Reconstructability-Aware Filtering and Forwarding of Time Series Data in Internet-of-Things Architectures

Time series stemming from monitored Internet-of-Things devices are expected to become one of the main Big Data enablers. If literally almost every object becomes connected, Internet-of-Things platforms will require systems that reduce the incoming data close to the data sources, e.g., On Gateway devices. Otherwise many systems will face problems with storage costs, bandwidth and energy consumption, and database I/O throughput limits. This paper presents a framework and a mechanism for reducing time series data in Internet-of-Things environments based on their reconstruct ability. A two-phase mechanism is described for analyzing, selecting, and enforcing appropriate data reduction handlers in a way that reduces load while maintaining a requested degree of reconstruct ability of the original data. The approach has been evaluated upon real data from publicly available time series. Along with a proof-of-concept that the reconstructed time series have 85% - 99.9% similarity to the original data sets despite having forwarded no more than 20% of the data, the evaluation has shown that our solution can estimate the reconstruct ability of the reduced time series during an analysis phase which usually does not need to last longer than some tens or hundreds of seconds.