Dimension and Partition Processing - Microsoft SQL Server 2008 Analysis Services - page 379

affects processing performance, too, because every error causes Analysis Services to execute
an additional code path.

When the process buffer becomes full, Analysis Services passes it to the last stage of the
attribute processing: the Write Data job. Then, it starts to fill the second process buffer as
shown in Figure 21.2. If the second process buffer is still in use by the Write Data thread,
the Process Data thread becomes blocked. The minimum size of the process buffer is 64K
(65,536) records.

Write Data Job The last stage of attribute processing is the Write Data stage. By this time,
Analysis Services has already created and saved the key store, property store, BLOB store,
and unary operator store. In this stage, it needs to create a related attributes map store
that contains information about attributes related to the current one. For more informa-
tion about attribute data structures, see Chapter 20.

The Write Data thread performs three main operations:

. Analyzes the data in the process buffer

. Calculates the compression ratio and compresses the data

. Writes the compressed data to disk

The Write Data stage is typically a “light” process in that it doesn't require many system
resources. However, if you have many related attributes and the related attributes map
store becomes large, Analysis Services might spend noticeable amount of time in the Write
Data stage of attribute processing.

Finishing Attribute Processing
The last stage of attribute processing is checking the consistency of processed data and
saving the attribute metadata file. The attribute metadata file has the extension
.info.N.xml , where N stands for the version number of the attribute. The appearance of
this file in the data folder usually indicates that the processing for that attribute is
complete.

During this stage, Analysis Services builds the hash table of the property store and verifies
that names of all members in the attribute are unique. If the definition of the attribute
requires all member names to be unique and Analysis Services detects duplicate names, it
produces an error and terminates processing.

Hierarchy Processing

To achieve fast access to the members of the dimension through hierarchies, Analysis
Services builds special data structures. To process dimension hierarchies, Analysis Services
creates a Process Hierarchies job, which is marked as dependent on the job responsible for
processing attributes. Analysis Services starts a Process Hierarchies job only after all attrib-
utes are processed. The processing of a hierarchy doesn't depend on the processing of
other hierarchies; therefore, Analysis Services processes hierarchies in parallel. The Process