Intelligent Router for LLM Workloads: Improving Performance Through Workload-Aware Load Balancing

TL;DR

This paper introduces an intelligent, workload-aware routing system for LLM inference that significantly reduces response latency by considering workload phases and leveraging reinforcement learning.

Contribution

It proposes a novel heuristic-guided reinforcement learning router that improves load balancing and end-to-end latency for LLM inference workloads.

Findings

Over 11% latency reduction on public datasets

7.8% latency reduction on real workload data

Framework serves as a benchmark for LLM inference schedulers

Abstract

Large Language Model (LLM) workloads have distinct prefill and decode phases with different compute and memory requirements which should ideally be accounted for when scheduling input queries across different LLM instances in a cluster. However existing scheduling algorithms treat LLM workloads as monolithic jobs without considering the distinct characteristics of the two phases in each workload. This leads to sub-optimal scheduling and increased response latency. In this work, we start by characterizing factors affecting the response latency during LLM inference serving. We establish that better load balancing of inference requests across the available LLM instances can improve the end-to-end latency to a larger extent than merely focusing on optimizing the instance-level scheduler. Motivated by our findings, we propose a heuristic-guided reinforcement learning-based intelligent router for data-driven and workload-aware scheduling. Our router schedules queries across LLM instances by leveraging a trainable response-length predictor, and a novel formulation for estimating the impact of mixing different workloads and achieves over 11% lower end-to-end latency than existing approaches on a mix of public datasets and 7.8% lower end-to-end latency on real workload data with diverse input and output trends from Cloud Provider X. Additionally, the proposed framework can also serve as a standard for benchmarking different LLM inference schedulers since it provides the best latency for a given model, hardware, and instance-level scheduler combination.