Francois,
> Anyway, the journalist had a question about tablespaces. I told him it
> enabled clustering on individual disks or arrays of disks. The
> journalist then said that this seemed like a rather basic feature and
> was surprised that PostgreSQL wasn't already able to do that in the
> previous versions. Is that indeed the case or was there another
> clustering mechanism before?
Well, there was the ad-hoc method.
Mostly, it wasn't done before because there wasn't much demand for it.
Tablespaces really aren't useful unless you have really large
(multi-gigabyte) databases and/or large arrays with lots of disks. Three
years ago, I think you could have counted the number of PG installations over
10GB on your fingers and toes.
Some databases, which really don't need tablespaces from a performance
perspective ... like MSSQL ... were forced to develop them because of their
use of raw disk partitions to store the database. This means that you can't
just move the database files when it outgrows the disk, you have to allocate
a new partition.
Also, our Tablespace feature is a bit more sophisticated than just dropping
the tables in a new location. That part I think (correct me if I'm wrong,
Gavin) was easy. The tough part was all the administrative tools and
recovery stuff to make it robust; to make sure, for example, that you can
restore a tablespace from backup even if the disk ceases to exist, and that
transactions are robust across tablespaces.
--
--Josh
Josh Berkus
Aglio Database Solutions
San Francisco
