Designing Real-Time Cubes - Professional SQL Server Analysis Services 2005 with MDX - page 950

One of the solutions for the long latency scenario is to use proactive
caching feature in Analysis Services 2005. In the proactive caching solu-
tion, you set proactive caching to kick in as soon as the data changes
using the option scheduled MOLAP. For the schedule MOLAP option you
need to specify a query that is to be run at scheduled time intervals to
determine if there has been a change to the source data. Here is how it
works: the first time Analysis Services sends the specified query to the
relational data source, it collects and stores the response. That stored
response provides a baseline against which subsequent query results
can be compared. When a subsequent query returns a result set that
does not match the baseline, it is presumed there has been a data up-
date and proactive caching will start the process of incremental update.
Depending on the other proactive caching settings such as latency the
cache will be updated. The latency setting is a property associated with
proactive caching; specifically, it tells Analysis Services how long to wait
between cache updates. This is what provides that real-time feeling to
the end user.
Figure 18-10 shows the proactive caching option where you specify a
polling query that will detect the change in source data. This could be as
simple as a count of rows in the relational table or as complex as a hash
value of the entire result set. For the long latency scenario you would
need to click on the Enable incremental updates option so that dimen-
sion and partitions are processed incrementally only with the data that
has been added. If this option is enabled, Analysis Services processes
the dimension or partition object by sending a Process Add statement. If
you do not specify this option, Analysis Services will automatically send
a Process update statement to the dimension or the cube partitions. Pro-
cess updates on dimensions could be expensive based on the size of
the dimensions and the partitions and aggregations built for the parti-
tions. For tradeoffs on which processing option (process update or pro-
cess add) would be good for your cube, please refer to the performance
chapter ( Chapter 13 ) in this topic. After specifying the polling query, you
need to specify the processing query that will retrieve appropriate data
from the relational data source for processing.