Orangutan: a compression algorithm for precision-redundant floating-point time series (Withdrawal Notice)
Zhenshan Zhao, Youli Qu · 2024
IoT devices and various sensors generate a large amount of time-series data every moment, and the cost of transmitting and storing this data is high. Compact, efficient, and lossless compression of time series is a common solution. Current state-of-the-art lossless compression algorithms for time series are based on XOR operations but do not take advantage of the fact that most floating-point sequences in real life have precision redundancy. In this paper, we propose a unique trailing bit zeroing scheme called Orangutan, which can zero as many trailing bits as possible in O(1) time without incurring any additional space overhead, resulting in a higher compression ratio. In addition, we also designed the compression and decompression methods specifically for the case of more zeroes in the tail bits and can obtain good compression results. Comparing Orangutan with 9 current state-of-the-art compression algorithms on 19 datasets shows that Orangutan is more space-efficient than the best compression algorithms currently used for time series, and the compression speed is at the same level as them.