Wednesday, December 22, 2010

What makes Exadata different from regular RAC?:

What makes Exadata different from regular RAC?
We recently migrated 6 of our 10+ TB production databases to Exadata server.

1. Smart scan:

To disable smart scan set cell_offload_processing=FALSE (default TRUE) for that database.

List of cell DB parameters:
NAME TYPE VALUE
------------------------------------ ----------- ------------------------------
cell_offload_compaction string ADAPTIVE
cell_offload_decryption boolean TRUE
cell_offload_parameters string
cell_offload_plan_display string AUTO
cell_offload_processing boolean TRUE
cell_partition_large_extents string DEFAULT

2. Flash cache:

- Frequently accessed data and index are automatically cached.
- control file read and write are cached.
- File header read and writes are cached.
- DBA's can pin objects in flash cache.

- table scan data is not cached.
- backup data is not cached.
- DB pump data is not cached.
- data file formating data is not cached.
- IO to mirror copy is not cached.

- DBA can cache an object by setting CELL_FLASH_CACHE setting to KEEP. The default value of this setting is DEFAULT.

3. Exadata Hybrid Columnar Compression:

EHCC is not pure columnar, it is hybrid columnar

Multiple Levels of compression
- Query Low
- Query High
- Archive Low
- Archive High

Algorithms:
LZO: Faster compression speeds, but lower ratios (Query Low)
ZLIB: Decent speeds and good ratios (Query High and Archive Low)
BZ2: Slow, but the highest ratios (Archive High)

4. Exadata storage index:

Transparent I/O Elimination with No Overhead.

- Exadata storage indexes maintain summary information about table data in memory.
* Store MIN and MAX values of columns.
* Typically one index entry for every MB of disk.
- Eliminates disk I/O if MIN and MAX can never match "WHERE" clause of a query.

5. DBFS (database file system) - Rest of the 10 SYSTEMDB disks can be used as DBFS.

6. IORM (I/O resource manager - Exadata specific)


Monitoring Exadata:

- Install GC agent on each compute node.
- No GC agent on cell nodes.
- Agent on one compute node talks with all cell nodes and collects data.
- Grid control plugin is installed in OMS.