A Two-Layer Framework for Joint Online Configuration Selection and Admission Control

We study online configuration selection with admission control problem, which arises in LLM serving, GPU scheduling, and revenue management. In a planning horizon with $T$ periods, we consider a two-layer framework for the decisions made within each time period. In the first layer, the decision maker selects one of the $K$ configurations (ex. quantization, parallelism, fare class) which induces distribution over the reward-resource pair of the incoming request. In the second layer, the decision maker observes the request and then decides whether to accept it or not. Benchmarking this framework requires care. We introduce a **switching-aware fluid oracle** that accounts for the value of mixing configurations over time, provably upper-bounding any online policy. We derive a max-min formulation for evaluating the benchmark, and we characterize saddle points of the max-min problem via primal-dual optimality conditions linking equilibrium, feasibility, and complementarity. This guides the design of **SP-UCB–OLP** algorithm, which solves an optimistic saddle point problem and achieves $\tilde{O}(\sqrt{KT})$ regret.

Many computing and business systems must make two decisions for each incoming request: first choose how the system should be configured, and then decide whether to accept the request after seeing its reward and resource needs. Examples include serving large language models, scheduling GPU jobs, and managing limited inventory or capacity. This paper studies how to make these decisions online when the system does not know in advance which configurations will be most useful. We propose a benchmark that fairly measures performance when mixing configurations over time can be better than committing to only one. We then design an algorithm that learns from observed requests, balances exploration with resource conservation, and provably achieves near-optimal performance over time.