I think I am running out of room for sequential write throughput. What's interesting is that this performance is blown away for parallel writers, because the DB can ack all write threads with the same flush.