• Floating Point
• Compression
• Algorithm
• Time Series

The widespread use of IoT devices has led to reconsider many aspects of the current data management systems due to the enormous amount of data that this technology produces. The state-of-the-art IoT data format is called time series. Those can be seen as a list of key-value pairs, where the key defines the exact time of measurement, whereas the value is the actual measure. These data are widely used in many Machine Learning applications such as Anomaly Detection, Classification and Regression. Given that the ICT industry has a significant impact on our environment, traditional ways of handling data should be adapted to allow the saving of resources like data storage, power consumption and network bandwidth.
In this work, we present TSXor a novel lossless time-series compression algorithm that achieve up to 6x compression ratios and decompression speeds up to 1GB/s. The main idea is to exploit the redundancy within consecutive floating-point values through a window that acts as a cache. Our method allows to obtain high compression ratios and excellent decompression speeds.