Dimension and Partition Processing - Microsoft SQL Server 2008 Analysis Services - page 393

Building aggregations and indexes for HOLAP partitions is an expensive operation because
to build aggregations, partition data should be loaded into Analysis Services. Even though
a HOLAP partition doesn't store processed data (because the storage of the partition itself
is ROLAP), it needs data to build indexes and aggregations. Therefore, for HOLAP parti-
tions, the ProcessData operation loads data, builds aggregations and indexes, and then
disposes of the data. We've found that HOLAP partitions require a lot of management
overhead as compared to MOLAP partitions, and have few benefits.

Building Aggregations and Indexes
Analysis Services builds aggregations and indexes when you issue the process command
with the ProcessFull , ProcessAdd , or ProcessIndex option. In addition, Analysis Services
can build indexes and aggregations as part of its lazy processing.

Lazy processing enables you to process only partition data using the ProcessData process-
ing option to enable your users to see data in the partition right away and to postpone
building of indexes and aggregations until a later time. Later on, Analysis Services auto-
matically starts building indexes and aggregations for your partition in the background.
When it completes these operations, your users will see improved performance.

Analysis Services has a lazy processing thread that periodically wakes up and iterates over
all the partitions on the server to decide which indexes and aggregations are missing.
Analysis Services starts lazy processing according to the priority specified by the
ProcessingPriority property of the Partition object. The lazy processing thread analyzes
CPU utilization and starts to build indexes and aggregations for more partitions to maxi-
mize utilization of the CPU.

Analysis Services processes indexes by internally issuing the Process command with the
ProcessIndex option. If during the lazy processing operation you issue any Data
Definition Language (DDL) command that affects the partition, Analysis Services cancels
the lazy processing. To enable lazy processing, use the Analysis Services server configura-
tion properties shown in Table 21.1.

TABLE 21.1

Lazy Processing Properties

Name Description

A Boolean property that controls
whether lazy processing is enabled

OLAP\LazyProcessing\Enabled

The percentage of the machine's CPU to
dedicate to background lazy processing

OLAP\LazyProcessing\MaxCPUUsage

The maximum number of partitions to
process in parallel by lazy processing

OLAP\LazyProcessing\MaxObjectsInParallel

The number of seconds that the lazy
processing thread sleeps between itera-
tions over partitions

OLAP\LazyProcessing\SleepIntervalSecs

The lazy processing mechanism has disadvantages; the main one is that you cannot
predict when Analysis Services will finish building indexes and aggregations, so you don't