Serverless architecture efficiency: an exploratory study

Abstract: Cloud service provider propose services to insensitive customers to use their platform. Different services can achieve the same result at different cost. In this paper, we study the efficiency of a serverless architecture for running highly parallelizable tasks to compare theses services in order to find the most efficient in term of performance and cost. More precisely, we look at the compute time and at the cost per task for a given task. The tasks studied is the count of the occurrence of a given word in a corpus. We compare the serverless architecture to the Apache Spark map reduce technique commonly used for this type of task. Using AWS Lambda for the serverless architecture and Amazon EMR for the Apache Spark map reduce, with similar compute power, we show that the serverless technique achieve comparable performance in term of compute time and cost. We observed that the lambda function is a great approach for real time computing, while EMR is preferable for task that require long compute time.