Concepts¶

Drove has two main goals: predictable container performance, and simple operations as the cluster grows. This section explains how Drove works and why.

Each page below explains one concept:

• Cluster Anatomy - the parts of a Drove cluster, and what each part does.
• Scheduling - the steps Drove takes, from picking an executor to starting a container.
• Placement Policies - how Drove picks the executor for an instance.
• CPU, NUMA and Memory - how Drove reserves, pins and scales CPU and memory.
• GPU Scheduling - how Drove finds GPUs and hands them to containers.
• Security Profiles - what a container can do on its host.

These pages explain concepts only. The options, defaults and setup steps are in the Setup section. Each concept page links to the page that documents the related options.