[Mondrian] Adding Grouping Set support for Distinct Count measures

Matt Campbell mkambol at gmail.com
Fri Jan 25 15:03:40 EST 2008


I had been hoping Ajit would answer your 2nd question.  I *believe* that
this will also work when in aggregate().

I've asked one of our Teradata DBAs for more info on the potential risks of
GROUPING SETS w/ count distinct.  I'll let you know what he says.

On Jan 24, 2008 11:27 PM, Julian Hyde <jhyde at pentaho.org> wrote:

>  Matt,
>
> That rationale is useful - thanks.
>
> Did you try against Teradata? I am a little hazy on what combinations of
> grouping sets/distinct-count/AJIs they support, but it's possible that
> distinct-count + grouping sets may prevent usage of an AJI, and that would
> be unfortunate because AJIs give a big boost to performance.
>
> You didn't answer my 2nd question: Would this work with distinct-count
> measures applied to aggregate members? (For example unit sales over [CA plus
> OR].)
>
> Julian
>
>  ------------------------------
> *From:* mondrian-bounces at pentaho.org [mailto:mondrian-bounces at pentaho.org]
> *On Behalf Of *Matt Campbell
> *Sent:* Thursday, January 24, 2008 7:38 AM
> *To:* jhyde at pentaho.org; Mondrian developer mailing list
> *Subject:* Re: [Mondrian] Adding Grouping Set support for Distinct Count
> measures
>
> We're interested in this change not so much for performance (although we
> have seen some cases where it seems to improve performance) but because:
>
> 1) It will allow correct totaling in cases where a dimension table does
> not join to every fact table row.  We can get over-counting in these cases
> because the total query is fired separately without joining to the dimension
> table.  I know in general dimension rows should be all-inclusive, ie have a
> member value for each fact table row.  We have some good reasons for
> violating this rule, though.
>
> 2) It will also bring greater consistency to how count distinct and
> summable measures are evaluated.
>
>
>
> On Jan 22, 2008 7:52 PM, Julian Hyde < jhyde at pentaho.org> wrote:
>
> >  No objections in principle. Have you hand-generated such queries to see
> > whether they are any improvement? It is possible that the queries do not
> > yield any performance improvement, because distinct-count is difficult for a
> > database to evaluate and therefore it might be difficult for the DBMS to
> > find commonality between the different grouping sets.
> >
> > Would this work with distinct-count measures applied to aggregate
> > members? (For example unit sales over [CA plus OR].)
> >
> > Julian
> >
> >  ------------------------------
> > *From:* mondrian-bounces at pentaho.org [mailto:
> > mondrian-bounces at pentaho.org] *On Behalf Of *Ajit Vasudeo Joglekar
> > *Sent:* Tuesday, January 22, 2008 9:33 AM
> > *To:* mondrian at pentaho.org
> > *Subject:* [Mondrian] Adding Grouping Set support for Distinct Count
> > measures
> >
> >
> > We would like to introduce grouping sets support for distinct count
> > measures in mondrian. We have analyzed this and it is a small change. Most
> > of the work is around changing and adding test cases.
> >
> > It will be great if we can make this a part of the upcoming release.
> > Requesting comments
> >
> > Thanks
> >
> > -Ajit
> >
> >
> > _______________________________________________
> > Mondrian mailing list
> > Mondrian at pentaho.org
> > http://lists.pentaho.org/mailman/listinfo/mondrian
> >
> >
>
> _______________________________________________
> Mondrian mailing list
> Mondrian at pentaho.org
> http://lists.pentaho.org/mailman/listinfo/mondrian
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.pentaho.org/pipermail/mondrian/attachments/20080125/ba418ad6/attachment.html 


More information about the Mondrian mailing list