Hello again Tatsuo,
> For your information, included is the patch against git master head to
> implement the lag in a way what I proposed. With the patch, I get more
> consistent number on Linux (and Mac OS X).
I must disagree with your proposal: At least, it does not provide the
information I want, but another one.
ISTM that this patch measures the lag which is due to pgbench thread
coming around to deal with a transaction after sleeping. I would expect
that to be quite small most of the time, so I agree that it must be
reassuringly consistent.
However it does not provide the information I want, which is the measure
of the health of pgbench with respect to the stochastic process scheduled
transactions.
Basically you propose a partial lag measure, which will be smaller, but
which does not tell whether pgbench is able to follow the schedule, which
is the information I find useful in this context to appreciate if the
throttling is going well.
--
Fabien.