Short-Term QoS Forecasting at the Edge for Reliable Service Applications
Gary White, Siobhán Clarke
Abstract
Accurate short-term forecasts allow dynamic systems to adapt their behaviour when degradation is forecast e.g., transportation forecasting allows for alternative routing of traffic before gridlock. This rationale can be applied to service-oriented computing when creating and managing service applications. Recent approaches to improve reliability in service applications have focused on reducing the time to recovery of application using collaborative filtering-based approaches to make QoS predictions for similar users. In this article, we focus on reducing the time to detection of a failure by forecasting when a service is about to degrade in quality. Previous approaches that have focused on QoS forecasting have used traditional time-series methods that are not designed for sudden peaks caused by network congestion or battery-powered IoT devices that can reduce processing capabilities to extend battery life. More modern recurrent neural network-based approaches such as GRUs and LSTMs have long training times, which are unsuitable for dynamic environments. We propose a noisy echo state network-based approach that has been designed to reduce training time allowing the model to incorporate recent QoS values on devices at the edge. Our results show increased response time forecasting accuracy compared to state of the art approaches when tested on IoT and web services datasets.