Hardware requirements and recommendations

Review the minimum CPU, Memory, RAM, and disk space requirements for setting up and running IBM® Cloud Private clusters.

Note: Ensure that you review and verify that you meet the increased memory requirements. For more information, see the Hardware requirements section.

The following tables list the minimum system requirements per node for running IBM Cloud Private. The minimum requirement for IBM Cloud Private is one master (and proxy) node, one management node, and one worker node.

• Hardware requirements
  • Single node requirements
  • Multi-node requirements
• Disk space requirements
  • Installation-time disk space requirements
  • Runtime disk space requirements
  • Installation-time CPU and memory requirements
  • Runtime CPU and memory requirements
• PowerVM environment requirements
• Linux on IBM Z and LinuxONE environment requirements
• IBM Edge Computing for Servers requirements

Hardware requirements

Single node requirements

Note for CPUs:

• For a Linux® x86_64 cluster, use a CPU that supports SSE 4.2.
• For a Linux® on Power® (ppc64le) cluster, use a CPU that is version Power8 or higher.
• For a Linux® on IBM® Z and LinuxONE cluster, use a CPU that is either version EC12 or later or any LinuxONE system.

Multi-node requirements

Note: If you do not use a management node in your multi-node cluster, ensure that the master node meets the requirements of the management node plus the master node.

Notes:

1. For CPUs:

   • For a Linux x86_64 cluster, use a CPU that supports SSE 4.2.
   • For a Linux on Power (ppc64le) cluster, use a CPU that is version Power8 or higher.
   • For a Linux on IBM Z and LinuxONE cluster, use a CPU that is either version EC12 or later or any LinuxONE system.

2. A virtual processor core (VPC) is a unit of measurement that is used to determine the licensing cost of IBM products. It is based on the number of virtual cores (vCPUs) that are available to the product. A vCPU is a virtual core that is assigned to a virtual machine or a physical processor core if the server isn’t partitioned for virtual machines. A vCPU is equivalent to a Kubernetes CPU. For more information, see Kubernetes Meaning of CPU .

3. If you disable logging or monitoring, or both, during installation, you can save some RAM and CPU. If you want to enable logging or monitoring, or both, see the sample deployment sizes in Sizing your cluster.

4. By default, systemReserved and kubeReserved reserve 0.2 GHz of CPU processing and 512 MB of memory. You can reserve more resources to make the Kubernetes Platform more stable, especially on the Power platform. Remember: The additional reserved resource must be considered when you are planning your hardware requirement.

5. For the Power platform, the following example contains the suggested values. See Reconfiguring Kubelet in a live cluster for the steps that are required for setting the value. For a single management node, which requires 16 GB of memory, you must expand your management host resource to use at least 20 GB of memory before reconfiguring Kubelet in a live cluster.

   systemReserved:
    cpu: "500m"
    memory: "1500Mi"
    ephemeral-storage: "1Gi"
   kubeReserved:
    cpu: "500m"
    memory: "1500Mi"
    ephemeral-storage: "1Gi"
   

Disk space requirements

Installation-time disk space requirements

Note: Installation-time disk space requirements are required for successful installation. In Installing IBM Cloud Private, Directory for placing offline images is /opt/ibm-cloud-private-3.2.0 and Directory for loading offline images is the directory that you place the installation file.

Runtime disk space requirements

Important: The /var directory is the default storage location for most Docker images and the containers that are used in your IBM Cloud Private cluster. The following directories are used by the installer but do not require significant amounts of disk space:

• /etc/cfc - this directory stores the IBM Cloud Private configuration and certification key file.
• /opt/ibm/cfc - this directory stores the IBM Cloud Private license files.

To prevent disk space issues, mount the default storage directories on separate paths that have larger disk capacities. For more information about mounting the Docker storage directory (/var/lib/docker), see Specifying a default Docker storage directory by using bind mount. You can also use this bind method to mount the other IBM Cloud Private default storage directories. To prevent disk space issues in your cluster, you might want to use a bind mount to mount the following directories:

• Etcd - /var/lib/etcd
• VA - /var/lib/icp
• Kubelet service - /var/lib/kubelet

For more information about mounting the default storage directories, see Specifying other default storage directories by using bind mount.

