About Host Memory ProvisioningΒΆ

For each NUMA node on a host, you can adjust the amount of memory reserved for platform use, and the size and number of memory pages allocated for use by applications and the virtual switch (vSwitch). vSwitch is only applicable on an openstack-compute labeled worker node, when running the StarlingX OpenStack OpenStack application.

Due to limitations in Kubernetes, only a single huge page size can be used per host.

The amount of platform memory reserved differs between AIO and standard configurations.

Note

For StarlingX Simplex deployments implemented on an Intel Xeon D-based platform, the default platform memory is reduced automatically to limit the number of worker threads, reducing the platform resource usage. You can increase the allotment if required to restore the default maximum number of worker threads.

Memory not reserved for platform and vSwitch use is made available as application memory. By default, no application memory is reserved. You can provision individual NUMA nodes to use either 2 MiB or 1 GiB hugepages. Using larger pages can reduce page management overhead and improve system performance for systems with large amounts of virtual memory and many running instances.

For an openstack-compute labeled worker node, the vSwitch memory is partitioned using 1 GiB huge pages by default. You can change this for individual NUMA nodes to use 2 MiB huge pages for systems without 1 GiB huge page support.

You can use the system host-memory-list and system host-memory-show commands to see how much memory is available for applications. This information is also shown on the Memory tab of the Host Inventory page (see Memory Tab).

For individual containers (or hosted applications if you are running OpenStack), you can specify which page size is required.