CN-122001764-A - Efficient power management for network devices

Abstract

The present disclosure relates to efficient power management of network devices. Techniques for improved energy efficiency of network devices of a network system are disclosed. For example, a computing system obtains time-series data that includes information about computing devices in a computer network. The computing device accesses other devices in the computer network through network devices that exchange network traffic for the computing device. The computing system applies a machine learning system trained with historical time series data to the obtained time series data to predict a need for one or more of the network devices to exchange network traffic of the computing device in a next time interval. The computing system adjusts operation of one or more of the network devices based at least in part on the predicted demand for the next time interval.

Inventors

• Raja Komula
• Ganesh Biagotti Matad Sankada
• Timnawan Srida
• Rajendra Shivaram Yawatkar
• Murugan Kaniapan

Assignees

• 瞻博网络公司

Dates

Publication Date

20260508

Application Date

20251107

Priority Date

20241108

Claims (20)

1. 1. A computing system, comprising: storage medium, and Processing circuitry in communication with the storage medium, the processing circuitry configured to: obtaining time-series data comprising information about computing devices in a computer network, wherein the computing devices access other devices in the computer network through network devices exchanging network traffic of the computing devices; Applying a machine learning system trained with historical time series data to the obtained time series data to predict a need for one or more of the network devices to exchange network traffic of the computing device in a next time interval, and The operation of the one or more of the network devices is adjusted based at least in part on the predicted demand for the next time interval.
2. 2. The computing system of claim 1, Wherein the computing device includes a server hosting one or more application workloads, Wherein the information about the computing device indicates a power throttling status of each of the servers, and Wherein the historical time series data indicates a historical power throttling state of the server and a historical network bandwidth usage of the network device corresponding in time to the power throttling state of the server.
3. 3. The computing system of claim 1, Wherein the computing device includes a server hosting an application workload, Wherein the information about the computing device indicates a central processing unit CPU usage or a graphics processing unit GPU usage of each of the servers, and Wherein the historical time series data indicates a historical CPU usage or GPU usage of the server, and a historical network bandwidth usage of the network device that corresponds in time to the CPU usage or GPU usage of the server.
4. 4. The computing system of claim 1, Wherein the computing device includes a server hosting an application workload, Wherein the information about the computing device indicates a resource utilization of each of the servers, and Wherein the historical time series data indicates a historical resource utilization of the server and a historical network bandwidth utilization of the network device corresponding in time to the resource utilization of the server.
5. 5. The computing system of claim 1, Wherein the computing device includes a server hosting an application workload, and Wherein the information about the computing device indicates a network traffic intensity for each of the application workloads, and Wherein the historical time series data indicates a historical network traffic intensity of the application workload hosted by the server and a historical network bandwidth usage of the network device corresponding in time to the network traffic intensity of the application workload.
6. 6. The computing system of claim 1, Wherein the computing device comprises a user equipment, UE, device, and wherein the network device comprises a wireless access point, AP, Wherein the information about the computing device indicates an operating channel frequency of each of the UE devices, and Wherein the historical time series data indicates an operating channel frequency of the UE device and a network usage pattern of a wireless AP that corresponds in time to the operating channel frequency of the UE device.
7. 7. The computing system of claim 1, wherein to adjust the operation of the one or more of the network devices, the processing circuitry is configured to adjust one or more operating parameters affecting energy consumption of the network device.
8. 8. The computing system of claim 1, wherein to adjust the operation of the network device, the processing circuit is configured to adjust at least one of: An operational state of a packet processing unit of the network device; a power budget for one or more packet processing units of the network device; clock frequency of central processing unit CPU of the network device, and The power level of the antenna or radio of the network device.
9. 9. The computing system of claim 1, wherein the network device is configured to use one or more of a first operating channel operating at a first frequency comprising a frequency band of approximately 2.5 GHz and a second operating channel operating at a second frequency comprising a frequency band of approximately 5 GHz, and Wherein, to adjust the operation of the network device, the processing circuit is configured to: enabling the first operating channel and disabling the second operating channel, or Both the first operating channel and the second operating channel are enabled.
10. 10. The computing system of any of claims 1 to 9, wherein the demand includes a network performance demand, and Wherein the processing circuitry is configured to adjust the operation of the one or more network devices by adjusting one or more operating parameters that increase network bandwidth throughput of the one or more network devices based on a prediction by the machine learning system of an increase in the network performance requirement for the next time interval compared to a past network performance requirement for a previous time interval.
11. 11. The computing system of any of claims 1 to 9, wherein the demand includes a network performance demand, and Wherein the processing circuitry is configured to adjust the operation of the one or more network devices by adjusting one or more operating parameters that reduce network bandwidth throughput of the one or more network devices based on a prediction by the machine learning system of a decrease in the network performance requirement for the next time interval compared to a past network performance requirement for a previous time interval.
12. 12. The computing system of any one of claims 1 to 9, wherein the demand comprises a network performance demand, Wherein the processing circuit is configured to: Adjusting operation of a first network device of the one or more network devices based on a prediction by the machine learning system of an increase in demand for network traffic by the first network device to exchange a first computing device of the computing devices in the next time interval over a past demand by the first network device in a previous time interval, so as to increase energy consumption of the first network device, and Based on a prediction by the machine learning system of a decrease in demand for network traffic of a second network device exchanging the computing devices within the next time interval compared to a past demand of the second network device at the previous time interval, operation of the second network device of the one or more network devices is adjusted so as to reduce energy consumption of the second network device.
13. 13. The computing system of any of claims 1 to 9, wherein the machine learning system is trained with historical time series data of the computing device and the network device.
14. 14. A computer network method, comprising: Obtaining, by processing circuitry of a computing system, time-series data comprising information about computing devices in a computer network, wherein the computing devices access other devices in the computer network through network devices that exchange network traffic of the computing devices; applying, by the processing circuitry, a machine learning system trained with historical time series data to the obtained time series data to predict a need for one or more of the network devices to exchange network traffic of the computing device in a next time interval, and The operations of the one or more of the network devices are adjusted by the processing circuitry and based at least in part on the predicted demand for the next time interval.
15. 15. The computer network method of claim 14, Wherein the computing device includes a server hosting one or more application workloads, Wherein the information about the computing device indicates a power throttling status of each of the servers, and Wherein the historical time series data indicates a historical power throttling state of the server and a historical network bandwidth usage of the network device corresponding in time to the power throttling state of the server.
16. 16. The computer network method of claim 14, Wherein the computing device includes a server hosting an application workload, Wherein the information about the computing device indicates a central processing unit CPU usage or a graphics processing unit GPU usage of each of the servers, and Wherein the historical time series data indicates a historical CPU usage or GPU usage of the server, and a historical network bandwidth usage of the network device that corresponds in time to the CPU usage or GPU usage of the server.
17. 17. The computer network method of claim 14, Wherein the computing device includes a server hosting an application workload, and Wherein the information about the computing device indicates a network traffic intensity for each of the application workloads, and Wherein the historical time series data indicates a historical network traffic intensity of the application workload hosted by the server and a historical network bandwidth usage of the network device corresponding in time to the network traffic intensity of the application workload.
18. 18. The computer network method of claim 14, Wherein the computing device comprises a user equipment, UE, device, and wherein the network device comprises a wireless access point, AP, Wherein the information about the computing device indicates an operating channel frequency of each of the UE devices, and Wherein the historical time series data indicates an operating channel frequency of the UE device and a network usage pattern of a wireless AP that corresponds in time to the operating channel frequency of the UE device.
19. 19. The computer network method of any of claims 14 to 18, wherein adjusting the operation of the one or more of the network devices comprises adjusting one or more operating parameters affecting energy consumption of the network device.
20. 20. A computer-readable storage medium encoded with instructions for causing one or more programmable processors to become configured as the computing system of any of claims 1-13 or to perform the method of any of claims 14-19.

