Cedana
Cedana Docs
Cedana
Cedana Docs
  • Welcome to Cedana
  • Get started
    • Authentication
    • Deploying on Kubernetes
    • Using the Cedana Platform
    • Deploying locally
    • Supported container runtimes
  • ARTICLES
    • Performance of Cedana's GPU Interception
    • Cedana vs. CRIU CUDA for GPU Checkpoint/Restore
  • Examples
    • GPU Save, Migrate & Resume (SMR) on Kubernetes
    • Redis Save Migrate & Resume (SMR) on Kubernetes
    • LLaMA Inference GPU Save, Migrate & Resume (SMR)
  • References
    • API
      • API reference
    • Cedana Daemon
    • Cedana CLI
    • GitHub
Powered by GitBook
On this page
  • Quickstart
  • Auth
  • Supercharging Scheduling

Was this helpful?

Edit on GitHub
  1. Get started

Using the Cedana Platform

PreviousDeploying on KubernetesNextDeploying locally

Last updated 7 days ago

Was this helpful?

Quickstart

After you've set up Cedana on Kubernetes, you can immediately start taking advantage of what the system offers by using our UI to manually live migrate or checkpoint/restore, create policies and manage your stateful workloads.

Auth

The same authentication flow as mentioned in Authenticationgets you through to the managed UI interface.

Supercharging Scheduling

While Cedana can work independently of other schedulers or Kubernetes tooling, we've built some custom integrations into existing schedulers and workload management systems to give them an extra Cedana boost.

Currently, we support:

  • for batch workloads, HPC/AI compute and more. Kueue provides a clean queuing experience (similar to SLURM), with fair-sharing, SLAs and quota allocations built-in. With a Cedana integration however, Kueue can now safely preempt workloads when higher priority work comes in or SLAs are shifted, allowing for elastic and safe rebalancing of your cluster's resource allocations.

  • for running on Spot Instances! With Cedana you can safely run long-running workloads on EKS+Spot, allowing for up to 80% savings in compute costs while ensuring that your workloads are safely migrated on revocation.

For a demo of the platform with both features enabled, check out the following video:

Kueue
Karpenter