[Mondrian] Re: Mondrian mp factory classes

Eric McDermid mcdermid at stonecreek.com
Thu Jul 16 18:14:28 EDT 2009

I ran the whole suite, dumping stats for every Map.get().  While there  
are some cases where using Flat3Map is suboptimal (map size > 3),  
they're outnumbered nearly 10 to 1 by the cases where it is used  
optimally.  Should be a net win overall.

If we really want to get serious about it I'm sure we can add some  
information in those cases where we put the extra properties into the  
map, but I'm not going to take that on just yet.  I'll log a JIRA  
issue to remind us.

As a side note, 99.98% of the suboptimal choices were associated with  
[Measures].[MeasuresLevel].  All of them had zero declared properties,  
but either 4 or 5 "extra" properties.

  -- Eric

On Jul 16, 2009, at 1:47 PM, Julian Hyde wrote:

>> Eric wrote:
>> Since the member is now passed as an argument to the create()
>> factory
>> method, my intent was to default to Flat3Map only where
>> member.getProperties().length < 4, and stick with a HashMap for
>> anything larger.  That will avoid the extra overhead of
>> Flat3Map when
>> there are more than three entries, and should give results
>> equivalent
>> to what we have now.
> Ah, not so. We may put properties into the map that are not explicitly
> declared as member properties.
>> I'll do the experiment anyway, though.
> Julian

More information about the Mondrian mailing list