Notes:

• The disk space requirements that are mentioned in table 4 include the space of the subdirectories and can be reduced if subdirectories are located elsewhere.
• Minimum disk space is the minimum space for running. It is recommended to follow ideal disk space requirements in the production environment.
• If multiple cluster roles are installed on one node, the disk requirement is the sum of the disk requirement for each role. In the production environment, it is not recommended to install multiple cluster roles on one node.
• If the etcd node is separated, the /var/lib/etcd directory is on the etcd node.
• In worker nodes, the /var/lib/docker directory requires more disk space because the production images are placed inside.
• The /var/lib/registry directory is a shared mount from an external shared file system and needs at least 50 GB if the cluster is a mix cluster. It must be large enough to host all of the Docker images that you plan to store in your private image registry.
• The /var/lib/kubelet directory needs at least 10 GB of disk space. If you enable the Vulnerability Advisor, the VA node needs >=100 GB of disk space.

• As etcd is sensitive to disk write latency, consider mounting the following directories to a dedicated disk:

  • /var/lib/etcd
  • /var/lib/etcd-wal

  A reasonably fast disk can typically satisfy the disk speed requirements. As a best practice, you are recommended to use a solid-state drive (SSD). To check whether your disk speed is sufficient, you can use fio, a popular I/O tester. If you do plan to use fio, use fio version 3.5 of later. As an example, the following command uses fio to check the speed for the etcd directory /var/lib/etcd-wal:

  fio --rw=write --ioengine=sync --fdatasync=1 --directory=/var/lib/etcd-wal --size=220m --bs=2300  --name=mytest
  

  Check the 99th percentile of the fdatasync durations within the command output. The 99th percentile of the fdatasync durations should be less than 10 ms. For more information, see Using fio to tell whether your storage is fast enough for etcd . For more information about etcd disk requirements, see etcd disks .

• For the /var/lib/icp/logging directory, the SSD drives are known to perform better than spinning disks. Always use local storage. Avoid remote filesystems such as NFS or SMB. For more information, see Elasticsearch reference .

Installation-time CPU and memory requirements

Note: The jobs in the following table run one time during the installation process.

Runtime CPU and memory requirements

PowerVM environment requirements

The values in the following table apply specifically to PowerVM environments. They do not apply to the Kernel-based virtual machine (KVM) environments or the bare-metal environments.

Recommendations for PowerVM environments:

• At least four virtual machines (VMs – also called LPARs) are recommended; with the master node, management node, vulnerability advisor node, and worker nodes on separate virtual machines. In high scale environments, the etcd node must also be on a separate VM.

• Using shared, uncapped processors is recommended to allow for overcommitting CPUs as needed. If dedicated processors are used, you need to follow the guidelines for Cores (vCPUs). For shared processor pool recommendations, see: Shared processors .

• Large-scale environments can require more vCPUs and processor units. However, try to maintain the vCPU-to-processor unit ratio that is listed in the previous table. For example, if you use eight vCPUs for your management node, use four processor units.

• For more information about configuring etcd nodes, see Setting the node roles in the hosts file.

Linux on IBM Z and LinuxONE environment requirements

The values in the following table apply specifically to Linux on IBM Z and LinuxONE environments.

Note:

• You must use a separate s390x architecture Linux® LPAR or zKVM guest to build Docker images for your applications.

Table 8. Minimum hardware requirements for a multi-node cluster on IBM® Z and LinuxONE environments

Requirement	Boot node	Master node	Proxy node	Worker node	Management node
Number of hosts	1	1, 3, or 5	1 or more	1 or more	1 or more
Cores (IFLs) For more information about IFLs, see Integrated Facility for Linux (IFL)	1	2	1 or more	1 or more	1 or more
Number of CPUs	6 or more	6 or more	3 or more	4 or more	5 or more
CPU speed	>= 2.4 GHz	>= 2.4 GHz	>= 2.4 GHz	>= 2.4 GHz	>= 2.4 GHz
RAM	>=24 GB	>=24 GB	>=16 GB	>=8 GB	>=32 GB
Free Disk space to install	>=100 GB	>=200 GB	>=150 GB	>=150 GB	>=200 GB