Celoxica, a company which was spun out of Oxford University in 1996, is applying its accelerator technology, which combines hardware, firmware, APIs and services to reducing the latency in market data. Lee Staines, the company’s CEO, claims the solution can capture market data faster than any alternative available today.

“That means that our clients can react to changing market conditions more quickly and trade faster than their competitors. Early performance testing has shown that our solution can deliver under 2 microsecond latency and extremely high bandwidth — over 10 million messages per second.”

The company uses field programmable gate arrays (FPGAs) that can fit onto a board with Intel processors without requiring more space or power in a data center.

“We use the hardware to accelerate specific functionality that lends itself to these techniques . We also offer the distinct advantage that our boards can be re-purposed so that they are adaptable and extensible to meet changing requirements.”

That’s important because one key complaint has been that FPGAs are difficult to program and relatively inflexible. If the gains are high enough, there will be no barriers in this area, which was born out in April by several participants at the Intel fasterMARKET DATA meeting Intel organized in London, who said they expect hardware solutions such as FPGAs to play a role in the mix going forward.

Commenting on this, Nigel Woodward who chaired the London event, “increasingly we are seeing high speed infrastructures for trading being designed and engineered for optimized performance. Given a clean sheet of paper – rarely available – a mix of technologies could be deployed. Accelerators for particularly high workloads acting as offload engines to mainstream CPUs supporting the firms key trading, risk and analytics applications. We have to work closely with our partners such as Celoxica, Xilnx, Nallatech Xtreme Data for example to achieve this. Until recently it was accelerators or CPUs, looking forward its now an architected combination.

Offload of specific functions to accelerators can also help the power consumption issue.
Lee Staines of Celoxia is placing some new bets, targeting the company at the market data area – leveraging experience on tap from exchanges and market service providers.

Adding to Woodward’s point regarding architecture – using tools such as Intel’s V - Tune, mainstream applications are now being tuned and optimized, which closes the gap massively on some early distorted comparims between the performance of respective emerging and mainstream technologies.

(No Ratings Yet)

 Loading ...

Filed under: Issue 5 - Summer 08, Technology, Trading