Relational Query Co-Processing on Graphics Processors
Speaker: Qiong Luo
When: 2009-07-30 10:00:00
Venue: 78-420
Host: Xiaofang Zhou
Abstract:Microprocessors consisting multiple processing cores, or multicore processors, have doubled their core counts at approximately two-year cycles since the emergence of dual-core CPUs in 2005. Such growth trend in hardware demands software architectures and programming models to adapt to massive parallelism. Graphics processors, or GPUs, are specialized co-processors traditionally dedicated to gaming applications and currently become commodity parallel processors for general-purpose computing. Since GPUs are widely available and are a representative manycore architecture, we study how to utilize them to accelerate relational query processing. To exploit the GPU's massive SIMD (Single-Instruction Multiple-Data) parallelism and high-latency, high-bandwidth memory, we propose and optimize a small set of data-parallel primitives such as scan, scatter, and gather, and use these primitives to implement relational query operators. Our experimental results show that query processing can be accelerated up to an order of magnitude on a recent GPU over a quad-core CPU. We also discuss issues in developing a query processor on the GPU and contemplate on the prospect of adapting it to manycore CPUs.
Biography:Qiong Luo is an Associate Professor of Computer Science and Engineering at the Hong Kong University of Science and Technology (HKUST). Her research interests are in database systems, with a focus on architecture-conscious databases, query processing in sensor networks and pervasive computing, and web data management. She has published over 60 technical papers at international conferences and journals, and has been involved in the organization and program committees of major conferences and workshops. She has graduated four PhD students and has co-supervised several international award-winning undergraduate theses. Qiong received her PhD from the University of Wisconsin-Madison, and her MS and BS from Peking (Beijing) University, all in Computer Science.
Type: Job application seminar for the DKE academic position
Contact:Xiaofang Zhou