Dave Vellante and John Furrier have set up theCUBE in Palo Alto for Hewelett-Packard’s Project Moonshot Press Event. As a guest in the Cube, Niall Dalton, the Director of High Frequency Trade at Cantor Fitzgerald, came in to speak about his company’s outlook on HP’s new Project Moonshot ARM low-power processors and the Calxeda module.

Dalton says that Cantor Fitzgerald is evaluating the HP technology. He mentions that usually high frequency trade looks to lowest latency and the fastest machine possible (in order to keep up with trading metrics and moving units); however, that there’s another angle to trading that involves pushing big data and that’s where the new ARM processors would fit in.

High Frequency Trade is locked heavily into real-time, but analysis and memory is extremely important. Not only do they need something that can deal with extremely low latency; but they also archive a great deal of data. After the days are done and they want to look back at what happened and understand why, the solution shifts from latency to throughput. An entire year of trading data represents a gigantic set and that’s where Big Data comes into play for simulations and what-if scenarios.

Cantor Fitzgerald cares about if they can be cost effective about analyzing their big data at scale. Anything that allows them to lower power usage while raising throughput can save them a great deal of money.

Dalton mentioned that Facebook probably faces extremely similar challenges in analyzing their data offline, after-the-fact, partitioning off their analytics and running big data. They too have giant data centers and look to both low latency real-time applications; but also do a great deal of analysis.

He says that Cantor Fitzgerald hasn’t just yet moved to ARM, although it’s obvious that when you’re looking at low-power with high throughput. However, they are currently looking into them as a very likely solution. They’ve done some experiments and spoken with Calxeda about building a server-class chip. Current ARM processors don’t have the I/O capacity for high bandwidth like Calxeda are seeking to build into the chips.

Power, energy, cooling, and infrastructure may represent almost one-third of the total cost of supporting processors—so performance and efficiency represents a huge windfall for every watt cut from the total power overhead.