DBAs managing production databases that support Oracle Data Mining should follow standard administrative practices as described in Oracle Database Administrator's Guide.
Building data mining models and batch scoring of mining models tend to put a DSS-like workload on the system. Single-row scoring tends to put an OLTP-like workload on the system. (Scoring operations are discussed in Scoring and Deployment of this manual.)
Database memory management can have a major impact on data mining. The correct sizing of PGA memory is very important for model building, complex queries, and batch scoring. From a data mining perspective, the SGA is generally less of a concern. However, the SGA should be sized to accommodate real-time scoring, which loads models into the shared cursor in the SGA. In most cases, you can configure the database to manage memory automatically. To do so, specify the total maximum memory size in the tuning parameter MEMORY_TARGET. With automatic memory management, Oracle Database dynamically exchanges memory between the SGA and the instance PGA as needed to meet processing demands.
Most data mining algorithms can take advantage of parallel execution when it is enabled in the database. Parameters in INIT.ORA control the behavior of parallel execution.
See Also:
"Database Performance Fundamentals" and "Tuning Database Memory" in Oracle Database Performance Tuning Guide
"Using Parallel Execution" in Oracle Database VLDB and Partitioning Guide