Predicting Temporal Aspects of Movement for Predictive Replication in Fog Environments

TL;DR

The paper addresses predictive data replication in fog environments by adding temporal context to spatial location prediction. It introduces the Temporal Fusion Multi Order Markov Model (T-FOMM) with independent temporal predictors, notably Holt-Winters Exponential Smoothing, to forecast stay durations and guide timely replication. In simulations using real GeoLife trajectories, T-FOMM-HWES achieves about a 15% reduction in excess data with only about a 1% decrease in data availability, outperforming baseline FOMM and naive replication schemes. The work highlights online adaptability, simple training, and the value of separating temporal and spatial components, offering practical gains for low-latency fog services. Potential future work includes exploring lifelong learning and attention-based DNNs under online constraints.

Abstract

To fully exploit the benefits of the fog environment, efficient management of data locality is crucial. Blind or reactive data replication falls short in harnessing the potential of fog computing, necessitating more advanced techniques for predicting where and when clients will connect. While spatial prediction has received considerable attention, temporal prediction remains understudied. Our paper addresses this gap by examining the advantages of incorporating temporal prediction into existing spatial prediction models. We also provide a comprehensive analysis of spatio-temporal prediction models, such as Deep Neural Networks and Markov models, in the context of predictive replication. We propose a novel model using Holt-Winter's Exponential Smoothing for temporal prediction, leveraging sequential and periodical user movement patterns. In a fog network simulation with real user trajectories our model achieves a 15% reduction in excess data with a marginal 1% decrease in data availability.

Figures (7)

• Figure 1: A fog architecture with a moving client connected to the closest edge device paper_bermbach2017_fog_vision.
• Figure 2: Example spatio-temporal predictive replication in the fog where a client moves through the fog. A spatial prediction has to be made about the future node to which the data should be replicated to (bright red node). Temporal prediction is necessary to know when movement will occur.
• Figure 3: A time series of stay durations from the trajectory data of a single user from the evaluated dataset. The periodical temporal pattern is illustrated with the rolling mean.
• Figure 4: Holt-Winter's Exponential Smoothing time series forecasting smoothed by a moving average of 20, with the user data split for a single user.
• Figure 5: T-FOMM(PCTL) model results show that availability and excess data increase with lower percentiles. Baseline FOMM model included for comparison.