PASCAL: A Phase-Aware Scheduling Algorithm for Serving Reasoning-based Large Language Models
This addresses a specific serving challenge for reasoning-based LLMs, improving deployment efficiency for users and providers, though it is incremental as it builds on existing scheduling frameworks.
The paper tackles the problem of serving reasoning-based large language models, which have extended reasoning phases that delay output and increase Time-To-First-Token (TTFT), by introducing PASCAL, a phase-aware scheduling algorithm that reduces tail TTFT by up to 72% while maintaining answering phase performance.
The emergence of reasoning-based LLMs leveraging Chain-of-Thought (CoT) inference introduces new serving challenges, as their extended reasoning phases delay user-visible output and inflate Time-To-First-Token (TTFT). Existing LLM serving frameworks fail to distinguish between reasoning and answering phases, leading to performance degradation under GPU memory constraints. We present PASCAL, a phase-aware scheduling algorithm that prioritizes reasoning to reduce TTFT while using controlled preemption and token pacing during answering to preserve Quality-of-Experience (QoE). Our hierarchical scheduler combines instance-level placement with intra-instance execution and enables dynamic migration at phase boundaries to balance load and reduce interference. Across benchmarks using DeepSeek-R1-Distill-Qwen-32B, PASCAL reduces tail TTFT by up to 72% while maintaining answering phase SLO attainment, demonstrating the importance of phase-aware scheduling for reasoning-based LLM deployment.