Range-based clustering?

Hi,


I have an application that gathers data in a table (who doesn't around
here? ;-) ). The behaviour is roughly as follows:

- on any single day, a lot of information is inserted. Inserts are more
or less in order of increasing timestamp; there may be inserts for data
as far back as an hour or two, but not further than that.

- inserted data may be updated several times, but only for data from the
last couple of hours. Older data is always left alone.

- data is inserted on a 24-hour basis (there is no downtime).

- to improve read-access speed, the table is clustered on timestamp.

- read access is over the entire set, and always returns all records
from a timerange (this is why clustering is useful to me).

- clustering and vacuuming takes place every 24 hours.

Over time, this table has grown considerably, making the
cluster-operation take longer and longer. This seems inefficient, since
most of the data (basically everything that is older than 24 hours) is
already perfectly fine as it is; it is only the last 24 hours, i.e. the
bit that has seen numerous inserts and updates, that needs to be
straightened out.

Is there some way to do limit the cluster-operation to only a subset of
the table? In my case that would be "all data with a timestamp in the
last 24 hours".

Alternatively, is there a way to add this as a feature to a new version
of PostgreSQL? I believe the use-case I have presented is in fact fairly
common, and might be interesting as a feature to others as well.


Regards,

Hans Guijt

TERMA B.V.
Schuttersveld 9
NL-2316 XG, Leiden
The Netherlands
Phone: +31 (0)71 52 40 835
Telefax: +31 (0)71 51 43 277
E-mail: hg@terma.com