77. DISTRIBUTED STORAGE AND INTERACTIVE ANALYTICS FOR GRAPH-STRUCTURED DATA

Department: Computer Science & Engineering
Research Institute Affiliation: Center for Networked Systems (CNS)
Faculty Advisor(s): Yuanyuan Zhou

Primary Student
Name: Michael Mihn-Jong Lee
Email: mmlee@ucsd.edu
Phone: 213-505-6695
Grad Year: 2015

Abstract
A growing number of applications require low-latency and interactive processing over large, graph-structured data. Moreover, many of these applications exhibit a variety of additional requirements, such as support for complex graph queries, transactional semantics, and scalability with respect to both data size and client load. Existing solutions such as relational databases, NoSQL-based stores, batch-graph processing systems, and specialized graph databases are significantly lacking in at least one of these dimensions. We present Concerto, a new graph store based on distributed, in-memory data structures. Concerto scales to hundreds of machines, and provides concurrency-safe updates and parallel graph processing. In our experiments Concerto is an order of magnitude faster than MySQL and Neo4j for graph insertions and queries like k-hop and k-core. Unlike other stores, Concerto is extensible and allows users to register functions on subgraphs and perform server side event-driven analysis. We demonstrate the utility of these features by implementing a real-world use-case: real-time traffic incident impact analysis on a road network.

« Back to Posters or Search Results