Nasdaq Tackles Bursts with FPGA TotalView Feed

Nasdaq OMX has rolled out a hardware-enabled version of its TotalView-ITCH full-depth equities datafeed, designed to make latency more deterministic for latency-sensitive firms during peak message volumes.

The exchange launched TotalView-ITCH FPGA (field-programmable gate array) at the start of August, and has rolled out the feed to more than a dozen trading firms among agency brokers, algorithmic brokers and buy-side firms, many of whom had been involved in the development and testing at an early stage, says Brian Hyndman, senior vice president of global data products at Nasdaq. “We have been in communication with all of our clients along the way, and some of those were also beta testers,” he says. In fact, Nasdaq envisages that the total addressable market for this service may only be around 25 firms that are sufficiently sensitive to need this level of determinism.

According to Hyndman, the exchange’s traditional, software-based feeds would sometimes see spikes in queuing—which introduces additional latency—around important market events or around the release of market-moving news. “The outliers have always been there, and have been an issue,” he says. By capturing the data and generating the feeds using FPGA processors, which can handle greater volumes of data without introducing queuing or latency, traders can have greater confidence in the timeliness of data around market events that create microbursts of quote and trade message volumes, he adds. Clients simply connect to the feed from Nasdaq’s datacenter in the same way as for any other feed.

As well as greater determinism around latency, the FPGA feed offers other benefits both for clients and Nasdaq, such as making it easier for the exchange to create customized versions of TotalView for individual client needs—for example, by only including a subset of symbols to ease client-site processing requirements—and to, over time, reduce the technology footprint required to support Nasdaq’s internal feed distribution infrastructure, Hyndman says.

In addition, Nasdaq can apply its new FPGA expertise to other suitable feeds in future. “Some of the work you can port over to other feeds, though in some cases, you would have to do fresh development for each feed. We haven’t definitively identified what the next feed to use this will be … [but] there are plenty of potential applications for this FPGA technology,” Hyndman says. Over the next quarter, the exchange will narrow down which feed to apply FPGAs to next, before rolling out hardware-enabled versions in Q1 of next year, he adds. “It’s fair to say that a lot of our feeds would work well with FPGA capabilities… but I wouldn’t go so far as to say that we will eliminate the software versions of all our feeds in future.”

Development of the FPGA feed took around 15 months, and was performed mostly in-house by existing staff and new hires, with the assistance of an unnamed development partner and an unnamed FPGA supplier, Hyndman says. “It’s not new technology, but it is still relatively new in the financial world, so we had to bring in some outside expertise… [to complement] some in-house staff who had dabbled with FPGAs in previous jobs,” he says.

Nasdaq will charge $25,000 per month for trading firms using the FPGA feed internally, and $2,500 for vendors to redistribute the feed—both in addition to the existing fees for TotalView data, which can cost up to $100,000 per month for an enterprise license for professional use.