Kubernetes is the main control plane and central API for practically everything built further up the stack. There are a few reasons for this decision:

• Kubernetes architecture is heavily inspired by the concept of declarative application management, which gives us the ability to manage different common off-the-shelf (COTS) components.
• Though Kubernetes is one of many solutions for orchestration, its biggest benefit is being a universal API for resource management.
• Another important aspect is its ability to extend its API using custom resources and operators without increasing complexity or creating tech debt.
• The final reason is that Kubernetes is already widely adapted across the industry, so no special skills are required to operate it

Figure 11.          Layers of a Kubernetes grid

To make it even easier to adapt Kubernetes - VMware released a Kubernetes support in the vSphere 7.0 release. Along with the deep vSphere integration for configuration and management, VMware provides required Kubernetes integration services like container networking or persistent storage support for containers. Kubernetes adoption is also streamlined by providing a Docker® image registry (based on open-source Harbor project). Support for user authentication using vSphere or any

SSO provider makes it ready for integration into existing enterprise systems.

To speed deployment and time to solution of an on-premises data science platform, the reference architecture from VMware is a perfect starting point to build your own solution.

Figure 12.          Software and versions table