Simulation-as-a-Service with Serverless Computing

Abstract

Simulations are the greatest means for evaluating systems and producing knowledge related to their optimal configuration for production. Simulation systems support the execution of simulations. These can be installed and executed internally to an organisation or can be offered as a service in the cloud. Current simulation-as-a-service (SimaaS) offerings rely on VM or container-based deployments which lead to additional costs due to the charging in an hourly basis. Further, such offerings cannot be easily adapted at runtime to still be able to sustain their promised service level. To resolve these issues, this paper proposes a novel SimaaS architecture and solution which exploits the serverless computing paradigm for reducing the simulation cost based on the actual usage of resources as well as accelerating the simulation time through the limitless, parallelised invocation of functions. Further, this solution relies on the MELODIC/Functionizer multi-cloud platform which enables adapting the simulation execution at runtime in order to sustain the right service level according to the user requirements and preferences. For the validation of our solution, a real business application provided by AI Investments has been used. It aims to optimise investment portfolio using the most advanced AI-based methods and requires heavy computational power to accomplish the respective tasks.