Oracle Data Warehouse Configuration Don Burleson

Configuring an Oracle Data warehouse is different than an OLTP Oracle database.

Oracle data warehouses have special needs, based on the types of Oracle data warehouse use:

1 - Summarization and aggregation
2 - Lots of large-table full-table scans

Oracle data warehouse configuration parameters

RAM region parameters

• db_keep_cache_size = You need to define a KEEP pool for small-table full-table scans and frequently referenced tables and indexes.  Click here for details.
   

• db_block_size = Most data warehouses set the default to 32k, and define smaller blocksizes for non-full table scan objects.  Click here for details.

Multiblock parameters

• sort_multiblock_read_count = you can control sort IO to disk by using SORT_MULTIBLOCK_READ_COUNT. It defaults to 2, and generally I suggest setting it up to, but not greater than, the DB_FILE_MULTIBLOCK_READ_COUNT value if you have sorts going to disk.
   

• hash_multiblock_read_count = set HASH_MULTIBLOCK_READ_COUNT to a similar value. A setting of zero for HASH_MULTIBLOCK_READ_COUNT allows the system to calculate it.
   

• db_file_multiblock_read_count = Click here for details

Materialized views - Oracle data warehouse configuration requires enabling MV's

query_rewrite_enabled  = true
query_rewrite_integrity  = trusted

CBO parameters - You must configure the Oracle data warehouse for STAR transformation joins and hash joins:

optimizer_mode = "all_rows"
star_transformation_enabled  = true
always_anti_join = hash
always_semi_join = hash
hash_multiblock_io_count = xx - Used to quickly build hash tables in the TEMP tablespace
 

Parallel query - Oracle data warehouse configuration requires enabling fast full-tale scans with paralel processes:

parallel_automatic_tuning = true
parallel_threads_per_cpu = 4
cpu_count = xx - Always verify that this is set properly by the Oracle installer