RpDelta: Supporting UCR-Suite on Multi-versioning Time Series Data.

In real applications, various cleaning strategies are adopted to repair a specific time series several times for better effects. These multiple versions of the repaired time series, along with the raw time series, are often stored directly in the system for the users. However, as the scale of data explodes, high storage cost becomes a non-negligible problem. To address this problem, we propose RpDelta, a repaired time series storage strategy, under which a repaired time series can be represented as the combination of the raw time series and a differential file to use the storage space more efficiently. Meanwhile, we design a sequential reading strategy based on a finite state machine to make RpDelta adaptive to practical uses, which will almost not introduce additional time and space overheads. We also take the UCR-Suite algorithm as an example to introduce our optimizations on a simultaneous-operation circumstance with the help of RpDelta’s properties. The extensive experiments show the effectiveness and efficiency of our work.