2018 Symposium Posters

Primary Investigator:
Bharat Bhargava

Project Members
Servio Palacios, Byoungyoung Lee, Ananth Grama, Bharat Bhargava

Many modern big-data applications work on top of graph-structured datasets. For instance, graph databases such as Neo4j, TitanDB, OrientDB, and others handle large-scale graphs that model biological interactions and social networks. Graphs are semi-structured, as opposed to the relational model. Thus, accessing network-structured data from/to secondary storage necessitates new techniques. These new methods must ensure access locality for graphs. Moreover, in multi-tenant environments, protecting sensitive information from unauthorized access and leakage is imperative. Unfortunately, although existing solutions for the graph processing model include secure communication and authentication (for independent graphs and authenticated entities); there is no solution for multiple authenticated entities (data owners) sharing a diversity of data graphs. Also, current solutions do not take into consideration graph access locality and authenticated multi-tenant clouds at the same time. In this paper, we propose a novel authenticated-blocked graph processing model. The project includes a set of unique algorithms that model the authentication scheme and graph structure into a series of disk blocks with high-locality. Our contributions include: (1) the definition of the practical and formal security model of authentication for the blocked graph, (2) three new algorithms for (i) block construction with high locality, (ii) disk layout for authenticated and blocked graphs, (iii) new data structures for graph access. (3) An open-source implementation and evaluation of the proposed techniques. Through a set of benchmarks and applications, we demonstrate the practicality of our solution.