Exploiting Internal Parallelism of SSD for Hash Join

By regarding a Solid-state drive(SSD) as a black box and observing its external behavior instead of peeping into its internal details, we investigate how the factors of I/O granularity and I/O queue depth influence the throughput of an SSD through a series of experiments and relate to the internal parallelism of an SSD, and then propose the concept of Combination equivalence class (CEC) as the set of combination pairs of these two factors. A novel buffer allocation scheme for hash join over SSDs is invented by taking both factors into account. Extensive experiments demonstrate the effectiveness of our scheme.