Investigate how to collect information about GPU usage on AWS

[demeringo]

Problem

In frame of the future integration of GPU impacts in Boavizta API, cloud scanner may need to provide GPU usage data.

Related API issues:

• https://github.com/Boavizta/boaviztapi/issues/150
• https://github.com/Boavizta/boaviztapi/issues/65

Solution

Investigate which kind of information would be usefull to gather and how to do it.

In particular which AWS SDK, API can be called to retrieve GPU usage data (or how to instrument GPU instances to collect data).

Alternatives

Additional context or elements

• https://aws.amazon.com/blogs/machine-learning/monitoring-gpu-utilization-with-amazon-cloudwatch/

[demeringo]

https://docs.aws.amazon.com/dlami/latest/devguide/tutorial-gpu-monitoring-gpumon.html

[demeringo]

A recent (2024) article with uses cases for EKS or Deep Lerning AMIS: https://blog.devops.dev/monitor-gpu-nodes-in-aws-21c8b68a095c