Graph Processing on an "almost" Relational Database
Ramesh Subramonian · 2014
It would be hard to disagree with the contention that graph processing (whether it be of connections between people, shopping habits,...) allows the creation of valuable data-driven products and insights. There is less consensus on the systems that make it easy to analyze these graphs. In this paper, we argue that a relational database (actually, a close approximation to one) is well suited for many graph processing applications. We restrict our claims to the following case, which, we believe, dominates the nature of much data analyses --- the data does not change during the analysis in response to external events. We present representative examples from our work at LinkedIn, the world's largest professional network. We present performance results for these examples using Q, a single-node, analytical database.