Optimizing LLM Inference Throughput via Memory-aware and SLA-constrained Dynamic Batching

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Pang, Bowen, Li, Kai, Wang, Feifan
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929747026509824
author Pang, Bowen
Li, Kai
Wang, Feifan
author_facet Pang, Bowen
Li, Kai
Wang, Feifan
contents The increasing adoption of large language models (LLMs) necessitates inference serving systems that can deliver both high throughput and low latency. Deploying LLMs with hundreds of billions of parameters on memory-constrained GPUs exposes significant limitations in static batching methods. Current inference serving systems often treat batch sizes as fixed hyper-parameters, hindering real-time adaptation to varying system conditions. In this paper, we propose a dynamic batching method that continuously monitors memory utilization and adheres to service-level agreements (SLAs) to enable real-time batch size configuration adjustment. The method comprises two core components: a memory-aware batch scheduler that dynamically allocates GPU resources and a latency feedback mechanism that optimizes decoding processes under SLA constraints. The numerical experiments demonstrate throughput gains of 8% to 28% and capacity improvements of 22% compared to traditional static batching methods, while maintaining full compatibility with existing inference infrastructure. These results highlight the effectiveness of dynamic batching in balancing computational efficiency and quality-of-service requirements for contemporary LLM deployment scenarios. The source code of this work is publicly available at https://github.com/KevinLee1110/dynamic-batching.
format Preprint
id arxiv_https___arxiv_org_abs_2503_05248
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Optimizing LLM Inference Throughput via Memory-aware and SLA-constrained Dynamic Batching
Pang, Bowen
Li, Kai
Wang, Feifan
Distributed, Parallel, and Cluster Computing
The increasing adoption of large language models (LLMs) necessitates inference serving systems that can deliver both high throughput and low latency. Deploying LLMs with hundreds of billions of parameters on memory-constrained GPUs exposes significant limitations in static batching methods. Current inference serving systems often treat batch sizes as fixed hyper-parameters, hindering real-time adaptation to varying system conditions. In this paper, we propose a dynamic batching method that continuously monitors memory utilization and adheres to service-level agreements (SLAs) to enable real-time batch size configuration adjustment. The method comprises two core components: a memory-aware batch scheduler that dynamically allocates GPU resources and a latency feedback mechanism that optimizes decoding processes under SLA constraints. The numerical experiments demonstrate throughput gains of 8% to 28% and capacity improvements of 22% compared to traditional static batching methods, while maintaining full compatibility with existing inference infrastructure. These results highlight the effectiveness of dynamic batching in balancing computational efficiency and quality-of-service requirements for contemporary LLM deployment scenarios. The source code of this work is publicly available at https://github.com/KevinLee1110/dynamic-batching.
title Optimizing LLM Inference Throughput via Memory-aware and SLA-constrained Dynamic Batching
topic Distributed, Parallel, and Cluster Computing
url https://arxiv.org/abs/2503.05248