Large Cluster
-
Scale the medium design further, taking into consideration that we may consider non-converged storage based on use cases and cluster architectures:
-
Control plane running on three (3) nodes
-
Storage running on a minimum of 3 nodes with compute
-
Compute running on all nodes
-
Scale storage nodes in increments of four, which aligns with ODF entitlements of 256TB raw storage
-
Scale compute nodes in increments of one (1)
-
| We recommend that when deploying OpenShift Virtualization, failure domains are also defined at the physical architecture layer when possible to provide better resiliency. Topology labels can be applied to nodes and used by VMs and containerized applications to spread across failure domains. A few examples of spreading across failure domains are: |
-
Use a multirack topology (at least 3), distributing the roles across the multiple racks
-
Each rack is a failure domain
-
Deploy a CLOS Leaf-Spine architecture, where each rack is contained within a Leaf
-
For more information, check the following references: https://kubernetes.io/docs/concepts/scheduling-eviction/topology-spread-constraints/#node-labels https://kubernetes.io/docs/reference/labels-annotations-taints/#topologykubernetesioregion
| The financial company’s final cluster design will fit into the Large category. |