Difference between revisions of "SAP HANA"
From Suhrid.net Wiki
Jump to navigationJump to search| Line 11: | Line 11: | ||
* Data locality in memory | * Data locality in memory | ||
* Columnar structures | * Columnar structures | ||
| + | * Row store : in memory : Transactions are quicker | ||
| + | * Column store : for analytics | ||
| + | * Projections, Dynamic Aggregation, Integrated Compression (From columnar structure) | ||
| + | * Insert-only : When something changes in a colstore, insert it and invalidate the old data | ||
| + | * Partitioning : distribute data across machines. rows/cols can be distributed. | ||
| + | * Hot/cold : When semantics are known, then some data can be kept in hot. | ||
Revision as of 07:32, 6 April 2014
Motivation
- Earlier: CPU, Memory expensive
- New DB needs to take advantage of multi-core, large RAM
- and Columnar technologies : Transactional vs Analytical
- Better for retrieval performance (analytical workloads)
Technology
- Multi-core parallelism : massively parallel. Every operator is in parallel.
- Data locality in memory
- Columnar structures
- Row store : in memory : Transactions are quicker
- Column store : for analytics
- Projections, Dynamic Aggregation, Integrated Compression (From columnar structure)
- Insert-only : When something changes in a colstore, insert it and invalidate the old data
- Partitioning : distribute data across machines. rows/cols can be distributed.
- Hot/cold : When semantics are known, then some data can be kept in hot.
