Tibco Unveils Sub-Microsecond Messaging

Tibco will this week release a new ultra-low-latency messaging system, designed from the ground up to leverage new multi-core servers in order to meet the nanosecond-level latency requirements of high-frequency traders for their internal data distribution.

The vendor has tightly integrated the new platform, dubbed Tibco FTL, with its Rendezvous middleware product, so firms with current RV deployments in their less latency-sensitive middle and back-office areas will be able to standardize on Tibco messaging across their enterprise by rolling out Tibco FTL to support the low-latency requirements of their front-office trading operations as well, says Rourke McNamara, senior director of global product marketing at Tibco.

“[Tibco FTL was] built specifically to run on the massively multi-core Intel-based servers, to run on a co-located style environment and to meet the needs of folks who really need something that gets their message there within a microsecond,” McNamara says. For communication within the same physical hardware using shared memory, Tibco FTL delivers intra-host latency of 384 nanoseconds, while transmitting data between two servers using RDMA (remote direct memory access) over InfiniBand yields latency of 3.1 microseconds, he says.

McNamara says that while the product was developed entirely by Tibco, the vendor worked closely with Intel to optimize FTL for Intel’s 64-core processors, including its Xeon 5600 and 7500 series—for example, to allow the processing load to be spread over all 64 cores—which included writing a large portion of Tibco FTL’s code in the Assembly programming language, which uses Intel syntax.

By doing so, the vendor has been able to reduce latency without stripping out functionality such as metadata support and content-based addressing, which enables users to route messages based on information contained within the message, as opposed to subject-based addressing which relies on a single data field, McNamara says.

FTL is targeted at hedge funds, high-frequency trading firms and algorithmic traders that co-locate in exchange datacenters to achieve the lowest possible latency, as well as the exchanges and execution venues themselves.

In fact, more than two dozen firms—mainly hedge funds in New York and Chicago, but also firms in London, Paris, Tokyo and Hong Kong—have been beta testing Tibco FTL since last summer, while two unnamed hedge funds in New York and London have already signed on to purchase the technology, McNamara says.

In particular, the vendor expects the system to appeal to firms with homegrown middleware solutions, since these are typically the most concerned with being as fast as possible. “They get to focus on what’s unique to them—the stuff that gives them their value proposition and advantage in the marketplace—rather than focusing on the underlying plumbing, which is really what our technology does,” he says.