GENSERVE: Efficient Co-Serving of Heterogeneous Diffusion Model Workloads

TL;DR

GENSERVE is a system designed to efficiently co-serve heterogeneous diffusion model workloads like T2I and T2V on shared GPU clusters, optimizing resource management and meeting latency SLOs.

Contribution

It introduces step-level resource adaptation leveraging the predictability of diffusion inference, enabling better handling of diverse workloads.

Findings

GENSERVE improves SLO attainment rate by up to 44%.

It effectively manages heterogeneous diffusion workloads on shared GPU clusters.

The system utilizes step boundary preemption and dynamic batching for efficiency.

Abstract

Diffusion models have emerged as the prevailing approach for text-to-image (T2I) and text-to-video (T2V) generation, yet production platforms must increasingly serve both modalities on shared GPU clusters while meeting stringent latency SLOs. Co-serving such heterogeneous workloads is challenging: T2I and T2V requests exhibit vastly different compute demands, parallelism characteristics, and latency requirements, leading to significant SLO violations in existing serving systems. We present GENSERVE, a co-serving system that leverages the inherent predictability of the diffusion process to optimize serving efficiency. A central insight is that diffusion inference proceeds in discrete, predictable steps and is naturally preemptible at step boundaries, opening a new design space for heterogeneity-aware resource management. GENSERVE introduces step-level resource adaptation through three coordinated mechanisms: intelligent video preemption, elastic sequence parallelism with dynamic batching, and an SLO-aware scheduler that jointly optimizes resource allocation across all concurrent requests. Experimental results show that GENSERVE improves the SLO attainment rate by up to 44% over the strongest baseline across diverse configurations.