Administering compute nodes

Administering the compute nodes in the system includes day to day operations such as: starting, initializing, and maintaining compute nodes.

About this task

Compute nodes contain components such as microprocessors, memory, and Ethernet controllers. The components receive power and network connections from the rack.

The number of virtual machines (VMs) that can be deployed on a compute node depends on the vCPU and memory allocations for the VMs.

When high availability (HA) is not enabled:
  • Each regular compute node (16 cores) can accommodate 128 vCPU and approximately 600 GB of memory.
  • Each regular expanded compute node (32 cores) can accommodate 256 vCPU and approximately 1500 GB of memory.

If sufficient storage is available, Cloud Pak System continues to deploy to the compute nodes until either the vCPU or the memory resource is exhausted.

Another factor to consider is the level of performance that you want for your VMs. For example, you can deploy 128 1-vCPU (regular compute node) VMs with 1.5 GB, but each VM gets only 1/8 of a physical core when CPU ratio is set to default 1:8.

When HA is enabled and there are multiple compute nodes in a cloud group, vCPU and memory resources are reserved on each compute node and are not available for customer deployment.

Specific amounts of memory and CPU resources are reserved for migrating virtual machines when a compute node is suspended. The reserved memory and CPU resources are not available for new deployments, but might be used when a compute node is suspended; for example, when a compute node is removed from a cloud group or is placed into maintenance mode. The memory and CPU resources that are reserved on a compute node is based on the number of compute nodes within the cloud group. If there is a single compute node in a cloud group, no memory and CPU resources are reserved. If there are two or more compute nodes in a cloud group, the amount of memory and CPU resources for a single compute node are reserved and divided evenly among all of the compute nodes in the cloud group minus the memory and CPU resources that are used by the hypervisor.

For each compute node, an overhead of approximately 10 - 13% cores is allocated for CPU and an overhead of approximately 3% is allocated for memory.