Property Graph Benchmarking and Storage
Professor Margo Seltzer
Harvard's School of Engineering and Applied Sciences
Oracle Fellowship Recipient
Daniel Margo, Elaine Angelino, Peter Macko
Oracle Principal Investigator
Jeffrey Alexander, Consulting Member Technical Staff
Karl Haberl, Senior Director, Oracle Labs East
Prof. Seltzer’s research group at Harvard has developed a framework for understanding and evaluating property graph implementation and a collection of wrappers for the most widely used property graph implementations and more conventional systems (e.g., SQL, RDF). The framework provides for both introspective analysis as well as inter-database analysis. It uses models to contrast empirical performance to theoretical performance to identify aspects of system implementation that are either under- or over-performing. The results identify implementation techniques that are effective in leveraging knowledge of graph-structure and typical access patterns as well as highlighting places where existing systems do poorly, providing an opportunity for future product differentiation. Based on this work, the team has several concrete ideas for research, leading to production of a high-performance property-graph database.
The work proposed here extends prior work in two directions. The first area of work is to extend the existing suite to evaluate concurrency and parallelism and to add wrappers for Oracle’s current SQL, RDF, and NDM stores and ultimately, the upcoming property-graph implementation. The second area applies the lessons learned from this measurement effort to produce prototype on-disk (and SSD) and in-memory graph representations, query-oriented indexing structures, and graph-specific caching algorithms.