← 研踪 PaperTrailarXiv:2607.00466 · v3
ELDR: Expert-Locality-Aware Decode Routing for PD-Disaggregated MoE Serving
首次提交 2026-07-01 13:34
版本更新 2026-11-06 16:44
雷达观测 2026-11-09 11:19
摘要
Equal load does not mean equal decode latency in prefill-decode (PD) disaggregated mixture-of-experts (MoE) serving. In bandwidth-bound decode, more distinct experts mean more weight loads and higher latency. We present ELDR, a lossless, expert-locality-aware decode router. Prefill signatures predict decode expert usage. Balanced $K$-means fits per-decoder centroids with background refitting. Locality-band routing selects the least-loaded worker among close matches. A block-granular cache keeps signatures exact under prefix caching. We evaluate ELDR in vLLM on up to 40 GPUs. Across three MoE models on 12 public task benchmarks and multilingual WildChat, it reduces median time-per-output-token (TPOT) by 5.4-14.5% on average across request rates and up to 17.8% over the best of four load-balancing baselines, preserving model outputs and token-to-expert assignments.
作者备注：15 pages. Accepted to ACM SIGOPS ATC 2026. Updated to the final conference version
观测到的版本
相邻记录可能跳过中间版本；首次发现 v2 及以上也会标为版本更新。
v3 · 2026-11-06 16:44 · 最新观测版本
ELDR: Expert-Locality-Aware Decode Routing for PD-Disaggregated MoE Serving
Equal load does not mean equal decode latency in prefill-decode (PD) disaggregated mixture-of-experts (MoE) serving. In bandwidth-bound decode, more distinct experts mean more weight loads and higher latency. We present ELDR, a lossless, expert-locality-aware decode router. Prefill signatures predict decode expert usage. Balanced $K$-means fits per-decoder centroids with background refitting. Locality-band routing selects the least-loaded worker among close matches. A block-granular cache keeps signatures exact under prefix caching. We evaluate ELDR in vLLM on up to 40 GPUs. Across three MoE models on 12 public task benchmarks and multilingual WildChat, it reduces median time-per-output-token (TPOT) by 5.4-14.5% on average across request rates and up to 17.8% over the best of four load-balancing baselines, preserving model outputs and token-to-expert assignments.
查看此版本解读 →查看此版本原文 ↗