Performance has been measured on 2.5 GHz desktop PC and using common desktop hard drive. All data was stored in a single flat file, and RAM memory was limited to around 300MB of in-memory cache.
Speed obviously would vary depending on the number of changes made in a single commit. Therefore, tests were made with different batch sizes, starting with 10000 inserts/updates/reads at once and testing down to 1.
Performance was measured using ObjectBenchmark application from Examples folder, running from zero to 1 million objects, with different batch sizes, and taking the average speed during the run:
Looking at speed how it behaves when starting from zero objects up to 1 million objects
, for different batch sizes, it is noticeable that read speed drops after a while, due to
limited read-cache of 300MB
and random read access, while insert/update has fairly