Overview of Lingjun node pools
Container Service for Kubernetes (ACK) manages Lingjun nodes through Lingjun node pools, which support node pool lifecycle management, batch node addition and removal, node configuration, scheduling and orchestration, monitoring and diagnosis, and automated O&M, among other capabilities. They also provide enhancements for AI and high-performance computing (HPC) scenarios, such as Remote Direct Memory Access (RDMA) networking and GPU scheduling.
Binding relationship
A Lingjun node pool in ACK is bound one-to-one to a node group of Intelligent Computing Lingjun (a Lingjun bare metal cluster). The node group of a Lingjun cluster corresponds to a single Lingjun node pool in an ACK cluster, and each Lingjun node can belong to only one Lingjun node pool. Partition Lingjun nodes in an ACK cluster into different node pools for differentiated management.
AI and HPC enhanced capabilities
After you Deploy the cloud-native AI suite, Lingjun node pools provide the following enhanced capabilities for AI and high-performance computing (HPC) scenarios:
GPU scheduling: supports topology-aware scheduling for multiple GPUs and, together with GPU container virtualization, provides shared GPU scheduling and isolation.
Job scheduling policies: supports scheduling policies such as Gang, Capacity, and Binpack.
Data acceleration: supports dataset orchestration and access acceleration.
Feature overview
Node lifecycle management
Feature | Description |
Manage Lingjun node pools in the ACK console. | |
Add nodes from a Lingjun node group to a node pool in batches so that ACK can manage them centrally. The operating system, system disk, and data disks of the nodes are not replaced. | |
Remove a node from a Lingjun node pool to stop ACK from managing the node. After a node is removed, the node is not scaled in from the Lingjun node group, and is not automatically released or unsubscribed. |
Node configuration and O&M
Feature | Description |
Configure kubelet startup parameters, such as resource reservation and the maximum number of Pods per node, for all nodes in a node pool to meet the requirements of specific workloads. | |
Configure operating system kernel parameters, such as network stack parameters and file descriptor limits, for all nodes in a node pool to optimize the underlying runtime environment of the nodes. | |
Upgrade the kubelet of nodes in a node pool to the same version as the cluster control plane to maintain version compatibility of the cluster. Only in-place upgrades are supported. Disk replacement upgrades are not supported. | |
ACK automatically monitors abnormal events and, when a fault caused by an underlying event occurs on a Lingjun node, invokes the Lingjun ApproveOperation capability to complete repairs. |
AI and HPC enhanced capabilities
Feature | Description |
Provides RDMA network communication for Pods on Lingjun nodes to meet the requirements of AI training and HPC scenarios for high-bandwidth and low-latency networking. | |
Enables sharing and isolation of GPU resources on Lingjun nodes. Multiple Pods share the same GPU based on quotas, which improves GPU utilization. | |
In model training scenarios, enable the Binpack policy when scheduling Pods. This prioritizes consolidated placement of Pods to reduce cross-node communication latency. | |
Optimizes Pod placement based on the network topology of Lingjun nodes. Pods with intensive communication are preferentially scheduled to nodes close to each other in network distance, reducing cross-node communication latency. |
Billing
When you use Lingjun node pools in ACK managed Pro cluster, the fees consist of the following three parts:
Cluster management fee: ACK managed Pro cluster incurs the fees described in Cluster management fees.
Lingjun node management fee: When a Lingjun node pool contains Lingjun nodes, you are charged the fees described in Lingjun node management fees.
Cloud product resource fees: You are charged the Cloud resource costs incurred by the resources used by the cluster and nodes, such as compute, storage, and network resources.
The free invitational preview for the Lingjun node management feature ends on August 05, 2025, and commercial billing will begin. For more information, see [Billing Announcement] Lingjun Node Management Billing Announcement. During the invitational preview period, the Lingjun node pool feature is free to use. However, other fees for ACK managed Pro clusters, such as cluster management fees and cloud product resource fees, are charged as usual. For more information, see Billing overview.