Graph database study finds large differences in data loading speed

A preprint comparing eight database systems found that the speed of loading data for graph queries varied by three orders of magnitude, from 5.0k to 4.3M rows per second. The researchers tested a synthetic biomedical-style graph and a range of query types to examine the cost of preparing data for AI agent memory. No engine was fastest across the workload. Ladybug performed better on narrow neighborhood queries, while Corvic AI was faster on queries that scanned or joined much of the graph. The results come from synthetic data, so production workloads may differ.