October 16 OTN Developer Day – 12c InMemory Database Highlights

Mike Appleyard presentation.

Can be used to change from batch processing to more realtime processing.

In memory is supported with multi tenant.

12c option goals

– 100x Faster Queries – Real-time Analytics

– Faster mixed workload oltp

– Transparent – no application changes

Transactions run faster on row format, analytics run faster on column format.

Both row and column format at the same time and transactional consistent.

2x to 20x compression for in-memory

In memory columnar format can be enabled at table or partition level and per column level by excluding columns.

Loading into the in memory area is first fit and once full there is no aging/LRU algorithm.

In memory enabled tables/partitions have priorties – critical is loaded before accessed.

Complex OLTP slowed by Analyic Indexes – potentially can remove indexes used only for analytical queries depends on your workload – less io/storage needed.

In-memory is supported with RAC.- 2 modes, either duplicate data on each node (engineered systems only) or partitional across nodes with multi-instance parallel query.

In-memory is supported with Dataguard but not Active Data Guard.

Flashback query/flashback transaction?

Instance Parameters

–  inmemory_query enabled/disable

– inmemory_clause_default

– inmemory_force

Use v$im_segments to monitor in-memory.

In-memory is a licensable option.

5 types of compression of in-memory, can be configured with alter table on a per partition level.

Currently release in-memory does query/bloom filtering but aggregation is done in the pga.

In-memory storage indexes – built on the fly in-memory segments which contain min/max values per compression units.

This are not the same as storage indexes for exadata which are per 1MB chunk.

Optimizer hint can be used to turn on/off im-memory.

Bloom filter – filters on multiple predicates in one pass rather than hash join on 1 predicate and rescanning the same data 4 times to apply each predicate.

RAC leaf nodes – no ORACLE_HOME or voting disk just compute nodes.thin install.- no locking LMS/LMON. readonly but transactional consistent in 11gr2.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s