| 摘要: |
| 服务器无感知计算是一种新兴的云计算范型, 它允许开发者专注于应用逻辑的开发, 而不需要负责底层复杂的任务管理. 通过这种范型, 开发者可以快速构建更小粒度的应用, 即函数级别的应用. 随着服务器无感知计算的日益流行, 各大云计算厂商相继推出各自的商业服务器无感知平台. 然而, 这些平台的特点尚未得到系统的研究和可靠的比较. 全面分析这些特点可以帮助开发者选择合适的服务器无感知平台, 并以正确的方式开发和执行基于服务器无感知计算的应用. 为此, 开展了面向主流的商业服务器无感知平台特征的实证研究. 涵盖的主流服务器无感知平台包括亚马逊Lambda、谷歌Cloud Functions、微软Azure Functions和阿里巴巴Function Compute. 研究内容主要分为两大类: 特征总结和运行时性能分析. 在特征总结中, 通过对这些服务器无感知平台的官方文档进行探究, 从开发、部署和运行时3个方面的关键特征进行总结和比较. 在运行时性能分析中, 使用代表性的基准测试程序, 从多个维度分析了这些服务器无感知平台提供的运行时性能. 具体而言, 首先分析了影响应用冷启动性能的关键因素, 如编程语言和内存大小. 其次, 探究了服务器无感知平台执行各类任务的执行性能. 基于特征总结和运行时性能分析的结果, 总结了一系列发现, 并为开发者、云计算厂商和研究者提供了具有现实指导意义的启示和潜在的研究机会. |
| 关键词: 度量研究 服务器无感知计算 软件工程 云计算 平台比较 |
| DOI：10.13328/j.cnki.jos.007191 |
| 分类号:TP311 |
| 基金项目:国家重点研发计划(2022YFB4500700); 国家杰出青年科学基金(62325201) |
|
| Measurement Study on Performance of Serverless Platforms |
|
WEN Jin-Feng1, CHEN Zhen-Peng2, LIU Yi3, LIU Xuan-Zhe1
|
|
1.School of Computer Science, Peking University, Beijing 100871, China;2.Department of Computer Science, University College London, London WC1E 6BT, UK;3.Advanced Institute of Big Data, Beijing 100091, China
|
| Abstract: |
| Serverless computing is an emerging paradigm of cloud computing, allowing developers to focus only on application logic development without the need to manage complex underlying tasks. This paradigm allows developers to quickly build smaller-granularity applications, the one at the function level. With the increasing popularity of serverless computing, major cloud computing vendors have introduced their commercial serverless platforms one after another. However, the characteristics of these platforms have yet to be systematically studied and reliably compared. A comprehensive analysis of these characteristics can help developers choose an appropriate serverless platform while developing and executing serverless applications in the right way. To this end, an empirical study is conducted on the characteristics of mainstream commercial serverless platforms. This study involves such mainstream serverless platforms as AWS Lambda, Google Cloud Functions, Microsoft Azure Functions, and Alibaba Function Compute. This study is divided into two major parts: feature summarization and runtime performance analysis. In the feature summarization, the official documents of these serverless platforms are discussed and their key features are summarized and compared in terms of development, deployment, and runtime. In the runtime performance analysis, representative benchmarks are applied to analyze the runtime performance offered by these serverless platforms on a multidimensional basis. Specifically, key factors for the cold-start performance of the applications are first analyzed, such as programming languages and memory sizes. Furthermore, the tasks-executing performance of serverless platforms is discussed. Based on the results of feature summarization and runtime performance analysis, this study sums up a series of findings and provides practical insights and potential research opportunities for developers, cloud computing vendors, and researchers. |
| Key words: measurement study serverless computing software engineering cloud computing platform comparison |
