Going in Production - Expert Cube Development with SSAS Multidimensional Models - page 289

Partitioning in Analysis Services is different in several respects:

• Partitions in the same measure group can, and often do, have different slices.

For example, as in our case, older data may be partitioned at the year level,
while more recent data may be partitioned at the month level. As we said,
in the relational database, partitioning is normally done with only one
type of slice.

• Similarly, different partitions may have different aggregation designs. Older

data tends to be queried less often and at a higher level, and so partitions
containing older data may need different aggregations to partitions
containing newer data.

• Sometimes, not all the data from the relational database needs to be loaded

into the cube. Very old data might be useful in the relational database for
relational reporting purposes but dropped from the cube, for example, to
make it faster to process, query, or back up.

• We may change our partitioning strategy if our requirements change. If, for

example, it turns out that many queries are being run at the month level for
old data, we might change our partitioning strategy so that we use monthly
partitions for all of our data. This, of course, would not require any changes
to the relational database, although it would mean that we would have to
process all of the newly-created partitions, which might take a long time.

Building a template partition
The first step towards implementing a dynamic partitioning strategy is to create
a template partition in our measure group. Since in a partitioned measure group
all of the partitions are identical except for the slice of the data they contain and
possibly the aggregation design they use, what we're going to do is to create a single
partition that will be used as a blueprint by our Integration Services package. When
the package creates a real partition in the measure group, it will make a copy of the
template partition and change some of its properties so that it will contain, when
processed, the appropriate slice of the fact data.