ACK Lingjun is a cluster type in Container Service for Kubernetes (ACK), built for AI training, inference, and High Performance Computing (HPC) workloads. It pairs fully-managed, highly-available Kubernetes control planes with Lingjun computing nodes, and can be used as the cloud-native base of Platform for AI.
This document is for platform engineers and AI/HPC infrastructure teams evaluating a managed Kubernetes solution for large-scale GPU workloads.
Prerequisites
Before you begin, ensure that you have:
-
A Lingjun cluster with ACK activated For more information, see Create a Lingjun cluster with ACK activated., created in the Intelligent Computing LINGJUN console
How it works
ACK Lingjun decouples software from hardware and integrates with Alibaba Cloud services to deliver stable, secure infrastructure for cloud-native AI workloads.
Control planes are deployed across three availability zones by default, giving you high availability without manual setup. ACK manages the control plane lifecycle — you focus on your workloads.
Features
Cluster management
ACK Lingjun provides the same cluster management capabilities as ACK Pro clusters. ACK creates and manages the control planes. Lifecycle operations — granting permissions, monitoring, updating, and managing components — are all available through the console or API.
Node management
Lingjun node pools let you deploy and manage Lingjun computing nodes with the same experience as Elastic Compute Service (ECS) node pools. Supported operations include:
-
Add or remove nodes in batches
-
Configure, maintain, and diagnose nodes
-
Use fully-managed nodes
-
Schedule workloads to specific nodes
-
Monitor nodes and run automatic O&M tasks
Cloud-native AI
ACK Lingjun ships with components tuned for AI and HPC workloads:
| Capability | What it does |
|---|---|
| Topology-aware multi-GPU scheduling | Topology-aware multi-GPU scheduling (enabled by default) |
| eGPU-based GPU scheduling and isolation | GPU scheduling and isolation based on eGPU, a GPU virtualization component for GPU-accelerated containers |
| Gang scheduling | Gang scheduling for distributed jobs |
| Capacity scheduling | Capacity scheduling for resource quota management |
| Binpack scheduling | Binpack scheduling policy for workload placement |
| Dataset orchestration and access acceleration | Dataset orchestration and access acceleration |
Why ACK Lingjun for AI/HPC
ACK Pro clusters handle general-purpose Kubernetes workloads well. ACK Lingjun adds what AI and HPC jobs specifically need: native Lingjun node pool integration, GPU-aware scheduling policies, and pre-installed AI components — without configuring them yourself.
Security and stability
ACK Lingjun includes the same enterprise-class features as ACK Pro clusters: managed control planes, cross-zone high availability, and SLAs with compensation clauses. No manual cluster provisioning is required. The setup meets the reliability requirements of large-scale production AI environments.
Simplified operations
Deep integration with Intelligent Computing LINGJUN and Alibaba Cloud services means cluster operations and node O&M are largely automated. Managing Lingjun nodes follows the same workflow as ECS nodes, significantly reducing adaption and O&M costs.
GPU efficiency and scheduling
GPU sharing, GPU scheduling, topology-aware GPU scheduling, and priority-based job queue management work together to improve utilization and execution efficiency across AI training and inference workloads. A unified scheduling layer across AI resources and jobs provides a standard method to manage and deliver workloads.