Christian Mock <cm@coretec.at> writes:
> I started the insert on the real database to copy some 3.6 million records
> between the tables, and plotted the timings -- now (with about half a
> million records done) you can see the time it takes to insert a row
> actually goes in a sawtooth pattern -- cf.
> http://www.coretec.at/~cm/insert-trigger.gif
> which has msec/inserted row on the Y axis and number of rows in the
> "output" table on the X axis.
Interesting ... I'd have to guess that there is some external influence
causing that. Do you by any chance have a cron jon set up to do
"vacuum"s periodically, and if so do the vacuums correspond to the
points where the insertion time drops?
regards, tom lane
