Moving Recursion Out of the RDBMS for Transactional Graph Workloads

This paper presents work in progress that focuses on querying transactional graph data that is stored in a relational database system (RDBMS). We focus on transactional workloads where there are frequent insert and update operations. Although these types of workloads are common in social network, scientific, and business applications, much of the prior work has focused on graph analytics workloads where there is little to no change to the data over time. We introduce an approach that combines simple database queries with parallel programming, and compare our approach to the recursive SQL operations that are known to have poor performance. Our initial experiments and results provide guidance for the future directions of this project where we will refine the parallel programing approach and structure of the experiment in order to better compare these two approaches.