It's Just Graph
Bryan B. Thompson · 2019
Graph has emerged as a hot topic in many different communities (RDF/SPARQL, Property Graph (Gremlin / Cypher / PGQL), graph embeddings, graph learning, linear algebra, etc. I will look back over 20 years of Graph at different conceptual approaches and discuss how they touch and whether they will or should converge and why Graph should and does stand apart from various other models (e.g., Relational, Object databases, linear algebra on unattributed graphs). I will discuss why Graph is intrinsically a harder problem than Relational (e.g., lack of explicit schema and the impact this has on query planning), why Graph is a challenging computational problem (non-locality, data-dependent parallelism, bandwidth limited compute, wildly varying workloads), challenges in Graph languages (pattern matching and graph algorithms), architectural and computational challenges in scaling to large graphs (user space algorithms, data parallel runtime, latency hiding), and opportunities for combining Graph and ML techniques.