Iteration time is of course interesting, but let's also use throughput estimation on MVCC operations such as read(), begin_tx(), etc..