Galaxy

Introduction

Galaxy is a serverless control plane that implements independent scalability of each component in the control plane, and provides elastic scaling capabilities for the control plane itself.

Tutorial

1. Build and upload images

cd script/
sh buildall.sh # You need to manually modify the image repository address

2. Modify the image address in the yaml file

apiVersion: apps/v1
kind: Deployment
metadata:
  name: dispatcher-deployment
spec:
  replicas: 16
  selector:
    matchLabels:
      app: dispatcher
  template:
    metadata:
      labels:
        app: dispatcher
    spec:
      containers:
        - name: dispatcher-rpc
          image: dispatcher-rpc:latest # Modify image address
          imagePullPolicy: IfNotPresent
         resources:
           limits:
             memory: "4Gi"
             cpu: "2"
          ports:
            - containerPort: 16444

3. Launch all control plane components

kubectl apply -f yaml/

4. Wait until all components are fully started

5. Run the test script to send scaling requests

cd benchmark/coordiantor_test
go run main.go # Default virtual node memory size is 4GB

6. Collect logs of all components

cd script/
sh getlogs.sh

7. Data analysis

cd script/
python analyze.py