S&P Real-Time Debuts FPGA OPRA Feed

Officials say S&P Capital IQ developed the FPGA OPRA feed after realizing that projected volume growth on the OPRA feed would quickly outstrip the ability of software-only feed handler and book-building solutions to process the amount of options data being generated─which is expected to reach peak rates of 100 million messages per second (mps) within five years, and hit 120 million mps by 2020.

QuantFeed FPGA for OPRA can process 40 million mps per FPGA board─enough to handle July's projected rates of just under 23 million mps on a single FPGA─with sub-microsecond message decoding and sub-10 microsecond overall processing latency.

"The options trading market is becoming more and more intense every year, and is about to reach an inflection point," says Stephane Leroy, vice president and head of global real-time solutions at S&P Capital IQ. "We think that using a pure software solution for this kind of demand is not possible anymore, because these kinds of volumes would force firms to stack up a lot of servers.... So that's the reason we went down the FPGA route."

In comparison, Leroy says, most other FPGA solutions use FPGA processors for the "brute force" they can bring to data intake and decoding, but still use software to provide value-added functions such as reconstructing order books. Moving to hardware also allows the vendor to future-proof against the need to rewrite software and how it handles market data, and against the need for consumer firms to rewrite their internal applications that consume the data.

The vendor has already been using the FPGA solution in-house for processing OPRA data, and went live with an undisclosed number of quantitative trading clients in recent weeks, Leroy says.

"We have two business models: clients can use our own DMA (direct market access) infrastructure via our API, which provides access to the full universe of OPRA data. But you might not trade that whole universe, and might not want all that data, so we can switch to a dedicated solution where we can put the "white list" and blacklist onto the FPGA chip, and can run it wherever they want-either in our datacenter or in the client's own Rackspace," he says.

Target clients include all types and sizes of options trading firms needing to handle increasing volumes of OPRA data-not just firms that are highly latency-sensitive. "The challenge is not only to provide the data as fast as possible to automated trading applications-the challenge is to deal with the volume... so that we are able to satisfy clients and also keep up with the source," Leroy says.

The vendor will also extend its use of FPGA-supported feeds to other high-volume markets, having already rolled out FPGA feed support for Nasdaq OMX around a year ago, Leroy says, though he declines to identify upcoming projects.