CALF: Comparison of Attribute Layouts on Flash

Solid state drives(SSDs) are increasingly being used for database applications, due to their low seek latency and low power consumption. Solid state drives differ fundamentally in their characteristics from disk drives, which mandates that some core database design decisions, such as column layout, needs to take into account the specific characteristics of SSDs for optimal performance. This paper reexamines the question of column layout models for Flash based databases. We propose a flexible data storage model which partitions attributes based on a given workload, taking into account both reads and writes. We evaluate the performance of this intelligent partitioning against NSM and C-Store storage models.