WX2 produces the world’s most powerful database server
How long does it take to read 23 billion rows of customer data? Getting one’s head around being able to read 23 billion of anything takes some concentration. Realising that it can be done in only 1 second halts the thinking process in favour of jaw dropping astonishment. Nonetheless, that is the performance that Kognitio, application vendors of the Massively Parallel Processing WX2 database platform, have achieved during trials with their blade server manufacturing partners, Hewlett Packard.
To put the performance of this multi-terabyte, in memory, data search into context it would be like being able to read a stack of printed A4 paper that stood 60 kilometres high in 1 second. Even when considering that, using the WX2database, an individual processor was reading a 240 metre high stack of paper – That’s still very fast reading.
To be fair the test was completed using a simple scanning query matching only one piece of data against each of the 23 billion rows. The test team then considered something more realistic, introducing a more complex query that demanded the production of temporary results – that was bound to affect performance. They were right, it did – but the WX2database environment brought back the results of these more complex queries in just 7 seconds.
The trials run during October 2005 at Hewlett Packard’s European test centre in Boblingen, were designed not only to test the performance of the database environment but also it’s scalability. Starting from around 25 HP Proliant BL25p blades, the test team simply kept adding servers and processors until they figured that at 125 and 250 respectively, they had already built Europe’s largest single database server and needed to go no further.
With all 250 processors acting as a single server using WX2 Massively Parallel Processing capability, the task of reading 23 billion records was simply divided up and made more manageable.
For Kognitio and HP’s clients however, with ever increasing demands for data analysis, this is proof that they can continue to scale up database environments without expensive step changes in storage and processing capacity. Blade servers have a relatively low cost compared to the proprietary hardware platforms associated with Massively Parallel Processing. This means expansion costs can be smoothed in line with demand.
The performance was achieved using 250 2.6 GHz AMD Opteron Processors on 125 HP Proliant BL25p blades. This provided 18 Terabytes of disk storage and 2 Terabytes of system memory for Kognitio’s WX2 software to perform its in memory data search.
For further information on the performance of Kognitio’s WX2 and its relations with Hewlett Packard Proliant blades please contact Roger Gaskell at Kognitio on +44(0)1344 788628.