Evaluation of FaaS as an alternative to build HPC environments

Abstract

Running HPC applications in the cloud has proven to be a viable option to conventional parallel or distributed architectures, which require a high degree of management as well as poor resource scaling. The traditional approach for a user is to usually use the Cloud provider to provision resources to virtual machines (VMs), using them in a similar way to an on-premises infrastructure, with the consequent problem of resource management coupled with the degradation of application performance due to contextualization of virtualized environments. Serverless computing allows a user to run code written in the programming language of their choice, without first having to provision a virtual machine. On the other hand, elasticity, availability, scalability, and fault tolerance are provided transparently by the cloud provider. This way it is possible to reduce the complexity of infrastructure management for the developer, allowing him to focus on the logic of the application. Also, economic advantages arise, when just paying for usage time. The work focuses on the challenge of evaluating the cost, not only monetary but also of performance, of migrating HPC applications to serverless environments. This evaluation will allow the decision to be made REGARDING which infrastructure will be used, in order to obtain the best performance benefit.