FogMLTA: um algoritmo com aprendizado de máquina para alocação de recursos em ambientes de computação em névoa

Abstract

Fog computing emerges as an alternative to cloud computing to address latency issues and improve resource allocation in distributed environments. By bringing data processing closer to end devices, this paradigm offers greater flexibility and efficient utilization of resources at the network edge, benefiting applications that require rapid responses. This work proposes a novel task allocation algorithm for fog computing environments, called FogMLTA (Fog Computing Machine Learning-based Task Allocation), which leverages machine learning and considers multiple criteria including bandwidth, processing capacity, and latency to select the most suitable nodes for task execution. The performance of FogMLTA was compared with two classical algorithms: Round Robin, which allocates tasks in a circular and equitable manner, and an algorithm inspired by Dijkstra’s method, focused on selecting the path with the highest bandwidth. All three algorithms were implemented and evaluated using the YAFS (Yet Another Fog Simulator), which allows configuring different scenarios and monitoring relevant parameters for performance analysis. Various topologies and workloads were tested, enabling a comprehensive comparison. The results demonstrate that FogMLTA is capable of enabling a more intelligent and efficient task allocation, improving both the use of computational resources and communication paths.