As Published In
Oracle Magazine
November/December 2008


Data Warehousing Gets Extreme

By Jeff Erickson

Oracle delivers the HP Oracle Database Machine and the HP Oracle Exadata Storage Server.

Oracle is moving into the hardware business with partner HP to deliver a performance boost to Oracle data warehouses. The move, says Oracle CEO Larry Ellison, is a reaction to a trend that has seen databases triple in size every two years. While processor speed and disk storage capacity have grown to keep up, disk throughput has not. In traditional data warehouse configurations, this creates a bottleneck between storage and database servers that limits query performance. ”Our customers’ disk storage systems can store 10, 100, 200 terabytes of data, but they can’t move that data off the disks and into the database servers fast enough,“ says Ellison.

With the introduction of the HP Oracle Database Machine and its key component, the HP Oracle Exadata Storage Server, Oracle has engineered a way to overcome this bottleneck.

The HP Oracle Database Machine is a high-performance system that includes a grid of eight database servers featuring 64 Intel processor cores, and Oracle Enterprise Linux. The machine also includes a storage grid of 14 HP Oracle Exadata Storage Servers with 112 processor cores of its own. The storage grid provides up to 168 terabytes of raw storage and delivers 14GB-per-second data bandwidth between the storage servers and the database servers.

Intelligent Storage Servers

A traditional database query that uses a large table scan moves all data blocks from the storage array to the database servers for processing. Moving data off storage disks, through a network, and into database servers can be a slow process. At large data volumes, the process creates a bottleneck that slows query performance. There are two ways to solve this problem, explains Ellison: ”One is to reduce the amount of data moving from the storage to the servers; the other is to increase the bandwidth between them. The HP Oracle Exadata Storage Server does both.“

First, the HP Oracle Exadata Storage Server reduces the amount of data traveling between storage servers and database servers by doing much of the query processing in the storage layer. Each server includes two Intel quad-core processors and runs the Oracle Database fast parallel query software. ”We’ve taken capabilities you would normally find in the database server and moved them into the storage server right next to the storage disk drive,“ says Ellison. The HP Oracle Exadata Storage Server ”doesn’t pass disk blocks back to the database server; it passes only query results,“ he adds.

”Next, we added wider pipes and more of them,“ says Ellison. Each HP Oracle Exadata Storage Server includes two InfiniBand connectors, which allow the storage server to move 1GB of data per second between storage and servers.

Scale-Out Architecture

The HP Oracle Database Machine provides a fault-tolerant, scalable architecture based on Oracle Grid technology. The database servers are preconfigured with Oracle Database 11g and Oracle Real Application Clusters running on Oracle Enterprise Linux. The HP Oracle Exadata Storage Servers are preloaded with Oracle Automatic Storage Management for dynamic load balancing and data mirroring. Because each storage server comes with its own processing, cache, and network bandwidth, customers can add storage while keeping the data warehouse in balance. ”With this architecture, as your database gets larger your performance stays the same,“ says Ellison. 

Next Steps

LEARN more about
The HP Oracle Database Machine
The HP Oracle Exadata Storage Server

In almost a year of beta testing the HP Oracle Database Machine, Oracle and its customers in the finance, retail, and telecommunication industries experienced performance improvements of 10 times and more over their current data warehouses. ”Beta testers have experienced astonishing results,“ says Ellison.

The HP Oracle Database Machine is a complete system preinstalled and preconfigured to deliver extremely fast query performance for business intelligence and data warehousing applications. ”If you don’t want to configure your own grid, your own interconnect, and your own hardware and software,“ says Ellison, ”we’ve got it all in a package for you.“

Jeff Erickson is a senior editor with Oracle Publishing.

Send us your comments