TurboServe: Serving Streaming Video Generation Efficiently and Economically
2026-06-17 • Distributed, Parallel, and Cluster Computing
Distributed, Parallel, and Cluster Computing
AI summaryⓘ
The authors address a new challenge in generating streaming video, where videos are made bit by bit while keeping each user's session active. They created TurboServe, a system that smartly schedules video sessions on multiple GPUs to deliver video chunks quickly and manage resources efficiently. TurboServe moves sessions between GPUs and adjusts GPU usage based on demand, reducing delay and saving costs. The authors tested TurboServe using real data and showed it improves speed and lowers GPU costs compared to typical methods.
streaming video generationGPU schedulingsession state managementautoscalingload balancingNCCLGPU migrationbatch processinglatency optimizationonline scheduling
Authors
Youhe Jiang, Haoxu Wang, Haotong Bao, Kai Jiang, Jianfei Chen, Jun Zhu, Fangcheng Fu, Jintao Zhang
Abstract
Streaming video generation is emerging as a new serving workload in which users interact with long-lived sessions that generate video progressively, chunk by chunk. Unlike offline video generation or typical LLM serving, streaming video generation must preserve session state across active and idle periods, repeatedly schedule ongoing sessions, and deliver each chunk under a tight latency target. This creates two key serving challenges in multi-user, multi-GPU environments: session duration heterogeneity, where long-running sessions make placement decisions suboptimal over time, and temporal user-demand heterogeneity, where the number of active sessions fluctuates sharply across bursts and idle periods. We present TurboServe, the first serving system designed specifically for streaming video generation workloads. TurboServe formulates serving as an online scheduling problem that jointly coordinates session placement and GPU provisioning. Its closed-loop scheduling algorithm combines a migration-aware placement controller, which rebalances sessions across GPUs to reduce the maximum per-chunk latency, with a load-driven autoscaling controller, which adapts the GPU budget to workload variation for improved cost efficiency. To support these decisions at runtime, TurboServe implements coalesced chunk processing for batching concurrent active sessions on the same GPU, GPU-CPU offloading for session suspension and resumption, and NCCL-based GPU-GPU migration for online rebalancing. We evaluate TurboServe on real-world production traces from Shengshu Technology across multiple model sizes and GPU clusters with up to 64 NVIDIA B300 GPUs. Compared with baseline serving configurations, TurboServe reduces worst-case per-chunk latency by 37.5% and total GPU operating cost by 37.2% on average. Our code is publicly available at https://github.com/shengshu-ai/TurboServe.