A Ray-based horizontally scalable in-memory graph computational platform developed at BASF, in cooperation with DerwenAI, is presented.
Loading time for a 2 billion entity graph into 2.1Tb of RAM with 145 workers is 3 min and a two-layer (Python/Ray, C++) breadth-first search traversal retrieves BASF's network of the bill of materials structures (~1000 entities) in real-time (0.5 s) via a client API.
Business impact datasets are modeled on Knowledge Graphs (KG) with an emphasis on rich relationship context. The KGs vary in scale from millions to billions of entities and require parallel processing techniques for building, traversing, and computation.
Visualizations of relationship context are presented using Graphistry.
Traversals of BASF's network of the bill of materials structures using the Ray-based graph platform are real-time because the platform runs in memory.
Fast loading times for billions of data points in parallel are achieved due to horizontal scalability.
Trillion-scale industrial datasets await due to hundreds of thousands of multilevel bill of materials structures, having a few tens of levels, creating a natural backbone for manufacturing, operations, financial, and planning data.
Janez Ales built his first graph back end in the late 1980s for a heuristic search for Hamilton cycles in cubic graphs, has all three degrees in algebraic/algorithmic graph theory, and has been working on KGs at scale at BASF since 2018. In 2019 he built a 55 billion entity graph on the world's journal and patent metadata with detailed annotations, including a 300 million entity chemistry ontology, and BASF reports. He contributes to the development of a Ray-based horizontally scalable in-memory graph platform (BASF, DerwenAI), focusing on real-time traversals and efficient memory structures, has programmed (optimization) algorithms in over a dozen of computer languages, and works on business impact KGs at BASF.
Come connect with the global community of thinkers and disruptors who are building and deploying the next generation of AI and ML applications.