SAS achieves this by using dual 8-core commodity blades with 128GB of RAM and two 600GB hard drives.
At least eight of these blades are used, with up to 16 in a chassis, and up to three chassis in a rack.
So a fully populated rack has 1536 cores, over 6TB of RAM, and over 57TB of disk space - and costs around $US500,000. Bigger systems are possible.
Mr Schabenberger presented a scenario where an $11 million marketing budget was to be allocated to reach 20 million customers with certain constraints (eg, no customer was to receive more than two offers, if two offers were made to the same customer they had to be of different types and delivered by different channels, and the total number of offers of each type was limited).
What would have taken six or seven hours to process on a conventional system was done in around 90 seconds, giving a $US19.2 million outcome.
But the results showed that the constraints on two of the offer types were limiting the expected return, and allowing more of those and fewer of others increased the expected outcome to $US19.5 million.
Running the analysis a second time only took another 90 seconds or so, and generated an outcome that would go a long way to paying for the analytics system, he pointed out.