Minimum Node Counts¶
This page lists the minimum number of servers required for each Omnia deployment scenario, categorized by functional role and architecture.
Slurm + Kubernetes -- x86_64 and aarch64¶
| Role | Architecture | Quantity |
|---|---|---|
| Omnia Infrastructure Manager (OIM) | x86_64 | 1 |
| Service Kubernetes Control Plane | x86_64 | 3 |
| Service Kubernetes Node | x86_64 | 1 |
| Slurm Control Node | x86_64 | 1 |
| Slurm Node | aarch64 | 1 |
| Login Node | aarch64 | 1 |
| Login Compiler Node | aarch64 | 1 |
Total: 9 nodes
Slurm + Kubernetes -- x86_64¶
| Role | Architecture | Quantity |
|---|---|---|
| Omnia Infrastructure Manager (OIM) | x86_64 | 1 |
| Service Kubernetes Control Plane | x86_64 | 3 |
| Service Kubernetes Node | x86_64 | 1 |
| Slurm Control Node | x86_64 | 1 |
| Slurm Node | x86_64 | 1 |
| Login Node | x86_64 | 1 |
| Login Compiler Node | x86_64 | 1 |
Total: 9 nodes
Note
The Service Kubernetes Node quantity in the Slurm and Kubernetes tables is the baseline minimum. Size the service Kubernetes cluster for the required availability and workload capacity.
Slurm -- x86_64¶
| Role | Architecture | Quantity |
|---|---|---|
| Omnia Infrastructure Manager (OIM) | x86_64 | 1 |
| Slurm Control Node | x86_64 | 1 |
| Slurm Node | x86_64 | 1 |
| Login Node | x86_64 | 1 |
| Login Compiler Node | x86_64 | 1 |
Note
One of either Login Node or Login Compiler Node is required.
Total: 5 nodes
Kubernetes + Telemetry -- x86_64¶
| Role | Architecture | Quantity |
|---|---|---|
| Omnia Infrastructure Manager (OIM) | x86_64 | 1 |
| Service Kubernetes Control Plane | x86_64 | 3 |
| Service Kubernetes Node | x86_64 | 1 |
Total: 5 nodes
Role descriptions¶
| Role | Functional Group | Description |
|---|---|---|
| OIM | -- | Management node. Runs Pulp, OpenCHAMI, provisioning services, and local MinIO when selected. Always exactly 1. Cannot be co-located with cluster roles. |
| Service K8s Control Plane | service_kube_control_plane_<os>_<version>_x86_64 |
Runs Kubernetes API server, etcd, scheduler, and controller-manager. 3 required for HA quorum. |
| Service K8s Node | service_kube_node_<os>_<version>_x86_64 |
Kubernetes worker node. Hosts telemetry pods and application workloads. |
| Slurm Control Node | slurm_control_node_<os>_<version>_x86_64 |
Runs slurmctld, slurmdbd, and MariaDB for job accounting. |
| Slurm Node | slurm_node_<os>_<version>_<architecture> |
Compute nodes running slurmd. Scale out as needed. |
| Login Node | login_node_<os>_<version>_<architecture> |
Interactive SSH access for users to submit jobs. Runs slurmd. |
| Login Compiler Node | login_compiler_node_<os>_<version>_<architecture> |
Login node with compiler toolchain. |
Kubernetes rows in the PXE mapping must use catalog-qualified names, such as
service_kube_control_plane_rhel_10_0_x86_64 or
service_kube_control_plane_rhel_10_2_x86_64. Use the same catalog-selected
RHEL minor version for every Kubernetes control-plane and worker node. Other
supported roles can use the naming form defined by their selected catalog.
Note
The OIM must remain a dedicated, standalone server. Do not co-locate Slurm or Kubernetes roles on the OIM.
Info
- Disk Space -- Disk and memory requirements per node role.
- Ports -- Network ports required per role.
- HA Config -- Kubernetes HA settings.