Building low latency FX trading architectures with customised network solutions

For firms trading with high-frequency or utilising high-speed event driven strategies, low latency is crucial, but with so much in the hands of network providers, what can they fine tune? Dan Barnes investigates

First Published: e-Forex Magazine 55 / Networks, Hosting and Connectivity / January 2014

FX trading firms increasingly need low-latency access to data and execution. Rather than relying on the plain vanilla network offerings, some are taking network control into their own hands, or seeking expert customisation.

“Some true high frequency FX traders have begun to dip their toe into the market,” says Joe Hilt, vice president sales and marketing for North America at Hibernia Networks. “We are seeing more and more algorithmic trading in FX. The momentum is starting to build. These  customers have expertise that is not only in the finance arena. They are experts  at running networks, and most build the actual networks themselves. In some cases where speed is not the absolute goal, they are less likely to build the network and will seek a carrier or service provider to outsource their network. These providers help them to deliver protection on their routes between a couple of points or to hand over their entire network and allow someone else to handle and monitor their gear for them.”  

Many big institutions are changing their approach to infrastructure. Where a few years ago the standard approach was to work in-house, with large, capable teams, the model offers little flexibility on cost and as budget pressures mount that changes the dynamic in the market, which means that that approach is less favoured. 

Joe Hilt“We are seeing more and more algorithmic trading in FX. The momentum is starting to build. These  customers have expertise that is not only in the finance arena. They are experts  at running networks.”

Joe Hilt

“We are seeing more and more algorithmic trading in FX. The momentum is starting to build. These customers have expertise that is not only in the finance arena. They are experts at running networks.”

“Typically now, firms are dealing with infrastructure providers such as BT. We look at how we can customise what we do and wrap a solution around their business problems,” says Mark Akass, chief technology officer at BT Radianz. “That is a change in behaviour as well as a change in execution. We have to get fairly intimate to do that effectively.”

The path of maturity in FX differs to that of other asset classes as there are a limited number of centralised places to trade, when compared with equities and derivatives trading. In FX trading there is a diaspora of buyers and sellers, so the first problem firms face is that they need to be ‘everywhere’ in a sense. 

“You do see community effects taking place, which helps,” says Hugh Cumberland, solution manager at Colt Technology Services. “There are loads of different liquidity centres. Players and firms have to interconnect at data centres, which provides a good low-latency service for firms that are co-located there.”  

Akass says, “The clustering of clients into large data centres allows the network provider to adapt cloud and network solutions. Instead of having separate wires into different buildings, you can put network switching fabrics into that building, with high capacity in and out. You can then provide local switching within the building to all of the participants. You get a concentration of clients because the capacity in and out of those datacentres tends to be pretty high; that is where you start getting into the fibre path and the 40-100 gbps pipe. That is where scalability hits today, it is providing capacity with diversity and resilience.”

Hugh Cumberland “The rule of thumb we see for deterministic low latency is that it is somewhere between 2-4 times slower than ultra-low latency.”

Hugh Cumberland

“The rule of thumb we see for deterministic low latency is that it is somewhere between 2-4 times slower than ultra-low latency.”

“The bulk of the liquidity is in the LD4, NY4 and TY3 data centres,” says Emmanuel Carjat, managing director of TMX Atrium, “however some very large players are still calculating prices outside of those locations which creates a potential distortion of price formation,” he states. 

“Although the majority of trading may be done in the location you are in, if you don’t have your pricing engine hooked up at one of those other locations you have to go to your pricing engine and back,” he says. “It is also the case that those external locations are not very well connected to the rest of the world. If you were to track latency on a map, with the amount of latency based on the scale of the map, it would look extremely distorted due to the big players. That is changing, a number of our clients are moving their engines into our environment at those locations. Once at those locations, they are one step away from those liquidity providers in LD4, NY4 and TY3.”  

Mark Akass“The clustering of clients into large data centres 
allows the network provider to adapt cloud and 
network solutions,”

Mark Akass

“The clustering of clients into large data centres allows the network provider to adapt cloud and network solutions,”

Getting the network right

For the client, determining the right level of service is the starting point, including the latency required, the range of contact points needed, reliability requirements and of course, cost. 

