A Two-Layer Framework for Joint Online Configuration Selection and Admission Control
Owen Shen, Haoran Xu, Peter Glynn, Yinyu Ye, Patrick Jaillet · 2026
Read A Two-Layer Framework for Joint Online Configuration Selection and Admission Control by Owen Shen, Haoran Xu, Peter Glynn, Yinyu Ye, Patrick Jaillet (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-07663.txt.
Description
Serving language models and scheduling GPUs both involve picking a configuration and then deciding whether to accept each incoming request, a two-layer online problem. Owen Shen and four co-authors build a method for joint online configuration choice and admission control over a planning horizon.
Each period, a configuration choice induces a reward-resource spread for the next request, which is then accepted or rejected. Benchmarking needs care, so the paper presents a switching-aware fluid oracle that values mixing configurations over time and upper-bounds any online policy, with a max-min form and saddle-point characterization. Operations teams get a principled benchmark and policy method for a ubiquitous serving problem.
Catalog record
| CREATOR | Owen Shen, Haoran Xu, Peter Glynn, Yinyu Ye, Patrick Jaillet |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 124 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-07663