Define test scenarios for GPU recommendations

[chandrams]

• Validate the GPU recommendation generated for a valid gpu experiment / workload

  • Workload - deployment, statefulset or job
  • Validate the recommendations json for gpu to check for gpu recommendations
    Why is box plots with cpu usage present?

• Validate the GPU recommendation generated for multiple valid gpu experiments / workloads

  • Validate the recommendations json for gpu to check for gpu recommendations

• Invalid simulation in metric profile

  • Invalid parameter value say for device, for eg. B100 in GPU queries in metric profile and validate if no recommendations are not generated

• Generate recommendations when GPU is not present

• Are there any GPU notifications?

• Create an invalid experiment (NA)

  • Target is remote
  • Datasource is not reachable
  • Do we have a list of supported GPUs? How do we specify this? How to make it invalid?