Controlling an optical bypass switch in a data center based on a neural network output result

Patent metadata

Field	Value
Publication number	US-9755948-B1
Application number	US-201514841723-A
Country	US
Kind code	B1
Filing date	Sep 1, 2015
Priority date	Sep 1, 2015
Publication date	Sep 5, 2017
Grant date	Sep 5, 2017

Abstract

A flow of packets is communicated through a data center including an electrical switch, an optical switch, and multiple racks each including multiple network devices. The optical switch can be controlled to receive packet traffic from a network device via a first optical link and to output that packet traffic to another network device via a second optical link. One network device includes a neural network that analyzes received packets of the flow. The optical switch is controlled to switch based on a result of the analysis performed. In one instance, the optical switch is controlled such that immediately prior to the switching no packet traffic passes from the first optical link and through the optical switch and to the second optical link but such that after the switching packet traffic does pass from the first optical link and through the optical switch and to the second optical link.

First claim

What is claimed is:

1. A method involving an optical switch, wherein the optical switch can be controlled to receive packet traffic from a first network device via a first optical link and to output that packet traffic to a second network device via a second optical link, the method comprising:
(a) using a neural network to analyze packet traffic and determine a number of elephant flows passing through the first network device; and
(b) controlling the optical switch based at least in part on a result of the analyzing of (a) to switch such that immediately prior to the switching of (b) no packet traffic passes from the first network device via first optical link and through the optical switch and via the second optical link to the second network device but such that after the switching of (b) packet traffic does pass from the first network device via the first optical link and through the optical switch and via the second optical link and to the second network device, wherein the optical switch performs circuit switching, wherein the electrical switch performs packet switching, and wherein the neural network is implemented on an Island-Based Network Flow Processor (IBNFP) included in the first network device.

2. The method of claim 1 , wherein the controlling of (b) causes a flow that was being communicated out of a third network device via a first output port of the third network device to an electrical switch to be communicated out of the third network device via a second output port of the third network device and across the first optical link to the optical switch.

3. The method of claim 2 , wherein the first network device is a host server device in a rack, wherein the analyzing of the packet traffic of (a) occurs in the host server device, and wherein the third network device is a top-of-rack switch disposed in the same rack with the host server device.

4. The method of claim 1 , wherein neither the first network device nor the second network device is a top-of-rack switch.

5. The method of claim 1 , wherein at least one of the first and second network devices is a top-of-rack switch.

6. The method of claim 1 , wherein the neural network comprises a multi-layer perceptron circuit, wherein the multi-layer perceptron circuit is an amount of combinatorial digital logic that includes a processor having no instruction counter.

7. The method of claim 1 , wherein the analyzing of (a) involves identifying elephant flows.

8. The method of claim 1 , wherein the analyzing of (a) occurs in the first network device, wherein the method further comprises:
(c) using a neural network in a third network device to analyze packet traffic;
(d) communicating first information from the first network device to a fourth network device, wherein the first information indicates the number of elephant flows traversing the first network device;
(e) communicating second information from the third network device to the fourth network device, wherein the second information indicates the number of elephant flows traversing the third network device;
(f) generating an optical switch configuration instruction on the fourth network device, wherein the generating of the optical switch configuration instruction of (f) is based at least in part on the first information and at least in part on the second information; and
(g) sending the optical switch configuration instruction generated in (f) from the fourth network device to the optical switch thereby controlling the optical switch to switch in (b).

9. A hybrid electrical/optical switch system, comprising:
a plurality of network devices, wherein one of the network devices includes a neural network and uses that neural network to generate information indicative of a number of elephant flows passing through the first network device that are passing through the network device or that have passed through the network device;
an electrical packet switch that is coupled by one or more network cable links to each of the network devices;
an optical circuit switch that is coupled by one or more network cable links to each of the network devices; and
means for generating an optical switch configuration instruction and for sending the optical switch configuration instruction from the means and to the optical circuit switch, wherein the means is for receiving the information indicative of a characteristic and for using the information indicative of a characteristic to generate the optical switch configuration instruction, and wherein the optical switch performs circuit switching, wherein the electrical switch performs packet switching, and wherein the neural network is implemented on an Island-Based Network Flow Processor (IBNFP).

10. The hybrid electrical/optical switch system of claim 9 , wherein the means is also for determining a number of flows that have the characteristic and that pass from said one network device to each of the other network devices, wherein the information indicative of a characteristic that is generated by said one network device comprises one such number for each of the other network devices.

11. The hybrid electrical/optical switch system of claim 9 , wherein the means is also for determining a number of flows that have the characteristic and that pass from each of the other network devices and into said one network device, wherein the information indicative of a characteristic that is generated by said one network device comprises one such number for each of the other network devices.

12. The hybrid electrical/optical switch system of claim 9 , wherein the means is also for determining a number of flows that have the characteristic and that pass between said one network device and each of the other network devices, wherein the information indicative of a characteristic that is generated by said one network device comprises one such number for each of the other network devices.

13. The hybrid electrical/optical switch system of claim 9 , wherein the optical switch configuration instruction is a network packet that is communicated across the network cable to the electrical packet switch.

14. The hybrid electrical/optical switch system of claim 9 , wherein the means is a part of the electrical packet switch.

15. The hybrid electrical/optical switch system of claim 9 , wherein the optical switch has a housing, and wherein the means is disposed in the housing along with the optical switch.

16. The hybrid electrical/optical switch system of claim 9 , wherein the means is not a part of the electrical packet switch and is not a part of the optical circuit switch but rather is a part of another network device of the hybrid electrical/optical switch system.

17. The hybrid electrical/optical switch system of claim 9 , wherein the means comprises a set of processor-executable instructions that is stored in a non-transitory processor-readable medium.

18. A method involving a system, wherein the system comprises a plurality of network devices and an optical circuit switch and an electrical packet switch, wherein the system includes a neural network that analyzes packet traffic somewhere in the system and that outputs information indicative of a result of the analysis, wherein the result of the analysis includes a number of elephant flows, the method comprising:
(a) receiving the information; and
(b) using the information to maintain a log of packet traffic in the system, wherein the log includes a value for each unidirectional network communication path in the system between each pair of the network devices, wherein (a) and (b) are performed

Assignees

• Netronome Systems Inc

Inventors

• Viljoen Nicolaas J

Classifications

• H04L47/2483

  involving identification of individual flows · CPC title

• H04L41/083

  for increasing network speed · CPC title

• H04Q2213/13523

  bandwidth management, e.g. capacity management · CPC title

• H04Q11/0066

  Provisions for optical burst or packet networks · CPC title

• H04Q2011/0079

  Operation or maintenance aspects · CPC title