connections that support multiple channels to allow dynamic channel reallocation between connection points on the strand; and

a set of server ports operable to connect with one or more servers, or one or more external networks; and

processing logic operable to:

transmit traffic demand information of ingress traffic and egress traffic of the network device to a control plane;

receive channel allocation information from the control plane based on the traffic demand information;

configure the reconfigurable multipoint optical connections with the channel allocation information; and

distribute traffic for a destination node to network nodes along corresponding strands of the fabric ports, the traffic being routed to the destination node via the network fabric.

2 . The network device of claim 1 , wherein the multiple channels supported by the strand are implemented on respective optical wavelengths.

3 . The network device of claim 1 , wherein at least half of the bandwidth capacity of the network device is allocated to the set of fabric ports.

4 . The network device of claim 1 , wherein the network nodes connected along each strand include at least four network nodes.

5 . A network device operating as a network node in a network fabric, the network device comprising:

a plurality of connection ports providing a bandwidth capacity for the network device, the plurality of connection ports including:

a set of fabric ports operable to connect with respective strands using multipoint optical connections, each strand connecting network nodes of the network fabric according to a harmonic specifying a node distance between adjacent connection points on the strand,

wherein each strand supports multiple channels, and the multipoint optical connections are reconfigurable multipoint optical connections that allow dynamic channel reallocation between connection points on each strand; and

a set of server ports operable to connect with one or more servers, or one or more external networks; and

processing logic operable to distribute traffic for a destination node to network nodes along corresponding strands of the fabric ports, the traffic being routed to the destination node via the network fabric.

6 . The network device of claim 5 , wherein the network device is part of a logical grid of network nodes, and the strands connecting network nodes of the network fabric include vertical strands connecting network nodes along a same column, and horizontal strands connecting network nodes along a same row.

7 . The network device of claim 6 , wherein half of the fabric ports are allocated for the vertical strands, and another half of the fabric ports are allocated for the horizontal strands.

8 . The network device of claim 5 , wherein at least half of the bandwidth capacity of the network device is allocated to the set of fabric ports.

9 . The network device of claim 5 , wherein the multipoint optical connections are implemented with breakout optics to connect a fabric port of the network device to fabric ports of respective network nodes along a strand.

10 . The network device of claim 5 , wherein the reconfigurable multipoint optical connections are implemented with breakout optics to connect a fabric port of the network device to multiple ports of an optical circuit switch that is connected to other network nodes along a strand.

11 . The network device of claim 5 , wherein the multiple channels along a strand are implemented on respective optical wavelengths.

12 . The network device of claim 11 , wherein the reconfigurable multipoint optical connections are implemented with optical couplers and splitters to merge and split the optical wavelengths along a strand.

13 . The network device of claim 11 , wherein the reconfigurable multipoint optical connections are implemented with optical add-drop multiplexers (ROADM) to add and drop optical wavelengths along a strand.

14 . The network device of claim 11 , wherein the reconfigurable multipoint optical connections are implemented with silicon photonics using ring resonators to modulate and detect the optical wavelengths along a strand.

15 . The network device of claim 5 , wherein the processing logic is operable to transmit traffic demand information of ingress traffic and egress traffic of the network device to a control plane.

16 . The network device of claim 15 , wherein the processing logic is operable to receive channel allocation information from the control plane based on the traffic demand information.

17 . A method, comprising:

transmitting traffic demand information of ingress traffic and egress traffic of a network device to a control plane that manages a network fabric having network nodes interconnected by strands according to harmonics, each harmonic specifying a node distance between adjacent connection points on a corresponding strand, and each strand implemented using reconfigurable multipoint optical connections that support multiple channels to allow dynamic channel reallocation between connection points on the strand;

receiving channel allocation information from the control plane based on the traffic demand information;

configuring the reconfigurable multipoint optical connections of the network device with the channel allocation information; and

distributing traffic for a destination node to network nodes along strands connected to the network device, the traffic being routed to the destination node via the network fabric.