“Are you looking for low deterministic latency or ultra-low latency?” asks Cumberland. “The rule of thumb we see for deterministic low latency is that it is somewhere between 2-4 times slower than ultra-low latency. It varies on a whole load of factors including distance etc. Our deterministic low latency around Europe is somewhere between 10-20 milliseconds, and for a lot of people that is plenty fast enough, but the questions is, is it deterministic enough? There are layer-three network designs that would not be deterministic, they might be ten milliseconds one minute and fifty milliseconds the next, like the public internet.” 

Effective connectivity can involve many different technology layers and counterparties, which an FX trader has to take into account when customising their network, says Carjat.

“You don’t just have electronic communication networks (ECNs), you also have single bank portals, which, depending upon how you are trading, you may also want to access. The single bank portal is another layer of the network which often is not very optimised. First generation extranets have been around for quite some time, they do provide a lot of connectivity, but are not as optimised as the new generation of extranets.”

For trading across long distances, firms also need to look at the major cable providers to ensure they have selected the fastest route. Hibernia’s biggest project at present is a new transatlantic cable, called ‘Project Express’, due to be finished in September 2014. It is expected to have a significant latency decrease, roughly 10% lower latency on a fibre by fibre basis. 

“The decrease is really a result of the route,” says Hilt. “Most of the transatlantic cables in the past were built for diversity into shallow water and the objective is to get your cable from shallow water to deep water as fast as possible. That is for two reasons. The first is that in a shallow-water environment you need to double armour the cable, so providing security on it is more difficult and also for cost. When you are double armouring the cable and digging it into the surface it is definitely more difficult. Cables typically leave Long Island and head south before they start to move up north. If you flew on an aeroplane from New York you would fly out of Halifax, Nova Scotia and then over into the UK.” 

He continues, “We will be staying in shallow water, spending extra money to trench our cable in across to the Flemish cap in that region then into the UK. From a route perspective it works and we added a few things to make sure the cable could not be duplicated, and we will be the fastest.”

Jay Lawrence“If you take advantage of it, microwave or millimetre wave once in the air are moving at the actual speed of light, just shy of what it would be in a vacuum,”

Jay Lawrence

“If you take advantage of it, microwave or millimetre wave once in the air are moving at the actual speed of light, just shy of what it would be in a vacuum,”

The global net

These macro network concerns are crucial as sophisticated trading strategies grow in developing markets, increasing the need for low latency connectivity. Colt has seen an increase in demand from these markets, particularly in Asia Pacific. 

“This is a demand we expect to be associated more with foreign exchange than with equities and derivatives as an asset class,” says Cumberland, “HFT continues in the listed derivatives and equity market but as the competition increases and as strategies get worn out HFT firms or shops with similar strategies look for new markets and asset classes.”

Akass says that firms have to identify their network partners with these concerns in mind, taking on board the reach and reliability of service in addition to the ability to deliver low-latency trading. 

“The market is very global, so the ability to execute consistent solutions with a scalable, global foot print, I think, is a necessity. What we can deliver in London we can deliver to pretty much to any key market around the world. In addition to experience you have the technology spectrum, are you a one trick pony connecting data centres or cities together? Or can you provide a range of solutions appropriate to the business?Some of our clients at the low end might only want 10 meg but at the high end they might want ten gig. It’s about providing a range of solutions at a cost effective scalable model.  Service operations and quality are key. Plugging it in and doing the plumbing is one part of the problem, but keeping it operating to a high standard, keeping the capacity running to the right level, having that level of investment and aggregation of client demand to justify that becomes quite critical.”

…there is a point which is reached where lower latency offerings can only be achieved with a move from fibre to microwave.

…there is a point which is reached where lower latency offerings can only be achieved with a move from fibre to microwave.

Hibernia also tries to control distance as much as possible, as that controls latency more than any other factor. 

“We always look at how many fibre miles we can take off,” says Hilt. “We also build cables for diversity purposes, so with things like Hurricane Sandy, that really affected a lot of cable in the New York and New Jersey area but Hibernia’s was one of the only ones not in threat because it leads from Halifax, Nova Scotia. We route our customers across eight different cables in the Atlantic, two of our own, and six others, such as AC1, so we give our customers not only a financial advantage through speed but also resilience through diversity.”

