With Wall Street betting its future on high-speed trading, yesterday Tibco Software announced an extreme-low latency messaging technology that is targeted at capital markets chasing the speed of light.
Built from the ground up, Tibco FTL—whose internal code name was “Faster Than Light”— is meant for electronic trading of high performance algorithms on massively multi-core machines. The new messaging technology is optimized for intra-host – applications —passing messages on the same machine or inter-host communications — applications passing messages between different machines.
“The trend we’ve seen is for trading firms to collapse the algorithm into a single machine,” said Rourke McNamara, senior director of product marketing in an interview with Advanced Trading in New York. With intra host communications, “The idea is you’re passing messages between different applications on the same machine,” explained McNamara. Based on Tibco’s internal testing, Tibco FTL had an average application latency of 384 nanoseconds for intra-host communications, whereas its closest competitor had results between 600 and 700 nanoseconds, noted McNamara.
When optimized for inter-host communications for RDMA over Infiniband, Tibco FTL’s latency was 3.1 microseconds, a little over 40 percent faster than its nearest competition, Informatica, which has claimed that its inter-host latency is 5.2 microseconds, according to McNamara.
While previous advances were incremental gains, McNamara said the latency reduction achieved with Tibco FTL was a significant margin. “We feel there is still a bit of optimization we can do” to get the latency below 1 microsecond, he said.
Rourke said that FPL was designed to live on top of massively multi-core Intel machines and that it was also designed based on Tibco’s 24-year expertise in the low-latency middleware messaging space.
“Most applications don’t need to go that fast, but obviously in high frequency trading they do,” observed Roy Schulte, distinguished analyst and VP, Gartner who follows the middleware software sector. “If you want to run under one microsecond, or one millionth of a second, then you’re talking about messaging within a computer, where the messages are in memory and you’re just sending them around from one section of the computer to another section of the computer,” said Schulte.