边缘内存占用测试

Why measure memory footprint?

• This platform is designed for a light-weight edge computing deployment, capable of running on devices with few resources (for example, 256MB RAM)
• It is important to know when deploying many pods that it showcases as little memory footprint as possible

KPI's measured

• %CPU
• %Memory
• Resident Set Size (RSS)

How to test

After deployment and provisioning of KubeEdge cloud and edge components in 2 VM's (supported and tested over Ubuntu 16.04) respectively, start deploying pods from 0 to 100 in steps of 5. Keep capturing above KPI's using standard linux ps commands, after each step.

Test setup

Fig 1: KubeEdge Test Setup

Creating a setup

Requirements

• Host machine's or VM's resource requirements can mirror the edge device of your choice
• Resources used for above setup are 4 CPU, 8GB RAM and 200 GB Disk space. OS is Ubuntu 16.04.
• Docker image used to deploy the pods in edge, needs to be created. The steps are:
  1. Go to github.com/kubeedge/kubeedge/edge/hack/memfootprint-test/
  2. Using the Dockerfile available here and create docker image (perftestimg:v1).
  3. Execute the docker command sudo docker build --tag "perftestimg:v1" ., to get the image.

Installation

• For KubeEdge Cloud and Edge:

  Please follow steps mentioned in KubeEdge README.md

• For docker image:

• Deploy docker registry to either edge on any VM or host which is reachable to edge. Follow the steps mentioned here: https://docs.docker.com/registry/deploying/
• Create perftestimg:v1 docker image on the above mentioned host
• Then push this image to docker registry using docker tag and docker push commands (Refer: Same docker registry url mentioned above) [Use this image's metadata in pod deployment yaml]

Steps

1. Check edge node is connected to cloud. In cloud console/terminal, execute the below command

root@ubuntu:~/edge/pod_yamls# kubectl get nodes
NAME                                   STATUS     ROLES    AGE     VERSION
192.168.20.31                          Unknown    <none>   11s
ubuntu                                 NotReady   master   5m22s   v1.14.0

2. On cloud, modify deployment yaml (github.com/kubeedge/kubeedge/edge/hack/memfootprint-test/perftestimg.yaml), set the image name and set spec.replica as 5
3. Execute sudo kubectl create -f ./perftestimg.yaml to deploy the first of 5 pods in edge node
4. Execute sudo kubectl get pods | grep Running | wc to check if all the pods come to Running state. Once all pods come to running state, go to edge VM
5. On Edge console, execute ps -aux | grep edgecore. The output shall be something like:

USER        PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
root     102452  1.0  0.5 871704 42784 pts/0    Sl+  17:56   0:00 ./edgecore
root     102779  0.0  0.0  14224   936 pts/2    S+   17:56   0:00 grep --color=auto edge

6. Collect %CPU, %MEM and RSS from respective columns and record
7. Repeat step 2 and this time increase the replica by 5
8. This time execute sudo kubectl apply -f <PATH>/perftestimg.yaml
9. Repeat steps from 4 to 6.
10. Now repeat steps from 7 to 9, till the replica count reaches 100