一种面向云函数的超轻量运行时环境构建方法

A Lightweight Runtime Environment Construction Method for Cloud Functions

摘要: 随着虚拟化技术和软件体系结构的不断发展，云函数(也称无服务计算)逐渐成为一种新的计算范式，该范式允许用户直接部署函数代码，而不必关心基础架构，同时能够提供极高的弹性和快速生灭能力。然而，现有的云函数通常运行在功能较为完备的容器中，存在启动延迟高、资源共享度低、镜像构建和分发时间长等问题。本文分析并验证了影响Linux容器启动时延和资源共享的关键因素，提出了面向云函数的超轻量运行时环境构建方法，该方法从资源隔离、资源限制、文件系统、网络通信等方面对运行时环境进行定义，基于操作系统虚拟化技术，通过进一步抽取共享层来提高资源共享度，通过降低网络隔离和控制组操作的性能瓶颈来加快启动速度。根据上述方法实现了超轻量运行时环境引擎FRE (Function Runtime Environment)。实验中，从顺序与并发启动时间、内存资源利用率、镜像大小等方面对FRE与当前主流的Docker容器引擎进行对比实验，验证了FRE在云函数场景下的有效性。

Abstract: With the development of virtualization technology and software architecture, cloud function (also known as serverless computing) has gradually become a new computing paradigm, which allows users to directly deploy function codes without having to consider the infrastructure, while providing high flexibility and rapid birth and destroy ability. However, existing cloud functions usually run in fully functional containers, with problems such as high startup latency, low resource sharing, and long image build and distribution time. This paper analyzes and verifies the key factors affecting Linux container startup latency and resource sharing, and presents a cloud function-oriented lightweight runtime environment construction method, which defines the runtime environment from four aspects: resource isolation, resource restriction, file system, and network communication. Based on the operating system virtualization technology, the resource sharing is improved by further extracting the shared layer, and the startup speed is accelerated by reducing the performance bottleneck of network isolation and control group operation. Based on these designs, we implement FRE (Function Runtime Environment), a lightweight runtime environment engine. Finally, we carry out a series of comparative experiments on sequence and concurrent startup time, memory re-source utilization, image volume etc. between FRE and Docker, and verify the effectiveness of the proposed method in the cloud function scenario.