Speeding ahead

With a network in place, the trader has to determine the best ways to reduce latency. Cumberland says that Colt has seen that there is a point which is reached where lower latency offerings can only be achieved with a move from fibre to microwave. 

“That is an expensive exercise but once Pandora’s box has opened you have no choice,” he says. “If you are first in to a new strategy, asset class or venue you have the luxury for a brief period of time. You are making profit based on being first to market. Once you are competing with other firms it becomes a race to the bottom. Our flagship microwave route is Basildon to Frankfurt and it is a very successful venture for us.  However microwave technology will be restricted to the few routes where you can justify the investment. There is no long term certainty in these investments, a business opportunity might be 12-24 months.”

“Fibre is a good medium but the physics of fibre has two things working against it,” says Jay Lawrence, CEO at Nexxcom, a microwave radio specialist. “Firstly there is latency inherent in the medium. Light has to travel through a physical medium of glass which is like trying to drink syrup. Secondly, by default you can’t always get the straightest line with fibre because you have things in the way so it is not the most efficient route. As an alternative to that we are using microwave architecture and we have used a unique radio engines.”

There are a range of model that allow the use of microwave says Cumberland, which avoid the risks of signal disruption halting trading.

“Some customers use microwave in conjunction with fibre  and prioritise some data to go via microwave,” he says. “If you connect two centres with microwave and the customer has colo in both then you can use the microwave to synchronise rather than send huge amounts of data.”

Lawrence explains that Nexcomm provides a holistic service which includes designing the engines to fit a particular level of performance, which is hard for trading firms to do. 

“A lot of folks have tried to buy boxes off the shelf, some have tried to retrofit boxes, but it hasn’t worked the way they hoped, as they try to stretch them to the edge of performance which affects reliability,” Lawrence says. “We design an engine to fit in the car to perform the way you want it and some of those key principles include being standards oriented, so all of our networks hand off either at a multiple fast ethernet or a gigabit Ethernet. That means you are not doing things to the data stream that are atypical or off standard, an important efficiency component in low latency. Our perpetual joke is that we don’t molest our bits.”

Emmanuel Carjat“If you were to track latency on a map, with the amount of latency based on the scale of the map, it would look extremely distorted due to the big players.”

Emmanuel Carjat

“If you were to track latency on a map, with the amount of latency based on the scale of the map, it would look extremely distorted due to the big players.”

Science and fiction

Every time a signal processing event takes place inside  an architecture, the transmission is slowed down. To avoid that there needs to be low noise inside of the system itself and in the radio itself. 

“To keep noise to a minimum you don’t want to be stretching those engines to the extreme because when you stretch to the far end you either get distortion from turning the power up all the way or you get noise from going the absolute maximum distance and something in the environment can get in the way,” Lawrence says. 

By putting the architecture together in a simple fashion without stretching anything, microwave can offer a very straight line, with a more efficient signal transmission than in fibre. 

“If you take advantage of it, microwave or millimetre wave once in the air are moving at the actual speed of light, just shy of what it would be in a vacuum,” says Lawrence. “You get a benefit of a 30% premium and then you can get other pickups by having a more efficient path and by minimising the amount of signal processing you have to have by having a well architected network.” 

Cumberland believes that while valuable, microwave will only happen on routes where there is the most traffic and opportunity, and where firms have the greatest belief that an opportunity will be longer term. 

“There is speculation about lasers and free space optics but the reality is the demand for broad bandwidth, ultra-low and deterministic low latency will persist,” he says. “The great thing about fibre is if you have a 10 gig connection you can flood it with every price from every asset class. If you have a much narrower microwave connection, maybe 10 meg, then you have to be much more careful, so have to be smarter about how to use the connection.”

Summary

High performance FX trading firms looking for access gateways to global currency trading venues and exchanges are now able to leverage the benefits of a new generation of tailored network solutions which are designed to reduce latency and which can deliver scalable on-demand connectivity,  guaranteed network availability, security, fiber diversity and redundancy.  So as the saying goes, they have never had it so good.