Plan Before You Execute: A Cost-Based Query Optimizer for Attributed Graph Databases

Abstract

Proliferation of NoSQL and graph databases indicates a move towards alternate forms of data representation beyond the traditional relational data model. This raises the question of processing queries efficiently over these representations. Graphs have become one of the preferred ways to represent and store data related to social networks and other domains where relationships and their labels need to be captured explicitly. Currently, for querying graph databases, users have to either learn a new graph query language (e.g. Metaweb Query language or MQL [6]) for posing their queries or use customized searches of specific substructures [14]. Hence, there is a clear need for posing queries using the same representation as that of a graph database, generate and evaluate alternate plans, develop cost metrics for evaluating plans, and prune the search space to converge on a good plan that can be evaluated directly over the graph database.

In this paper, we propose an approach for effective evaluation of queries specified over graph databases. The proposed optimizer generates query plans systematically and evaluates them using appropriate cost metrics gleaned from the graph database. For the time being, a graph mining algorithm has been modified for evaluating a given query plan using constrained expansion. Relevant metadata pertaining to the graph database is collected and used for evaluating a query plan using a branch and bound algorithm. Experiments on different types of queries over two graph databases (Internet Movie Database or IMDB and DBLP) are performed to validate our approach. Experimental results show that the query plan generated by our system results in exploring significantly fewer portions of the graph as compared to any other query plan for the same query.