DDC: Efficient Dynamic-Dictionary-Based Compression on Floating Time Series Data

Abstract

Growing demand for large-scale time-series data storage has rendered floating-point time-series compression a key technology for better storage efficiency and performance. General-purpose compression rely on exact matching to build dictionaries, making them effective for repetitive data but unsuitable for time-series data with small fluctuations and gradual changes. While time-series-specialized compression leverage correlations between adjacent data points, they ignore long-term similarities and periodic patterns. To address these limitations, we propose an efficient Dynamic-Dictionary-Based Compression (DDC) algorithm for floating-point time-series data. DDC uses three matching strategies to capture both local patterns and global trends. It dynamically updates the dictionary and adapts its compression strategy during processing. Moreover, DDC reduces precision redundancy by adjusting mantissa bit lengths and separating floating-point components. We evaluate DDC against state-of-theart time-series-specific (Gorilla, Chimp, AFC, TSXor, FPC) and general-purpose (LZ4, LZW, Snappy) compression algorithms. Experimental results demonstrate that DDC achieves superior compression ratios and faster processing speeds, highlighting its effectiveness and practicality.