Granted Patent B2

US 12,572,401 · App. 18/018,653 · Granted Mar 10, 2026

Inventors: Sree Nandan Atur (Newark, CA); Ravi Kumar Alluboyina (Santa Clara, CA)

Assignee: Rakuten Symphony, Inc.

G06F11/008G06F11/0709G06F11/0712G06F11/0793

Abstract

Claims (48)

1 . A computer implemented method comprising:

receiving network packets over time at a platform;

monitoring metrics derived from platform hardware components and derived from one or more workloads utilizing the platform hardware components and processing data contained in the network packets;

formulating model training data from the metrics;

training a health check model using the model training data;

automating execution of the health check model computing a probability that a monitored platform hardware component is on a path to failure;

determining that the probability exceeds a threshold;

relocating a workload from a pod containing the monitored platform hardware component to another pod;

receiving additional network packets over time at the platform; and

processing data contained in the additional network packets using the workload at the other pod.

2 . The method of claim 1 , wherein training the health check model comprises training a Recurrent Neural Network (RNN).

3 . The method of claim 1 , wherein training the health check model comprises training a Long Short-Term Memory (LSTM) model.

4 . The method of claim 1 , wherein computing the probability that the monitored platform hardware component is on the path to failure comprises detecting a probability that a network interface card is on a path to failure.

5 . The method of claim 1 , wherein computing the probability that the monitored platform hardware component is on the path to failure comprises detecting a probability that a storage device is on a path to failure.

6 . The method of claim 1 , wherein computing the probability that the monitored platform hardware component is on the path to failure comprises detecting a probability that one of: a processor unit, system memory or a field programmable gate array (FPGA) is on a path to failure.

7 . The method of claim 1 , wherein relocating the workload to the other pod comprises relocating the workload from one pod at a node to another pod at the node.

8 . The method of claim 1 , wherein relocating the workload to the other pod comprises relocating the workload from a pod at a current node to another pod at a destination node.

9 . The method of claim 8 , wherein relocating the workload from the pod at the current node to the other pod at the destination node comprises:

backing up a workload state of the workload from the pod at the current node;

deleting the workload from the pod at the current node;

selecting a destination pod;

importing the workload state to the destination pod; and

restoring the workload at the destination pod.

10 . A computer system comprising:

a processor; and

system memory coupled to the processor and storing instructions configured to cause the processor to:

receive network packets over time at a platform;

monitor metrics derived from platform hardware components and derived from one or more workloads utilizing the platform hardware components and processing data contained in the network packets;

formulate model training data from the metrics;

train a health check model using the model training data;

automate execution of the health check model computing a probability that a monitored platform hardware component is on a path to failure;

determine that the probability exceeds a threshold;

relocate a workload from a pod containing the monitored platform hardware component to another pod;

receive additional network packets over time at the platform; and

process data contained in the additional network packets using the workload at the other pod.

11 . The computer system of claim 10 , wherein the instructions configured to train the health check model comprise instructions configured to train a Recurrent Neural Network (RNN).

12 . The computer system of claim 10 , wherein the instructions configured to train the health check model comprise instructions configured to train a Long Short-Term Memory (LSTM) model.

13 . The computer system of claim 10 , wherein the instructions configured to compute the probability that the monitored platform hardware component is on the path to failure comprise instructions configured to detect a probability that a network interface card is on a path to failure.

14 . The computer system of claim 10 , wherein the instructions configured to compute the probability that the monitored platform hardware component is on the path to failure comprise instructions configured to detect a probability that a storage device is on a path to failure.

15 . The computer system of claim 10 , wherein the instructions configured to compute the probability that the monitored platform hardware component is on the path to failure comprise instructions configured to detect a probability that one of: a processor unit, a system memory, or a field programmable gate array (FPGA) is on a path to failure.

16 . The computer system of claim 10 , wherein the instructions configured to relocate the workload to the other pod comprise instructions configured to relocate the workload from one pod at a node to another pod at the node.

17 . The computer system of claim 10 , wherein the instructions configured to relocate the workload to the other pod comprise instructions configured to relocate the workload from a pod at a current node to another pod at a destination node.

18 . The computer system of claim 17 , wherein the instructions configured to relocate the workload from the pod at the current node to the other pod at the destination node comprise instructions configured to:

back up a workload state of the workload from the pod at the current node;

delete the workload from the pod at the current node;

select a destination pod;

import the workload state to the destination pod; and

restore the workload at the destination pod.

Assignments (2)

Continuity (1)

References Cited (16)