Towards Automating the AI Operations Lifecycle
Matthew Arnold, Jeffrey Boston, et al.
MLSys 2020
Despite multiple papers on autoscaling of stateless services, their adoption remains limited in production. We posit that a major reason is that most production deployments have significant stateful aspects, and yet the community has nei- ther laid out the challenges of autoscaling in the stateful domain, nor proposed any recipes. Motivated by discussions with multiple organizations’ SREs and DevOps personnel, we call for action in the systems community to build stateful autoscalers. We describe the challenges, and lessons learned from stateless scaling, and discuss key directions that may show the path forward
Matthew Arnold, Jeffrey Boston, et al.
MLSys 2020
Felix George, Harshit Kumar, et al.
ICSE 2026
Felix George, Divya Pathak, et al.
ICPE 2026
Genady Ya. Grabarnik, Filippo Poltronieri, et al.
CASCON 2023