jwjwjw3/kepler-powermodels

Scripts for automatically obtaining, collecting, and exporting Power models of Kepler K8s plugin.

User Guide

Overview

This repo contains several bash scripts used to obtain Kepler powermodel from a BARE METAL machine (Not VM!). User needs to run bash scripts one-by-one in a specified order on the machine to collect its power model, manually check outputs of each bash script, and re-run if some script fails.

Requirements

• Root or sudo access on the target machine is required. (target machine should be running a linux OS)
• Internet access of the target machine is needed.

K8s tools used under directory /libs

• Kepler-model-server

  • Using git commit 005852032f2fd03c2e84818ea92abc0541464f95
  • Using Kepler release-0.7.2
  • Both Kepler and kepler-model-server container images from my DockerHub are already provided in the scripts.

• Tekton

  • Already downloaded from https://storage.googleapis.com/tekton-releases/pipeline/latest/release.yaml at 2024-01-05.
  • Official guideline at https://tekton.dev/docs/installation/pipelines.
  • More details can be found at kepler-model-server Git Repo: https://github.com/sustainable-computing-io/kepler-model-server/tree/main/model_training

Scripts behavior description

All scripts the user needs to run are in:

/src/

User needs to cd to this directory, and then run all the following scripts one-by-one, starting from ./1_init_and_start_pipeline.sh. (Note that you may need to run chmod +x *.sh if you haven't granted the scripts execution permissions.) Switch to root account before running the following scripts is highly recommended.

• 0_settings.sh: This script does not need to be executed, user need to write OUTPUT_POWERMODEL_RELATIVE_PATH to the desired relative path from root of this Git Repo.

• 1_init_and_start_pipeline.sh

  • update some automated settings in 0_settings.sh
  • download Kepler-model-server and switch to corresponding commit
  • start a local kind cluster on the target machine
  • start running the tekton pipeline for several hours and obtain runtime data for power model

  After this script finishs, user can run kubectl on the local kind cluster. If seeing errors like kubectl: command not found, then please run: source 0_settings.sh.

  Note that this script finishes in several minutes, but the pipeline runs could take several hours. When checking if pipeline finishes, user can use command kubectl get pods. Prometheus WebUI should be available at http://localhost:9090 after this script finishes.

• 2_retrieve_results.sh

  • deploy in a new pod and download results from local kind cluster to the specified directory in 0_settings.sh.

• 3_cleanup.sh

  • remove the local kind cluster created by 1_init_and_start_pipeline.sh.