Beyond the Buzz: A Pragmatic Take on Inference Disaggregation

TL;DR

This paper systematically evaluates disaggregated inference at scale, revealing its benefits for specific traffic patterns and emphasizing the importance of dynamic rate matching and elastic scaling for optimal performance.

Contribution

It provides the first large-scale analysis of disaggregated inference, offering practical insights and strategies for deploying disaggregation effectively.

Findings

Disaggregation is most effective for prefill-heavy traffic and larger models.

Dynamic rate matching improves system throughput and interactivity.

Elastic scaling is critical for achieving Pareto-optimal performance.

Abstract

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.