Less is More: a Lossy Storage Middleware for Multivariate Time Series

Nowadays, devices are producers of big data: whether the device be a smartphone, a workstation or a dedicated server, they output time series that can be of value.It is however challenging to deal with highly multidimensional series: the compromise between disk space and access speed is not trivial to find.To solve this issue, we propose FLIND, a lossy storage middleware improving the storage cost of compressing multivariate series by leveraging their similitude.Thanks to a series selection strategy allowing to pick only matching series, we show FLIND provides a good balance between storage cost and I/O performances.