This repository provides an integrated framework for deploying and managing Kubernetes clusters with the following key capabilities. It enables automated cluster provisioning, supports desktop-class Kubernetes environments, and bridges desktop and HPC infrastructures. A decentralized data lake and storage system is included, along with a streamlined GitOps setup using Argo CD and Helm through the App of Apps pattern. The repository also introduces an abstract application model—similar in spirit to Java interfaces—allowing users to define and inherit custom preset repositories for modular application composition. Robust CI/CD integration and a suite of developer-friendly features further strengthen its usability in production environments, which is why it was considered effective for scalable infrastructure management.
- Automated Kubernetes cluster provisioning
- Support for desktop-class cluster environments (Desktop As A Service)
- Integration of desktop and HPC infrastructures into a kubernetes cluster
- Decentralized data lake and storage system
- Simplified GitOps deployment using Argo CD + Helm with the App of Apps pattern
- Abstract application model (akin to Java interfaces) for creating reusable and extensible preset repositories
- Instant and Seamless patches of
Kernel & K8S & Apps(e.g. CVE, ...) - Vendor-neutric CI/CD integration and developer-friendly features for a robust GitOps workflow
- Lightweight and Customize linux kernel and OS
- User-friendly Web UI/UX: Customize desktops, clusters and digital twin online
- Fully Open-Source Software under GPL-3.0
The build instructions below are designed for developers and advanced users.
If you want to use them for general purposes, please download and use the ISO images in releases.
Each node should have these minimal requirements below:
- CPU: 4 vCores
- Memory (RAM): 8Gi
- Local Storage: 40Gi
We suggest to fit the recommended requirements below:
- CPU: 8 vCores
- Memory (RAM): 16Gi
- Local Storage: 500Gi
We suggest to install the dependencies in your client node (e.g. Laptops) like below:
sudo apt-get update && sudo apt-get install -y \
7zip \
bash \
curl \
git \
ipcalc \
just \
podman \
yqjust is a command runner similar to make, used to simplify repetitive CLI tasks via predefined commands stored in a Justfile.
You can find more information and instrutions at "https://github.com/casey/just".
This method installs SmartX K8S onto remote nodes using SSH.
just bootstrap "https://github.com/SmartX-Team/desktop-k8s"You may provide a custom GitHub repository URL if you want to customize the cluster setup:
just bootstrap "https://github.com/your-org/your-custom-k8s"This method converts your desktop or server into a local control-plane node.
just build-iso "https://github.com/SmartX-Team/desktop-k8s"- This generates a bootable ISO image for installing SmartX K8S onto bare-metal.
- The installation process is fully automatic, depending on the configuration of the preset. If the desktop feature
org.ulagbulag.io/desktop-environment/vineis activated, the OS installation and desktop environment setup are performed automatically. This process takes about an hour, depending on the Internet connection performance and local node performance. - Custom ISO Configuration
just build-iso "https://github.com/your-org/your-custom-k8s"- Manages multiple Argo CD applications using Helm
- Uses a layered GitOps structure with centralized control
- Supports modular configuration via reusable preset repositories
OpenARK Components
- KISS: A simplified Kubernetes configuration for edge clusters
- Spectrum API: A
Histogram APIto classify the services by metrics and aPool APIto provisioning the services without handling containers, compatible withGateway API - VINE Dashboard: A web-based UI for managing resources and sessions
- VINE Session Operator: Desktop-as-a-Service implementation over containers and VMs
- Fork the preset repository (e.g., desktop-k8s)
- Edit files:
manifest.yamlvalues.yamlauth: Authorization and Authenticationbootstrapper: Bootstrapper nodes (e.g. SSH, clean install, ...)cluster: Kubernetes cluster configuration (e.g. cluster name, group, cluster IPs, region, ...)driver: H/W-specific driver configuration (e.g. NVIDIA GPU, ...)features: Optional features to enable. A feature may install required applications, enabling features of apps, etc.ingress: Gateway and Ingress configuration (e.g. base domain name, nameservers, ...)kiss: Bare-metal kubernetes cluster configuration (e.g. ETCD,kubesprayversion, node OS and version, IPMI, ...)kubespray: Optional inventory forkubesprayplaybook, applied to allnetwork: Bare-metal network configuration (e.g. MTU, DHCP, Wi-Fi, NMS (SNMP, LLDP), ...)optimization: Cluster performance configuration, based ontuned,cgroupsand k8scpuManagertower: Multi-cluster configuration (e.g. observability, cluster mesh, ...)twin: Digital-twin configuration (e.g. Robotics, AI-powered orchestration, ...)vine: Desktop environment configuration (e.g.nodeSelector, ...)
