A Comparative Study of the Performance of Serverless Functions and Kubernetes Containers

Abstract

Serverless computing is gaining popularity among software developers and companies as a deployment architecture choice in recent times. This is facilitated by the recent shift in the adoption of containers and microservices in enterprise application architectures. Serverless is a software deployment paradigm that aims to provide developers means to deploy their software without having to manage physical servers or virtual machines themselves. In this approach, the application is developed as individual functions that perform independent tasks and thus can be separately managed and executed. This provides application developers the freedom to write different parts of the application in different programming languages that are best suited for the task. There are several serverless platforms where the code can be deployed, the most popular ones by public cloud providers such as AWS Lambda, Azure Functions, and Google Cloud Functions. However, there are some limitations on these platforms. Application developers are concerned about vendor lock-in when using these platforms since it requires considerable effort to port apps from one cloud platform to another. Furthermore, the applications usually use other services provided by the cloud provider such as data storage or message queuing, which makes it more difficult to make the application work on multiple clouds. Open source serverless frameworks such as Fission, Knative & OpenFaas are an alternative for running serverless functions on-premises that promise to address some of these limitations. While these frameworks have been compared to each other and have been evaluated against the platforms provided by public cloud providers, there have also been extensive cost comparisons of running applications on serverless there has been limited research into how the performance provided by applications running on these platforms compares to running the same applications on managed docker containers.