Holger Pirk was awarded a PhD today (2015-05-01) for his on work on investigating the management of relational data in modern hardware and how data processing performance can be improved. He developed a new processing model using the basic MonetDB BATs, which in this case are not treaded as memory mapped files, but each value is decomposed into relevant (leading) bits and residual bits. The fast memory (usually attached to a GPU) holds the relevant bits, while the residual bits are stored in the slower CPU memory. The partial data that resides in the GPU is used for approximate representation of data and the CPU for refinement. The developed Bitwise Decomposition with Approximate & Refine model makes efficient use of the asymmetries in the underlying data and memory components.
Dr. Holger Pirk defended his dissertation “‘Waste not, want not!’- Managing Relational Data in Asymmetric Memories”, at the University of Amsterdam. Promotors were Prof. Dr. Martin Kersten and Prof. Dr. Stefan Manegold.