Performance Tuning and Troubleshooting Tabular Models - SQL Server Analysis Services 2012 Cube Development - page 402

Specific issues will require investigation and systematic troubleshooting steps, but
consider these general guidelines that may help to prevent or resolve many common
issues:
• If the Tabular server is used to run other services such as the SQL Server
relational engine, restrict the memory allocated to those services so the Tab-
ular instance doesn't have to compete for the required memory.
• Monitor the rate of compression for each table and database. Investigate and
optimize objects that are not being compressed effectively by eliminating un-
used columns, and use conservative data types, especially for columns that
contain sparse and redundant values.
• After estimating the size of each Tabular database on the server, once the
data is compressed, allow at least 20 percent of more memory as a contin-
gency buffer.
• Allow additional memory to facilitate object processing. The total memory
that is required will largely depend on the processing strategy. For databases
that use the Process Full option, allow twice the amount of memory that is
required when the database is at rest, while the database is being processed.
• Use a partitioning strategy to reduce the memory overhead on the server
and control the use of memory on development machines. If you can process
a single partition containing added rows or process partitions for groups of
rows one at a time, this will reduce the overall use of memory during the pro-
cessing time.