>
> Look at the pgbench utility. I can't run that program without a +- 10%
> variation from run to run, no mater how many times I run vacuum and
checkpoint.
>
It's pgbench's fault, TPC-B was replaced with TPC-C because it is not
accurate enough, we run a pseudo TPC-H and it has almost no variations from
one run to another.
Regards