Preble: Efficient Distributed Prompt Scheduling for LLM Serving
This addresses the challenge of high latency in production LLM serving systems for complex prompts, representing a novel distributed optimization rather than an incremental improvement.
The paper tackles the problem of inefficient distributed LLM serving due to repetitive prompts by proposing Preble, a distributed scheduling system that optimizes KV state reuse and load-balancing, achieving 1.5X to 14.5X lower average latency and 2X to 10X lower p99 latency compared to state-of-the-art systems.
Prompts to large language models (LLMs) have evolved beyond simple user questions. For LLMs to solve complex problems, today's practices are to include domain-specific instructions, illustration of tool usages, and/or long context such as textbook chapters in prompts. As such, many parts of prompts are repetitive across requests. Recent works propose to cache and reuse KV state of prompts. However, they are all confined to a single-GPU optimization, while production LLM serving systems are distributed by nature. This paper proposes Preble, the first distributed LLM serving platform that targets and optimizes for prompt sharing. We designed a distributed scheduling system that co-optimizes KV state reuse and computation load-balancing with a new scheduling algorithm and a hierarchical scheduling mechanism. Our evaluation of Preble with real workloads and request arrival patterns on two open-source LLMs shows that Preble outperforms the SOTA serving systems by 1.5X to 14.5X on average latency and 2X to 10X on p99 latency.