DSJul 20

Mixture-of-Experts Serving

arXiv:2607.178808.72 citations
Predicted impact top 29% in DS · last 90 daysOriginality Highly original
AI Analysis

For system designers deploying MoE models, this work provides the first principled analysis of dynamic GPU allocation to experts, balancing latency and reconfiguration cost.

The paper introduces a formal model for serving Mixture-of-Experts models and provides a polynomial-time online algorithm with O(√log k) competitive ratio, along with a matching lower bound, and offline results including NP-hardness and a constant-factor approximation.

Mixture-of-Experts (MoE) models route each token to only a few expert networks, distributing the serving load across experts whose popularity shifts over time. A serving system must therefore dynamically decide how many GPUs to assign to each expert, trading off service latency against the cost of reconfiguring the assignment. We introduce a formal model of MoE Serving and initiate a principled study of online and offline algorithms for it. Our main result is a polynomial-time $O(\sqrt{\log k})$-competitive online algorithm, where $k$ is the number of GPUs beyond one per expert. We complement it with a matching $Ω(\sqrt{\log k})$ barrier for the online dual problem underlying our analysis. In the offline setting, we give a constant-factor approximation, show that MoE Serving is NP-hard, and rule out an FPTAS assuming ETH.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes