In the new version of Exadata, the X3, Oracle validates flash's value, improves performance and freezes hardware costs. Oracle customers will need to buy additional licenses.
On 30 September, 2012, Oracle announced Exadata X3, which will be available in two models, X3-2 and X3-8. In X3-2, important changes include:
An upgradable entry-level one-eighth rack that enables an Oracle Real Application Clusters (RAC) environment with 50% of the capabilities of a quarter rack at 60% of the price.
Flash Cache Write-Back (FCWB) — a new feature in the Exadata Storage Server Software (ESSS) that allows data written to flash to remain in cache. It increases write speed by about 20x for the X3 and 10x on older V2 and X2 systems.
The list price of the X3-2 hardware and the storage servers and the storage sizes remain unchanged. (For further X3-2 specifications, see www.oracle.com/us/products/database/exadata/overview/index.html )
X3-2 hardware comes with larger, faster components, including 32 GB more memory per compute node — with no price increase. The change in Intel processors and increase in base memory are predictable upgrades following Moore's law. Gartner is focusing on the X3-2 because we see few changes to the X3-8 at this time.
In the X3-2, the most significant change is the four-fold increase in flash, to a maximum of 22.4 TB, coupled with the FCWB feature. With this increase, Oracle makes a statement — it's time to increase reliance on flash because its performance has improved while the failure risk has dropped. FCWB, which allows data to remain in flash without the need to write it to disk, is available on the X3 plus existing V2 and X2 systems. The increase in flash helps Oracle refute marketing from flash vendors such as Fusion-io and Violin Memory.
The value of larger flash and FCWB depends on the use case for Exadata:
For data warehousing, with the Hybrid Column Compression of the ESSS (averaging about 10x), flash can now hold more than 200 TB of data. This will allow a large portion of the most used data to remain in flash for improved read performance.
For online transaction processing (OLTP) with the Oracle Advanced Compression Option (averaging about 3x), flash can hold more than 60 TB of data. Gartner estimates that for more than 80% of organizations, all the transaction data will fit in flash and with FCWB, will not be written to disk. This offering will greatly increase OLTP performance.
But these advantages have some added costs. Customers will need to buy additional Oracle database management system (DBMS) licenses because each server (compute node) has four additional cores requiring two additional DBMS licenses — approximately $750,000 list price on a full rack (8 nodes) of X3-2, plus related option licenses. Another concern involves the use of many generations in a cluster (e.g., X2 with X3). While the performance of multi-generational configurations will be better than an older, homogenous configuration, some tuning may be required to optimize performance. Balancing workloads across a multi-generational cluster is part manual and part automatic (with RAC) and will be increasingly difficult. As a result, some customers will consider replacing the entire set at significant cost.
Current Oracle customers:
Budget for increased DBMS costs because with X3-2, two additional licenses are required for each server.
If you use Exadata V1 or V2 and are planning a hardware replacement, consider a smaller configuration because speed, performance and storage size have increased in X3-2.
If you are using any version of Exadata for OLTP applications, install the free upgrade of the ESSS to take advantage of FCWB for increased performance.
If you wish to add an X3-2 to existing X2-2 or V2 infrastructure, consider the implications of multigenerational infrastructure and the potential need for tuning to extract maximum performance.
Some documents may not be available as part of your current Gartner subscription.
"Magic Quadrant for Data Warehouse Database Management Systems" — Gartner research examines the leaders, challengers, visionaries and niche players. By Mark Beyer and others
"IT Market Clock for Database Management Systems, 2012" — Our research examines the market maturity and commoditization levels for major DBMS technology asset classes. By Donald Feinberg and Eric Thoo