wao-scheduler

WAO-Scheduler plugin

Kubernetes is a portable, extensible, open source platform for facilitating declarative configuration management and automation, and managing containerized workloads and services. Kubernetes has a huge and fast-growing ecosystem with a wide range of services, support and tools available.

This Document shows the steps to build and deploy Workload Allocation Optimizer (WAO)-Scheduler with power minimizing policy.

Architecture overview

WAO-scheduler gets the CPU usage of each node with Metrics-server, also gets the ambient temperature and CPU temperature with ipmi_exporter. Then WAO-Scheduler predicts power increases with Tensorflow-serving and score the nodes. Finally, WAO-Scheduler selects the node that is expected to the power increases is minimum.

Prerequisites

• Go-lang v1.15.1
• Kubernetes v1.19.7
• ipmi_exporter
  Running on each node
• local docker repository
• docker image of tensorflow serving containing power consumption model of each node.

Build WAO-Scheduler with power minimization policy

1. Download the Kubernetes source code

   *Since the directory structure is different in the latest version, use Kubernetes v1.19.7

   curl -L -o kubernetes.tar.gz https://github.com/kubernetes/kubernetes/archive/v1.19.7.tar.gz
   tar xvzf kubernetes.tar.gz
   

2. Add WAO-Scheduler source code

$ git clone https://github.com/kaz260/WAO-Scheduler
$ cp plugins ~kubernetes/pkg/scheduler/framework/plugins

*Overwrite the file with the same name

1. Add the requied packages Add the prom2json package

   go get github.com/prometheus/prom2json
   

2. Build a WAO-Schduler Build result is output to kubernetes/cmd/proxy/

   cd ~kubernetes/cmd/kube-scheduler/
   CG0_ENABLE=0 go build -mod=mod scheduler.go
   

Deploy to Kubernetes

1. Create a Docker image for scheduler

   Create Dockerfile with the following contents. The oiginal image has been confirmed and may be up to date.

   FROM busybox
   ADD ./kube-scheduler /usr/local/bin/kube-scheduler
   

   Copy the proxy built in the above steps to the same directory as the Dockerfile. Create an image and push it to your local repositoty.

   docker build -t [repository-address]/[image-name] .
   docker image push [repository-address]/[image-name]
   

2. Preparing to start WAO-Scheduler

   • Create configmap

   kubectl create -f wao_configmap.yaml
   

   • Launch metrics-server

   kubectl apply -f https://github.com/kubernetes-sigs/metrics-server/releases/latest/download/components.yaml
   

   • launch tensorflow serving

   kubectl create -f tensorflow-server-dep.yaml
   

   • Labeling Give each node the following label;
   • ambient/max: Maximum ambient temperature in celsius.
   • ambient/min: Minimum ambient temperature in celsius.
   • cpu1/max: Maximum CPU1 temperature in celsius.
   • cpu1/min: Minimum CPU1 temperature in celsius.
   • cpu2/max: Maximum CPU2 temperature in celsius.
   • cpu2/min: Minimum CPU2 temperature in celsius.
   • tensorflow/host: IP address of tensorflow serving.

3. launch MinimizePower scheduler

   kubectl create -f wao-scheduler-deployment.yaml
   

   Success if you can confirm the startup on each node with the following command (Successful if the pod status is [Running])

   kubectl get pod -n kube-system -o wide | grep wao-scheduler
   

   If you want to use wao-scheduler, please refer test.yaml.

   kubectl apply -f test.yaml
   

License

Apache License 2.0, see LICENSE.