Granted Patent B2

US 12,399,742 · App. 18/672,892 · Granted Aug 26, 2025

Inventors: Naorem Khogendro Singh (San Jose, CA); Sandeep Reddy Goli (San Jose, CA); Satyam B. Vaghani (San Jose, CA); Shyan-Ming Perng (San Jose, CA)

Assignee: Nutanix, Inc.

G06F9/5011G06F9/3869G06F9/5077G06F9/542H04L67/125

Abstract

Claims (55)

1. At least one non-transitory computer-readable storage medium including instructions that, when executed, cause a system to:

store a map associating multiple edge systems with respective communication channels, wherein each of the communication channels are associated with a respective instance of a centralized manager, the instances of the centralized manager forming a distributed centralized manager providing a control plane for the edge systems;

access the map to identify a selected communication channel of the communication channels, the selected communication channel associated with a specified edge system, the specified edge system including an edge stack having a machine learning (ML) service configured to provide inference data;

publish an update message using the control plane on the selected communication channel to update the specified edge system including the ML service; and

provide the inference data to a data plane of a computing system.

2. The at least one non-transitory computer-readable storage medium of claim 1 , the instructions, when executed, further cause the system to:

store a map associating one or more channel numbers with one or more communication channels, including the respective communication channels, the one or more channel numbers indicative of which of the one or more communication channels are subscribed to publishing of messages associated with the edge systems, including update messages.

3. The at least one non-transitory computer-readable storage medium of claim 2 , wherein providing the inference data to the data plane of the computing system is based at least on using a data plane communication interface, including an application programming interface, wherein the data plane communication interface is configured to communicate with one or more other data planes.

4. The at least one non-transitory computer-readable storage medium of claim 1 , the instructions, when executed, further cause the system to:

scale the instances of the centralized manager to increase based at least on a determination that a number of tenants, a number of edge systems, or a combination thereof, increases.

5. The at least one non-transitory computer-readable storage medium of claim 1 , wherein the respective instance of the centralized manager is communicatively coupled to the specified edge system including the edge stack having the ML service, via a WebSocket.

6. The at least one non-transitory computer-readable storage medium of claim 1 , the instructions, when executed, further cause the system to:

receive a request, at the ML inference service, for an ML model inference;

based at least on receipt of the request, load the requested ML model inference in an inference engine;

output the inference data; and

provide the inference data to the data plane of the computing system.

7. The at least one non-transitory computer-readable storage medium of claim 1 , wherein the data plane connects one or more sensors to the edge system.

8. The at least one non-transitory computer-readable storage medium of claim 1 , wherein the data plane connects the respective edge system to a cloud.

9. The at least one non-transitory computer-readable storage medium of claim 1 , wherein the data plane is configured to store the inference data, process the inference data, provide access to the inference data, or combinations thereof.

10. A method comprising:

storing a map associating multiple edge systems with respective communication channels, wherein each of the communication channels are associated with a respective instance of a centralized manager, the instances of the centralized manager forming a distributed centralized manager providing a control plane for the edge systems;

accessing the map to identify a selected communication channel of the communication channels, the selected communication channel associated with a specified edge system, the specified edge system including an edge stack having a machine learning (ML) service configured to provide inference data;

publishing an update message using the control plane on the selected communication channel to update the specified edge system including the ML service; and

providing the inference data to a data plane of a computing system.

11. The method of claim 10 , further comprising storing a map associating one or more channel numbers with one or more communication channels, including the respective communication channels, the one or more channel numbers indicative of which of the one or more communication channels are subscribed to a publishing of messages associated with the edge systems, including update messages.

12. The method of claim 11 , wherein providing the inference data to the data plane of the computing system is based at least on using a data plane communication interface, including an application programming interface, wherein the data plane communication interface is configured to communicate with one or more other data planes.

13. The method of claim 10 , further comprising scaling the instances of the centralized manager to increase based at least on a determination that a number of tenants, a number of edge systems, or a combination thereof, increases.

14. The method of claim 10 , wherein the respective instance of the centralized manager is communicatively coupled to the specified edge system including the edge stack having the ML service, via a WebSocket.

15. The method of claim 10 , further comprising:

receiving a request, at the ML inference service, for an ML model inference;

based at least on receipt of the request, loading the requested ML model inference in an inference engine;

outputting the inference data; and

providing the inference data to the data plane of the computing system.

16. The method of claim 10 , wherein the data plane connects one or more sensors to the edge system.

17. The method of claim 10 , wherein the data plane connects the respective edge system to a cloud.

18. The method of claim 10 , wherein the data plane is configured to store the inference data, process the inference data, provide access to the inference data, or combinations thereof.

19. A system comprising:

a central computing system comprising at least one processor and memory storing instructions which when executed by the at least one processor, cause a respective instance of a centralized manager to perform operations, the central computing system hosting one or more instances of the centralized manager, including the respective instance of the centralized manager, the one or more instances of the centralized manager forming a distributed centralized manager providing a control plane for one or more edge systems; and

the one or more edge systems, wherein the respective instance of the centralized manager is configured to:

store a map associating the one or more edge systems with respective communication channels, wherein each of the communication channels are associated with respective instances of the centralized manager;

access the map to identify a selected communication channel of the communication channels, the selected communication channel associated with a specified edge system of the one or more edge systems, wherein the specified edge system includes an edge stack having a machine learning (ML) service configured to provide inference data;

publish an update message using the control plane on the selected communication channel to update the specified edge system including the ML service; and

provide the inference data to a data plane of a computing system.

20. The system of claim 19 , wherein the respective instance of the centralized manager is further configured to store a map associating one or more channel numbers with one or more communication channels, including the respective communication channels, the one or more channel numbers indicative of which of the one or more communication channels are subscribed to a publishing of messages associated with the one or more edge systems, including update messages.

21. The system of claim 20 , wherein providing the inference data to the data plane of the computing system is based at least on using a data plane communication interface, including an application programming interface, wherein the data plane communication interface is configured to communicate with one or more other data planes.

22. The system of claim 19 , wherein the central computing system is further configured to scale the instances of the centralized manager to increase based at least on a determination that a number of tenants, a number of edge systems, or a combination thereof, increases.

23. The system of claim 19 , wherein the respective instance of the centralized manager communicatively coupled to the specified edge system including the edge stack having the ML service, via a WebSocket.

24. The system of claim 19 , further comprising:

provide a request, by the respective instance of the centralized manager and to the ML inference service, for an ML model inference, the request causing the ML inference service to:

load the requested ML model inference in an inference engine,

output the inference data, and

provide the inference data to the data plane of the computing system.

25. The system of claim 19 , wherein the data plane connects one or more sensors to the edge system.

26. The system of claim 19 , wherein the data plane connects the respective edge system to a cloud.

27. The system of claim 19 , wherein the data plane is configured to store the inference data, process the inference data, provide access to the inference data, or combinations thereof.

Assignments (2)

Continuity (4)

References Cited (400)