Beyond the Buzz: A Pragmatic Take on Inference Disaggregation
This work addresses the problem of practical deployment for disaggregated inference in multi-node systems, offering incremental insights to improve throughput and interactivity trade-offs.
The paper tackled the challenge of deploying disaggregated inference at scale by conducting a systematic study across diverse workloads and hardware, finding it most effective for prefill-heavy traffic and larger models, with dynamic rate matching and elastic scaling being key for Pareto-optimal performance.
As inference scales to multi-node deployments, disaggregation - splitting inference into distinct phases - offers a promising path to improving the throughput-interactivity Pareto frontier. Despite growing enthusiasm and a surge of open-source efforts, practical deployment of disaggregated serving remains limited due to the complexity of the optimization search space and system-level coordination. In this paper, we present the first systematic study of disaggregated inference at scale, evaluating hundreds of thousands of design points across diverse workloads and hardware configurations. We find that disaggregation is most effective for prefill-heavy traffic patterns and larger models. Our results highlight the critical role of dynamic rate matching and elastic scaling in achieving Pareto-optimal performance. Our findings offer actionable insights for efficient disaggregated deployments to navigate the trade-off between system throughput and interactivity.