Latency Top Priority, but Firms Cool on Proximity Hosting

However, firms currently find themselves having to make a trade-off between speed and functionality, according to James Watson, executive director at Lehman Brothers. Comparing trading systems to cars, Watson suggested that the ideal vehicle for his firm’s trading would combine both speed and comfort, whereas now, “the lower the [latency], the more you have to strip out the functionality.”

User firms are still examining a range of alternatives to solve latency issues, including physically locating trading hardware closer to exchanges’ matching technology—known as proximity hosting—to reduce data latency. “If there’s demand for people to put their servers into our datacenter... that’s something we would look at,” said David Lester, chief information officer at the London Stock Exchange.

However, a straw poll revealed that only 15 percent of audience members thought that proximity hosting was very important, with 45 percent responding that it was not at all important to their trading strategies.

But Reuters officials say this response may be misleading because the user firms present were not of the size that would benefit most from proximity hosting. “If you are sitting in London and you want to trade in New York, you want to put the application in New York,” said Peter Moss, global head of enterprise solutions at Reuters. “The reason why I think a lot of people in the audience don’t care [about proximity hosting] is because a lot of the audience are from large firms, which by their very nature are present in all of these [global] markets and therefore don’t have to worry about proximity hosting because they actually have branches in those centers,” Moss said.

Panelists also viewed hardware as an important element in reducing latency, suggesting that multi-core processor technology can produce significant benefits, though this can require software re-engineering to adopt the parallel processing techniques that make full use of the extra processing power available.

When asked about hard-wiring data processing functions into silicon, Moss said, “There’s no doubt that hardware acceleration will have a part to play.” But he was unconvinced of the argument for burning the capabilities of Reuters’ RMDS data platform onto a chip, as low-latency competitors are doing with their systems, preferring to strike a “middle ground” that would comprise an optimized combination of both hardware and software.

However, one of the largest contributors to latency is exchange matching, which is beyond users’ control, the panel said. “The biggest issue with regard to latency… is the exchange,” Watson said. “But the good news is that [the latency of exchanges] is the same for everybody.”

Jean-Paul Carbonnier