More about HKUST
Adding Support for New Hardware in Database Systems
Speaker: Dr. Stratis D. VIGLAS School of Informatics University of Edinburgh Title: "Adding Support for New Hardware in Database Systems" Date: Monday, 1 September 2008 Time: 11:00am - 12 noon Venue: Room 3315 (via lifts 17/18), HKUST Abstract: Database systems developers have always had to adjust their implementations to the changing hardware environment. After a long period of relative stability, both in terms of storage and also in terms of processor design, the environment has radically changed. In terms of storage, solid state drives are rapidly becoming standard storage solutions, while the current multi-core processor design effectively dictates a shared memory design for high-performing applications. In this talk, I will (a) focus on our work on making better use of solid state drives in database applications and (b) briefly present some of our ongoing work on making better use of current processor designs. As far as solid state drives go, we target hybrid systems that contain both solid state and magnetic disk drives. We propose that instead of treating the solid state drive as a cache for the magnetic disk, we put them both at the same level of the memory hierarchy. Then, we take advantage of the asymmetric properties of the solid state drive's read and write speeds and using a workload-driven approach we decide on the optimal placement of each database page: pages with a read-intensive workload are placed on the solid state drive, while pages with a write-intensive workload are placed on the magnetic disk drive. Our theoretical results and our experimental evaluation prove this to be the optimal usage in such hybrid systems. In terms of supporting new processor architectures, we argue that the bottleneck is in fact the complexity of the database engine in terms of its use of the iterator model. We propose that instead of composing operators for a given database query, one is better off generating custom, query-specific, and processor-friendly code and then compiling and executing it. Our results in this area show that such an approach dominates both the iterator-based design, and other hardware-friendly designs. ******************** Biography: Dr Stratis D. VIGLAS is a Lecturer in Database Systems at the School of Informatics of the University of Edinburgh. He received his BSc and MSc degrees in Computer Science from the University of Athens, Greece, and his PhD degree in Computer Science from the University of Wisconsin-Madison in 2003. His main interest is query processing and optimization and in the past years the bulk of his work has been on data stream processing and XML. His current research interests are in the areas of enhancing the performance of database systems in contemporary hardware setups, and peer-to-peer database systems.