Data center networks using bottleneck structures

Patent metadata

Field	Value
Publication number	US-12413481-B2
Application number	US-202418653829-A
Country	US
Kind code	B2
Filing date	May 2, 2024
Priority date	Feb 17, 2021
Publication date	Sep 9, 2025
Grant date	Sep 9, 2025

Abstract

A network is designed based on its topology and the expected flow patterns in the network. The use of the latter can lead to efficient use of network resources and can reduce or even minimize waste. Non-interference properties of the expected flows can yield an improved or even optimal design.

First claim

What is claimed is:

1. A processor-implemented method for selecting a network parameter, the method comprising:
obtaining an expected traffic pattern for a network, the network comprising a plurality of levels of switches and corresponding links;
selecting a network parameter corresponding to a switch or link at a selected level based on, in part, a portion of the expected network traffic pattern associated with that switch or link; and
calculating a wasteless network design, based on the network parameter, that minimizes flow completion times and maximizes network throughput, the wasteless network design configured to route the expected traffic pattern with all bandwidth of each link used throughout transmission of the expected traffic pattern.

2. The method of claim 1 , further comprising:
determining from the expected traffic pattern skewedness of the traffic, wherein selecting the network parameter is based on, in part, the skewedness.

3. The method of claim 2 , wherein selecting the network parameter is based on, in part, a tapering parameter that is based on, in part, the skewedness.

4. The method of claim 1 , wherein:
the network comprises a fat-tree network;
the network parameter comprises capacity of links at a particular level of switches; and
selecting the capacity of links at one level of switches is further based on, in part, capacity of links at another level of switches.

5. The method of claim 1 , wherein:
the network comprises a folded-clos network;
the network parameter comprises a number of spine blocks of the folded clos network; and
selecting the number of spine blocks is further based on, in part, a radix of spine-level switches.

6. The method of claim 1 , wherein:
the network comprises a Dragonfly network;
the network parameter comprises a capacity of interpod links and a capacity of intrapod links; and
selecting the capacity of interpod links is further based on, in part, the capacity of intrapod links.

7. The method of claim 6 , further comprising:
iteratively partitioning pods in the Dragonfly network into a plurality of groups, wherein selecting the capacity of interpod links is further based on, in part, a total number of groups and a total number of pods in a last group.

8. The method of claim 1 , wherein a capacity of a link at one of the plurality of levels is selected to be proportional to an expected size of flows traversing that link.

9. The method of claim 1 , wherein a capacity of a link at one of the plurality of levels is selected based on a specified time of completion of all expected network flows.

10. The method of claim 1 , further comprising:
obtaining an updated traffic pattern; and
updating the network parameter based on, at least in part, the updated traffic pattern.

11. The method of claim 1 , wherein:
the network comprises a sliced network having a plurality of virtual network slices; and
selecting the network parameter comprises selecting a corresponding network parameter of at least one network slice.

12. A system for selecting a network parameter, the system comprising:
a first processor; and
a first memory in electrical communication with the first processor, and comprising instructions that, when executed by a processing unit that comprises one or more computing units, wherein one of the one or more computing units comprises the first processor or a second processor, and wherein the processing unit is in electronic communication with a memory module that comprises the first memory or a second memory, program the processing unit to:
obtain an expected traffic pattern for a network, the network comprising a plurality of levels of switches and corresponding links;
select a network parameter corresponding to a switch or link at a selected level based on, in part, a portion of the expected network traffic pattern associated with that switch or link; and
calculate a wasteless network design, based on the network parameter, that minimizes flow completion times and maximizes network throughput, the wasteless network design configured to route the expected traffic pattern with all bandwidth of each link used throughout transmission of the expected traffic pattern.

13. The system of claim 12 , wherein the instructions further program the processing unit to:
determine from the expected traffic pattern skewedness of the traffic; and
select the network parameter is based on, in part, the skewedness.

14. The system of claim 12 , wherein:
the network comprises a fat-tree network;
the network parameter comprises capacity of links at a particular level of switches; and
the instructions program the processing unit to select the capacity of links at one level of switches based on, in part, capacity of links at another level of switches.

15. The system of claim 12 , wherein:
the network comprises a folded-clos network;
the network parameter comprises a number of spine blocks of the folded clos network; and
the instructions program the processing unit to select the number of spine blocks based on, in part, a radix of spine-level switches.

16. The system of claim 12 , wherein:
the network comprises a Dragonfly network;
the network parameter comprises a capacity of interpod links and a capacity of intrapod links; and
the instructions program the processing unit to select the capacity of interpod links based on, in part, the capacity of intrapod links.

17. The system of claim 12 , wherein the instructions program the processing unit to select a capacity of a link at one of the plurality of levels: to be proportional to an expected size of flows traversing that link, and/or to select the capacity based on a specified time of completion of all expected network flows.

18. The system of claim 12 , wherein the instructions program the processing unit to:
obtain an updated traffic pattern; and
update the network parameter based on, at least in part, the updated traffic pattern.

19. The system of claim 12 , wherein:
the network comprises a sliced network having a plurality of virtual network slices; and
to select the network parameter, the instructions program the processing unit to select a corresponding network parameter of at least one network slice.

20. A network comprising:
at least one processor;
memory coupled to the at least one processor;
a plurality of levels of hardware switches and corresponding links, the hardware switches coupled to the at least one processor and the memory, wherein
a network parameter corresponding to a switch or link at a selected level is proportional to an aggregate expected network traffic pattern associated with that switch or link, the network parameter implementing a wasteless network design for routing the expected network traffic pattern while minimizing flow completion times and maximizing network throughput with all bandwidth of each link used throughout transmission of an expected traffic pattern.

21. The network of claim 20 , wherein the network parameter is based on, in part, a skewedness of the traffic.

22. The network of claim 21 , wherein the network parameter is based on, in part, a tapering parameter that is based on, in part, the skewedness.

23. The network of claim 20 , wherein:
a topology of the network is fat-tree;
the network parameter comprises capacity of links at a particular level of hardware switches; and
the capacity of links at one level of hardware switches is based on, in part, capacity of links

Assignees

• Qualcomm Inc

Inventors

• Ros-Giralt Jordi
• Amsel Noah
• Lethin Richard

Classifications

• H04L49/1515

  Non-blocking multistage, e.g. Clos · CPC title

• H04L47/127 Primary

  by using congestion prediction · CPC title

• H04L45/48

  Routing tree calculation · CPC title

• H04L41/145 Primary

  involving simulating, designing, planning or modelling of a network · CPC title