Open Source Pivot table (OLAP cube?)

[Jon Clements]

On 27 Apr, 12:16, Laszlo Nagy <gand... at shopzeus.com> wrote:
> > Please note: this is not a direct answer to your question.
>
> > I would personally go for a client-server model; not worrying what the
> > server is written in or how it works. For some reason I have a scary
> > thought of your widget pulling in 100million records and cross-tabbing
> > two dimensions :) Don't worry about python bindings or such, why not
> > try and find something that can take an OLAP query, maintain
> > aggregates/computations on a DB that can possibly return in XML/JSON/
> > WHATEVER then just have your widget render visually the response and
> > send out XML/JSON/WHATEVER on click events.
>
> Regarding the implementation, I was already thinking about separation of
> the data query and the rendering part. Your answer was particularly
> useful, because I was not thinking about MDX and a real client server
> model. Until now. :-)
>
> Can you tell something about the licence of the MDX language? Can I just
> make a new engine/library that uses MDX? Or do I need to develop a new
> query language in order to make it open source?

IIRC it was MS' "analysis services", but it seems so wide spread --
I'm not sure if there's an ISO standard on it or anything...
I'm not a solicitor/lawyer-- GIYF.

>
> It seems strange that several open source RDBMS exist (sqlite,
> postgresql, mysql) but none of them support multi dimensional queries.

Not strange at all! When was MD part of the SQL standard? :)

One method would be to create a load of triggers that keep summary
info.
Depends on the DB and other stuff... Although then you're just
duplicating
work, and making the DB store meta-data, when something else should
be
doing that for it (in another DB maybe :) )

> Is this a very difficult thing to implement? Even if we could only have
> a not-so-efficient implementation that works with smaller datasets (100
> 000 records or so), it would be nice to have it.

Implement properly for small things (or at least acceptability),
probably not. To do it properly
waste your time and everyone else's -- oh yes.

Let's put it this way -- would you write your own RDBMS from scratch?

It also depends if you want "immediate" information, or if you're
happy for the queried DB to be slightly out of sync with the main DB.
So many options -- but I think this is getting OT for the python list.

>
> Thanks
>
>    Laszlo

Again, my humble opinion,

Jon.