Transformer-Based Model for Cold Start Mitigation in FaaS Architecture

IEEE TRANSACTIONS ON XXX 1 Transformer-Based Model for Cold Start Mitigation in FaaS Architecture Alexandre Savi Fayam Mbala Mouen, Jerry Lacmou Zeutouo, Vianney Kengne Tchendji Member, IEEE Abstract --Serverless architectures, particularly the Function as a Service (FaaS) model, have become a cornerstone of modern cloud computing due to their ability to simplify resource management and enhance application deployment agility. However, a significant challenge remains: the cold start problem. This phenomenon occurs when an idle FaaS function is invoked, requiring a full initialization process, which increases latency and degrades user experience. Existing solutions for cold start mitigation are limited in terms of invocation pattern generalization and implementation complexity. In this study, we propose an innovative approach leveraging Transformer models to mitigate the impact of cold starts in FaaS architectures. Experimental evaluation using a public dataset provided by Azure demonstrates a significant reduction in cold start times, reaching up to 79% compared to conventional methods. I NTRODUCTION T HE rapid emergence of cloud computing has transformed the digital industry by enabling immediate and flexible access to virtualized computing resources. Services such as storage, computing, and networking allow organizations to reduce costs [1] while benefiting from scalable, on-demand infrastructure managed by external providers. At the core of this evolution is the Function as a Service (FaaS) model, an innovative solution that allows functions to be deployed in response to specific events without requiring the management of underlying resources. This model, foundational to serverless architectures, provides increased flexibility and usage-based billing, reducing both costs and operational complexity for developers. Serverless architectures, built upon the FaaS model, abstract infrastructure management, allowing developers to focus solely on the code and features of their applications [2]. This approach optimizes application scalability and simplifies infrastructure management, which is a notable advantage in an era of massive automation.