Latency: Why Care?


Citihub’s Paul Chew and Richard Donaldson lay out the latest in Asia Pacific latency reduction and discuss how firms should prioritize their technology investments.
No one in AsiaPac used to care much about latency. However, when the Tokyo Stock Exchange (TSE) launched arrowhead in 2010 reducing their matching engine latency from 1 second to 5 milliseconds (a 200 fold improvement) it created a paradigm shift in the trading landscape by eliminating the exchange as the chief cause of latency and shifting the focus back onto market participants.
What’s more, this was not an isolated event – August 2011 will see the culmination of a US $200m investment by the Singapore Stock Exchange (SGX) to create the world’s fastest matching engine (SGX REACH) with average response times of 90 microseconds. The Australia Stock Exchange (ASX) invested US $35m to drive down their latency from 30 milliseconds to 300 microseconds; at the end of this year the Hong Kong Stock Exchange (HKEx) is expected to launch its new matching engine reducing average order response times from  130 milliseconds to 9 milliseconds (a 15 fold improvement).
At the same time, volumes are rising. TSE’s daily average equity order volume jumped 22% to 8.239 million orders in 2010 after the launch of arrowhead. Even before the proposed latency improvements by HKEx, they recorded a 41% increase in volume during the same period.So how will increased volumes and a reduction in exchange matching latency impact buy/sell-side firms? Surely it’s a benefit to doing business? Actually, with market participants now contributing to the majority of order processing latency, it creates both a challenge and an opportunity.
Increased stress will be placed on market participants’  trading systems because message volumes are growing and the time interval between messages from the exchange is falling. Conversely, market participants that can support growing volumes and drive down their own latency will create a competitive advantage. So what will this really mean for market participants and where should they target their limited investment dollars?
eTrading Platform Maturity
Our industry experience in Asia Pacific across buy/sell-side firms and vendors indicates a broad range of capability and focus. This is evident from contrasting client feedback and has given rise to what we have termed eTrading Platform Maturity:

  • Tier One: Platform Stability – “We care about stability, availability and reliability, not latency.”
  • Tier Two:  Instrumentation – “We care about platform latency but we need to improve the way we measure and analyze it.”
  • Tier Three: Platform Optimization – “We don’t care about absolute latency as long as we’re first on the order book.”

Fundamentally, latency is one of the key barometers of system health. Significant increases in measured latency are indicative of a stressed platform which can lead to outages impacting reputation and resulting in lost revenue.
We believe all firms should first establish a reliable platform that copes with daily business demands with predictable and consistent levels of latency before chasing the next tier of eTrading Platform Maturity.
Of course this is all a balancing act, often  requiring business and technology teams to prioritize stability over new product development and increased functionality. Smart investment in platform stability can be achieved through simple measurement and analysis of latency to target improvements providing these are supported with the appropriate post-implementation controls.

In order to address the balancing act of how and where to invest we have defined the Latency Framework (see Figure 1) and Instrumentation Capability Curve (Figure 2). These frameworks are used to determine the impact of volume and latency on platform stability and performance, to relate instrumentation to capability maturity and also to establish where to target platform improvements for greatest impact. For example, a key to determining the inherent capacity and performance of a system is through statistical profiling of changes in latency as volumes increase to the point at which the system becomes unstable.

Latency measurement
Since latency occurs at every node across the entire order flow (see Figure 3), instrumentation, and the ability to accurately measure and monitor latency, is important to understand platform behaviour. The evolution of technology has reduced ‘best in class’ latency within nodes to micro-seconds; consequently, the ability to dissect the latency of complex trading platforms (Tier 2 in eTrading Platform Maturity) non-intrusively is essential so as not to degrade performance. It often requires the procurement of tools and the ability to assign latency a ‘monetary value’ in order to secure the required funding.
Understanding the latency at each point in the trade flow allows intelligent Return On Investment decisions to be made. The latency framework (Figure 1) is used to categorise the business and IT latency requirements against potential solutions to ensure a cost-effective approach using the best-fit technology and process.
For example, when reviewing the statistical profile of message latency through an application component (see Figure 4) it is imperative to look at a standard frequency distribution instead of taking an average of latency across the day. Frequency distribution will demonstrate both the ‘jitter’ and the ‘tail’ of the message latency.

In summary
We believe the Asia Pacific markets will continue to change significantly over the next few years. New entrants to the market, market deregulation, fragmentation and continued improvements in the performance of the major regional exchanges will all play their part in driving up volumes and forcing down the latency benchmarks that firms need to attain in order to remain competitive. With a solid foundation in place, market participants can be confident that they have a trading platform that responds to peaks in market volumes with predictable and consistent levels of latency (Latency Maturity Tier One).
Targeted investment in appropriate instrumentation and monitoring, combined with continued measurement and analysis into the cause and effect of platform latency are the advances necessary to reach Tier Two in our maturity stack. Firms attaining this level of maturity are already typically amongst the most competitive in the market (Latency Maturity Tier Two). Market participants who have developed a comprehensive understanding of their platform latency and its tolerance levels can then start to think about the third tier in the Latency Maturity Stack and the ‘race to zero latency’.
Platform optimization requires the forensic analysis of every layer: network, hardware operating system, messaging and the application to yield the biggest reductions in latency. In many cases the application layer itself can yield the greatest potential for improvements; for example, by reducing the number of components and requirements for messaging and protocol translations between internal systems. However, this can be a complex and costly undertaking, so while the benefit may be clear it may be easier to drive down latency in other areas first; for example, through the replacement of aggregated feeds with direct exchange feeds.
Whether a market participant’s focus is stability, instrumentation or to establish their own internal latency benchmark and beginning the ‘race to zero’, never has there been a greater imperative to care about latency.