This setting (with trim.timespan_unit) controls how the date range to gather will be split.
In order to speed up the crawl and improve reliability the crawler will split the date range to gather in smaller chunks. Smaller chunks means less records to select for each chunk, reducing the time spent in the initial query for each chunk, but generating more chunks.
The HP Records Manager / TRIM connection will be disconnected and reconnected again between each chunk to restart with a clean connection.
The default is to select records by 3 month chunks: