Serverless Computing: An Investigation of Deployment Environments for Web APIs

Abstract

Cloud vendors offer a variety of serverless technologies promising high availability and dynamic scaling while reducing operational and maintenance costs. One such technology, serverless computing, or function-as-a-service (FaaS), is advertised as a good candidate for web applications, data-processing, or backend services, where you only pay for usage. Unlike virtual machines (VMs), they come with automatic resource provisioning and allocation, providing elastic and automatic scaling. We present the results from our investigation of a specific serverless candidate, Web Application Programming Interface or Web API, deployed on virtual machines and as function(s)-as-a-service. We contrast these deployments by varying the number of concurrent users for measuring response times and costs. We found no significant response time differences between deployments when VMs are configured for the expected load, and test scenarios are within the FaaS hardware limitations. Higher numbers of concurrent users or unexpected user growths are effortlessly handled by FaaS, whereas additional labor must be invested in VMs for equivalent results. We identified that despite the advantages serverless computing brings, there is no clear choice between serverless or virtual machines for a Web API application because one needs to carefully measure costs and factor-in all components that are included with FaaS.