Thursday, May 27, 2010
12:00 pm - 1:00 pm
PLACE: Gates Center 8102
TITLE: LazyBase: Trading Freshness for Performance at Scale
Many of today's web and enterprise applications wish to perform a variety of data-mining and information-retrieval-style queries on rapidly changing data sets. In most traditional systems, this workload creates a tension between update and query performance, however, we observe that many of these applications can perform correctly on varying degrees of stale data. We have developed LazyBase, a system which allows applications to trade result freshness for query latency while maintaining update throughput. In our experiments, we found that applications that can support up to 30 minutes of result staleness can achieve update rates unachievable with the systems we tested against while simultaneously maintaining query performance on-par with unloaded commercial database systems.
Craig Soules is a researcher at HP Labs who is interested in the many varied problems of information management and storage, with work ranging from versioned file systems up through context-based information retrieval. He received his Ph.D. in Computer Science from Carnegie Mellon University in 2006.
SDI / LCS Seminar Questions?
Karen Lindenfelser, 86716, or visit www.pdl.cmu.edu/SDI/