=================
Basic Terminology
=================
The following definitions describe key concepts and terminology that are
commonly used in the |org| community and in this documentation.
.. glossary::
:sorted:
All-in-one Controller Node
A single physical node that provides a controller function, worker
function, and storage function.
Bare Metal
A node running without Kubelet or hypervisors (for example, application
workloads run directly on the operating system which runs directly on
the hardware).
Worker
A node within a |prod| edge cloud that is dedicated to running application
workloads. There can be 0 to |max-workers| worker nodes in a |prod| edge
cloud.
In a configuration running OpenStack, a worker node:
- is labeled as 'compute'
- may be referred to as a compute node.
- runs virtual switch for realizing virtual networks.
- provides L3 routing and NET services.
Controller
A node within a |prod| edge cloud that runs the cloud management software
(*control plane*). There can be either one or two controller nodes in a
|prod| edge cloud.
- Runs cloud control functions for managing cloud resources.
- Runs all Kubernetes control functions such as kube-apiserver,
kube-controller-manager and kube-schedule
- Runs all OpenStack control functions, such as managing images, virtual
volumes, virtual network, and virtual machines.
- Can be part of a two-node |HA| control node cluster for running control
functions either active/active or active/standby.
Data Network(s)
Networks attached to pci-passthrough and/or sriov interfaces that are made
available to hosted containers or hosted |VMs| for pci-passthrough and/or |SRIOV|
interfaces.
Networks attached to data interfaces of the OpenStack vSwitch, on which the
OpenStack / Neutron provider networks are realized and become the |VM| tenant
networks.
Only worker-type and all-in-one-type nodes, which host end-user containers
and/or |VMs| would have data network(s) attached.
In the case of openstack-compute labelled worker nodes, a data network MUST be
attached to at lease one 'data' (vSwitch) interface.
Deployment Tools
Tools that make the process of distributing, installing, and managing
updates.
Edge Computing
The delivery of computing capabilities to the logical extremes of a
network in order to improve the performance, operating cost and
reliability of applications and services. By shortening the distance
between devices and the resources that serve them, and also reducing
network hops, edge computing mitigates the latency and bandwidth
constraints of today's Internet, ushering in new classes of applications.
From `Open Glossary of Edge Computing <https://github.com/State-of-the-
Edge/glossary/blob/master/edge-glossary.md#edge-computing>`_
Infra Network
A deprecated optional network that was historically used for access to the
storage cluster.
If this optional network is used, all node types are required to be
connected to the INFRA network.
IoT (Internet of Things)
A system of computing devices that can operate with little/no human
interaction.
IPMI Network
An optional network on which |IPMI| interfaces of all nodes are connected.
The network must be reachable using L3/IP from the controller's OAM
interfaces.
You can optionally connect all node types to the |IPMI| network.
Kubernetes Cluster
A set of machines that has a common control plane for running orchestrated
applications.
Management Network
A private network (that is, not connected externally), typically 10GE,
used for the following:
- Internal StarlingX Infrastructure management monitoring and control
- in the case of openstack, it is also used by |VM| I/O access to
virtualized disks in Ceph Storage Cluster
All nodes are required to be connected to the management network.
Node
A computer that is usually a server-class system.
Node Interfaces
All nodes' network interfaces can, in general, optionally be either:
- Untagged single port.
- Untagged two-port LAG and optionally split between redundant L2 switches
running vPC (Virtual Port-Channel), also known as multichassis
EtherChannel (MEC).
- VLAN on either single-port ETH interface or two-port LAG interface.
|OAM| Network
The network on which all external |prod| platform APIs are exposed, (that
is, REST APIs, Horizon web server, |SSH|, and |SNMP|), typically 1GE.
Only controller type nodes are required to be connected to the OAM
network.
.. only:: starlingx
Open Source StarlingX
A set of services that can be used to build cloud infrastructure. The
source code of the services is available under an open source license
that allows access and re-distribution of the codebase. The software
components are created and maintained with an open development process
through collaboration.
PXEBoot Network
An optional network for controllers to boot/install other nodes over the
network.
By default, controllers use the management network for boot/install of
other nodes in the OpenStack cloud. If this optional network is used, all
node types are required to be connected to the PXEBoot network.
A PXEBoot network is required for a variety of special case situations:
- Cases where the management network must be IPv6:
- IPv6 does not support PXEBoot. Therefore, you must configure an IPv4
PXEBoot network.
- Cases where the management network must be |VLAN| tagged:
- Most server's BIOS do not support PXEBooting over tagged networks.
Therefore, you must configure an untagged PXEBoot network.
- Cases where a management network must be shared across regions but
individual regions' controllers want to only network boot/install nodes
of their own region:
- You must configure separate, per-region PXEBoot networks.
|prod|
|prod| is an open source, complete cloud infrastructure software stack for
the edge used by the most demanding applications in industrial |IoT|,
telecom, and other use cases. The platform creates a fusion between
OpenStack and Kubernetes to provide a robust and flexible environment for
all kinds of workloads, let them be containerized, virtualized or bare
metal.
Storage
A node within a |prod| edge cloud that is dedicated to providing file and
object storage to application workloads. There can be 0 or more storage
nodes within a |prod| edge cloud.
- Runs CEPH distributed storage software.
- Part of an |HA| multi-node CEPH storage cluster supporting a replication
factor of two or three, journal caching, and class tiering.
- Provides |HA| persistent storage for images, virtual volumes (that is,
block storage), and object storage.
Virtualization
The act of creating a virtual version of CPU, network or storage device.
Virtual Machines (VM)
An instance of a node provided by software (a hypervisor), which runs
within the host operating system and hardware.