Glenn Galvizo: Removing the ‘A’ in DAG: Navigational Queries in Hyracks

The need to “view” existing data under different models (e.g. JSON to graph) is a requirement seen in many modern applications. A naive solution involves utilizing narrow-purposed systems to handle each model, however, this multi-DBMS architecture significantly increases the cost of owning one’s data. For Apache AsterixDB users, we offer Graphix as a way to issue synergistic document-graph queries on their existing Big Data, in-situ (i.e. in partition-parallel).

In this talk, we’ll be walking through how we modified Hyracks, the runtime platform for AsterixDB, to execute a recursive Graphix query. We’ll first talk about how tuple-pipelineable recursion occurs in a non-distributed setting. We will then extend our discussion for the distributed setting, and conclude with optimizations we take to handle dense graphs.