I'm looking around for a Relational OLAP solution based on
PostgreSQL (and liberally licensed) to contribute to. If there
is one out there, great, please forward. If not, is there anyone
interested in joining a new project?
I was thinking of a simple "star schema" with a single fact table
to start. The fact table would have two sorts of columns, a
"dimention" column or an "aggregate" column. These columns
could be marked... perhaps using extra schema information
on the relname/attname pair. Once this base schema was defined,
we'd have to define what a "query" looks like, it appears that
there is an "XML Analysis" which Hyperion and Microsoft use...
although YAML would be cleaner. Once there is a query format
(something simple at first) we could then implement a stored
procedure which returns a result set when passed a query. The
first implementation of the proceure would work directly against
the fact table, doing the necessary aggregations. This bootstrap
would then be given a nice regression test (with base data and
expected query results).
With this bootstrap in place, we could then focus on various
query and/or cacheing optimizations. Further, we could then
extend the scope to handle more than one fact table (multicube)
which are attached via one or more shared dimentions, allowing
the user to bounce between cubes. In parallel to this, someone
could then be working on a user interface; I was thinking Python
via Webware or mod_python using PIL.
Anyway, please ping if you are working in this direction or would
like to work in this direction. The result would be liberally
licensed (BSDish) and ideally put into the "contrib".
Best,
Clark