Serverless Computing Platforms Performance and Scalability Implementation Analysis

Abstract

Serverless computing and Platform as a Service (PaaS) are similar in that they are both tiny runtime containers that enable infrastructure-free execution of individual lines of code. In 2014, Amazon launched Lambda functions, an event-driven computing platform with a 25 concurrent maximum. Like Platform as a Service (PaaS), but without the overhead of managing the underlying infrastructure, with serverless computing, programs may be executed in a small runtime environment. In 2014, Amazon introduced event-driven computing with the Lambda functions. Functions are often designed for microservices and light workloads, although they are connected to concurrent distributed data processing. We claim that existing serverless computing platforms may allow dynamic applications to operate simultaneously if a partitioned operation can be finished on a small function instance. Results for compute performance in terms of throughput, network bandwidth, file I/O, and concurrent invocations are shown. Even though our solution is yet a prototype, we believe the suggested strategy has a lot of potential. We also talk about potential next steps for running scientific procedures on serverless infrastructures. This paper provides a cost and runtime throughput of several cloud service providers and computes with our prototype model. We do a cost analysis, time conception and performance analysis in the end and discuss the implications for scientific applications' resource management in general.