ELDR: Expert-Locality-Aware Decode Routing for PD-Disaggregated MoE Serving
For LLM serving systems using PD-disaggregated MoE models, ELDR addresses the overlooked issue of expert activation-induced latency variation, achieving significant latency reduction without changing model outputs.
ELDR reduces median time-per-output-token (TPOT) by 5.9-13.9% over the best load-balancing baseline for PD-disaggregated MoE serving on up to 40 GPUs, by routing requests based on expert locality.
In prefill-decode (PD) disaggregated LLM serving, each request is assigned to a decode worker after prefill. Existing decode routers balance only load; for mixture-of-experts (MoE) models this is incomplete: equally loaded workers can differ in latency, since each decode step loads the weights of every distinct expert its batch activates. We present ELDR, an expert-locality-aware decode router for PD-disaggregated MoE serving. From a request's prefill expert activations, ELDR builds an expert signature predicting the experts it will activate during generation. Offline, balanced K-means partitions signature space across decode workers; online, locality-band routing sends each request to the least-loaded worker among those best matching its signature. A signature cache, co-indexed with the KV cache at KV-block granularity, keeps signatures exact under prefix caching. Implemented in vLLM and evaluated on deployments of up to 40 GPUs, ELDR reduces median TPOT by 5.9-13.9% over the strongest of four load-balancing baselines across three MoE models and two workloads, with model outputs unchanged.