In-Memory
Automatic In-Memory Enhancements for Improving Column Store Performance
Automatic In-Memory (AIM) has been enhanced to automatically enable creation and removal of Database In-Memory performance features based on an enhanced workload analysis algorithm. These features include Join Groups, caching of hashed dictionary values for join key columns and In-Memory Optimized Arithmetic.
Automatic In-Memory (AIM) has been enhanced to identify and enable or disable Database In-Memory features that can improve performance. It enables features either selectively or globally, depending on which adds the most benefit. This improves application performance and also conserves space in the In-Memory column store without requiring manual intervention.
Automatic In-Memory Sizing for Autonomous Databases
The In-Memory column store will now automatically grow and shrink dynamically based on workload. This allows the In-Memory column store to be available on Autonomous Database. Exadata scan performance is further improved for objects that are partially populated.
With Automatic In-Memory sizing, there is no longer a need to manually resize the In-Memory column store to accommodate different database workloads. This reduces the administrative effort of enabling Database In-Memory. Automatic In-Memory sizing also allows the In-Memory column store to be enabled on Autonomous Database (ADB), enabling applications running on ADB to also take advantage of faster analytic query performance.
In-Memory Optimized Dates
To enhance the performance of DATE-based queries DATE components (i.e. DAY, MONTH, YEAR) can be extracted and populated in the IM column store leveraging the In-Memory Expressions framework.
This enhancement enables faster query processing on DATE columns which can significantly improve the performance of date based analytic queries.
In-Memory RAC-Level Global Dictionary
Database In-Memory Join Groups now support Global Dictionaries across RAC nodes. With Join Groups a common dictionary is shared by columns that are joined together. In-Memory RAC-level global dictionaries now synchronize these common dictionaries across nodes within a RAC database.
In a RAC environment, this feature further improves database In-Memory performance for distributed hash joins.
Selective In-Memory Columns
With Selective In-Memory columns, it is now easier to add or exclude columns for in-memory. An ALL sub-clause has been added so that all columns can be either enabled or disabled from in-memory. This reduces the need to have very long strings of included or excluded columns.
With Selective In-Memory columns, the ability to specify ALL columns to be enabled or disabled from in-memory reduces the need for very long strings of columns, which reduces the chance for errors and makes configuring in-memory enabled tables easier.
Vectorized Query Processing: Multi-Level Joins and Aggregations
This feature enhances the In-Memory Deep Vectorization framework by fully exploiting SIMD capabilities to further improve hash join and group by aggregation performance. New optimizations include incorporating multi-level hash join support, full In-Memory group by aggregation support, and support for multi-join key and additional join methods.
This feature adds improvements in the performance of joins and aggregations, which are the foundations for analytic queries. This enables faster real-time analytic performance and requires no application SQL changes. This feature is automatically used when enabled, which is the default.