SkyPilot and Other Systems

SkyPilot is a framework for running AI and batch workloads on any infrastructure. SkyPilot can be used in conjunction with some existing system to provide additional benefits.

SkyPilot vs Vanilla Kubernetes

SkyPilot layers on top of your Kubernetes cluster:

1. deliver a better developer experience
2. allow you to scale your infra beyond a single Kubernetes cluster
   • If the Kubernetes cluster is full, SkyPilot can get GPUs from other regions and clouds to run your tasks at the lowest cost
   • Kubernetes cluster is typically constrained to a single region in a single cloud. This is because etcd, the control store for Kubernetes state, can timeout and fail when it faces highers latencies across regions
3. provide faster iteration for interactive development
   • With Kubernetes, a single iteration is a multi-step process involving building a Docker image, pushing it to a registry, updating the Kubernetes YAML and then deploying it.
   • With SkyPilot, a single command (sky launch) takes care of everything.
4. provide simpler YAMLs

SkyPilot is designed to scale across clouds and regions:

1. it allows you to run your tasks on your Kubernetes cluster,
2. and burst to more regions and clouds if needed.
3. In doing so, SkyPilot ensures that your tasks are always running in the most cost-effective region,
4. while maintaining high availability