Green Data: Impact Versus Low Latency?

By Professor Michael Mainelli
Published by Inside Market Data, Volume 24, Number 32, Incisive Media Limited (11 May 2009), page 9.

When milliseconds matter, so does the speed of light. Equally, so does public opinion. Professor Michael Mainelli, Chairman of London-based think-tank Z/Yen Group, examines how trading can remain competitive while dealing with issues more important than latency.

As the time between an input being processed and the corresponding output emerging approaches zero, low-latency trading approaches real-time trading. As recently as five years ago, latency was measured in three digits, from 250 to 500 milliseconds. Post-MiFID and with increasing competition for liquidity, exchange venues brag about their processing times while information technology suppliers slavishly trump each other with dubious speed claims.

Data centre solar

Today latency is measured in one or two digits. But, as latency approaches zero, the speed of light becomes important. A 150km roundtrip is a minimum of one millisecond. Trading competition is a race for speed and proximity. In London, equities traders are moving their trading operations and servers closer to the London Stock Exchange’s server farms. Elsewhere, other traders are finding that the location of their primary exchanges is geographically important, particularly in algorithmic trading where mathematical engines attempt to exploit tiny time differences bound up in the architecture of exchange structures. We may soon measure latency in basis point increments.

Public sentiment is clearly against many of the operations of financial markets, and previously unregulated entities are coming under the spotlight. A recent Knowledge@Wharton article by Professor Sidney Winter says there are "clear reasons why individual traders want to be able to move fast, but whether there is a public interest in knowing something an hour earlier or a day earlier is not apparent," adding that low-latency is not proven to benefit ordinary investors in the same way as addressing "the social stakes involved."

What public interest could be more important to investors than speed? Climate change. If financial services accounts for around 8 percent of global GDP, and Information and Communication Technologies (ICT) account for about 4 percent of total energy consumption, then the financial services industry is responsible for at least 0.32 percent of global carbon emissions. Call it 0.5 percent, given the industry’s fascination with and intense use of ICT. The global population will rise to 150 percent of current levels over the next 40 years. If most people, particularly in China and India, reach European levels of affluence, then we would require three planet Earths to sustain that lifestyle, or five planets if people attain American affluence levels.

A simple equation, "environmental impact = population x affluence x technology" (I-PAT) implies a 450% increase in environmental impact. If we wish to keep our planet in turn-of-the-century ‘mint’ condition, we must reduce our environmental footprint over the next 40 years by 78 percent. This is not simplistic belt-tightening, but radical change—four-fifths less travel, four-fifths fewer boiled drinks, four-fifths less travel, cars that last 25 years rather than five, and four-fifths less energy consumption in financial services.

One change underway is roving grid computing for heavy MIP (million instructions per second) applications. Value-at-risk (VAR) or portfolio optimization applications could be dynamically scheduled to run where carbon emissions are lowest. Nigel Woodward, Global Director of Financial Services at Intel points out that, "Grids, virtualisation, clouds and energy management are the new vocabulary. Technologies, such as Intel’s VPro, can remotely switch on or off and redeploy computing." Smart grids and energy sourcing strategies mix. A New York bank could take advantage, for example, of cheap French overnight nuclear (low carbon) power or Danish wind energy. As carbon and power markets converge, roving processing correlates with low power prices.

A second change is to optimise the efficiency of applications and operating systems. Reduce MIPs. Today’s ICT departments can easily and materially reduce usage through simple expedients such as software that shuts down unused machines. However, years of organic growth have resulted in grossly inefficient ICT infrastructure that requires replacement. ICT departments have been profligate with scores of unused features now lurking on desktops, gobbling processor time and energy, waiting aimlessly to be called into action. Computational efficiency used to be a badge of pride among assembler programmers. Project managers, systems analysts and programmers can radically reduce MIP usage through redesign at all levels - applications, middleware, operating systems, hardware and grid.

Of course, reducing MIPs also reduces latency, so a third, more radical change would be to challenge the excessive importance of latency itself. Low latency, with its knock-on implications for market data rates and gross power, has been on the agenda of buy and sell side firms for a few years. Today multi-lateral trading facilities (MTFs) and suppliers talk about "ultra low" latency. By co-locating servers close to exchanges, banks, investment managers and traders subject themselves to local power sources. Financial services firms need to rethink markets. Can we redesign markets so that traders using an Icelandic geothermally-powered, low-carbon emission server farm trade on equal terms with a London-based server farm near the London Stock Exchange’s server farm running on electricity produced by a coal-fired power station? Should the London Stock Exchange server farm be based near New Zealand geothermal energy?

Financial services could break loose of latency binds if firms convinced exchanges to use periodic auctions every few seconds, easing today’s latency pressures by several orders of magnitude. Most algorithmic and time-sensitive trading occurs in a limited number of typically highly liquid instruments. On the London Stock Exchange, it’s a score of stocks led by Vodafone. Periodic block auctions have attracted academic interest as exhibiting potentially more efficient allocation, pricing and efficiency across the market, as opposed to existing simplistic "first-in, first-out" models. There are numerous variations on periodic auctions that deserve examination. As buy-side firms, sell-side brokerages and exchanges trumpet their environmental credentials, it seems hypocritical to structure markets in environmentally damaging ways when there might be more environmentally benign structures that also equate with more efficient markets.

Ultimately, financial services will have to align market structures with social goals. Roving processing and energy sources, slashing MIP requirements, and redesigning exchange architectures will all matter when seeking an 80 percent reduction in environmental impact. Financial markets must lower the importance of low latency if they want to prove they’re saving the planet too.

[An edited version of this article first appeared as "Green Data: Impact Versus Low Latency?", Inside Market Data, Volume 20, Number 19, Incisive Media Limited (11 May 2009), page 9.]

About the Author

Professor Michael Mainelli is Executive Chairman of Z/Yen, the City of London’s leading commercial think-tank serving financial services, technology and government. Michael is a former scientist who has worked in financial services for 25 years ranging from accountancy firm partner to director of Western Europe’s largest R&D firm.