Efficient Exploratory Searches With an Incremental R-Tree Under Oscillating Workloads

Traditional spatial index structures face challenges in efficiently managing dynamic and irregular workloads, which are common in real-time spatial data analysis. In particular, oscillating workloads can significantly hinder data access performance due to the continuous influx of spatial records in varying volume and frequency. To address these challenges, this paper presents an Incremental R-tree index structure optimized for real-time storage and querying of large-scale spatial datasets under dynamic conditions. The proposed method employs a two-stage overflow mechanism, where newly arriving data batches are first distributed into a grid-based structure and then selectively integrated into the tree leaves based on query activity. This approach delays full indexing until it becomes necessary, thereby improving responsiveness and avoiding unnecessary tree restructuring. Furthermore, the Incremental R-tree adaptively incorporates principles from both STR-tree and ${\mathrm{R}}^{\ast }$ -tree to enhance indexing efficiency. Experimental results demonstrate that the Incremental R-tree significantly reduces cumulative query response latency compared to traditional ${\mathrm{R}}^{\ast }$ -tree and STR-tree structures under oscillating workloads, and consistently outperforms the widely used GiST-based R-tree implementation in PostGIS.