"Competitors love to describe Exadata as just an Oracle database running on Sun hardware, but it's much, much more than that," Tim Shetler, vice president of product management at Oracle, said in an interview. "It's a computing system designed to run the Oracle database as well as any platform could."
This means that Exadata storage devices include some of the Oracle database code to help them understand exactly what an Oracle 11g database running on the Exadata server needs, speeding up access times and reducing load on the network and the server's processor. "Most storage just responds to requests for blocks, but Exadata can employ storage processing power to offload database requests," said Shetler. To further reduce latency, it also includes solid-state flash caches, with decisions on what to cache again helped by the onboard Oracle code.
Exadata has implemented these proprietary techniques since 2009, but until now only on the hard drives within its database machines so Big Data customers faced an awkward workaround. "Previously, we'd tell them to buy another database machine and turn off the processors," said Shetler.
The new system offers the Exadata storage in dedicated racks, with each one using 2-TB hard drives to support up to 432 TB of storage when fully loaded. Up to seven of the racks can be connected to a single database machine. Customers can start out with a 96-TB quarter-rack system. However, the actual capacity that users will see varies depending on the type of data. Shetler said that Exadata's proprietary version of RAID reduces the full rack figure to about 200 TB, then compression increases it to about 2 PB. To boost performance, the same compression is used throughout the Database Machine, even for data held in RAM.
Whereas Exadata was originally positioned as a competitor to data warehousing appliances such as those from Teradata and IBM's Netezza, Oracle's ambitions for it are now much bigger. "Exadata supports standard Oracle databases and any data workload that customers can throw at it," said Shetler. In particular, Oracle is aiming it at online transaction processing (OLTP) applications that have a lot of reads and writes, whereas data warehousing is mostly reads. Teradata argues that the two types of applications are fundamentally different and require different database systems, whereas Oracle argues that they can all run on the same system.
Oracle also takes issue with other vendors' descriptions of their products as appliances. "We don't call it an appliance because that suggests something that's relatively small or that requires very little administrative support," said Shetler. While an Exadata machine is easier to get up and running than a traditional Oracle system, it can still take days of programming. In contrast, software databases take weeks or months, something Shetler blames on hardware rather than Oracle software.
The Exadata philosophy of designing server and storage hardware to fit a specific application runs counter to the virtualization trend, in which applications are decoupled from hardware and can run anywhere in the cloud. However, very high performance almost always requires dedicated hardware, and the time savings offered by pre-integrated appliances (or database machines) can provide many of the flexibility benefits often touted for virtualization.
Data centers face increased resource demands and flat budgets. In this report, we show you steps you can take today to squeeze more from what you have, and also provide guidance on building a next-generation data center. Download it now.