Database Reference
In-Depth Information
Figure 7.8
Data loaded for cube AS2sTc55 (top) and AS2Tc55z (bottom). (From Oracle Essbase Administration Services.
With permission.)
The estimated cube size in mb (Level-0 mb) in Figure 7.7 can be calculated as:
[CompressedrecordCount][AvgCompressedrecordSize]
1024
∗
∗∗
2
(note: this is to convertthe result to mb)
Where
[numberofInput-Level Cells]
[ABF]
[CompressedrecordCount]
≡
[Avg CompressedrecordCount] [max.key Length(bytes)]) ABF] [AvL]4)
≡
+
∗
+
note that only two of the five fact data values have been changed. The three unchanged
fact data members hide, to some extent, the impact of the change from three to two dig-
its after the decimal point. other tests (results not shown here) on cubes with only one
member [original Price] loaded in the [measures] dimension show a reduction from
AvL = 7.52 with three decimals data loaded to AvL = 3.83 with two decimals loaded.
to summarize the discussion of AvL, remember it only matters if you are using com-
pression and, as in the results noted just above, not storing unnecessary digits can result
in significant reductions to the input-data Size. remember, a smaller input-data Size
means a smaller box of cards and less memory and, therefore, improved performance.
7.4.3.3.3 The Compression Dimension—Will It Help?
The description of the
Compression dimension certainly was not short or, even with all the help and sugges-
tions I received from the co-authors of this topic, easy to understand. In the end, what
have we learned and how will those lessons improve performance? So far, we know only
that using a Compression dimension will (virtually always) reduce the size of the data-
base on disk and in memory. That will certainly help performance. Counterbalancing
this is the requirement that the Compression dimension
must
be a Dynamic dimension.
In the ASosamp cube the Dynamic [measures] dimension, as designed, does not
have any examples of additive (+) consolidation that could take advantage of a Stored
hierarchy. The only calculations within [measures] are ratios that could not take advan-
tage of Stored consolidation. Thus, there is nothing to be lost in making [measures] a
