This post came out of the difficulty to find the information needed to prepare myself for a week long Exadata proof of concept I have scheduled next month at the Oracle Technology Center in Menlo Park.
The topics listed below should confirm your suspicion that there is significant learning and study required to understand Exadata. Outside of the Oracle Database software (which is the same) there is a lot more complexity to operate an Exadata Machine over a tradtional database server running on Linux, UNIX, Windows. Now I know why Oracle had to change the Exadata marketing from an “appliance” to a “machine”.
I plan to update these notes once the proof of concept is complete and I fully understand all the nuances of Exadata. Until then, consider this a list of topics and source of suggested resources for further study.
Exadata Infrastructure Components
- Database Servers
- Storage Servers
- InfiniBand
- Flash Cache
- CELLBOOT USB Flash Drive
Exadata Software Components
- Oracle Automatic Storage Management (ASM)
- Oracle Real Application Clusters (RAC). Not required but sales will say it is.
- Oracle Database
- Linux Logical Volume Manager (LVM)
Connectivity
- Database Servers connect to Storage Servers using the Intelligent Database protocol (iDB) implementing the “function shipping” architecture.
- iDB use the Reliable Datagram Sockets (RDS) protocol. RDS is designed for low latency and low overhead using less significantly less CPU when compared to UDP and TCP.
- iDB is based on libskgxp.
Exadata Database Servers
- No storage devices are presented to the OS on the Exadata database servers.
Exadata Storage Servers
- Storage Indexes contain min and max values for up to eight columns.
- Storage Indexes are 1MB (default) units. Identify locations where the data is not “Reverse Index”. Best for sorted data.
Exadata Smart Scan (Offloading)
- Column Projection
- Predicate Filtering
- Storage Indexes
- Bloom Filters
- SQL Functions (most single row functions can be offloaded)
- Compression and Decompression
- Encryption/Decryption
Smart Scans not available on:
- Clustered tables
- Index Organized Tabes (IOTs)
- When ROWDEPENDENCIES is enabled
Compression Types
Basic
- Compression unit is a single Oracle block. 0% PCTFREE
OLTP (Advanced Compression)
- Compression same as basic but uses symbol table to replace repeating values. 10 % PCTFREE for updates
Hybrid Columnar Compression (Exadata Storage Only)
- QUERY LOW uses LZO compression (4x)
- QUERY HIGH uses ZLIB compression (6x)
- ARCHIVE LOW uses ZLIP compression (7x)
- ARCHIVE HIGH uses bzip2 compression (12x)
Compression should not be used on tables or partitions where data is updated due to contention.
Exadata Backup and Recovery
- RMAN incremental backups benefit from the storage servers filtering out unchanged blocks.
Recommended Reading:
- Expert Oracle Exadata
- Achieving Extreme Performance with Oracle Exadata (Osborne ORACLE Press Series)
- A Technical Overview of the Oracle Exadata Database Machine and Exadata Storage Server
- Oracle Exadata Price List
No comments yet.