My PhD research focuses on Function-as-a-Service (FaaS), a computing paradigm\u2014applicable in both cloud and edge environments\u2014that enables developers to deploy their code as functions without managing the underlying infrastructure, as seen in platforms like AWS Lambda.<\/p>\n
In the edge environment\u2014where the research focus lies\u2014computing resources are positioned closer to data sources to minimize latency, enhance responsiveness, and support real-time processing for applications such as IoT, autonomous systems, and smart infrastructure. The most critical challenge in FaaS is the efficient processing of requests on edge devices\u2014an aspect often overlooked in the literature\u2014despite its potential to reduce CO\u2082 emissions significantly.<\/p>\n
In the initial phase of this research journey, the focus was on minimizing node usage in the FaaS platform by leveraging an auto-scaling mechanism to power down idle devices during periods of low request load. This was achieved through a Mixed-Integer Programming (MIP) formulation designed to solve the problem of minimizing active node usage while maintaining service availability. The details of this work can be found in the corresponding conference article:<\/p>\n
S. Vahabi,<\/strong> F. Righetti, C. Vallati, N. Tonellotto. Energy-Efficient Resource Management for Real-Time Applications in FaaS Edge Computing Platforms. IEEE\/ACM International Conference on Utility and Cloud Computing<\/em>, 2024.<\/p>\n In the second phase, to further optimize energy consumption during function execution on the platform, various time-series prediction models were utilized to estimate the number of function invocations. These estimates enabled the resource manager to proactively allocate resources\u2014specifically processors and memory\u2014on edge devices, aiming for greater efficiency. Out of the ten models considered, three were selected from different architectures and accuracy levels to investigate their impact on overall efficiency. Additionally, a MIP formulation was developed, specifically focusing on the allocation of critical resources, namely CPU and memory, from small to larger scales of resources and workloads of requests. The details of this work can be found in the corresponding journal article:<\/p>\n S. Vahabi,<\/strong> F. Righetti, C. Vallati, N. Tonellotto. The Impact of Prediction Models on Energy-Aware Resource Management in FaaS Platforms. IEEE Access<\/em>, 2025.<\/p>\n In the third phase is the development of a predictive hybrid scheduling framework that couples a zero-shot forecasting model (Google TimesFM) with the proposed LRU+LL (Least Recently Used + Least Loaded) policy to improve function scheduling in heterogeneous FaaS platforms. This framework proactively anticipates future invocation patterns and pre-warms containers accordingly, enabling efficient reuse of warm instances while balancing memory utilization across worker nodes. By combining recency-based eviction with load-aware placement decisions, the scheduler effectively minimizes cold starts, improves resource utilization, and enhances overall system responsiveness. This contribution highlights the benefit of integrating lightweight prediction with adaptive scheduling to achieve more consistent and efficient execution in Edge FaaS environments. This work was presented as a conference paper at:<\/p>\n