TVStore: Automatically Bounding Time Series Storage via Time-Varying Compression

A pressing demand emerges for storing extreme-scale time series data, which are widely generated by industry and research at an increasing speed. Automatically constraining data storage can lower expenses and improve performance, as well as saving storage maintenance efforts at the resourceconstrained conditions. However, two challenges exist: 1) how to preserve data as much and as long as possible within the storage bound; and, 2) how to respect the importance of data that generally changes with data age.

To address the above challenges, we propose time-varying compression that respects data values by compressing data to functions with time as input. Based on time-varying compression, we prove the fundamental design choices regarding when compression must be initiated to guarantee bounded storage. We implement a storage-bounded time series store TVStore based on an open-source time series database. Extensive evaluation results validate the storageboundedness of TVStore and its time-varying pattern of compression on both synthetic and real-world data, as well as demonstrating its efficiency in writes and queries.