Low latency is a hot topic which is being widely debated at
the moment. This is hardly surprising given the steadily rising volumes of
market and trade data which exchanges and financial institutions need to
process. The New York Stock Exchange trades and quotes (TAQ) volumes have
rocketed from 100 million per day in 2006 to an astonishing 475 million in
2007. There does not appear to be a slowdown in the growth rate; if anything,
the rate of growth appears to be increasing.
There are numerous factors driving the increases in volumes
including growth in derivatives, algorithmic trading, new assets and new asset
classes, and the ever more active participation of the emerging economies in
the global financial markets.
This results in institutions being forced constantly to look
for better ways of processing large quantities of real-time data but on
occasions when volumes surge some have difficulty coping. One such time was
during the extremely busy day in February 2007 when some US exchanges
experienced difficulties finding that their systems had ground to a halt and
they were unable to publish and distribute market and price data in a timely
manner.
At one time it was thought that providing limitless bandwidth
was the obvious solution but, even though most of the financial markets
communication networks now provide “speed of light” communication, the
financial markets community is still saying, “faster please”.
A simple analysis of the problem is that speed of light
comms are pointless if all it means is information getting to the next
bottleneck really quickly. The route forward within institutions is thought to
be the constant review, analysis and management of all “internal” comms,
systems and applications in order to optimise each component. The holy grail of
low latency is to have a real-time view of critical systems and data and use
predictive analysis to monitor capacity levels, potential system and software
failures. This will then optimise the disparate systems and inputs into a
cohesive management system in order to reduce or remove bottlenecks and
roadblocks, thus reducing latency to the lowest possible level.
There are many ways of trying to cope with ever-higher
volumes: there are specialist software vendors for processing high volumes of
complex data and various forms of hardware acceleration to name but two. While
these work, whatever hardware or software is in place, an institution still has
to know whether its systems are keeping up, slowing down, or are close to being
overwhelmed.
A recent survey [May 2007 sponsored by Reuters] by low-latency.com
revealed some interesting facts. The first question was whether system
latency was being monitored; 30% of those who responded said they didn’t monitor
latency, so they would have no idea of whether and by how much their systems
were falling behind. Another question asked whether their systems would be able
to cope with increasing market data volumes - 39% said that they would - and
while another 31% had a plan in place, they said they would suffer in unusually
heavy trading conditions. This left 30% whose systems would be unable to cope,
and who did not have a plan.
When answering a question on what tools and procedures were
in place for measuring and/or monitoring latency 11% said that they were
unaware that such tools even existed.
The survey was completed by a wide cross-section of
participants in the financial industry, including directors and heads of market
data at global firms, and technical and IT departments for buy- and sell-sides.
The results of the survey are alarming in that they indicate
that there are many – far too many – institutions in the global market that
will most probably face major problems on an unusually busy day in the markets,
or just at some point in the not-too-distant future when they suddenly discover
that their mission critical systems are falling behind.
Meanwhile, according to forecasts from Tabb Group the market
for low latency messaging middleware will rise from $95 million in 2007 to $168
million in 2010 as investment banks increase spending on technology to cope
with rapidly increasing message volumes. Tabb Group estimates that the overall
spend on low latency infrastructures will have reached $300 million in 2007. So
all evidence points to a need for institutions to decrease latency to the
lowest possible levels and being prepared to spend large amounts of money to
get there.
So there is much to be done and a seeming willingness to pay
for solutions. However, in order to reduce latency, an organisation will first
need to have a very clear indication of problem systems and processes in
real-time.
This is where the Geneos management and monitoring system
comes into its own. Even with hundreds of applications and systems in use
around the world an institution can locate the bottlenecks in real-time and
predict which applications are close to failing or will have difficulty in
coping as data volumes rise. More importantly, Geneos can analyse latency at
the application level at different stages and help pinpoint the potential
problems which require attention. It is this ability to provide detailed
hierarchical analysis that means that resources can be focused on the weakest
links and latency kept to an absolute minimum. The resulting benefits won’t
just keep you competitive, they could also keep you in business.
