Could we remove lines 552-560 of pgbench.c? The behavior that guardedagainst is long gone, and forcing a checkpoint every few thousand tuplesseems to be putting a huge crimp in the speed of pgbench -i ...
regards, tom lane