Serving Chain-structured Jobs with Large Memory Footprints with Application to Large Foundation Model Serving

TL;DR

This paper addresses the challenge of serving large foundation models with high memory demands by formulating a novel server chain composition problem, proposing scalable algorithms, and demonstrating significant response time improvements.

Contribution

It introduces the fundamental problem of server chain composition for large memory footprint jobs, proves its NP-hardness, and develops scalable algorithms with performance guarantees.

Findings

Significant reduction in response times using the proposed algorithms

NP-hardness of the server chain composition problem established

Effective load balancing improves large model serving efficiency

Abstract

As a current trend in Artificial Intelligence (AI), large foundation models are increasingly employed as the core of AI services. However, even after training, serving such models at scale remains a challenging task due to their heavy resource footprints, particularly in terms of GPU memory. While recent works revealed unique characteristics of systems serving foundation models that distinguish them from traditional distributed computing systems, there is still a lack of fundamental understanding of the underlying system management problems. This work aims at addressing this gap by extracting a novel problem of "server chain composition" via block placement and cache allocation for serving chainstructured jobs with large memory footprints, which models a fundamental problem in serving large foundation models through pipeline parallelism. After showing the NP-hardness of the optimal solution, the focus is turned to developing scalable algorithms with guaranteed performance under state-of-the-art load balancing. Application of the proposed solution to a distributed large language model (LLM) serving system shows significant reduction of response times compared to state-of-the-art solutions.