Download the EDBT 2026 paper: Dynamic prefiltering for massive-scale equi- and geospatial joins
Modern hyperscale data warehouses execute joins across trillions of rows, where performance is often dominated by I/O, network shuffle, and CPU-intensive join processing. This paper presents Ocient’s dynamic join prefiltering framework, designed to reduce unnecessary data movement and computation by generating runtime filters from the build side of a join and pushing them directly into storage-adjacent I/O operators.
Presented at EDBT 2026 in Tampere, Finland, the paper details how Ocient applies a unified prefiltering architecture to both high-throughput equi-joins and complex geospatial joins at massive scale.