This repository contains the test scripts and raw data used in the RunD paper. The paper has been accepted by ATC 2022.
The hardware we used for experiment is an ecs.ebmg6.26xlarge instance from Aliyun. The specification is as follow:
| Item | Spec |
|---|---|
| CPU | Intel Xeon(Cascade Lake) Platinum 8269CY |
| Cores | 104 |
| Memory | 384 GiB |
| Storage | Two ESSDs: 100GB + 500GB |
| OS | Aliyun Cloud OS 2, with Linux kernel 4.19.91 |
The source code of RunD is submitted to the Kata community, and most of them are still under review. The related commits and pull requests to Kata community are here:
Considering that some related binary packages are tightly integrated with our internal system, we provide a screencast of the tool along with the results in our artifact.
Refer to video: ATC_RunD_AE.mp4.
Test scripts used in the video are also provided in the script directory. Logs and analysis results are provided in the ae_data directory.
We are using containerd 1.3.10, kata-containers 2.2.3, and kata-containers 1.12.1 for the baseline experiments. Follow the instructions to install them:
-
Download and unzip.
wget https://github.com/containerd/containerd/releases/download/v1.3.10/cri-containerd-cni-1.3.10-linux-amd64.tar.gz wget https://github.com/kata-containers/kata-containers/releases/download/2.2.3/kata-static-2.2.3-x86_64.tar.xz wget https://github.com/kata-containers/runtime/releases/download/1.12.1/kata-static-1.12.1-x86_64.tar.xz tar xf cri-containerd-cni-1.3.10-linux-amd64.tar.gz -C / tar xf kata-static-2.2.3-x86_64.tar.xz -C / mv /opt/kata /opt/kata2 tar xf kata-static-1.12.1-x86_64.tar.xz -C /
-
Create symbol link to the executables.
ln -sf /opt/kata/bin/containerd-shim-kata-v2 /usr/local/bin/containerd-shim-kata-v2 ln -sf /opt/kata2/bin/containerd-shim-kata-v2 /usr/local/bin/containerd-shim-kata2-v2
-
Copy our config files.
mkdir -p /etc/containerd /etc/kata-containers cp ./config/config.toml /etc/containerd/config.toml cp ./config/configuration-* /etc/kata-containers/ -
Setup volume pool for device mapper.
./script/dmsetup.sh
-
Start containerd.
systemctl start containerd
-
Install dependencies
mkdir result pip3 install pandas yum install smem
Scripts time_kata_test.sh, time_katafc_test.sh, and time_katatemplate_test.sh can run the high-concurrency test for kata-qemu, kata-fc and kata-template.
-
Run high-concurrency tests.
./script/time_kata_test.sh ./script/time_katafc_test.sh ./script/time_katatemplate_test.sh
They may takes several hours to finish. To shorten the time, some concurrency tests can be removed by removing the corresponding concurrency setting in file time_test.conf. The scripts will create a directory named like time_kata_05120948 to store the logs.
We provide python scripts to analyze logs from the tests.
-
Analyze results from high-concurrency tests.
python3 data/time.py python3 data/cpu.py
The python script will create two csv files in the result directory: time.csv and cpu.csv. Each line in the csv file indicates the average cold-start latency and cpu time of a container runtime.
Scripts mem_kata_test.sh, mem_katafc_test.sh, and mem_katatemplate_test.sh can run the high-density test for kata-qemu, kata-fc and kata-template.
-
Run high-density tests.
./script/mem_kata_test.sh ./script/mem_katafc_test.sh ./script/mem_katatemplate_test.sh
Density and memory capacity of containers in the tests can be changed in the file mem_test.conf. The scripts will create a directory named like mem_kata_05120948 to store the logs.
We provide python scripts to analyze logs from the tests.
-
Analyze results from high-density tests.
python3 data/mem.py
The python script will create a csv file for each runtime, named like mem_kata.csv, containing the average memory consumption of containers with different memory capacity in different density.
Scripts density_kata_test.sh, density_katafc_test.sh, and density_katatemplate_test.sh can run the high-density test for kata-qemu, kata-fc and kata-template.
-
Run high-density tests.
./script/density_kata_test.sh ./script/density_katafc_test.sh ./script/density_katatemplate_test.sh
The background density and the concurrency of the tests can be changed in the file density_test.conf. The scripts will create a directory named like density_kata_05120948 to store the logs.
We provide python scripts to analyze logs from the tests.
-
Analyze results from high-density tests.
python3 data/density.py
The python script will create a csv file for each runtime, named like density_kata.csv, containing the average cold-start latency under different background densities and concurrencies.
script: test scripts.
data: python scripts for data analysis.
config: configuration file for containerd and kata-containers.
ae_data: raw logs and analysis results in the RunD AE video.
paper_data: raw logs and analysis results used in the paper.