EXCLUSIVE: ‘Just the ticket’ – Daniel Brown, Cisco; Milan Dvorak, Netcope and Adam Lister, Blackcore in ‘The Fintech Magazine’

Lightning quick tick-to-trade is a must-have for today’s trading platforms. Daniel Brown, Milan Dvorak and Adam Lister discuss the pursuit of speed and the minute margins that separate profit from loss

Time is money across all financial markets. But none more so than in the high-octane, high-speed world of online trading, where platforms must deliver ever-faster executions for traders whose success depends on speed and agility. So, how do you create a trading platform that gives you that all-important edge?

We asked three industry experts to explain what makes a great platform and what today’s traders look for.

Few organisations have better technology credentials than Cisco Systems, where Daniel Brown is a technical solutions architect with a strong focus on electronic trading. Many of his solutions are used by tier 1 banks, proprietary traders and hedge funds, and they range from ultra-low latency (ULL) platforms to global exchanges and multicast data distribution networks.

As part of the ULL division, Brown is responsible for SmartNICs – network interface cards that offload processing tasks normally handled by a central processing unit (CPU), accelerating traffic and boosting agility. SmartNICs are based on field-programmable gate array (FPGA) platforms – ideal for designs requiring complex logic combined with high-speed processing capability.

“A primary objective of a trading system is to take as much as possible away from the CPU,” says Brown. “With SmartNICs and FPGAs, we can send orders directly from the FPGA, without having to do any calls to the CPU. That’s one of the biggest advances we’re seeing today. “Traditionally, software apps were all running in the CPU,” adds Brown, “which meant quite a big delay (by which he means a matter of microseconds) between the application and the NIC card itself. We’ve done away with that through the technologies we’ve been developing at Cisco and other organisations.”

One of those organisations is Czech-based Netcope Technologies, which is also driving the use of FPGA technology in trading. Netcope CEO Milan Dvorak explains that it develops FPGA-based tick-to-trade applications. This lowers the latency of trading systems because the whole tick-to-trade pipeline is implemented in the FPGA.

Tick-to-trade is the time gap between receiving a market ‘tick’ (a price movement) and executing the buy or sell order (the trade). And time is vital – picoseconds, nanoseconds, microseconds: these are the fractions that make the difference between success or failure, profit or loss. Put simply, the lower the latency, the better your trading prospects. Dvorak agrees with Brown that the way to achieve low latency is to move out of the CPU space and into FPGAs.

“FPGAs are so fast,” says Dvorak, “that by the time the standard software even learns about a new message, a new exchange event, the FPGA can process the incoming data, execute the strategy, prepare the response to the exchange and send it onto the wire. If you have the full tick-to-trade pipeline in the FPGA, you’re always faster than the guys staying in software.”

However, Dvorak believes that CPUs are still important, because if you move everything to the FPGA, it will become clogged and unnecessarily slow. Sometimes you need to finetune on the fly, he says, and the lower the latency between the FPGA and the software, the more you can do and the more aggressive you can be with your strategies. Dvorak says the FPGA environment has proved its worth during spike periods and, notably, the unsettling market conditions during the pandemic.

“Today, when most exchanges are still running at 10 gigabits per second, there’s no buffering with FPGA. It’s just streamlined data processing and you know that your latency is going to be constant and deterministic, even during microbursts, or whatever is happening in the market.” Adam Lister expands on the latency challenge. As a system architect at Blackcore, a technology vendor for high-frequency trading, Lister is at the forefront of trading innovations, developing and deploying trading platforms and exchanges worldwide.

“While CPU core frequencies are extremely important, we also optimise everything around that CPU socket,” he says. “So we’re tuning the cash frequencies and we’re optimising the PCIe (peripheral component interconnect express) bus to ensure the whole solution is as low-latency as possible, not just what’s on the FPGA, or what’s in memory.”

Lister says the industry has changed in the last few years and that ‘raw clock frequency’ has levelled off. “We’ve had to optimise everything around the socket,” he says, “not just the core frequencies, and we’re seeing great results for latency reduction in areas that might not have been addressed before.”

It’s been challenging, especially with some clients wanting to integrate bigger and more power-hungry devices. Naturally, given environmental priorities, energy consumption is a key concern and must be balanced with the need for speed.

Blackcore has made a name for itself in liquid cooling systems for servers, which enable users to maintain more consistent temperatures across the system  and improve performance. If you have a stable platform for your FPGA and software, it’s a much better environment for trading, adds Lister. Heat transfer in liquids is also in orders of magnitude more efficient than air systems, leading to less energy consumption, even as the processors work harder.

“In recent customer testing, we had a 15°C reduction, moving from air-cooled to liquid-cooled, in the same datacentre, in the same rack unit, which enabled higher frequencies. Being cooler lifts you above the competition. The cooler you are, the faster you can run. And if you have workloads such as algorithmic trading and automated trading, you want those single threads to be as fast as possible,” says Lister.

Turning to industry collaboration, Dvorak says it’s encouraging the way organisations like Cisco, Netcope and Blackcore are working together to promote  innovation, to address this and other issues.

“The FPGA industry has been through a commoditisation phase,” he says, which improves time-to-market. “We’ve seen a big shift in the industry and the fact that Cisco offers firmware development kits (FDKs), which foster compatibility between platforms and systems, means we have flexibility and can quickly bring our intellectual property (IP) onto a new platform and push it to clients sooner. We’ve just deployed a new card that will be at least10-to-15 per cent faster and clients won’t have to wait for us to finish development and move our IP onto the new platform, because the work’s done.”

Brown agrees that a big advantage of a SmartNIC is futureproofing because its FPGA can be rewritten, whereas with an application-specific integrated circuit  ASIC), changes are complicated and costly. Being efficient and economical is a major consideration for all CTOs, says Dvorak. As an example, he cites Netcope’s methodology.

“We use a technology called high-level synthesis,” he says. “It can make the programming of FPGAs faster and more efficient, so a single developer can do more work and put more code into the FPGA. It can also grow your FPGA team, because you can add software people to the mix.”

One thing Brown, Lister and Dvorak firmly agree on is the power partnership. Creating a best-in-class trading system requires teamwork and cross-fertilisation, with each vendor or supplier contributing skills and services that complement each other and make the system greater than the sum of its parts.

“Cisco has traditionally been the best in class,” says Brown, “in terms of network performance and bandwidth. And now we have the SmartNIC line-up. But, in partnership with vendors such as Blackcore, for their servers, and Netcope, with their tick-to-trade and gateways, you can go to a different level. So I think the future is definitely about partnerships.”

“At Blackcore, collaborating and integrating with partners such as Cisco and Netcope is certainly the way forward,” says Lister. “We can deploy a whole trading solution, rather than being just one building block. If you take a low-latency FPGA and install it in our server, you’ve added Netcope in the background and created a turnkey solution.”

“The bottom line,” says Dvorak, “is for everyone to focus on what they do best, and that applies equally to clients. Let vendors provide their know-how and technology in partnership, and let traders concentrate on making money trading on the exchanges.”

This article was published in The Fintech Magazine #21, Page 107-108