Cost efficient function execution in the cloud: a comparison between an academic and a commercial solutions

• Ano de defesa: 2025
• Autor(a) principal: Marques, Luciana da Costa
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Cloud computing; Computação em nuvem; Function as a service; Máquinas virtuais; Spot instances; Virtual machines
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/45/45134/tde-17112025-040120/

The advent of the Cloud enabled customers to scale their services and applications due to not being required anymore to maintain their own datacenters. Ever since then, cloud providers have been offering different types of Compute products that allow customers to run their applications remotely. Since 2009, Amazon Web Services (AWS) offers their spot instances, which are virtual machines with attractive discounts as opposed to their on-demand equivalent, but without a Service Level Agreement (SLA), meaning they can be terminated by AWS at any time. Since 2014, AWS also offers their serverless commercial solution named AWS Lambda, which allows users to execute short, distributed applications in a pre-configured environment infrastructure, which allows them to focus on the business logic and coding of their use case. Function as a Service (FaaS) solutions like AWS Lambda are quite popular for offering a more cost-effective and straightforward solution for executing short tasks, however an investigation of the feasibility of using spot instances, which also offer a generous discount, for distributed function execution has not been explored before. We here present a research conducted to compare the use of an academic FaaS platform, Globus Compute, running on spot instances with a commercial solution, AWS Lambda. Finally, we demonstrate that setting up Globus Compute to run on spot instances is more cost effective than AWS Lambda.