Analyzing In-Memory Hash Joins: Granularity Matters

Predicting the performance of join algorithms on modern hardware is challenging. In this work, we focus on mainmemory no-partitioning and partitioning hash join algorithms executing on multi-core platforms. We discuss the main parameters impacting performance, and present an effective performance model. This model can be used to select the most appropriate algorithm for different input data-sets for current and future hardware configurations. We find that for modern systems an optimized no-partition hash join often outperforms an optimized radix partitioning hash join.