A few hours ago, Larry did his keynote here at Oracle Open World 2008. The big announcement that had loomed over the conference has been made. Oracle – in joint partnership with HP – introduces the world’s fastest hardware for running databases and especially data warehouses: the Exadata Storage Server. Proud as father who shows his new born child to the world for the very first time, Larry was positively beaming – extremely pleased with the announcement and the achievement. And the fact that he could throw punches at different competitors this year – not Microsoft – but companies like Teradata, Netezza and to some extent other hardware vendors like IBM and NetApp.
The concept of the Exadata sounds pretty simple:
with data warehouses growing far beyond the 1 Tb limit, queries simply run for too long. A table scan can take up to 30 minutes, which is just not acceptable. The problem is for a large part that the volume of data retrieved from disk and pushed to the database server for processing is so huge that the pipes from storage to database server cannot handle it timely. Note that today’s databases process blocks of (raw) data retrieved from disk – even though most of the data may not even be required for the query at hand.
Exadata alleviates this problem in two ways: it uses much wider pipes for streaming data (20 Gb/s InfiniBand connections that in actual Exadata practice will process about 1 Gb/s as it is limited by the speed of the disks. And much more importantly: it puts intelligence into the storage server. With every disk comes a multinode processor that performs Oracle Parrallel Queries. This processor receives the instruction to retrieve a certain result set (not raw data blocks). The volume of data being returned through the much wider pipes is actually much smaller, as only the requested data from the already filtered records needs to be returned. The database server combines the resultsets from the various nodes in the Exadata Storage Server and performs any aggregations.
The HP Oracle Database Machine consists of a grid of Oracle Database
Servers and a grid of new OracleÂ® Exadata Storage Servers packaged in a
single rack ordered as a complete system from Oracle. No changes are required to existing queries or business intelligence
applications to deliver extreme performance for large Oracle data
warehouses. The first customers – pretty serious corporations such as Amazon, Yahoo, British Telecom, ABSA, Giant Eagle – started testing just under a year ago. On average, their queries were sped up some 30 times. Queries that ran for 30 minutes were back to under 1 minute. The Exadata Database Machine had been up against some serious IBM, Teradata and NetApp infrastructures, and beat them hands down.
HP Oracle Database Machine is pre-configured for performance, pre-
tuned, and certified for Oracle Business Intelligence Enterprise
Edition tools and Oracle Real Application Clusters. Complete
configurations can be ordered from Oracle, with hardware support by HP.
The HP Oracle Database Machine is a high-performance system configured
for data warehousing that includes a grid of eight database servers
featuring: 64 Intel processor cores, and Oracle Enterprise Linux; and a
grid of 14 Oracle Exadata Storage Servers that include up to 168
terabytes of raw storage and 14 GB/sec data bandwidth to the database
servers. HP Oracle Exadata Storage Servers are key performance enablers for the
database machine and can be ordered separately if customers have an
existing data warehouse and merely require the storage enhancements.
Customers can build data warehousing solutions using HP Oracle Exadata
Storage Servers, which feature industry-standard components including
two Intel processors, up to 12 TB of raw storage and InfiniBand
connectivity delivering 1 GB/sec of data bandwidth per storage server.
Oracle sells the machines and provides system support; delivery, manufacturing and on-site hardware support comes from HP. The Exadata server is the result of a three year joint effort with HP. It sells at interesting rates:
It is hard to read but it says: $650.000 for the hardware + $4000 per TB of data + $1.680.000 for the software licences – database licences for 14 nodes. The total prices comes at under $14k per Tb and apparently that is much cheaper than Teradata and Nietezza and has much better performance results as well.