From A-Team Group’s Market Data Insight, December 2006: Benchmarking tests suggest that the new quad-core Intel Xeon processing 5300 series chips launched by the vendor in considerable style at the so-called ‘Gherkin’ building in London late last month will facilitate vast latency reductions in market data delivery.
As the panel discussion during the event explored, though, roundtrip latency is determined by a number of interdependent factors. Moreover, ongoing technology improvements will be required to keep up with the ever-intensifying drive for lower latency as the shift to algorithmic trading continues and external influences like MiFID begin to bite. Intel’s strategy to encourage the uptake of its quad-core chips – which deliver four computing ‘brains’ inside a single microprocessor to enable better performance while consuming less power – has been to work closely with partners, including market data vendors, to ensure they are ready to take advantage of the new technology and to prove its capabilities in real-world scenarios through benchmarking tests.The results are impressive. For example, in an evaluation of the Linux version of the Reuters Market Data System (RMDS) running with Red Hat Enterprise Linux 4.4 on a two-socket HP ProLiant server with dual-core Intel Xeon processors, RMDS 6.0 surpassed previous two-way benchmarks and set a record of 4 million updates per second with a multi-core optimized server configuration.
Speaking on the panel, Peter Moss, global head of enterprise solutions at Reuters, emphasised the growing complexity of its clients’ requirements, and the need to accommodate the requirement for both low latency and breadth. “I don’t see this complexity disappearing – rather it will increase under MiFID,” he said. The onus is on providers such as Reuters to enable access to direct feeds alongside consolidated feeds, to meet the need for “a mix of some low latency requirements and some breadth requirements” – and that “will be pretty standard going forward”, he said.
P.J. Di Giammarino, CEO of JWG-IT, predicted that one of the key impacts of MiFID would be to force institutions to improve market data handling and latency. “The arrival of new execution venues will drive a tremendous uptick in market data volumes, and this in itself will lead to fine-tuning of systems,” he said.
Lehman Brothers’ head of direct access trading sales, James Watson, pointed out that the extent of the demand for low latency “has to be put into context, and depends on what the client’s trying to do.” He continued: “For stat/arb strategies, the most important thing is low latency, and roundtrip latency is a vital factor. For algorithms, particularly the more sophisticated ones, latency becomes less of an issue and there is a cross-over into computational needs. It depends on the client.”
Asked whether the drive for lower latency would ever end, he said: “We are looking at 130-135 milliseconds now for the roundtrip, and the next dimension or step will be 125, and there will certainly be a fight to get there. There is also an interest in having the computational ability to make algorithms deeper, and to bring several algorithms together.”
The importance of optimising all the components within the technology stack to ensure lowest possible latency was highlighted by Ray Mulligan, global head of FX IT at Credit Suisse. “Although the performance of the hardware is very important,” he said, “while a hardware vendor can make things four times faster, a software vendor can easily make them 100 times slower.”
Indeed, the industry is facing a significant challenge as it seeks to get the right balance of software and hardware as competing suppliers continue apace to make superior claims on the low-latency performance of their products.