An Agent-Based Fog Computing Architecture for Resilience on Amazon EC2 Spot Instances

Abstract

Cloud computing providers have started offering their idle resources as transient servers. Spot instances are transient servers offered by Amazon, whose prices dynamically change over time based on supply and demand. By using appropriate strategies and fault-tolerant mechanisms, users can effectively use spot instances to run applications at a lower price. This paper presents a resilient agent-based fog computing architecture that combines machine learning and a statistical model to predict time to instance revocation and helps to refine fault tolerance parameters and reduce total execution time. The experiments demonstrate that our model predicts with high levels of accuracy reaching 94% success rate what indicates the model is effective under realistic working conditions.