If there are partitions split across time, and there are measures that span
across time (i.e. Average, Median, etc),

how does that affect the partitioning aggregations?

and

how does that affect merging?

My customer would like to have a historical partition and a current
partition such that only the current parition would need to be reprocessed.


-=Steve

--
RDA Corp
Business Intelligence Evangelist Leader
www.rdacorp.com

All depends on how the aggregations are set up on the partitions.

We had a cube with three different types of partitions:

1. Last 30 days - lots of aggregations
2. 31 days to 180 days - some aggs
3. 181 + days - low amount of aggs

We partitioned daily so we had lots of partitions. Since the users were
mainly querying the last 30 days we had a lot of aggregations set up so the
queries would return quickly. They rarely queried from 180 days out so those
partitions were lacking on the aggregations. The only difference was the
speed in which the queries came back. Different aggregations does not affect
anything besides query speed.

Be careful when merging. Since you do have multiple partitions it's possible
for data to get doubled up between them if you're not careful. When merging
current into the historical make sure you process the current right
afterwards. Of course the current shouldn't overlap with any of the time
frame in the historical.





"Steve Mann" <SteveMann (AT) discussions (DOT) microsoft.com> wrote