NASA jettisons Neo4j database for Memgraph citing costs
NASA’s people analytics group has swapped its Neo4j graph database for Memgraph due to costs.
David Meza, senior data scientist on the team, told a recent webinar that despite using Neo4j for around ten years, cost had become an issue.
Meza had previously spoken to The Register about the benefits of using the Neo4j graph database system at NASA to bring together data from the space agency’s various enterprise applications to understand the relationship between knowledge, skills, abilities, tasks and technologies (KSATTs), and occupations, roles, and training.
But at a recent webinar for Memgraph, he explained the switch to the in-memory graph database.
“The biggest thing with Neo4j is that it is very costly for me. I can’t afford that within my current environment,” he said.
Last week, the Trump administration proposed slashing NASA’s annual budget by 24 percent from $24.8 billion to $18.8 billion amid efforts to cut government spending.
Memgraph uses the same Cypher query language as Neo4j. However, it is written in C++ and integrates better with Python than Neo4j, which uses Java to build applications.
“There were a lot of benefits. [We could] utilize the same tool without having to relearn a whole lot, because we’ve done a lot of work with Neo4j. And then [Memgraph] showed me the cost. That kind of sold me,” he said. “It was more about ease of transition as well as cost.”
NASA is integrating Memgraph in its Human Capital Intelligent Query System to allow faster retrieval of relevant information for employees. “Its graph-based approach allows us to keep track of real-time updates, ensuring accurate connections between various policy documents and data sources. By incorporating Memgraph into our Retrieval-Augmented Generation process, we enhance our system’s responsiveness and better address NASA’s knowledge extraction without requiring extensive manual data coordination,” Meza said in a statement.
Speaking to The Register, Memgraph CEO Dominik Tomicevic said Neo4j relied on complex b-tree structures on disk, with in-memory caching. “So in order to run graph algorithms, one needs to have lots of really random jumps in the graph.”
Because disks are built for sequential reads, not random reads, the solution was to duplicate the data to run analytics workloads from a read-only format and then write back the results to the original data as needed.
“While you can do a lot of stuff with Neo4j, especially if you’re doing some batch workloads, if you need to do decisions kind of in real time, then it becomes very expensive to rebuild those in-memory indexes and reduplicate all of the data and run the algorithms and then write the data back into the original graph,” Tomicevic said.
In Memgraph, the data structures are built for the data science workloads first, he claimed. With snapshots, the in-memory system can support transactional workloads as well as real-time analytics.
Neo4j was invited to comment for this article. ®
