In the DoricStore project, our goal is to design an in-memory column-store for high-performance emerging hardware. Over the last decade Columnar database systems, or column-stores for short, take advantage of the decomposition storage model (DSM) to boost the performance of read-optimized databases. Many different systems can leverage column-stores, like Business Information Services (BIS), Customer Relationship Management (CRM) and electronic library catalogue. But now many of these systems are presenting real-time analysis requirements that together with emerging new hardware offer an opportunity to rethink the design of the column-stores. In this project, we give particular attention to multi-core machines and high-performance Hybrid Memory Cubes (HMC). We believe the HMC is particularly convenient for read-optimized databases as they glue multiple logic control chips to the memory stack. Thus, we run logic query operations within these chips to avoid going to CPU as much as possible. Otherwise, we seek efficient scheduling on multi-core machines. HMC can be built over DRAM or NAND Flash, but this flexibility may present different challenges that we are working on in our research agenda. In particular, we are investigating what happens to the current state of column-stores when running atop multi-core machines and HMC to present new algorithms and data structures in topics, such as scheduling, compressing, vectorization, and late materialization.



Please email