Compute plane node and HMC host scaling

Edit online

Starting PowerVC version 2.0.2.1, Compute plane nodes are introduced to share the compute services load and allow managing more HMC managed hosts.

When PowerVC manages a host through HMC interface, a new compute service that is responsible for all virtualization operations on that host will be started on the PowerVC management server itself. If you need to manage more than the defined number of hosts, you must maintain multiple PowerVC instances. With the new compute plane node feature that provides better scaling and reliability, you can avoid maintenance of multiple PowerVC instances.

Compute plane node

A compute plane node (CPN) can be any plain virtual machine with RHEL (ppc64le or x86) (SLES support to be provided in future releases) in the same subnet as the PowerVC management node. Users can register the VM that they want to use as a CPN in PowerVC UI with credentials. When a CPN is registered it will only be used for compute services startup. With the compute plane node registered in PowerVC, users have two different options to scale compute hosts in their environment.

Operational architecture for PowerVC - HMC scaling with single node and additional compute plane

This image shows the operational architecture for PowerVC HMC scaling with single node and additional compute plane that is described in the previous paragraphs.

Additional compute plane node(s) can be added to support more compute hosts scaling. Hosts are supported in single-node and extra compute plane (two node) PowerVC. For a single compute plane added to a single node PowerVC, Compute services running on PowerVC and compute plane (one node) do not have high availability (HA). But for two Compute planes added, Compute planes has active-passive configuration under Pacemaker, and hosts can be added on the (two node) compute plane cluster.

Operational architecture for PowerVC - HMC scaling with multinode and two compute nodes

This image shows the operational architecture for PowerVC HMC scaling with multinode and two compute nodes that is described in the previous paragraphs.

Hosts managed through HMC are supported with multinode feature. Each Compute service running within PowerVC cluster has active-passive configuration under Pacemaker. Optionally, Compute plane node(s) can be added as scaling is required.

Considerations

Consider the following while using the compute plane node feature.
  • After a compute plane node is registered, it can only be used for any new compute services (new hosts registered). If compute plane node reaches the maximum capacity, you must add more compute plane nodes. PowerVC management nodes is not used for the new services.
  • APIs and UI remain same for adding HMC hosts.
  • You must restart powervc nova services to re-balance load in any of the following cases.
    • When you add any new compute plane node.
    • When one of the source compute plane goes down, compute services will be balanced onto available compute planes. Once the source compute plane comes back from down state, the compute services will not be moved back automatically.

Limitation

Removal of compute plane nodes with active or failed compute services is not allowed. However, compute plane nodes with inactive service can be de-registered.