<Preference Name="MultiFileCubeThreshold" Value="500000"/>

[kado]

Hello Gurus,

Using Transformer in an 8.4.1 C8 environment on a Windows 32-bit OS with 4 Gigs of RAM ...
I adopted a cube that is taking about 7 hours to build 10.1 records and then fails with a Transformer Error:

"TR1901 The local data cache size exceeded the allowed 2147418112 bytes (2 gigs) in .... cube.mdc

I found an IBM board update about adding this to cogtr.xml:

<Preference Name="MultiFileCubeThreshold" Value="500000"/>

It almost finished but choked at 8500000 records (of 10m). Same error.

I'm convinced its because it was originally built to source only 1 data source that is one huge query (referencing 1 Fact and 8 Dims) from a Framework Manager package. And am intending to break up the data sources and hopefully partition the queries to return from the Oracle source quicker. Another complexity is that there are a close to 14 dimensions in the dimension map and some hierarchies with multiple drill paths. Fortunately though there are only about 8 measures.

A couple questions from general to specific:

1) Is an attempt to rebuild the data sources to essentially partition the cube a good approach?
2) Is there an easier way to get around this error so as to get a cube out there for our users?
3) Would a redesigned package specific to the complex hierarchies and dimensions be better for performance?

thanks,
kado