Java Industry News
Column Store, In-Memory, MPP Databases and Oracle
How Oracle implements latest developments in database world
By: Ranko Mosic
Dec. 27, 2012 02:02 PM
( For latest information on Oracle 12c database update please refer to the following article: Oracle 12c Database and How It Relates to SAP Hana )
RDBMSs are stable and mature products. While there is nothing radically new on horizon that would challenge Codd's relational theory and related advances in data processing there are some developments that force established vendors like Oracle to come up with new features and products.
Column Stores and Oracle
This is truly hybrid design, i.e., column store is implemented on top of a standard row store.
It is not in the scope of this article to discuss pros and cons of this implementation from performance, locking, compression and other points of view. I will just mention that HCC requires change in standard operating procedures and methods.
MPP (shared nothing) and Oracle
Oracle's implementation of this idea could be classified as asymmetric MPP. Oracle Exadata uses offloading to storage layer, Smart Scan, Storage Indexes and other techniques to improve performance.
Storage layer (Exadata cells) are tasked with as much work as possible to reduce load on database server and network. Each Exadata storage cell has the ability to perform some parts of data processing operations as well as decompression.
IMDB and Oracle
Oracle puchased Times Ten in-memory database, but marketed it mostly as caching layer to standard Oracle database. Times Ten is not marketed as stand alone IMDB the way SAP Hana is.
Oracle database can have Flash Cache devices configured as an extension of SGA for better performance (via database parameter), or for database logging purposes.
Exadata can be configured with terabytes of Flash Cache memory for database caching and to serve as solid state disk. This is not memory directly accessible by CPU though ( DRAM ), i.e., Oracle database accesses Flash Cache via PCI interface and IO operating system calls. In other words, Flash is treated the same as disk, with all negative consequences of such approach regarding code complexity and performance. The latest release of Exadata performs writes directly to flash cache first to improve performance. We should expect more optimizations that will try to better utilize abundance of various types of memory. Expected scenario could be similar to Microsoft Hekaton project is also about adding IMDB features to SQL Server ( tables can be loaded in memory and processed in IMDB fashion, with reduced latching and locking; perhaps choice between different storage engines will be possible).
SOA World Latest Stories
Subscribe to the World's Most Powerful Newsletters
Subscribe to Our Rss Feeds & Get Your SYS-CON News Live!
SYS-CON Featured Whitepapers
Most Read This Week