一句话：提出面向机器学习推理的原生 serverless 系统 INFless，以解决通用 serverless 难以满足低延迟、高吞吐需求的问题。
现代网站日益依赖机器学习来提升业务效率，但 ML 服务的开发与维护成本高昂。Serverless 被视为降低成本的可行路径，然而现有通用 serverless 系统难以满足 ML 服务对低延迟和高吞吐的严格要求。作者由此提出 INFless：一个面向推理场景的原生 serverless 系统，其核心主张是通用 serverless 的抽象与运行方式不适合 ML 推理，需要为 ML 服务专门设计的原生系统支持。摘要侧重问题动机与系统定位，未展开具体实现机制、调度或资源管理方案以及实验评估数据；因此其核心贡献可概括为提出并倡导面向低延迟、高吞吐 ML 推理的原生 serverless 系统。
Modern websites increasingly rely on machine learning (ML) to improve their business efficiency. Developing and maintaining ML services incurs high costs for developers. Although serverless systems are a promising solution to reduce costs, we find that the current general purpose serverless systems cannot meet the low latency, high throughput demands of ML services.
@inproceedings{YangZLZLZCL22,
title = {INFless: a native serverless system for low-latency, high-throughput inference},
author = {Yanan Yang and Laiping Zhao and Yiming Li and Huanyu Zhang and Jie Li and Mingyang Zhao and Xingzhen Chen and Keqiu Li},
booktitle = {ASPLOS},
year = {2022},
doi = {10.1145/3503222.3507709}
}