Neo Technology, Inc. is taking steps to move its Neo4J graph database into the real-time analytics world with updates that significantly improve read and write performance and the addition of a cost-based optimizer to improve query performance.

Neo4J, which is the world’s most popular graph database, according to DB-Engines Ranking, gets a new in-memory page cache in version 2.2, which the company claims improves overall read and write performance between 10- and 100-fold. Write performance is boosted with a fast-write buffer that “scoops up transactions as they’re coming in so they can commit multiple writes at once,” said Philip Rathle, vice president of products. “You gain throughput without losing speed.”

To improve read performance the company has added a proprietary cache that uses memory-mapped input/output. Previously, Neo4J used the operating system for caching. “In cases where you have a lot of concurrent threads reading from the database and a lot of cores and processors, we can get up to 10 times the write performance,” Rathle said.

The cost-based optimizer that is now part of the Cypher query language chooses the best query execution plan using built-in statistics containing information on graph size and shape.  Unlike the rule-based optimizer that preceded it, the cost-based optimizer changes rules depending on spread, layout, size and other statistics about the database. In contrast, rules-based optimizers generally process similar queries the same way. Users can get a visual preview (above) that they can use to fine-tune queries before execution.

Graph databases are type of NoSQL technology that have attracted a lot of attention recently for their distinctive ability to map relationships in ways that can yield rapid insights for real-time and predictive analytics. While not suited for all applications, they are particularly popular in use cases involving rapid discovery of patterns, such as retail recommendation engines and credit card fraud detection.

Neo has raised $45 million in venture funding, including a $20 million round in January.