More work on the synthetic read benchmark