Trovares Inc., a property graph analytics company started by the co-founder and former chief executive of Cray Inc., has raised $2 million to launch a new graph analytics engine.

The engine uses parallel processing to deliver what the company says is order-of-magnitude speed improvements compared to conventional graph databases. The Series A funding, announced Monday, brings the nine-person company’s total funding to $2.6 million.

Trovares xGT was developed under contract with the U.S. Department of Energy. It performs highly parallel graph analytics on symmetric multiprocessing platforms using a technique that can utilize 100% of each processor, even when more than 1,000 processors are applied, the company said. This enables xGT to traverse in-memory graphs consisting of billions of edge points to find complex patterns.

Graph analytics has been on a roll lately with cloud giants Amazon Web Services Inc. and Microsoft Corp. getting into the market and top graph database vendor Neo4j Inc. raising $80 million in a recent funding round. Graph engines can traverse complex relationships to discover correlations that would be impossible or impractical to find using conventional relational engines. While not appropriate for transaction processing, graph analytics is increasingly finding favor in areas like cybersecurity, fraud detection, social network influence analysis and recommendation engines.

Trovares xGT isn’t a database but a processing engine that works with any data that can be represented in a row-and-column format. “We can accommodate significantly larger datasets than traditional graph databases,” said Jim Rottsolk, founder and CEO of the Seattle-based company. “We approached the problem from the beginning as a high-performance computing problem.” The technology was developed at the Pacific Northwest National Laboratory and the U.S. Department of Defense is Trovares’ first customer, Rottsolk said.

Instead of rows and columns, property graphs use nodes, relationships and key-value pairs, the latter of which define linked data items using a unique identifier. Properties can be defined for each node and any node can easily be connected to any other. By traversing connections, data scientists can identify relationships that aren’t otherwise obvious.

Trovares goes one step beyond standard property graphs by applying attributes such as timestamps and geographic data to both nodes and connections. The result is that “where we see other tools not able to address datasets larger than a few millions of edges, we can handle billions of edges and terabytes of data,” Rottsolk said. The parallel architecture also significantly speeds up data ingestion, he said.

The software is available now for both on-premises data centers and cloud deployment. Pricing is listed on the company’s website.

Image: Flickr CC