Bringing serverless to private clouds

Identifier:  TFM:2339

Handle: https://hdl.handle.net/20.500.11797/TFM2339

Authors:  Martínez Martínez, Enrique

Abstract:
Serverless computing, especially Function as a Service (FaaS), offers an appealing execution model in which users focus on application logic while the platform manages infrastructure and scalability. Although serverless is usually associated with not very resource intensive use cases like APIs, previous works like PyWren [1] or Lithops [2] have demonstrated its suitability for data-intensive tasks. Despite the fact that public clouds have made serverless widely accessible, adopting serverless computing in private cloud environments remains a challenge. Many organizations operate their own data centers to meet requirements such as security, compliance, or cost. However, reproducing the simplicity and efficiency of serverless computing within these settings is not straightforward. Existing solutions often introduce additional orchestration layers or complex management systems that increase overhead, consume resources, and demand specialized expertise. As a result, private deployments are frequently less efficient, harder to maintain, and unable to fully capture the responsiveness and elasticity associated with the serverless model. The goal of this research is to explore how the advantages of serverless computing can be extended to private cloud environments in a more direct and lightweight manner. Specifically, the objectives are: • Design, implement and validate an approach to enable serverless execution in private cloud environments. • Investigate alternative virtualization and isolation techniques, assessing their impact on critical aspects of serverless performance such as cold start latency, concurrency, and memory usage. This thesis is structured as follows: first, it reviews existing approaches to running serverless workloads in on-premise environments. Next, it proposes and implements an alternative design, which is then validated. Finally, it examines lightweight virtualization technologies and evaluates how they can improve the performance and efficiency of the approach.