Granted Patent B2

US 12,743,317 · App. 18/474,658 · Granted Sep 22, 2026

Inventors: Giuseppe Coviello (Robbinsville, NJ); Kunal Rao (Monroe, NJ); Srimat Chakradhar (Manalapan, NJ); Priscilla Benedetti (Perugia, IT)

Assignee: NEC Corporation

G06F9/5055

Abstract

Claims (33)

1 . A computer implemented method for resource management of stream analytics at each individual node comprising:

computing a mean of output processing rate of microservices in a pipeline;

evaluating a state of each microservice of the microservices in the pipeline, wherein the evaluating of the state employs a Q-learning workflow of a reinforcement learning (RL) method;

selecting a single microservice from the pipeline for updating resources for an action that changes the state in the single microservice that is selected, wherein in exploitation mode the microservice with a maximum Q-value is selected from the Q-learning workflow, wherein a Q-value illustrates a positive change in state for an allocation change;

performing resource allocation updates for the selected microservice; and

updating the state of the selected microservice with a chosen resource allocation.

2 . The computer implemented method of claim 1 , wherein the evaluating of the state of each activation state comprising storing a best Q-value for the each microservice.

3 . The computer implemented method of claim 1 , wherein in exploration mode to select the microservice for updating resources, the microservice for updating the resources is selected randomly.

4 . The computer implemented method of claim 1 , wherein performing the resource allocation updates includes computing an expected output processing rate.

5 . The computer implemented method of claim 1 , wherein the updating the state of the selected microservice with the resource allocation includes a reward computation and updating of a Q-value for the selected microservice using a Bellman's equation.

6 . The computer implemented method of claim 1 , wherein the resources are selected from nodes by availability of CPU cores and random access memory (RAM) availability.

7 . The computer implemented method of claim 1 , wherein the microservices are directed towards object identification from a video stream.

8 . The computer implemented method of claim 7 , wherein the object identification is facial recognition.

9 . A system for resource management of stream analytics at each individual node comprising:

a hardware processor; and

a memory that stores a computer program product, the computer program product when executed by the hardware processor, causes the hardware processor to:

compute a mean of output processing rate of microservices in a pipeline;

evaluate a state of each microservice of the microservices in the pipeline, wherein the evaluate of the state employs a Q-learning workflow of a reinforcement learning (RL) method;

select a single microservice from the pipeline for updating resources for an action that changes the state in the single microservice that is selected, wherein in exploitation mode the microservice with a maximum Q-value is selected from the Q-learning workflow, where a Q-value illustrates a positive change in state for an allocation change;

perform resource allocation updates for the selected microservice; and

update the state of the selected microservice with a chosen resource allocation.

10 . The system of claim 9 , wherein the evaluate of the state of each activation state comprising storing a best Q-value for the each microservice.

11 . The system of claim 9 , wherein in exploration mode to select the microservice for updating resources, the microservice for updating the resources is selected randomly.

12 . The system of claim 9 , wherein the perform the resource allocation updates includes computing an expected output processing rate.

13 . The system of claim 9 , wherein the updating the state of the selected microservice with the resource allocation includes a reward computation and updating of a Q-value for the selected microservice using a Bellman's equation.

14 . The system of claim 9 , wherein the resources are selected from nodes by availability of CPU cores and random access memory (RAM) availability.

15 . The system of claim 9 , wherein the microservices are directed towards object identification from a video stream.

16 . A computer program product for resource management of stream analytics at each individual node comprising a computer readable storage medium having computer readable program code embodied therewith the computer readable program code executable by a hardware processor to cause the hardware processor to:

compute, using the hardware processor, a mean of output processing rate of microservices in pipeline of edge node;

evaluate, using the hardware processor, a state of each microservice of the microservices in the pipeline, wherein the evaluate of the state employs a Q-learning workflow of a reinforcement learning (RL) method;

select, using the hardware processor, a single microservice from the pipeline for updating resources for an action that changes the state in the single microservice that is selected, wherein in exploitation mode the microservice with a maximum Q-value is selected from the Q-learning workflow, wherein a Q-value illustrates a positive change in state for an allocation change;

perform, using the hardware processor, resource allocation update for the selected microservice; and

update, using the hardware processor, the state of the selected microservice with a chosen resource allocation.

Assignments (2)

Continuity (2)

References Cited (22)