DisagFusion: Asynchronous Pipeline Parallelism and Elastic Scheduling for Disaggregated Diffusion Serving
This work addresses the scalability and efficiency bottlenecks in production diffusion serving, enabling cost-effective deployment across heterogeneous hardware.
DisagFusion tackles the challenges of deploying large diffusion models across heterogeneous GPUs by introducing asynchronous pipeline parallelism and elastic scheduling. It improves throughput by 3.4x-20.5x and reduces latency by 18.5x compared to monolithic serving.
Diffusion-based generation is increasingly powering production content pipelines; however, deploying these models at scale remains a significant challenge. Model weights frequently exceed the memory capacity of commodity GPUs, while the encoder, diffusion transformer (DiT), and decoder stages exhibit highly imbalanced computational and memory footprints. A natural remedy is disaggregated serving-running stages as separate services on heterogeneous GPUs-yet this introduces new bottlenecks, including stage handoff overheads and fast-changing workloads that make cross-stage provisioning and scheduling brittle. This paper presents DisagFusion, enabling asynchronous pipeline parallelism and elastic scheduling for disaggregated diffusion serving. First, DisagFusion introduces asynchronous pipeline parallelism that overlaps computation and stage-to-stage communication to reduce pipeline bubbles and mitigate network jitter. Second, DisagFusion employs a hybrid instance scheduling strategy that combines lightweight performance prediction with runtime feedback to continuously rebalance instance ratio across stages under workload shifts. We implement DisagFusion and evaluate it with modern diffusion models. Compared to a monolithic baseline, DisagFusion improves throughput by 3.4x-20.5x and reduces end-to-end latency by 18.5x, while enabling flexible, cost-efficient deployment across heterogeneous GPUs.