[Initiative]: Cloud Native AI Scheduling Challenges Whitepaper

[raravena80]

Name

Cloud Native AI Scheduling Challenges Whitepaper

Short description

Whitepaper about the scheduling challenges for AI/ML workloads in Cloud Native environments

Responsible group

TOC

Does the initiative belong to a subproject?

Yes

Subproject name

Cloud Native AI Working Group

Primary contact

@raravena80

Additional contacts

@zanetworker
@ronaldpetty

Initiative description

https://docs.google.com/document/d/1KNmTKwI_cRXZ0KVBqdBhkO1EuS4PhLIUvT16Y2a5erU/edit?tab=t.0#heading=h.l5opvu2gvmzq

This paper aims to enumerate and educate the various challenges and opportunities regarding optimizing resource allocation (aka scheduling) for Cloud Native Artificial Intelligence (CNAI) workloads. Cloud Native allows easy scaling of resources, making it ideal for AI workloads of two types: training and inference. A standard Cloud Native scheduler like the one provided with Kubernetes is, by default, better suited for microservice-type workloads and not yet for AI-related workloads.

Deliverable(s) or exit criteria

Final draft version to be handed off to the CNCF publishing staff.

[angellk]

thanks @raravena80 ! Could you please add a timeline? Is it possible to breakdown this whitepaper into shorter initiatives that are 1 to 3 months in length to complete?

[leonardpahlke]

🙌 - browsing over this one. There are a lot of comments to the doc linked above. The process seems to be way underway, right? Most comments are a bit dated (a year old). Are meetings & progress happening or is it stale for a bit?

[Satarupa22-SD]

Hi, I would like to contribute as a co-author, could you please let me know the meetings schedule. Thanks!

[rajaskakodkar]

Started a discussion on slack to come up with a plan - https://cloud-native.slack.com/archives/C08Q78J65A7/p1760295102196879

Please chime in if you are willing to contribute

[rajaskakodkar]

The draft of the paper is ready for review. This has been broken down into 5 tech papers (check the tabs) which are aimed to be published.

cc @angellk @raravena80