NovaSparks Cuts Book-Building Latency FPGA Ticker Plant

NovaSparks will make its new architecture available for Nasdaq's TotalView ITCH feed next month, and will make it available for other order-based feeds─-as opposed to price- or level-based feeds that already include book-building features─supported by its NovaTick ticker plant in January.

"One of the most complex elements of our architecture is our book-building engine," says Olivier Baetz, chief operating officer at NovaSparks. "We knew for a while that there were a lot of improvements we could make to our firmware. But the biggest gain we identified was in the book-building engine.

Baetz says the vendor began redesigning the book-building architecture "from scratch" a year ago. "We optimized code at each step to be more efficient, and we moved some of the processing out of the critical path," he says. "To process data, you need to do A, B, C and D, each of which introduces latency. But if you can remove C or move it elsewhere, you can reduce that overall latency." However, Baetz declines to specify which processes the vendor was able to move out of the "critical path" of low-latency data.

For example, the 300-nanosecond reduction means that firms using 10 Gigabit Ethernet multicast distribution wire-to-wire will see latency fall from 1,100ns to 800ns. For those using the vendor's NovaLink service, which connects the NovaTick FPGA ticker plant to clients' own FPGA applications, latency will decline from 1,050 ns to 750 ns, and for those using PCIe direct memory access, latency will fall from 1,550 ns to 1,250 ns.

"Three hundred nanoseconds means different things to different customers. It depends on what kinds of strategies they're engaged in. For some, it will mean the difference between making money or not. For others, it will mean making money versus making more money," Baetz says. "Typically, our customers are improving their infrastructure on a constant basis─switches, servers, and telecoms─so it's important for us to ensure we are a part of that constant improvement."

In addition, the new architecture has allowed the vendor to increase its usage of QDR (Quad Data Rate) memory for book-building, instead of relying on limited internal memory, enabling it to increase the memory capacity of its ticker plant appliance and support more instruments per FPGA.

As a result, early next year, NovaSparks will be able to increase the density of instruments supported on each FPGA─of which there are four per appliance─from 9,000 at present to six times as much, making it easier to support the higher numbers of instruments in options markets than in equities markets.