18 . The method of claim 17 , wherein configuring the reconfigurable multipoint optical connections includes changing one or more optical wavelengths of the reconfigurable multipoint optical connections.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Nov 17, 2023
From: BERNARDI, GIACOMO; MAHAJAN, RATUL; KUMAR, SAURABH
To: AMAZON TECHNOLOGIES, INC.
Reel/Frame 065606/0826 →
Continuity (1)
Related Publication 20250016478A1 · Jan 9, 2025
References Cited (29)
US 10205654B2 · Choi et al. · 2019 [cited by applicant]
US 12132650B1 · Bernardi et al. · 2024 [cited by applicant]
US 20130083701A1 · Tomic et al. · 2013 [cited by applicant]
US 20140119728A1 · Zhang · 2014 [cited by examiner]
US 20170078191A1 · Choi et al. · 2017 [cited by applicant]
US 20210119938A1 · Pelekhaty et al. · 2021 [cited by applicant]
US 20210325623A1 · Islam · 2021 [cited by examiner]
US 20220053011A1 · Rao et al. · 2022 [cited by applicant]
US 20220247687A1 · Pelekhaty · 2022 [cited by examiner]
US 20240164044A1 · Henley · 2024 [cited by examiner]
EP 3648424A1 · 2020 [cited by applicant]
U.S. Notice of Allowance dated Jun. 10, 2024 in U.S. Appl. No. 18/346,439. [cited by applicant]
U.S. Non-Final Office Action dated Feb. 25, 2025 in U.S. Appl. No. 18/346,436. [cited by applicant]
Besta, M., et al., “Slim Fly: A Cost Effective Low-Diameter Network Topology,” SC '14: Proc. of the International Conference for High Performance Computing, Networking, Storage and Analysis, Nov. 2014, 12 pages, URL: ht… [cited by applicant]
Flajslik, M., et al., “Megafly: A Topology for Exascale Systems,” ISC High Performance 2018: High Performance Computing, 2018, pp. 289-310 (part of the book series: Lecture Notes in Computer Science, vol. 10876). URL: h… [cited by applicant]
Guo, C., et al., “BCube: A High Performance, Server-centric Network Architecture for Modular Data Centers,” ACN SIGCOMM Computer Communication Review, 2009, vol. 39(4), pp. 63-74. [cited by applicant]
Jouppi, N. P., et al., “TPU v4: An Optically Reconfigurable Supercomputer for Machine Learning with Hardware Support for Embeddings,” arXiv:2304.01433v3 [cs.AR], Apr. 20, 2023, pp. 1-14, URL: https://arxiv.org/abs/2304.… [cited by applicant]
Kim, J., et al., “Technology-Driven, Highly-Scalable Dragonfly Topology,” ACM SIGARCH Computer Architecture News, Jun. 2008, vol. 36(3), pp. 77-88. [cited by applicant]
Matsuoka, S., “You Don't Really Need Big Fat Switches Anymore—Almost,” IPSJ SIG Technical Report [Japan], Aug. 2003, pp. 157-162, URL: https://ipsj.ixsq.nii.ac.jp/ej/?action=repository_uri&item_id=23400&file_id=1&file_n… [cited by applicant]
Shpiner, A., et al., “Dragonfly+: Low Cost Topology for Scaling Datacenters,” 2017 IEEE 3rd International Workshop on High-Performance Interconnection Networks in the Exascale and Big-Data Era (HiPINEB), Feb. 2017, 8 pa… [cited by applicant]
Singla, A., et al., “Jellyfish: Networking Data Centers Randomly,” NSDI'12: Proc. of the 9th USENIX Conference on Networked Systems Design and Implementation, Apr. 2012, pp. 1-14, URL: https://www.usenix.org/system/file… [cited by applicant]
U.S. Ex Parte Quayle Action dated Mar. 14, 2024 in U.S. Appl. No. 18/346,439. [cited by applicant]
U.S. Appl. No. 18/346,436, inventors Bernardi G, et al., filed on Jul. 3, 2023. [cited by applicant]
U.S. Appl. No. 18/346,439, inventors Bernardi G, et al., filed on Jul. 3, 2023. [cited by applicant]
Valadarsky, A., et al., “Xpander: Towards Optimal-Performance Datacenters,” CoNEXT '16: Proc. of the 12th International Conference on Emerging Networking Experiments and Technologies, Dec. 2016, pp. 205-219. [cited by applicant]
Celik, A., et al., “Design and Provision of Traffic Grooming for Optical Wireless Data Center Networks,” IEEE Transactions on Communications, Mar. 2019, vol. 67(3), pp. 2245-2259. [cited by applicant]
International Search Report and Written Opinion dated Sep. 10, 2024 in PCT Application No. PCT/US2024/035431. [cited by applicant]
Poutievski, L., et al., “Jupiter Evolving: Transforming Google's Datacenter Network via Optical Circuit Switches and Software-Defined Networking,” SIGCOMM '22: Proceedings of the ACM SIGCOMM 2022 Conference, Aug. 2022, … [cited by applicant]
U.S. Notice of Allowance dated Jun. 27, 2025 in U.S. Appl. No. 18/346,436. [cited by applicant]