SOLAR: Efficient Spatial Queries on Real-Time LSM-Based Storage

The burgeoning volumes of spatial data demand spatial databases that support both fast ingestion and efficient queries. Existing LSM-tree-based systems handle high write throughput but remain suboptimal for spatial queries due to their general-purpose storage design. We present SOLAR, an LSMbased spatial data store that embeds spatial awareness as a firstclass design principle. SOLAR adopts a spatial-centric storage model where each level is organized into clustered sorted runs to improve spatial locality. Building on this layout, we design query processing algorithms that effectively prune away irrelevant LSM component files for both spatial range queries and KNN queries. To further enhance write performance, we propose a selective compaction paradigm with a cost model-based selection policy, which reduces write amplification while producing clustered sorted runs that optimize query performance. Experiments on real-world datasets demonstrate the superior performance of SOLAR over existing LSM-based and page-oriented spatial data systems across different dynamic workload scenarios.