Electronic trading operations must seize opportunities lasting only fractions of a second. The winners and losers are determined by how fast a trader can digest a market feed and place orders.
That means reducing tick-to-trade – the time interval between receiving a market tick showing opportunity to an algorithm and the buy/sell order being sent – is the holy grail of high-frequency trading (HFT).
 |
| Alex Viall, Behavox |
In a recent interview with Euromoney, Alex Viall, head of regulatory intelligence at compliance software firm Behavox, spoke about the capacity for operational issues – such as delays to the dissemination of pricing movements caused by latency on the network – to impact trading efficiency. The solution developed by Solarflare and high-performance FPGA-based platforms specialist LDA Technologies runs on a network interface card (NIC) from Solarflare that manages multiple programmable circuits.
FPGAs, or field-programmable gate arrays, are semiconductor devices that can be reprogrammed for specific applications or requirements after manufacturing.
The firms say the new solution has cut latency from 250 nanoseconds (ns) to 120ns.
NICs have been used to connect trading application servers to ethernet networks for years: Rival Systems, which develops technology solutions for front-end trading platforms, has been using Solarflare’s NIC since 2015.
The reduced latency that Solarflare and LDA Technologies now offer has been achieved by adding to the processing power of the NICs by implementing more efficient software.
“With any first-in, first-out matching engine there is an obvious advantage to being at the top of the queue,” says Rob D’Arco, CEO of Rival Systems.
“Any time our clients possess a two-times faster tick-to-trade time, they are trading with a significant competitive advantage.”
The obvious target market for the technology is those on the cutting edge of electronic trading where latency matters most, notably the high-profile HFT shops that are often viewed as market makers.
 |
| Ahmet Houssein, Solarflare |
Ahmet Houssein, Solarflare’s vice-president of strategic development, acknowledges it is impossible to quantify how much more likely trades are to be executed when latency is 120ns rather than 250ns. “However, the knowledge that orders are typically processed in the order in which they arrive means that someone trading in 120ns will arrive 130ns ahead of someone trading in 250ns, assuming they both received the same market signal at the same time, which is often the case,” he says.
Jason Mochine, commercial director at Fixnetix, reckons further reductions in latency are technically achievable and has referred to getting risk and credit close to the installation to further reduce proposition latency as a key element of an FX co-location service.
Providers have claimed that co-location reduces execution latency to such a level as to create an almost-level playing field for market participants.
The technology developed by Solarflare is not designed to replace co-location as a means of reducing latency, since being as close to an exchange as possible will always afford the fastest response time.
However, when asked how much lower tick-to-trade latency can go, Houssein suggests that middleware launched by Solarflare last year has moved it closer to the limits imposed by physics by reducing the time trading solutions based on programmable circuits have to spend managing the software that handles communications between network devices.
And according to Solarflare, the basic implementation costs are relatively modest, at below $10,000 for a client with a server platform.
