无服务器云计算模型提供了一个框架,其中服务提供商将底层基础设施管理从开发人员中抽象出来。在此无服务器模型中,FaaS提供了一种事件驱动的、面向函数的计算服务,其特点是细粒度、基于使用量的定价,消除了空闲资源的成本。像AWS Lambda、Azure Functions和Cloud Run Functions这样的平台要求开发人员为其函数配置最少的操作资源以确保成功执行。这种资源分配影响这些函数的运营费用和性能质量。然而,平台透明度的明显缺乏迫使开发人员依赖专家知识或基于经验的临时决策来请求所需的函数资源。这使得在遵守性能约束的同时,最优资源配置成为一项非平凡的任务。此外,虽然商业平台通常按比例扩展CPU和网络带宽等资源与内存,但开源框架允许独立配置函数资源,这给旨在优化其函数的开发人员带来了额外的复杂性。这些复杂性引导研究人员解决开发人员的挑战,并朝着高效的无服务器执行模型前进。在本文中,我们识别了FaaS环境中资源配置技术的不同方面,并提出了一个影响函数设计、配置、运行时成本和性能保证的因素分类法。我们对现有关于资源配置的文献进行了分析,以全面回顾当前关于函数配置的研究。我们还识别了现有的研究空白,并提出了未来的研究方向,以增强函数配置并加强无服务器计算环境的能力,推动其更广泛的采用。
英文摘要:
The serverless cloud computing model offers a framework where the service provider abstracts the underlying infrastructure management from developers. In this serverless model, FaaS provides an event-driven, function-oriented computing service characterised by fine-grained, usage-based pricing that eliminates cost for idle resources. Platforms like AWS Lambda, Azure Functions, and Cloud Run Functions require developers to configure their function(s) with minimum operational resources for its successful execution. This resource allocation influences both the operational expense and the performance quality of these functions. However, a noticeable lack of platform transparency forces developers to rely on expert knowledge or experience-based ad-hoc decisions to request desired function resources. This makes optimal resource configuration a non-trivial task while adhering to performance constraints. Furthermore, while commercial platforms often scale resources like CPU and network bandwidth proportional to memory, open-source frameworks permit independent configuration of function resources, introducing additional complexity for developers aiming to optimise their functions. These complexities have directed researchers to resolve developer challenges and advance towards an efficient server-less execution model. In this article, we identify different aspects of resource configuration techniques in FaaS settings and propose a taxonomy of factors that influence function design, configuration, run-time cost, and performance guarantees. We conduct an analysis of existing literature on resource configuration to present a comprehensive review of current studies on function configuration. We also identify existing research gaps and suggest future research directions to enhance function configuration and strengthen the capabilities of serverless computing environments to drive its broader adoption.
TOSCAdata: Modelling data pipeline applications in TOSCA
无服务器平台允许客户有效地使用云资源,并为所使用的资源的确切数量支付费用。许多专用的开源和商业云数据管理工具可用于处理大量数据。由于缺乏成熟和稳定的标准,这种现代云数据管理工具还不够成熟,无法将通用云应用程序与无服务器平台集成。TOSCA(Topology and Distraction Specification for Cloud Applications)是最流行和最成熟的标准之一,主要关注应用和服务的可移植性以及通用云应用组件的自动化管理。本文提出了TOSCA标准的扩展,TOSCAdata,专注于基于数据管道的云应用程序的建模。保持现代数据管道云应用的需求,TOSCAdata提供了许多TOSCA模型,这些模型可独立部署,可扩展,可重用,同时以管道方式有效处理数据的流动和转换。我们还展示了建议TOSCAdata模型的适用性,以基于Web的云应用程序的背景下,旅游推广作为一个用例场景。