Processing Options
The Type element in the Process command specifies the type of processing to be done. This is also referred to as the "processing option". Choosing the right processing option is very important since it typically affects large volumes of data. This section describes the different processing options for each type of object.
Following are the types of Analysis Services objects that can be processed:
- Database
- Dimension
- Cube
- Measure group
- Partition
- Mining structure
- Mining model
Before we look at the processing options, we first need to understand what exactly processing is doing. Processing generally sends queries to the relational data source and uses the results to populate the Analysis Services object. Following are the storage contents that are built by processing for each object type. A complete discussion of the Analysis server storage format is beyond the scope of this document.
- Database
- None (a database is just a container of dimensions, cubes, and mining structures; it has no data storage of its own)
- Dimension
- Attribute stores
- Hierarchy stores
- Bitmap indexes
- Cube
- MDX Script cache
- Measure group
- None (a measure group is just a container of partitions; it has no data storage of its own)
- Partition
- Fact data
- Aggregation data
- Bitmap indexes
- Mining structure
- Training data
- Mining model
- Algorithm content
When an object is processed, the processing option specifies what part of the storage content of the object is to be built and how. The following table contains the list of all processing options and which objects they apply to.
ProcessFull
ProcessFull applies to all objects. It discards the storage contents of the object and and rebuilds them. ProcessFull is recursively applied to all descendants of the object as well.
ProcessClear
ProcessClear applies to all objects. It discards the storage contents of the object. ProcessClear is recursively applied to all descendants of the object as well.
ProcessDefault
ProcessDefault applies to all objects. It does the bare minimum required to bring the object to a fully processed state. In other words, it builds only the storage contents that are not currently built. For example, if a partition has fact and aggregation data, then ProcessDefault will only build the bitmap indexes.
The only exception to the above rule is bindings. When you change the bindings of an object (e.g., KeyColumns of a dimension attribute), the object retains its data, but remembers that the bindings have changed. ProcessDefault on the object will discard all the storage contents and rebuild them.
The server handles ProcessDefault by analyzing the object and dynamically converting it to another processing option such as ProcessFull,ProcessIndexes, etc. ProcessDefault is recursively applied to all descendants of the object as well.
ProcessData
ProcessData applies only to the OLAP objects, i.e. dimension, cube, measure group and partition. It discards the storage contents of the object and rebuilds only the "data". For dimensions, it builds only the attribute and hierarchy stores. For partitions, it builds only the fact data.
Essentially ProcessData builds the bare minimum required for the object to be available for queries. Indexes are considered optional and affect only the query performance. ProcessData is recursively applied to all descendants of the object as well.
ProcessIndexes
ProcessIndexes applies only to the OLAP objects, i.e., dimension, cube, measure group, and partition. It requires that the object must already have its "data" built; otherwise, it raises an error. ProcessIndexes preserves the data and rebuilds the "indexes". For dimensions, it builds the bitmap indexes. For partitions, it builds the aggregation data and bitmap indexes. ProcessIndexes is recursively applied to all descendants of the object as well.
ProcessUpdate
ProcessUpdate applies only to dimensions. It is the equivalent of incremental dimension processing in Analysis Services 2000. It sends SQL queries to read the entire dimension table and applies the changes—member updates, additions, deletions.
Since ProcessUpdate reads the entire dimension table, it begs the question, "How is it different from ProcessFull?" The difference is that ProcessUpdate does not discard the dimension storage contents. It applies the changes in a "smart" manner that preserves the fact data in dependent partitions. ProcessFull, on the other hand, does an implicit ProcessClear on all dependent partitions. ProcessUpdate is inherently slower than ProcessFull since it is doing additional work to apply the changes.
Depending on the nature of the changes in the dimension table, ProcessUpdate can affect dependent partitions. If only new members were added, then the partitions are not affected. But if members were deleted or if member relationships changed (e.g., a Customer moved from Redmond to Seattle), then some of the aggregation data and bitmap indexes on the partitions are dropped. The cube is still available for queries, albeit with lower performance.
ProcessAdd
ProcessAdd applies only to dimensions and partitions.
ProcessAdd is a new processing option for dimensions that did not exist in Analysis Services 2000. It essentially optimizes ProcessUpdate for the scenario where only new members are added. ProcessAdd never deletes or updates existing members. It only adds new members. The user can restrict the dimension table so that ProcessAdd reads only the new rows.
ProcessAdd for partitions is the equivalent of incremental partition processing in Analysis Services 2000. The user typically specifies an alternate fact table or a filter condition pointing to the new rows. ProcessAdd internally creates a temporary partition, processes it with the specified fact data, and merges it into the target partition.
See the Out of Line Bindings section for details on how to specify the new rows for ProcessAdd.
ProcessStructure
ProcessStructure applies only to cubes and mining structures.
ProcessStructure for cubes is the equivalent of the Analysis Services 2000 processing option, processBuildStructure, in DSO. It discards the storage contents of the cube and its partitions. It implicitly does a ProcessDefault on all dimensions of the cube and marks the cube as processed. At this point, the cube is available to queries but it will not return any fact data. This is supported mostly for backward-compatibility reasons. It was useful in Analysis Services 2000 for parallel processing utilities. Once the "structure" of a cube is processed, its partitions can be processed in parallel by multiple client sessions without running into locking conflicts.
ProcessStructure on a mining structure discards its storage contents (training data) and rebuilds them. It does not affect the contents of the mining models under the mining structure.
ProcessScriptCache
ProcessScriptCache applies only to cubes. The MDX script in a cube can contain CACHE statements. ProcessScriptCache evaluates the script and persists the results for the CACHE statements.
ProcessClearStructureOnly
ProcessClearStructureOnly applies only mining structures. It clears the storage contents (training data) of the mining structure while preserving the contents of its mining models.
分区处理命令
在需要对分区执行处理操作时，将发出分区处理命令。 每个分区处理命令都创建一个或多个作业以便执行所需操作。
提供以下分区处理命令：
• ProcessFull
• ProcessData
• ProcessIndexes
• ProcessAdd
• ProcessClear
• ProcessClearIndex
ProcessFull 将放弃分区的存储内容并且重新生成它们。 在后台，ProcessFull 执行 ProcessData 和 ProcessIndexes 作业。
ProcessData 将放弃对象的存储内容并且只重新生成事实数据。
ProcessIndexes 要求一个分区以便生成其数据。 ProcessIndexes 保留在 ProcessData 期间创建的数据并且基于这些数据创建新的聚合和位图。
ProcessAdd 在内部创建一个临时分区，使用目标事实数据处理该分区，然后将该分区与现有分区合并。 请注意，ProcessAdd 是 XMLA 命令的名称，在 Business Intelligence Development Studio 和 SQL Server Management Studio 中，该命令作为 ProcessIncremental 公开。
ProcessClear 删除分区中的所有数据。 请注意，ProcessClear 是 XMLA 命令的名称。 在 Business Intelligence Development Studio 和 SQL Server Management Studio 中，该命令作为 UnProcess 公开。
ProcessClearIndexes 删除分区中的所有索引和聚合。 这使分区所处的状态就像在 ProcessData 后刚运行 ProcessClear 一样。 请注意，ProcessClearIndexes 是 XMLA 命令的名称。 该命令不可用于 Business Intelligence Development Studio 和 SQL Server Management Studio 中。