As the London Stock Exchange today postponed the launch of a Linux based system billed as breaking speed records, other trading venues – including in New York, Singapore, and elsewhere in the UK – expressed their desire to deliver faster technology.

The new LSE environment was developed in C++, and runs an Oracle database linked in to a customised version of Linux for its matching engine. Its network uses Cisco and Juniper switches.

A fortnight ago, the LSE announced that its trades were being conducted at an average of 126 microseconds – a figure largely accepted in the industry as a record for live use.

But only five days after its announcement the LSE faced a potential contest when the New York-based NASDAQ exchange, whose Genium INET platform runs on Linux in a reported C++ environment, said it had delivered an average 97 microsecond latency during a week in mid-October. In its Nordic markets, it said it was “capable” of continuing to deliver sub-100 microsecond latency on average.

Importantly, however, NASDAQ is understood to quote 250 microseconds to clients because of peak trading loads at the start and end of trading days, a challenge for all markets.

The Singapore Stock Exchange, which is buying in the INET platform from NASDAQ, has said in recent months it achieved 90 microsecond times in benchmark testing, on Voltaire switches and HP ProLiant blade servers. That system will go live next year and then accurate live measurements can be taken.


Other reported speeds at major venues include 250 microseconds for BATS and NYSE Euronext, and 350 for Chi-X.

The speed battle between exchanges has increasingly been termed a ‘technology arms race’. Last month, the LSE vowed to continue improving speeds, saying that it intends to “compete, and compete strongly”. NASDAQ said it is “committed to innovation through technology” to be the “driving force” among exchanges.

The Singapore Stock Exchange claimed instead that it was the venue “raising the bar for speed, capacity and scalability”.

But much faster speeds are reportedly available in one new system not yet running on the large exchanges. Algo Technologies, founded by Hirander Misra, former chief operating officer at LSE rival Chi-X, claims it can deliver 16 microseconds latency, a tenth of current LSE speeds. It also uses stripped down versions of Linux, crucially running on high-spec blade servers.

That speed was benchmarked by independent firm Corvil, but was measured in test conditions rather than a live environment. Algo Technologies was established this year, and its first major matching engine implementation will be in the UK’s Plus Markets venue over the next 12 months. It claims that the testing was conducted at volumes of messaging that will exceed even the heaviest loads exchanges tend to experience.

The standard process to calculate latency is to measure the time needed to accept, process, and acknowledge or fill an order. Nevertheless, after the LSE claimed it had the fastest speed, a number of readers questioned the claims, because of measurement transparency concerns.

“Hooray for Linux,” wrote one. “I think Linux is a good choice for truly reliable systems,” wrote another.

But in one comment, the writer said: “When numbers are coming from the venues they should not be taken seriously. They are just a number without a context.” Many of the exchanges guided on the figures for latency, but declined to provide much technological detail.

There was also a split opinion on what technology is really responsible for the speed. Hirander Misra, chief executive at Algo Technologies, said the customisation of systems is key to delivering speed. “Most exchanges using Linux have stripped it down to a bare bones version, designed to run fast under these conditions,” he said.

Another important factor in speed is the development environment. Observers largely agreed that over recent decades, most exchanges had effectively gone full-circle, from C environments to Java and back to C++.

The hardware is vital to performance, too. “Most of the traditional exchanges still run huge boxes in datacentres, and have to add units costing nearly a million dollars each to increase capacity,” said Misra. “I think in the future the move will be towards more blades, which are scalable, take up much less space and are more cost efficient.”

But speed is not the only question. The resilience and capacity of exchange systems are highly important, with Linux cited by many readers as handling heavy loads well. Many exchanges can handle between 50,000 and 300,000 messages per second, with some newer technology in development reportedly hitting a million messages per second.

In spite of the debate over numbers and capacity, “speed remains a very important measurement for exchanges”, said Ralph Silva, managing director at analyst firm SRN.

But he added: “We are quickly reaching the theoretical maximums on latency considering how things are done. Something else would have to change if it’s going to get faster, such as brokers sending in different parts of the messages through different pipes, rather than a single broadband connection, so the trades can be processed even faster.”

Chris Skinner, chief executive at think-tank Balatro, agreed that other technological changes would eventually have to take place, if speed is to be further increased. “An issue for brokers is that their own connections can add serious latency to what they’re doing,” he said, pointing out that those firms need their own cutting-edge connectivity to get the most out of the exchange systems.

“If the brokers who are co-locating don’t also have the best systems and connections, it’s not enough to deliver that kind of speed,” he said.

The quest for ever-greater speed, however, raises huge financial implications, observers and readers noted. The main concern was runs on the market, either as major items of news break, or as trading errors or rogue trading takes place. The risk is that panic trading can vastly affect share prices and even knock chunks off exchange indexes.

“I’m not keen on technology alone running exchanges,” said Silva. “There have to be checks and balances, and there have to be humans ready to halt trading immediately when problems occur, not minutes or hours later.”

As one reader put it: “I’m all for Linux making inroads, but... ‘The financial system can implode twice as fast’ would seem an equally valid headline.”