You may have stumbled on the news that Oracle is now in the hardware business or to be more precise, it is in the database engine business – and by database engine I’m talking about hardware specifically built for running big database applications – and in respect of big databases applications, I mean big data warehouses.

On Wednesday Oracle CEO, Larry Ellison, unveiled the company’s Exadata Storage Server and a Database Machine – shown in the adjacent illustration, with the words Extreme Performance written down the side in red. The hardware is made by HP and you can think of this machine as being an implementation of the Oracle 11g database implemented over Oracle RAC with a complete Oracle software stack, plus Oracle Enterprise Linux.

From a software perspective much of this is familiar territory. The smart part is that the hardware has been designed for lightning query performance. Oracle claims that the HP Oracle Database Machine will run queries 10x faster or more. You can think of the whole configuration as having two parts; up to 8 HP Database Servers running Oracle 11g connected to 14 Exadata Storage Servers. The Exadata Storage Servers marry Intel multi-core processors with blocks of memory to specific disk resources, so that query processing for each disk happens “over the disk”. That’s where the performance comes from.

Does The World Need This?

It’s a logical question to ask. Database engines have been tried before (remember Britton Lee), but the only one that saw much success was Teradata. Other ideas like ICL’s CAFS (Content Addressable File Store) delivered the performance. But performance is never the problem with devices like this, it’s whether the overall architecture has longevity.

The fact that this is Oracle makes a big difference of course. The database giant has a right to try to move the industry along a different path – and I’m sure that this machine will see some quick adoption. On the HP side of the equation some commentators may wonder whether there isnt a product clash here, with HP also offering its excellent Neoview – based on the Tandem architecture. But first of all, HP is simply providing the iron, it is not selling the database machine. Secondly, Neoview performs best when dealing with mixed workloads whereas the Database Engine specifically targets multi-terabyte data warehouses. It is not a head-to-head clash although they will doubtless meet in the market place.

A more interesting question is how the HP Oracle Database machine stacks up against Vertica. (See this posting for details of Vertica). My immediate impression is that Vertica is both faster and cheaper, but that’s just an impression – we’ll get an accurate knowledge of this once there have been some real world bake-offs. Technically though, I suspect that compressing the data and pushing as much of it into memory as possible, as Vertica does, will turn out to be a better long term solution for performance.

The Need

One thing that is beyond dispute is the need for better database technology. Data volumes keep on doubling every year and companies keep on wanting to analyze that data. We are now entering the era of petabyte databases and there’s no reason to believe that technology built for the gigabyte era is going to be adequate.