Sky computing

[hongbo-miao]

/kind feature

Why you need this feature:
This is just an early idea. It would be awesome to support sky computing - training a machine learning model cross multiple clouds (different vendors). It will help potentially lower the cost. ☺️

Describe the solution you'd like:
If possible, somehow integrating with SkyPilot would be awesome.
For example, a naive idea is somehow allowing different Kubeflow pipeline component can run on individual instances on different clouds (AWS, Google Cloud, Azure, etc.) and then collect the experiment results back to the cluster hosting Kubeflow for example AWS.

Here is screenshot from YouTube explaining how SkyPilot works for a cross-cloud machine job:

[andreyvelich]

Hi @hongbo-miao, I think this great idea!
Any thoughts on how SkyPilot can be integrated with Kubeflow Components in addition to KFP ?

[juliusvonkohout]

/transfer community