Maarten Boekhold <boekhold(at)tibco(dot)com> writes:
> This makes a difference, since a lot of processing is done on commit
> time. The second variant might thus be misleading.
Yup, but I did include the time to commit in the measurement. In fact,
I used the time(1) command to time the whole run of the client program.
-tih
--
Popularity is the hallmark of mediocrity. --Niles Crane, "Frasier"