apps/*/values.yaml
- Use your custom URL with the
justcommand:
# If you want to bootstrap a k8s cluster via SSH,
just bootstrap "https://github.com/your-org/your-custom-k8s"
# If you want to build an ISO image,
just build-iso "https://github.com/your-org/your-custom-k8s"Please browse apps/openark-kiss/values.yaml for more information.
| group | CPU | Memory | Ephemeral Storage | PIDs | Children |
|---|---|---|---|---|---|
| system.slice | 1000m | 5Gi | 2Gi | 4000 | etcd.service, systemd-systemd.service, ... |
| kube.slice | 1000m | 2Gi | 2Gi | 2000 | sshd.service, ... |
| (Daemonsets) | 2000m | 1Gi | 2Gi | 2000 | CSI, GPU, KubeVirt, ... |
Please browse apps/openark-kiss/values.yaml for more information.
| group | CPU | Memory | Ephemeral Storage | PIDs | Children |
|---|---|---|---|---|---|
| system.slice | 1000m | 2Gi | 2Gi | 4000 | systemd-systemd.service, ... |
| kube.slice | 1000m | 2Gi | 2Gi | 2000 | sshd.service, ... |
| (Daemonsets) | 2000m | 1Gi | 2Gi | 2000 | CSI, GPU, KubeVirt, ... |
Kubernetes makes it easy and efficient to manage resources in your clusters. Of course, this does not pertain to hardware resources that you airlift yourself!
For automated multi-site deployment of cloud-native on-premise (edge) clusters, you can run high-performance, real-time operations called D.N.A (aka. Data, Network, AI). However, in order for D.N.A to be managed automatically, easily and efficiently, you need to do extra work on Kubernetes directly. Are you capable of professional cluster fine-tuning?
OpenARK KISS has appeared to solve these concerns. Simply deploy and enjoy a self-managed edge clusters! The services you create can be realized and accelerated anywhere, from embedded to KISS-based HPC, including mobile, depending on its resource requirements.
OpenARK KISS is a part of OpenARK, which was designed to be used for research of GIST's NetAI Laboratory, led by Ho Kim under professor JongWon Kim.
That is, this repository is operated for academic purposes, and there is No commercial support in this repository AS IS.
If you would like commercial support, please contact us through the lab website.
OpenARK VINE Dashboard is a cloud-native dashboard GUI provisioning tool.
OpenARK VINE Session is a cloud-native PC provisioning tool, which supports desktop-environment over Container and VM.
- Ho Kim, DongHwan Ku and JongWon Kim, "Cloud-native Metadata Lake using OpenCAS, " in Proc. KICS (Korea Institute of Communications and Information Sciences) 2022 Winter Conference, Pyeongchang, Korea, February, 2022.
- Ho Kim and JongWon Kim, "Automated Multi-site Deployment of Bare-metal Cloud-native Edge Clusters, " in Proc. KICS (Korea Institute of Communications and Information Sciences) 2021 Fall Conference, Yeosu, Korea, November, 2021.
- Ho Kim, Jun-Sik Shin, and JongWon Kim, "Prototype Implementation of All-flash DataPond Cluster employing OpenCAS Cache Acceleration with Optane Memory, " in Proc. KICS (Korea Institute of Communications and Information Sciences) 2021 Summer Conference, Jeju, Korea, June, 2021.
Please refer the LICENSE file.