A Preliminary Study of Serverless Platforms for Latency Sensitive Applications

Abstract

Serverless computing is the new-age cloud delivery model wherein resources are provisioned only during event-triggered functions. It dramatically improves the flexibility and scalability of applications compared to virtual machine (VM)/container-based service delivery models. As serverless computing is gaining significant impetus, major cloud providers such as Amazon, Microsoft Azure, and Google have launched their respective serverless computing platforms. However, for a user selecting an appropriate service provider (SP), meeting its desired quality-of-services (QoS) is challenging. Moreover, there is not enough public information available to assist the users in making such accurate decisions. Hence, we provide preliminary analysis via real-time experimentation for the users in this work, acting as a stepping stone in selecting an appropriate SP. To evaluate, we consider execution time and execution cost as evaluation metrics to assess different real-world SPs’ performance by considering different workloads. Experimental results show that Azure functions achieved lower execution times than AWS Lambda and Google Cloud Functions, but in terms of execution cost, AWS Lambda costs much lower than the other two platforms.