一种基于动态可寻址会话的服务器无感知计算

Serverless computing based on dynamic-addressable session

服务器无感知计算作为云原生范式中快速发展的新兴技术,因其按需付费、自动资源伸缩和底层环境屏蔽等特点而受到越来越多的开发人员欢迎.FaaS(函数即服务)作为Serverless架构的主要实现方式,以函数粒度对应用进行解耦和执行.大多数云服务提供商也为应用开发人员提供了基于Serverless架构的应用搭建服务,这些服务允许开发人员以函数的形式部署代码,并根据实际的请求量进行自动扩缩容.然而,在部署有状态函数时,由于Serverless架构的无状态特性,管理其中的有状态数据变得复杂,往往无法满足Serverless中函数对有状态数据的访问性能要求.因此,本文提出了一种基于有状态和动态可寻址会话机制的服务器无感知计算系统XFaaS,实现了低开销的有状态数据访问和更高的应用吞吐.实验结果表明,通过采用XFaaS系统部署有状态函数的方式,可以降低有状态数据访问时延3个数量级,并提高2倍以上的函数最大吞吐量.

Serverless computing,as an emerging technology within the cloud-native paradigm,has gained increasing popularity among developers due to its features such as on-demand payment,automatic resource scaling,and offloaded management of underlying infrastructure.Function-as-a-service(FaaS)serves as the primary implementation approach in Serverless architecture,enabling the decoupling and execution of applications at the function level.Most cloud vendors also provide application development services based on Serverless architecture,allowing developers to deploy fine-grained functions and automatically scale based on actual workload.However,the management of stateful data within the Serverless architecture becomes complex when deploying stateful functions.Existing solutions,such as remote or shared storage,usually fail to meet the performance requirements for accessing stateful data in Serverless functions.This paper proposes a Serverless computing system based on a stateful and dynamic-addressable session called XFaaS,which achieves low-cost data access and higher throughput for stateful functions.Experimental results demonstrate that deploying stateful functions through the XFaaS system can reduce the latency of accessing stateful data by three orders of magnitude and increase the maximum throughput of functions by over 2X.