Performance Optimization - Professional SQL Server Analysis Services 2005 with MDX - page 662

role in getting the best performance. Analysis Services allows you to monitor re-
sources used during operations of the server by way of perfmon counters. There
are specific perfmon counters that show the memory utilization on the Analys-
is Services instance. You can monitor the server behavior during the process-
ing operation via perfmon counters of the objects \\<machinename>\Processor,
\\<machinename>\Memory, \\<machinename>\PhysicalDisk and specific counters
provided by Analysis Services \\<machinename>\MSAS 2005: Processing, \\ <ma-
chinename>\MSAS 2005: ProcIndexes, \\<machinename>\MSAS 2005: ProcAg-
gregations, and \\<machinename>\ProcIndexes. After identifying the bottlenecks
of the system, you can take appropriate action to relieve the server performance
hot spots.
Some simple hardware additions, such as increasing server memory, adding more
CPU, or using fast writing disks can potentially improve the system processing per-
formance. As mentioned earlier, if you have memory over 3G for a 32-bit machine,
you might consider using the /3-GB flag to allow Analysis Services to use memory
over the 3GB limit as discussed earlier in this chapter. If memory is the main bottle-
neck we highly recommend you to consider using 64-bit machines to increase the
performance and capacity of your system, especially for cubes with large dimen-
sions (>10 million dimension members).
In certain cases where the number of partitions is in the order of hundreds, pro-
cessing partitions in parallel on a 32-bit machine can result in partitions compet-
ing among themselves for memory resource. In such circumstances processing
can result in errors that system does not have sufficient memory. In such circum-
stances you can split the processing of the partitions by one of the following tech-
niques.
1. Process fewer partitions at a time and stagger them to complete process-
ing of all the partitions.
2. Use one of the techniques mentioned in the Parallelism during Processing
section.
3. Instead of doing a full process of the partitions which includes processing
data, indexes, and aggregations, split the processing as Process Data first
for all partitions followed by Process Indexes.
4. By default the server property OLAP\ProcessPlan\MemoryLimit is set to
65 which means 65% of the available memory. If the /3GB flag is not en-