Installation Requirements
Preparation Before Installation
- The installation package (not required in an FCP environment)
- At least one host that meets the requirements. In an FCP environment, this means at least one head node (management node), one compute node, and one login node.
- Network connectivity between all hosts used to install Fsched. Following common HPC practice, disabling all firewalls within the cluster is recommended.
- (Optional, recommended) Shared storage for data sharing between cluster nodes
- (Optional) Shared storage for head-node high availability
Node Requirements
Supported Architectures and Operating Systems
- x86: CentOS/RHEL 6/7, Rocky Linux 8, RHEL 8*, Ubuntu 18.04/20.04/22.04
- ARM: Amazon Linux 2*, UOS*, Kylin*
- LoongArch: Loongnix*, UOS*, Kylin*
*: Not publicly released yet. Contact the support team for more information.
Management Node Sizing
| Cluster Size | Active Jobs | Minimum Memory (Recommended Memory) | Recommended CPU Cores |
|---|---|---|---|
| Small (1 - 50 nodes) | 1,000 | 2 GB (16 GB) | 2 |
| 10,000 | 4 GB (32 GB) | 2 | |
| Medium (50 - 500 nodes) | 10,000 | 4 GB (32 GB) | 4 |
| 50,000 | 16 GB (64 GB) | 8 | |
| Large (more than 500 nodes) | 50,000 | 16 GB (128 GB) | 8 |
| 500,000 | 64 GB (256 GB) | 16 |
Login Node and Compute Node Sizing
Fsched has no specific sizing requirements for login nodes and compute nodes. Any node that meets the operating system and processor architecture requirements can be used as an Fsched login node or compute node. Size compute nodes based on your workload requirements. For login nodes, decide the sizing based on actual usage, including but not limited to whether access is terminal-only or also provides remote desktops, whether the node is used only for job submission or also runs graphical applications, and whether it is dedicated to a single user or shared by multiple users.
Partition Planning
Partition planning is a critical step when building an Fsched cluster. It can be done during cluster deployment or after the cluster is built. Good partition planning can reduce the learning and operational cost for users, improve resource utilization, and shorten job wait times. Fsched administrators should refer to the relevant sections of the administrator guide when planning and configuring partitions.