SMART POWER MANAGEMENT IN SWITCHES AND ROUTERS

• US 20160091913A1
• Filed: 09/30/2014
• Published: 03/31/2016
• Est. Priority Date: 09/30/2014
• Status: Abandoned Application

• Current Assignee: Cisco Technology, Inc. (Cisco Systems, Inc.)
• Original Assignee: Cisco Technology, Inc. (Cisco Systems, Inc.)
• Inventors: Pani, Ayaskant
• Application Number:
  US14/503,066
• Publication Number:
  US 20160091913A1
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G05F 1/66 Regulating electric power

Abstract

Various embodiments of the present disclosure provide methods for analyzing usage information at each of a plurality of network devices of a computing network according to one or more machine learning algorithms and predicting a usage pattern of a corresponding network device at a specific future time. In some embodiments, routing protocol information of a plurality of network devices and one or more corresponding upstream or downstream ports can be collected. Based upon the routing protocol information of the plurality of network devices and the corresponding upstream or downstream ports, or the predicted usage pattern at each of the plurality of network device, a reduced-power-consumption topology that scales with predicted demands at the plurality of network devices can be dynamically generated. An operation state of at least one of the plurality of network devices or at least one corresponding upstream or downstream port can be dynamically adjusted to achieve a power saving at the computing network.

Claims

1. A computer-implemented method, comprising:

collecting historical usage information at a network device and/or at least one of a peer node type identification, time of day, day of a year, port identifier, switch identifier, interface packet arrival rate, interface packet drop rate or packet queue statistic that is associated with the network device, the network device being one of a plurality of network devices at a computing network;

analyzing the usage information at the network device by using one or more machine-learning algorithms;

predicting a usage pattern of the network device at a specific future time based at least upon the historical usage information;

collecting routing protocol information of the network device and one or more corresponding upstream or downstream ports; and

based at least upon predicted usage pattern or the routing protocol information, dynamically adjusting an operation state of the network device or at least one of the one or more corresponding upstream or downstream ports to achieve a power saving at the computing network.

2. The computer-implemented method of claim 1, wherein collecting historical usage information at the network device comprises collecting the historical usage information at the network device within one or more predetermined time windows, each of the one or more predetermined time windows being a fixed time period.

3. The computer-implemented method of claim 1, wherein analyzing the usage information at the network device comprises analyzing the usage information at the network device by using a linear regression model according to a Gradient descent algorithm.

4. The computer-implemented method of claim 3, further comprising:

using the linear regression model on a sampled instance of an entire feature variable set; and

adjusting a linear regression parameter to minimize an error function of the linear regression model.

5. The computer-implemented method of claim 4, further comprising:

storing the linear regression parameter in a database;

in response to a reboot, retrieving the linear regression parameter from the database; and

using the stored linear regression parameter to predict the usage pattern of the network device.

6. The computer-implemented method of claim 3, further comprising:

analyzing usage information of the plurality of network devices; and

based upon the usage information, predicting an initial usage pattern for a newly deployed network device in the computing network by using a batch-gradient descent algorithm.

7. The computer-implemented method of claim 1, wherein analyzing the usage information at the network device comprises analyzing the usage information at the network device by using a support vector machine based model according to a linear kernel algorithm.

8. The computer-implemented method of claim 1, wherein analyzing the usage information at the network device comprises analyzing the usage information at the network device by using a neural network model or support vector machine based model, the usage information at the network device including correlation between time and traffic pattern at the network device.

9. The computer-implemented method of claim 8, further comprising:

reducing a cost function at the network device by using the neural network model together with one or more forward and backward propagation algorithms.

10. The computer-implemented method of claim 8, further comprising:

analyzing, by a network controller in the computing network, usage information of the plurality of network devices by using the neural network model together with one or more backward propagation algorithms; and

forwarding a machine-learned neural network model to each of the plurality of network devices.

11. The computer-implemented method of claim 1, further comprising:

bringing up a link associated with the network device by initially advertising a high cost metric for the link; and

in response to network devices associated with the link having their forwarding entries programmed, advertising a normal cost metric for the link.

12. The computer-implemented method of claim 1, further comprising:

increasing a cost metric of a link that is to be shut down without removing any programmed forwarding entries at network devices associated with the link; and

shutting down the link after a predetermined period of time.

13. The computer-implemented method of claim 1, further comprising:

dynamically switching the network device or at least one of the one or more corresponding upstream or downstream ports to a low speed mode based at least upon the usage pattern of the network device or the routing protocol information.

14. A computer-implemented method, comprising:

randomly shuffling usage information collected from each of a plurality of network devices at a computing network;

dividing randomly shuffled usage information into two or more subsets of historical usage information;

analyzing at least one subset of the usage information of a corresponding network device by using one or more machine-learning algorithms;

predicting a usage pattern of the network device at a specific future time based at least upon the historical usage information;

collecting routing protocol information of the plurality of network devices and one or more corresponding upstream or downstream ports;

dynamically generating a reduced-power-consumption topology that scales with predicted usage pattern at the plurality of the network devices; and

dynamically adjusting an operation state of at least one of the plurality of network devices or at least one of the one or more corresponding upstream or downstream ports to achieve a power saving at the computing network.

15. The computer-implemented method of claim 14, further comprising:

repetitively analyzing at least one subset of the usage information of the corresponding network device with two or more passes until the one or more machine-learning algorithms converge.

16. The computer-implemented method of claim 15, further comprising:

randomly re-shuffling the usage information collected from the corresponding network device between the two or more passes.

17. A system, comprising:

at least one processor; and

memory including instructions that, when executed by the at least one processor, cause the system to;

collect usage information at a network device, the network device being one of a plurality of network devices at a computing network;

analyze the usage information at the network device by using one or more machine-learning algorithms;

predict a usage pattern of the network device at a specific future time based at least upon the historical usage information;

collect routing protocol information of the network device and one or more corresponding upstream or downstream ports; and

based at least upon predicted usage pattern or the routing protocol information, dynamically adjust an operation state of the network device or at least one of the one or more corresponding upstream or downstream ports to achieve a power saving at the computing network.

18. The system of claim 17, wherein the one or more machine learning algorithms include at least one of linear regression model, neural network model, support vector machine based model, Bayesian statistics, case-based reasoning, decision trees, inductive logic programming, Gaussian process regression, group method of data handling, learning automata, random forests, ensembles of classifiers, ordinal classification, or conditional random fields.

19. The system of claim 17, wherein the instructions when executed further cause the system to:

increase a cost metric of a link that is to be shut down without removing any programmed forwarding entries at network devices associated with the link; and

shut down the link after a predetermined period of time.

20. The system of claim 17, wherein the predicted usage pattern includes a predetermined buffer to take into account unexpected increases of a future traffic rate at the network device.