From this post, I have concern on performance on graph behaviors like graph traversals or shortest path, etc., since dgraph should be classified to triple store as I understood.
My scenario focuses on finding relationship among large number of nodes in real time (I think most graph database users have similar scenarios, otherwise the will choose SQL/NOSQL) , so the performance on graph related behaviors are the most important for me.
Are there any performance report on these behaviors?
For your reference, I think arangodb’s report is more comprehensive
· Graph databases are better optimized for graph traversals (degrees of separation or shortest path algorithms). With RDF triple stores, the cost of traversing an edge tends to be logarithmic.
Dgraph isn’t an RDF triple store. We allow RDFs to be parsed, just because of the wide data availability – but are switching over to JSON data input; something that most developers are familiar with.
In terms of traversal, the way we store our data, the traversal happens in O(1) time. All the outgoing edges are stored in a posting list, which is stored as a value. So, to find all the nodes connected to a given node, is just a key lookup for us.
Shortest path uses this as well. In fact, super fast traversals and intersections are the key designs of Dgraph.