Description

Efficient power management for network devices RELATED APPLICATIONS The present application claims the benefit of U.S. patent application Ser. No. 19/352,233, filed on 7, 10, 2025, which claims the benefit of Indian provisional patent application No. 202441085871, filed on 8, 11, 2024, each of which is incorporated herein by reference in its entirety. Technical Field The present disclosure relates to computer networks and, more particularly, to improving energy efficiency in computer networks. Background A computer network is a collection of interconnected network devices that can exchange data and share resources. In packet-based networks, such as ethernet networks, network devices transmit data by dividing the data into variable length blocks, referred to as packets, which are individually routed across the network from a source device to a destination device. The destination device extracts the data from the packet and assembles the data into its original form. Some network devices or nodes, such as routers, maintain routing information describing routes through the network. Routers typically have many Central Processing Unit (CPU) cores and require a large amount of memory and energy usage to support various tasks such as management of the control plane and routing packets. In some cases, routers may have over a hundred CPU cores, and hundreds of gigabytes of random access memory. As enterprise networks, service provider networks, other types of networks, and data centers become larger, their overall energy usage increases. Some large data centers require a large amount of power-sufficient to power many households at the same time. Data centers may also run computing and data intensive application workloads, such as crypto-currency mining and machine learning applications, and consume large amounts of energy. To be more energy efficient, some networks may be derived from renewable energy sources. However, the configuration of networks, data centers, and/or applications running on such networks is constantly changing, and networks often cannot dynamically increase their energy efficiency. Disclosure of Invention The present disclosure describes techniques for improving and/or reducing power requirements and energy consumption of network devices exchanging network traffic of computing devices of a computing network. As an example, this may be useful in a data center network such that the data center consumes less energy while the devices of the network maintain a desired level of performance. In examples of the technology of the present disclosure, a power management controller of a computing system obtains time series data. The time series data includes information about computing devices in the computer network. In some examples, the information indicates, for example, a power throttling state of the computing device, a resource utilization (such as a Central Processing Unit (CPU) utilization or a Graphics Processing Unit (GPU) utilization), a network traffic strength of one or more applications executed by the computing device, an operating channel frequency at which the computing device operates, or a network usage mode of the computing device. The power management controller collects such metrics for each of the computing devices and for each of the plurality of time intervals. The power management controller applies a machine learning system trained with historical time series data of computing devices and network devices to the obtained time series data to predict a need to exchange network traffic for each of the computing devices in a next time interval. The power management controller adjusts operation of one or more of the network devices based at least in part on the predicted demand for the next time interval. For example, based on a prediction that a computing device may generate less network traffic in a next time interval than a previous time interval, a power management controller described herein may adjust operation of one or more of the network devices to reduce performance, such as by reducing network throughput, deactivating one or more radios, or reducing energy consumption of one or more network devices. In a similar manner, based on the prediction that the computing device may generate more network traffic in the next time interval than the previous time interval, the power management controller may adjust operation of one or more of the network devices to improve performance, such as by increasing network throughput, activating one or more radios, or increasing energy consumption of one or more network devices. The techniques of this disclosure may provide specific improvements to the computer-related art of computer networks, and more particularly, power management for networked devices that may have one or more practical applications. In particular, the techniques described herein may help manage power in a computing system to improve inefficiency due to differences between ove