Building Scalable Analysis Services Applications - Microsoft SQL Server 2008 Analysis Services - page 456

creates all the necessary data structures (see Chapter 20, “The Physical Data Model”). The
data structures remain empty until the first time a user queries the data. When Analysis
Services receives a query for the dimension data, it first determines whether the data is
available in the cache. If there is data in the cache, Analysis Services treats the linked
dimension as a multidimensional online analytical processing (MOLAP) dimension. If
Analysis Services can't locate data in the cache, it sends a request to the publisher server.
The publisher server then iterates over all the relevant data structures, gathers data for
only the requested dimension members, and sends the data to the subscriber server.

The subscriber server inserts the data it received into the data structures at exactly the
same ordinal position as it was stored in the publisher server and registers the data with
the dimension cache. At that moment, the dimension is considered populated; the
subscriber server starts to treat it as a MOLAP dimension.

Linked dimensions that are often used and relatively small might be fully cached on the
subscriber. You will see optimal query performance with these dimensions cached. As the
result of this caching, the key store and other dimension stores contain only the records
that appear in the dimension cache. Therefore, Analysis Services uses the dimension cache
as a registry of the linked dimension data. This leads to an unfortunate side effect. In
Analysis Services, the information about populated records is lost after the cache is
cleaned. The subscriber might have the entire dimension structure fully populated, but
the fact that it's populated is lost. As a result, the subscriber has to request data again from
the publisher.

When you have linked a dimension to the corresponding dimension on the publisher,
you can use it as if the dimension belonged to the subscriber database. You can include
the linked dimension in your cube and associate it with a measure group, and you can
load real data into your partitions. You can use a linked dimension in any measure group,
not only in a linked measure group. For example, you can link the Employee dimension
from your Human Resources server when your HR department needs information about
employees in your organization. The Geography dimension is another dimension that you
might use centrally in your organization to get a common view of the data on cities and
ZIP Codes across the United States. The advantages of such reuse are as follows:

. A single standard version of the dimension eliminates the need to duplicate the

dimension multiple times on each server.

. Updates to the dimension are centralized. A new version becomes available to all the

servers at once. There are no delays and no discrepancies between versions.

Nevertheless, there is a disadvantage. You can suffer some performance degradation
retrieving dimension data across the network from another server and recycling the
subscriber requires to bring dimension again.