Performance Optimization - Professional SQL Server Analysis Services 2005 with MDX - page 704

based on the Low Memory Limit. The default value is 75%. As soon
as the low memory limit crosses the 75% threshold then the cleaner
thread requests existing jobs to shrink memory and starts to reclaim
the memory. In certain instances while processing multiple partitions
in parallel the cleaner thread might not be fast enough to wake up
and reclaim memory. Due to this existing processing, jobs could fail.
By lowering the Low Memory Limit you can make sure cleaner
thread starts to reclaim memory earlier and hence all the processing
jobs will complete.
3. CoordinatorExecutionMode: This property is used for parallelism of
jobs and takes a negative or positive value. The value indicates the
number of coordinator jobs that can be in parallel at a given point in
time. If the value is negative then the number is multiplied by the
number of processors on the machines. Having a high value can de-
teriorate performance since multiple threads are competing for re-
sources. The default value is -4.
4. ThreadPool\Query: This node contains several properties such as
minimum and maximum number of threads to be allocated for a
query and the priority for these threads. The MinThreads property in-
dicates the number of threads created for each query and the
MaxThreads property indicates the number of maximum threads that
will be allocated for a query. A query can be split into multiple jobs
and these jobs can be executed in parallel. Based on the
MaxThreads, Analysis Services will have all the jobs execute in par-
allel (assume there are no dependencies).
5. ThreadPool\Processing: Similar to ThreadPool\Query this node con-
tains the same subproperties. However, the properties are applicable
for each process statement rather than the query.
6. OLAP\ProcessPlan\MemoryLimit: This value is represented as a per-
centage of available system memory. This value indicates the max-
imum amount of memory used for a specific processing job. As men-
tioned earlier in this chapter we recommend you increase this value
up to the Total Memory Limit if you encounter out of memory errors
while processing multiple partitions in parallel.