I always considered Cassandra's 'column' designation a guarantee that OLTP transactions will apply atomically to column groups; or that you'll minimize IO write

I always considered Cassandra's 'column' designation a guarantee that OLTP transactions will apply atomically to column groups; or that you'll minimize IO writes for a row to the column group (as opposed to writing a new version of the entire row). But this is actually just a wild guess. Anyone want to chime in and confirm/deny?