DuoServe-MoE: Dual-Phase Expert Prefetch and Caching for LLM Inference QoS Assurance

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhang, Yuning, Pinkert, Grant, Yang, Nan, Li, Yanli, Yuan, Dong
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911577732546560
author Zhang, Yuning
Pinkert, Grant
Yang, Nan
Li, Yanli
Yuan, Dong
author_facet Zhang, Yuning
Pinkert, Grant
Yang, Nan
Li, Yanli
Yuan, Dong
contents Large Language Models (LLMs) are increasingly deployed as Internet/Web services (LLM-as-a-Service) with strict latency Service-Level Objectives (SLOs) under tight GPU memory budgets. Mixture-of-Experts (MoE) models improve quality and throughput via sparse expert activation, but serving them efficiently is challenging because expert weights dominate memory footprint and incur costly host--device transfers when offloaded. Moreover, MoE serving exhibits a phase disparity: the prefill phase tends to activate experts densely across many tokens, while the decode phase activates only a few experts per step. A uniform expert loading/caching policy across phases leads to either peak-memory blowup (prefill) or tail-latency inflation (decode). We present DuoServe-MoE, a QoS-oriented MoE serving system that decouples prefill and decode and applies phase-specialized expert scheduling. For prefill, DuoServe-MoE uses a two-stream CUDA pipeline to overlap expert prefetching with non-MoE computation, reducing expert residency time and peak GPU memory. For decode, it employs a lightweight layer-level predictor trained offline from activation traces to prefetch only likely experts without model changes. Experiments on representative MoE LLMs show that DuoServe-MoE improves TTFT by up to $5.34\times$ and end-to-end latency by up to $7.55\times$ over representative baselines, while maintaining low runtime GPU memory usage under resource-constrained deployment.
format Preprint
id arxiv_https___arxiv_org_abs_2509_07379
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle DuoServe-MoE: Dual-Phase Expert Prefetch and Caching for LLM Inference QoS Assurance
Zhang, Yuning
Pinkert, Grant
Yang, Nan
Li, Yanli
Yuan, Dong
Distributed, Parallel, and Cluster Computing
Large Language Models (LLMs) are increasingly deployed as Internet/Web services (LLM-as-a-Service) with strict latency Service-Level Objectives (SLOs) under tight GPU memory budgets. Mixture-of-Experts (MoE) models improve quality and throughput via sparse expert activation, but serving them efficiently is challenging because expert weights dominate memory footprint and incur costly host--device transfers when offloaded. Moreover, MoE serving exhibits a phase disparity: the prefill phase tends to activate experts densely across many tokens, while the decode phase activates only a few experts per step. A uniform expert loading/caching policy across phases leads to either peak-memory blowup (prefill) or tail-latency inflation (decode). We present DuoServe-MoE, a QoS-oriented MoE serving system that decouples prefill and decode and applies phase-specialized expert scheduling. For prefill, DuoServe-MoE uses a two-stream CUDA pipeline to overlap expert prefetching with non-MoE computation, reducing expert residency time and peak GPU memory. For decode, it employs a lightweight layer-level predictor trained offline from activation traces to prefetch only likely experts without model changes. Experiments on representative MoE LLMs show that DuoServe-MoE improves TTFT by up to $5.34\times$ and end-to-end latency by up to $7.55\times$ over representative baselines, while maintaining low runtime GPU memory usage under resource-constrained deployment.
title DuoServe-MoE: Dual-Phase Expert Prefetch and Caching for LLM Inference QoS Assurance
topic Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2509.07379